Home » Undergraduate Research
Category Archives: Undergraduate Research
Floating Photovoltaics (FPVs): Impacts on Algal Growth in Reservoir Systems
By Benjamin Narwold, Environmental Science and Management major ’23
Author’s Note: I wrote this review paper to learn more about the environmental impacts of floating photovoltaics (FPVs) because this topic directly applies to my work as an undergraduate researcher position with the Global Ecology and Sustainability Lab at UC Davis. I wanted to focus specifically on the impacts of FPV on algae because of the biological implications of disturbing ecologically important photosynthesizers in reservoirs. I want readers to develop an understanding of FPVs as a climate change mitigation solution, how these systems may disturb algae, and the uncertainties in whether expected and observed changes in algae growth are beneficial or detrimental to the aquatic environment.
ABSTRACT
Floating photovoltaics (FPVs) are typical photovoltaics mounted on plastic pontoon floats and deployed on man-made water bodies. If FPVs are developed to cover 27% of the surface area of US reservoirs, they would provide 10% of the electricity in the US. Freshwater reservoirs are host to vulnerable ecosystems; therefore, understanding the water quality impacts of FPVs is necessary for sustainable development. This review aimed to fingerprint the impacts of FPVs on reservoir aquatic ecology in terms of algal growth and identify the uncertainties in FPV-induced algae reduction to present our current understanding of the environmental impacts of reservoir-based FPVs. The UC Davis Library database was searched for papers from peer-reviewed journals published from 2018 to 2022 that covered “floating photovoltaics”, “algae reduction”, and “environmental impacts”. A consistent result across studies was that FPVs reduce algal growth by reducing the sunlight entering the host waterbody, and this can disrupt phytoplankton dynamics and have cascading effects on the broader ecosystem. Modeling and experimental approaches found that 40% coverage of the reservoir by FPVs is optimal for energy production while maintaining the necessary algae levels to support the local ecosystem. The lack of research on the ideal percent coverage of FPVs to reduce algal growth but not disrupt ecosystem dynamics emphasizes the need for future research that addresses FPV disturbance of local microclimates, algae response to reduced sunlight, and the corresponding cascading impacts on other organisms dependent on the products of algal photosynthesis.
Keywords: floating photovoltaics, algae reduction, environmental impacts, water and ecology management, energy and water nexus
Caption: Floating photovoltaic (FPV) system in Altamonte Springs, Florida. One of four sites monitored by the Global Ecology and Sustainability Lab for water quality impacts of FPV.
INTRODUCTION
Climate change is a global problem of increasing intensity and poses challenges to food, water, and energy security. Global climate models predict a 2-4°C increase in global temperatures from now until 2100, which will degrade human health and threaten ecosystems [1]. Renewable energy is a critical component of reducing anthropogenic greenhouse gas emissions, and the widespread transition away from fossil fuels is becoming increasingly feasible with new technologies. One of these new renewable energy systems is floating photovoltaics (FPVs), standard photovoltaic (solar panel) modules mounted to a polyethylene pontoon float system, positioned off the water’s surface, and anchored to the bottom or shore of the host waterbody [2]. FPVs represent an intriguing and novel renewable energy solution because they can be deployed on human-constructed water bodies and improve land-use efficiency. Ground-mounted solar projects compete for land against agricultural and urbanization interests, whereas many artificial and semi-natural water bodies, such as wastewater discharge pools, have no conflicting human interests [3]. FPV development thus presents an opportunity to sustainably increase solar energy production without interfering with agricultural and urban development, which will continue to expand as world populations increase. In addition to optimizing land use, FPVs can produce up to 22% more power than conventional solar due to evaporative cooling [4]. The solar panels are located just above the water’s surface, so the local water evaporation contributes to a reduction in solar panel temperature, thus increasing efficiency. Generating electricity using FPVs is intended to augment solar power generation capacity and supply more renewable energy to the grid for households and industry.
Among the most abundantly available space to develop this pivotal land-use optimization and climate change mitigation solution are reservoirs, lakes formed from damming a river for water storage and hydropower production. A GIS analysis found that covering 27% of the surface area of reservoirs in the United States with FPVs would generate enough electricity to meet 9.6% (2116 Gigawatts) of the country’s 2016 energy demands [4]. But reservoirs and similar bodies of water nevertheless represent vulnerable freshwater ecosystems, so developing an understanding of the water quality and species impacts of FPVs represents the primary hurdle to informing sustainable development of these systems.
FPVs reduce the amount of sunlight reaching the surface of their host waterbody, which reduces the amount of evaporative water loss and results in significant changes to algae growth [5]. Several studies have found that FPVs alter phytoplankton dynamics and can have cascading effects on the other organisms in the ecosystem [6–8]. A key agent of uncertainty surrounding reservoir FPVs is determining the equilibrium range of algal growth needed to support reservoir food webs. In some reservoir systems, we see strong summertime algal blooms. An algal bloom is a rapid increase in or overaccumulation of an algal population that can result in oxygen-depleted waterbodies called “dead zones,” where the algae eventually die and decompose [9]. FPV-induced shading can counter harmful algal blooms, providing environmental benefits to augment renewable energy generation. Alternatively, in reservoirs that do not have problematic algal blooms, adding an FPV system may reduce healthy algal populations and cause adverse rippling effects to other species in the ecosystem. Developing an understanding of what percent of the total water surface area of the reservoir covered by FPVs is enough to reduce algal growth and bloom potential but not too large to disrupt ecosystem dynamics will require further research. Specifically, assessing the disturbance of local microclimates caused by FPVs, algae response to reduced sunlight conditions, and the impact on other aquatic species dependent on the ecosystem functioning provided by algae. Due to climate change, we predict an increase in temperature and shifting precipitation patterns; therefore, it is important to contextualize the water quality impacts of FPV and its influence on algae, given this variability.
Figure 1. Impact of FPVs on algal in reservoir ecosystems. FPV-induced shading can provide additional environmental benefits in reservoirs with algal blooms and may cause adverse effects in healthy reservoirs.
Methods
This review surveys what we know regarding the impacts of FPVs on algal growth in reservoir systems. The UC Davis Library database was searched for papers from peer-reviewed journals using the following keywords: “floating photovoltaics,” “algae reduction,” and “environmental impacts.” I looked at experiments on reservoir-based FPVs from 2018-2022 to analyze plot scale impacts on algal growth, quantified with chlorophyll-a monitoring data, and assessed global-scale changes in algal growth from a climate change perspective, with consideration of FPV materials and design. Although a study on crystalline solar cells incorporated in this review is from 2016 and falls outside the 5-year range of focus, it represents a necessary juxtaposition to the semitransparent polymer cell technology. Overall, I analyzed the methods and results of site-specific, laboratory, and global-scale studies to fingerprint the current state of knowledge on the impacts of FPVs on algae and algal blooms to inform reservoir management.
Algal Growth and FPV Coverage Scenarios
Algae are responsible for producing oxygen in the waterbody, and the impact of FPVs on algae growth depends on the percentage of the waterbody covered by the FPV and is measured by looking at chlorophyll-a (ch-a) differences. Ch-a, a pigment present in all photosynthetically active algae, is often used as a proxy measurement to assess algal growth dynamics within a waterbody [10]. Ch-a is measured using optical sensors and wavelengths of light, so it is an indirect measurement of algal concentration. FPVs reduce the amount of sunlight reaching the surface of their host waterbody and disrupt phytoplankton dynamics. Hass et al. (2020) and Wang et al. (2022) investigated different FPV coverage scenarios and used ch-a as a proxy for algal growth. Hass et al. used the ELCOM-CAED model to evaluate ten different FPV coverage scenarios, and Wang et al. simulated 40% coverage relative to 0% coverage control ponds using black polyethylene weaving nets as a proxy for an FPV array. Both the model output and experiment-based approach settled on 40% FPV coverage as an equilibrium development target [7, 11]. The results of these studies show continuity; however, Hass et al. did not consider the difference in absorption wavelength range for different microalgal taxa, and Wang et al. did not use actual solar panels in their experimental design. Additionally, Andini et al. (2022) investigated the difference in algae between 0% and 100% coverage at Mahoni Lake in Indonesia by experimenting with mesocosms, isolated systems that mimic real-world environmental conditions while allowing control for biological composition by taking samples at the same water depths. These researchers found that 100% FPV coverage reduced ch-a between 0 and 1.25 mg/L, average temperature between 0 and 2.5℃, dissolved oxygen between 0 and 1.5 mg/L, and electrical conductivity categorically in the waterbody. However, the researchers only considered directly measured water quality variables and did not assess the long-term trophic consequences of 100% FPV coverage [6]. Clearly, the study was designed to show the polarity between 0% and 100% coverage in terms of several water quality parameters; however, realistic intermediate FPV coverages incorporated into both Hass et al. and Wang et al. were absent from this study. Given these compiled results, future research can continue to work toward the broader question of determining what percent FPV coverage can be applied to a reservoir to maximize energy production and minimize environmental disturbance.
Algal Blooms and Mitigation Potential
Algal blooms are a product of high productivity conditions that favor rapid algae growth, and the shading provided by FPV systems could mitigate the intensity and negative impacts of summertime algal blooms. High productivity conditions include high water temperature, intense sunlight, and abundant nutrients such as nitrogen and phosphorus. The first two variables can be controlled by FPV coverage. In a study of the global change in phytoplankton blooms since the 1980s, Ho et al. (2019) found that most of the 71 large lakes sampled saw an increase in peak summertime bloom intensity over the past three decades, and the lakes that showed improvement in bloom conditions experienced little to no warming. Temperature, precipitation, and fertilizer inputs were the considered variables, and this study could not find significant correspondence of blooms to any of these variables exclusively [12]. This insignificant result suggests a diversity of causal agents on a per-lake basis. Thus, conducting site-specific studies and monitoring these water quality variables will help establish algal bloom causation and the relative intensity of the confounding variables and, therefore, whether FPV coverage would be an effective mitigation agent. If the algae in a reservoir are linked to less-controllable variables like carbon dioxide concentration in the water or nutrient loading from agricultural runoff, FPV-shading will have a negligible effect on algae [6, 7]. Such considerations are critical to informing the potential environmental co-benefits of an FPV installation.
FPV Solar Cell Design
The properties of solar cells within the photovoltaic panels themselves are instrumental in determining what wavelengths of light interact with the surface of the host waterbody under the panels. Crystalline silicon solar cells absorb radiation wavelengths from 300-1300 nm and have a thick active layer of about 300 µm, responsible for high photon absorption [13]. These properties result in opaque solar panels that do not allow photons to travel through the panel and interact with the waterbody. Conversely, semitransparent polymer solar cells (ST-PSCs) represent an alternative material and technological approach, and algae growth can be regulated by engineering the panels to provide specific transmission windows and light intensities. Zhang et al., 2020 found that the growth rate for the algal genus Chlorella was minimized under the opaque treatment; however, the changes in photosynthetic efficiencies did not significantly affect the growth rate of Chlorella during the 24-hour experimentation window. While the researchers were able to show the variability in the number of photons penetrating the panels from 300-1000 nm across three treatments of different layering of material within the ST-PSCs, they were unable to yield a significant result in their study [5]. These results have limited scope because this study was conducted in a lab and did not assess real-sized PV panels in the field; however, it highlights how algae species may prefer different light wavelengths for photosynthesis that may be discontinuous with the wavelengths an FPV system best controls. Therefore, it is vital to coordinate solar panel material design in order to reflect and absorb the primary wavelengths that support algal photosynthesis. The viability of prioritizing this component of FPV is uncertain; however, new materials and technologies are being developed and utilized, and this relationship must be considered as we work to maximize FPV coverage in reservoir systems with minimal ecological complications (Figure 2).
Figure 2. Relationship between FPV transparency, light profiles entering the waterbody and interacting with algae, and FPV coverage optimization. Solar cell design influences light transmission, and photosynthetic rates in algae vary with light wavelength and intensity, providing site-specific design opportunities.
CONCLUSION
FPVs are relatively untapped climate change mitigation solutions and can potentially reduce algae, benefitting water quality in freshwater ecosystems and reservoirs that suffer from strong summertime algal blooms. Algae are critical primary producers in reservoir ecosystems; therefore, areas for future research include microalgae response to the reduced sunlight conditions created by FPVs and the ecological role of algal taxa within the reservoir ecosystem. Further laboratory studies of solar panel designs in this context are needed. Future research on FPVs and water quality must also account for climate change, shifting baselines, and environmental variables. From a reservoir management viewpoint, this includes studying whether reservoirs have lower nutrient loading and whether the algae can be managed with FPV arrays, fingerprinting the inter-reservoir variability to determine where we should spatially place FPV arrays and localize impacts, and further modeling the relationship between warming and algal blooms to understand the long-term effectiveness of FPV-based algae management. Climate change will continue to operate in the background, and energy security issues will intensify. Our understanding of the environmental impacts of FPVs is currently limited to the point where we cannot safely approve and construct these systems on most reservoirs; therefore, future studies are needed to incorporate this modern technology into the global renewable energy portfolio.
REFERENCES
- Ara Begum R, Lempert R, Ali E, Benjaminsen TA, Bernauer T, Cramer W,Cui X, Mach K, Nagy G, Stenseth NC, Sukumar R, Wester P. 2022. Point of Departure and Key Concepts. In: Pörtner HO, Roberts DC, Tignor M, Poloczanska ES, Mintenbeck K, Alegría A, Craig M, Langsdorf S, Löschke S, Möller V, Okem A, Rama B (eds.). Climate Change 2022: Impacts, Adaptation and Vulnerability. Contribution of Working Group II to the Sixth Assessment Report of the Intergovernmental Panel on Climate Change. Cambridge (UK) and New York (NY): Cambridge University Press. 121–196. doi:10.1017/9781009325844.003.
- Energy Sector Management Assistance Program, Solar Energy Research Institute of Singapore. 2019. Where Sun Meets Water: Floating Solar Handbook for Practitioners. Washington, DC (USA): World Bank.
- Cagle AE, Armstrong A, Exley G, Grodsky SM, Macknick J, Sherwin J, Hernandez RR. 2020. The Land Sparing, Water Surface Use Efficiency, and Water Surface Transformation of Floating Photovoltaic Solar Energy Installations. Sustainability [Internet]. 12(19):8154. doi:10.3390/su12198154
- Spencer RS, Macknick J, Aznar A, Warren A, Reese MO. 2019. Floating Photovoltaic Systems: Assessing the Technical Potential of Photovoltaic Systems on Man-Made Water Bodies in the Continental United States. Environ Sci Technol [Internet]. 53(3):1680–1689. doi:10.1021/acs.est.8b04735
- Zhang N, Jiang T, Guo C, Qiao L, Ji Q, Yin L, Yu L, Murto P, Xu X. 2020. High-performance semitransparent polymer solar cells floating on water: Rational analysis of power generation, water evaporation and algal growth. Nano Energy [Internet]. 77:105111. doi:10.1016/j.nanoen.2020.105111
- Andini S, Suwartha N, Setiawan EA, Ma’arif S. 2022. Analysis of Biological, Chemical, and Physical Parameters to Evaluate the Effect of Floating Solar PV in Mahoni Lake, Depok, Indonesia: Mesocosm Experiment Study. J Ecol Eng [Internet]. 23(4):201–207. doi:10.12911/22998993/146385
- Haas J, Khalighi J, de la Fuente A, Gerbersdorf SU, Nowak W, Chen PJ. 2020. Floating photovoltaic plants: Ecological impacts versus hydropower operation flexibility. Energy Conversion and Management. Energy Convers Manag [Internet]. 206:112414. doi:10.1016/j.enconman.2019.112414
- Pimentel Da Silva GD, Branco DAC. 2018. Is floating photovoltaic better than conventional photovoltaic? Assessing environmental impacts. Impact Assess [Internet]. 36(5):390–400. doi:10.1080/14615517.2018.1477498
- Sellner KG, Doucette GJ, Kirkpatrick GJ. 2003. Harmful algal blooms: causes, impacts and detection. J Ind Microbiol Biotechnol [Internet]. 30(7):383–406. doi:10.1007/s10295-003-0074-9
- Pápista É, Ács É, Böddi B. 2002. Chlorophyll-a determination with ethanol – a critical test. Hydrobiologia [Internet]. 485(1):191–198. doi:10.1023/A:1021329602685
- Wang TW, Chang PH, Huang YS, Lin TS, Yang SD, Yeh SL, Tung CH, Kuo SR, Lai HT, Chen CC. 2022. Effects of floating photovoltaic systems on water quality of aquaculture ponds. Aquac Res [Internet]. 53(4):1304–1315. doi:10.1111/are.15665
- Ho JC, Michalak AM, Pahlevan N. 2019. Widespread global increase in intense lake phytoplankton blooms since the 1980s. Nature [Internet]. 574(7780):667–670. doi:10.1038/s41586-019-1648-7
- Battaglia C, Cuevas A, De Wolf S. 2016. High-efficiency crystalline silicon solar cells: status and perspectives. Energy Environ Sci [Internet]. 9(5):1552–1576. doi:10.1039/C5EE03380B
Inference on the Dynamics of COVID-19 in Kerala, India
By Darya Petrov
Author’s Note: I worked on this research project at the peak of the COVID-19 pandemic, while we were fully remote and on lockdown. I chose this topic because it was extremely relevant given the circumstances. I hope this report conveys the importance and value of the union of statistical modeling and public health in pandemic response efforts.
1 Introduction:
The coronavirus pandemic has been ongoing since the beginning of 2020. As of April 11, 2022, there have been 497 million confirmed cases of which 6 million resulted in death worldwide [1]; the recent World Health Organization (WHO) report on the pandemic indicates that this massive number itself is probably a significant underestimate [2]. Our understanding of the evolution of the coronavirus has dictated many large-scale social distancing measures including mask mandates, lock-downs, and travel restrictions that have had major impacts on society, economy, and public health. Conventional epidemiological models of infectious diseases, such as the SIR (Susceptible, Infected, Recovered) model which measures the spread of a disease through the change of the population in each of the three compartments listed, do not readily apply to COVID-19 dynamics; they do not utilize information on the count of asymptomatic individuals, an unobservable variable. It is well-known that asymptomatic but infected individuals have been the major spreaders of the COVID-19 pandemic, and therefore, it is imperative to obtain an estimate of such individuals in the population from available data. India is an excellent candidate for the analysis of disease dynamics because at one point during the pandemic, it had the worst COVID-19 crisis in the world. On May 6th, 2021, India had the largest worldwide single-day spike of over 400,000 new infections with shortages of hospital beds and ventilators [3]. We analyze publically available data from the state of Kerala in India to gain a better understanding of COVID-19 dynamics using a previously proposed methodology. The model is expressed through a system of difference equations, and incorporates information on social distancing measures and diagnostic testing rates to characterize the dynamics of the pandemic. The model’s key feature is its ability to estimate the unobservable count of asymptomatic individuals mentioned previously. This methodology has already been used to analyze COVID-19 dynamics in the United States [4].
2 Methods:
2.1 The Model
A graphical representation of the disease propagation model is depicted in Figure 1. The color of each box represents the observability of the compartment: red indicates unobserved, blue indicates observed, and purple indicates partially observed, meaning the compartments are observed together. Suppose at time t, Ct , Dt , Tt respectively represent the number of confirmed COVID-19 cases, number of deaths due to the disease, and number of tests performed up to time t. Let At denote the number of asymptomatic individuals, of which Ht denotes the number of persons hospitalized and Qt denotes those quarantined due to COVID-19 at time t. St denotes the number of susceptible individuals in the population at time t. Rt denotes the number of recovered individuals, of which RtQ, RtH, RtA respectively are those recovered from quarantine, hospitalization, or from being asymptomatic but never quarantined or hospitalized.
Figure 1: Graphical representation of the disease propagation model
2.2 Data Preprocessing
We consider the dynamics of the spread of COVID-19 in Kerala, India for a time window of April 1, 2020 to December 31, 2020. The proposed model is based on the observed state-wise daily counts of confirmed infections, deaths, hospitalizations and reported recoveries. Daily counts of the confirmed COVID-19 cases, recoveries, deaths, tests, and hospitalizations were obtained from an application programming interface (API) [5]. Daily hospitalizations were obtained from a state dashboard owned by the Kerala government [6]. Unfortunately, obtaining hospitalization data for India was arduous. We extracted the data manually for each district in Kerala for each day, and then combined the data into one data frame. The social mobility data was obtained from Google [7]. The data was preprocessed and cleaned, removing any irregularities present such as abnormally large counts for a given day that are clearly due to a mistake in the data reporting. Getting rid of such outliers is crucial because they can have a significant impact on the model and distort the actual relationships and patterns in the data. These irregularities or any missing observations were replaced using K-Nearest Neighbor (KNN) imputation with k=6 nearest neighbors, i.e. a missing observation for a given day was replaced by the average of the six closest observations by date. Inherent noise present in the daily counts, visually represented by frequent vertical spikes on a graph, were removed by pre-smoothing the trajectories using the Locally Estimated Weighted Scatterplot Smoothing (LOWESS) method with bandwidth 1/16. This fits smooth curves to the data points to capture general patterns in the data, with the bandwidth indicating how much of the data to use when smoothing at each point. The smaller the bandwidth, the rougher the smoothed curve will be, i.e. the graph will have more bumps.
3 Results:
3.1 Case Study for Kerala, India
We present our analysis based on the data from Kerala for the time window between April 1st, 2020 to December 31st, 2020. Figure 2 plots the daily number of people in hospitals. Note that no vaccine was available during this time period, and so any immunity from the virus could only be obtained through exposure. An important assumption our model makes is that once somebody is infected (either showing symptoms or otherwise), the person remains immune to reinfection. The black curve plots the observed values, and the red curve plots the fitted values from the model. It can be seen that the fitted values obtained from the model closely follow the observed values. This validates our proposed model and the estimation procedure. From the data and the fit, it is visible that a wave started early July 2020. The number of hospitalizations peaked early October 2020 and started decreasing afterwards.
Figure 2: Daily hospitalizations, fitted by the model (red ⎯) and observed (black ⎯)
3.2 Estimation of Latent Compartment
The estimated number of infected asymptomatic individuals (Figure 3) shows a similar pattern with a high point around the beginning of October, and dipping afterwards. There is also a local peak around the end of August. Estimation of this latent, i.e. unobservable, compartment across time is a key feature of our proposed methodology, since this information cannot naturally be obtained from the conventional epi-models.
Figure 3: Estimated number of asymptomatic individuals
3.3 Analogue of Basic Reproduction Rate
One large wave, i.e. a surge in new infections, can be observed from the plot of the proposed analogue of the basic reproduction rate (Figure 4). It measures the transmissibility of COVID-19 at time t and is influenced by spread mitigation efforts. The basic reproduction rate less than 1 indicates a decrease, while greater than 1 indicates a growth in the number of asymptomatic-infected individuals. Its estimate was mostly larger than 1 in the sub-interval, namely from the end of April to the beginning of October, indicating the singular large wave.
Figure 4: The basic reproduction rate ( R0 ) is the rate of growth of asymptomatic-infected individuals.
The plot of the number of daily new and daily reported infections (Figure 5) shows a maximum near October. The black curve plots Ct, the number of observed confirmed cases at time t+1. The red curve plots NI(t), the daily number of new infections at time t, which is calculated as the estimated number of susceptible individuals that become asymptomatic-infected at time t.
Figure 5: Daily new infections observed by the change in confirmed cases, Ct (black ⎯), versus the estimated number of new infections, NI(red ⎯) .
3.4 Transmission Rates
Figure 6 shows the plots of the crude infection rate (CIR) and net infection rate (NIR) . The red curve represents the CIR(t), the ratio of the daily change in the number of confirmed cases relative to the number of confirmed cases at time t+1. The CIR under-represents the infection rate, so the model estimates the infection rate with the NIR. This explains why the black curve represented by NIR(t) tends to be larger than the CIR(t) curve. The NIR(t) is the ratio of the daily change in the number of asymptomatic-infected individuals relative to the number of asymptomatic-infected individuals at time t.
Figure 6: The observed crude infection rate , CIR (red ⎯), and the estimated net infection rate, NIR (black ⎯)
The observed doubling rate obtained from the observed number of confirmed cases (Ct) and its estimate from the cumulative number of new infections (CNI) appear to be very close after mid July (see Figure 7). This implies reporting kept pace with the spread of the disease starting mid July. The doubling rate obtained from Ct is represented as the black curve, and the estimate obtained from CNI is represented by the red curve. It is the inverse of the doubling time at time t. The doubling time is the amount of time it takes to double the amount of infected individuals at time t. The higher the doubling rate, the faster the spread of the infection. The doubling rate reflects the effect of spread mitigation efforts, including social distancing campaigns, improved hygiene, and case tracking.
Figure 7: Doubling rate, Ct (black ⎯) and CNI(red ⎯)
Figure 8 shows the crude and net case fatality rates, CFR and NFR respectively. The black curve represents the CFR(t) and the red curve represents the NFR(t). CFR(t) is given by the percent of total deaths to the total confirmed cases up to time t. NFR(t) is given by the percent of total deaths to the cumulative number of infections up to time t estimated by the model. It is important to note that the formulas of the CFR and NFR are the same, except the denominator of the NFR is the CNI(t) while the denominator of the CFR is Ct . The observed number of confirmed cases Ct will be strictly less than or equal to the estimated cumulative number of new infections CNI(t), and likely much less, therefore the CFR is naturally much larger than the NFR.
Figure 8: The crude case fatality rate, CFR (black ⎯), and the net case fatality rate, NFR (red ⎯).
3.5 Testing and Hospitalization
The daily number of tests and its effect in quarantining asymptomatic but infected people can be judged from Figures 9 and 10. Figure 9 plots the number of tests performed per hospitalization. Tt represents the number of COVID-19 tests at time t+1. Ht represents the number of hospitalized persons for COVID-19 up to time t. This measure is an approximation of the contact tracing intensity. Figure 10 plots the RCCF, the relative change in confirmed fraction. The RCCF measures the change in the rate of currently asymptotic-infected individuals with COVID-19 that are detected through testing and quarantined relative to the rate of detection of currently infected individuals. This measure shows the dynamics of the effectiveness of detecting and isolating asymptomatic-infected individuals from the population through testing. Empirical comparison of Figures 2 and 9 reveals that although the number of daily tests could keep pace with daily number of hospitalized patients up to early July, the growing number of hospitalized people from July to October ultimately outpaced the number of daily tests. The daily number of hospitalizations beginning to decrease in early October was accompanied by the daily testing beginning to increase.
Figure 9: Tt / Ht
Figure 10: The relative change in confirmed fraction, RCCF
Discussion:
In comparison to conventional SIR models which model disease dynamics from the number of susceptible, infected, and recovered individuals, the proposed model also incorporates information about testing and quarantine. It is important to note the following assumptions the proposed model is based on:
- Only an asymptomatic individual who is not either in quarantine or in hospital can transmit the disease to a susceptible individual.
- People who recover from the disease are immune from subsequent infection.
- False positive rate for the test is negligible, so that if somebody is confirmed to be positive, then he/she is assumed to be infected.
- Anybody who shows significant symptoms, whether being in quarantine or not, is immediately hospitalized, and is tested to be positive.
- There is no effective treatment regime for the asymptomatic individuals, and so they recover or turn symptomatic at the same rate regardless of whether they are tested positive (and hence quarantined) or not.
These assumptions are quite general, however, the model could be modified if necessary to adjust for assumptions not met. For example, assumption 2 and 3 can be generalized by adding a fraction of recovered individuals to the susceptible population. Additionally, violations of some assumptions, such as assumption 1, are unlikely to have a significant impact on the disease dynamics. However, the current model does not incorporate impact of vaccination on the disease dynamics, which renders it applicable to the data being studied. Clearly, analyzing more recent data would require using a more enhanced version of the model that includes vaccination effects.
Smoothing was a crucial technique in this model because counts are rough. It was used in the data preprocessing to reduce the impact of anomalies, such as abnormally high counts likely due to incorrect data reporting. Additionally, it was used in the estimation of time varying parameters, which is intrinsic because of the locally weighted time window.
The goal of this study was to analyze how well the proposed model, which has already been used to model data from the United States, models disease dynamics of COVID-19 in Kerala, India. The performance of the model is validated by its ability to capture the large wave Kerala experienced between August and December of 2020, which is visible in the number of hospitalizations, estimated number of asymptomatic individuals, and the basic reproduction rate. The number of new infections estimated by the model appears reliable compared to the reported number of new confirmed cases. This reported number is an underestimate of the number of new infections since not all infections are reported. For example, an individual may be infected with COVID-19, but not reported as a confirmed case of COVID-19 if they are asymptomatic and did not get tested. This underestimate of the number of new infections worsens as the number of asymptomatic cases increases. The plots of testing per hospitalization and RCCF give us an idea of contact tracing intensity in Kerala, and how well it was coping with the pandemic. This model can help evaluate the effectiveness of measures used in hopes of reducing disease spread, such as social distancing, curfews, and mask mandate. The proper response to a pandemic is a controversial topic, and this model can help make informed actions in future pandemics. Just as this model was originally used on data from the United States and applied here to data from India, this model can also be applied to other regions, as long as the necessary data is available, preprocessed, and cleaned.
It is important to note that variants can have a strong influence on disease dynamics. For example, the omicron variant of the original SARS-CoV-2 strain is more infectious and spreads faster [8]. Additionally, the current state of this model is most applicable to a pandemic in which a vaccine has not yet been developed, which can be a big chunk of a pandemic since vaccine development takes time. The first cases of COVID-19 were detected in December of 2019, and a vaccine wasn’t approved until a year later in December of 2020 [9, 10], but even then the vaccine supply was limited and was distributed in phases, prioritizing those most at risk [11]. Because vaccinations impact disease dynamics, a potential next step is incorporating data about vaccinations into the proposed model [12].
Acknowledgements:
Thank you to Sruthi Rayasam for scraping the data from online, Satarupa Bhattacharjee for helping with the R code, and Dr. Debashis Paul for supervising this project.
References:
- World Health Organization. WHO coronavirus (COVID-19) dashboard. Accessed April 11, 2022. Available from: https://covid19.who.int/
- World Health Organization. 2022. Global excess deaths associated with covid-19, January 2020 – December 2021. World Health Organization. Accessed May 13, 2022. Available from: https://www.who.int/data/stories/global-excess-deaths-associated-with-covid-19-january-2020-december-2021
- Runwal, P. 2021. How India’s covid-19 crisis became the worst in the world. Science News. Accessed April 16, 2022. Available from: https://www.sciencenews.org/article/coronavirus-covid-india-crisis-social-distancing-masks-variant
- Bhattacharjee, S., Liao, S., Paul, D. Chaudhuri, S. 2022. Inference on the dynamics of COVID-19 in the United States. Sci Rep. 12(1): 2253. https://doi.org/10.1038/s41598-021-04494-z
- Babu, J., Shukla, A., & Bharath. Covid19-India API. Accessed April 21, 2021. Available from: https://data.covid19india.org/
- C-DIT. Kerala : COVID-19 Battle. GoK Dashboard. Accessed June 4, 2021. Available from: https://dashboard.kerala.gov.in/covid/index.php
- Google. Covid-19 Community Mobility Reports. Accessed August 9, 2021. Available from: https://www.google.com/covid19/mobility/
- Centers for Disease Control and Prevention. 2022. What you need to know about variants. COVID-19. Accessed April 16, 2022. Available from: https://www.cdc.gov/coronavirus/2019-ncov/variants/about-variants.html
- U.S. Food and Drug Administration.. FDA Approves First COVID-19 Vaccine. Accessed June 16, 2022. Available from: https://www.fda.gov/news-events/press-announcements/fda-approves-first-covid-19-vaccine
- Centers for Disease Control and Prevention. 2022. CDC Museum Covid-19 Timeline. Accessed June 16, 2022. Available from: https://www.cdc.gov/museum/timeline/covid19.html
- California Department of Public Health. Covid-19 vaccine prioritization recommendations for moving through vaccine phases and tiers. Accessed June 16, 2022. Available from: https://www.cdph.ca.gov/Programs/CID/DCDC/Pages/COVID-19/Vaccine-Prioritization.aspx?TSPD_101_R0=087ed344cfab2000b4d56b612d8caafb7743db82d9ff978793d54888b3c213b891734b5b4e6868b0087aa92767143000f5f590c2e053b6fe510cb7cedbfa962f4b18412db1954061d6f01003ae226244940e863d42ef53db7a5f6abb6d403af1
- Chen, X., Huang, H., Ju, J. Sun, R., Zhang, J. 2022. Impact of vaccination on the COVID-19 pandemic in U.S. states. Sci Rep. 12(1): 1554. https://doi.org/10.1038/s41598-022-05498-z
Recovering Individual Based Model Outcomes on Spatiotemporally Coarsened Data
By Sameerah Helal, Applied Mathematics, Under supervision of Stephanie Dodson
Author’s Note: Individual Based Models (IBMs) are commonly used to study animal migrations and foraging behaviors. These flexible models are powerful in identifying the mechanisms driving animal movement; however, when fed spatially or temporally coarse environmental data, IBMs can often produce inaccurate model outcomes. Here, we investigate how model adaptations can mitigate negative consequences of poor data using an IBM of blue whales. Specifically, we find that models running on coarse data lead the simulated whale agents to clump together in their foraging behaviors and migrations paths. Algorithm adaptations, like altering the rate at which the whales update their locations, can reduce the locational clustering effect incited by spatially coarse data, and introducing available fine data to coarse data can mitigate the behavioral inaccuracies caused by temporally coarse data. These changes improve verified utilization distributions of whale positions, behavioral state plots, and associated metrics.
1 Introduction
Recognizing how species interact and forage within their environment is an important aspect for understanding animal migrations. Predicting movement and behaviors given environmental conditions is crucial for understanding how changes in environment, like climate change, will affect future migrations. Though many species migrate, we focus on the foraging behaviors of Northern Pacific blue whales.
The largest animal to have ever lived, North Pacific blue whales are well known marine mammals and seasonal migrators [1]. In winter, they can be found in breeding grounds off the coast of Mexico and Baja California; during summer and fall, they migrate to foraging grounds off the coast of California. Blue whales forage only on krill and need to consume large amounts of prey to meet their energetic needs. Their seasonal foraging is timed with coastal upwelling in the California current system, which leads to high densities of krill. Blue whales are also a threatened species, and their population growth is hindered by human intervention in the form of shipping lanes, discarded fishing gear, and changing climate [1].
Modeling is a common method used to understand animal migrations and foraging behavior. Individual Based Models, or IBMs, treat individual animals as autonomous agents whose movements and behaviors are governed by a set of probabilistic rules. An example of this is the IBM in [2]. This model has individual, autonomous, simulated blue whale agents interact with and move through a domain: a restricted geographic area throughout which the agents are allowed to move; at any time they occupy either a transit or forage state. The state is selected based on krill density and sea surface temperature (SST) [2]. This model was used to evaluate the relative importance of prey and environmental factors in driving movement distributions of whales. Previously, the IBM was only run effectively on very fine spatiotemporal resolution Regional Ocean Modeling System (ROMS) data: computer generated sea data that is assimilative and hindcast, meaning that it gives realistic ocean conditions constructed from past observations. With ROMS being computationally expensive to generate and not available in real time, ROMS data is not feasible to use for current prediction purposes.
Our long-term goal is to predict real whale locations and behaviors by simulating whale agents on current environmental conditions using real-time satellite data. However, satellite data is messy and often of poor resolution, having spatiotemporal resolutions as coarse as 12 km grid cells and 8 day time-steps compared to the ROMS data’s 3 km and 1 day resolutions. A time-step is the length of time between each day for which the satellite has collected data. This poor resolution data yields inaccurate predictions and misleading results when applied to the blue whale IBM models.
Figure 1, created using output from our own models, highlights that models run on low resolution data return misleading predictions of whale positions, forecasting them to be clustered in much smaller areas than they actually are. In Figure 2, which shows the proportion of whale agents foraging over time and gives a sense of population-level behaviors, it can be seen that the models inaccurately recreate whale behavior by causing a behavioral update lag that corrects itself in jumps. Similarly unrealistic predictions are expected for IBMs of other animal populations.
In this study, our objective is to understand how to adapt the IBM developed by Dodson et al., as well as the data, to produce realistic predictions in spite of poor input resolution [2]. To do so, we will coarsen the original ROMS data to mimic poor satellite data, then compare the results of the IBM run on the coarse data to the results of the model run on the unaltered ROMS data. From there, we will formulate model adaptations that will yield accurate results from the IBM, even when run on coarse data.
In the methods section, we detail the functions of the IBM, our data manipulation methods, and
Figure 1: Contour map comparing the 95% utilization distribution map for the Gold Standard and 12 km resolution on September 1, 2008. Points are whales from the 12 km population.
our proposed adaptations, as well as the metrics to compare outputs of the IBM. In the results, we describe the efficacy of our solutions, showing quantitatively and qualitatively how they improve the model output. We analyze these results, their advantages, shortcomings, and future implications in the discussion section.
2 Methods
2.1 The ROMS Data
Throughout the study, we utilized ROMS (Regional Ocean Modeling System) Data, with two geospatial fields: sea surface temperature (SST) and krill [3] [4]. The domain is located off the California coast (32 − 42◦N Latitude and 116 − 128◦W Longitude) for the year 2008.
The datasets are identical in shape and structure, but contain different information: SST is sea surface temperature in Celsius and krill is the relative density with values between 0 and 1. The resolution of this data, which we will henceforth refer to as the ‘gold standard’, is 3 km spatial and 1 day temporal. Imagine the environmental data to be two-dimensional maps stacked vertically, one for each time interval (Figure 3).
2.2 Description of the IBM
Our IBM, or individual based model, takes inputs of sea surface temperature (SST) and krill data as environmental information that whale agents interact with and use to inform their movements around the domain. This choice of input is motivated by the fact that real-life whales partially decide their movements based on krill availability, which is influenced by sea surface temperature. Note that all mentions of whales from this point onward refer to the simulated whale agents in the IBM. As they progress through the model, the whales can occupy either a transit state (S1), where they simply travel through the domain, or a foraging state (S2), where they “forage” for krill; the movement distribution of each state is dictated by the step length and turning angle variables. These behavioral states are inspired by the analysis of blue whale tagging data in [5]. This type of model
Figure 2: Proportion of population foraging over time for the Gold Standard and 8 day resolution.
is also referred to as a “state-switching” or “semi-Markovian” model, respectively because the whales switch to random states and because the switch depends largely on their current and previous locations. In a fully Markovian model, the state variable would not depend at all on past information, only current.
Movement updates are parametrized by step lengths and turning angles. On any position update, the turning angle determines at what angle from its current trajectory a whale should turn, and the step length determines how far in that direction to travel. For each whale these parameters are drawn from respective Gamma and Von Mises, two probability distributions with their own parameters µ and σ. Distinct distributions are defined for whales in the foraging versus transit state. The distributions of the step length and turning angle are also functions of the time intervals given as parameters in the model. They have moments µ = h · 3700, σ = h · 2150 for the transiting state, and µ = h · 1050, σ = h · 970 foraging state [5]. Here, h represents the number of hours in one time-step (the amount of time between a location update) as follows:
Figure 3: Visualization of the SST and krill data as a 3-dimensional tensor.
(a) Forage state
(b) Transit state
Figure 4: Step length probability distributions for whales in forage or transit state for rate 1, 2, and 4.
Note that, while the step length and turning angle themselves are random, the absolute direction in which the whale turns depends on its current and previous location, which form the start of the zero degree angle from which the whale will turn. Hence, the “semi-Markovian” reference. At every update, each individual reads its local SST and krill conditions, which influence the probability of foraging through a distribution defined by
where z = 3 is a normalization parameter that ensures that the probability does not exceed 1, w1 = 1 and w2 = 2 are weights, and P SST and P krill are the respective probability distributions of foraging given SST and krill.
Lower SST and higher krill densities in a whale’s surroundings are associated with an increased likelihood of it either switching to or continuing in the foraging state (Figure 5).
(a) Based on surrounding SST
(b) Based on surrounding krill density
Figure 5: Probability of a whale foraging based on its environment.
To summarize the process: the IBM is initiated on May 1st in the southern portion of the domain, and at each update, the whales’ movement aligns with a movement distance based on the distributions in Figure 4 and their current location. Their foraging state is determined from their environmental conditions. This update cycle repeats until the end of the simulation, which, in this case, is the end of the year.
2.3 Data Coarsening in Space and Time
We manipulated the ROMS data to mimic the data collection shortcomings of satellites to achieve data coarsening. Specifically, satellite readings are plagued by coarse spatial and temporal resolutions. These are in part due to a satellite’s movement, type of readings, and presence of cloud cover. Making the ROMS data ‘bad’ in the way that satellite data is flawed required coarsening the data in space and time.
Spatial coarsening of the SST and krill ROMS data was accomplished through Matlab’s meshgrid and interp3 functions, for which the input data was the original 3 km spatial resolution, and returned it in lower resolutions of 6 km, 9 km, and 12 km. These coarser resolutions were selected to match available satellite data and test the limits of our methods.
To make up for missing data over time, satellites tend to use moving means: averaging the data over an interval, shifting the interval forward, and repeating the process; thus, our temporal coarsening was accomplished through averaging the data over time. To obtain coarsened data with an n−day resolution, we averaged the data in every n time-slices. The n slices were compressed into a single slice, resulting in a total of 1/n times the number of original time-slices. We performed this temporal averaging for 3 days and 8 days to mimic time resolutions consistent with satellite data.
Figure 6: SST data from September 1, 2008, coarsened temporally and spatially. The axes of each map show the number of grid cells from the bottom left corner.
Note that minor alterations to the IBM’s parameters, not mentioned because of their triviality, were required so that the model would run given the new dimensions of the data.
2.4 Issues from Coarser Data
Once we had coarsened the data, feeding it into the IBM no longer resulted in the same whale behavior or locations as the gold standard, which used data with 3 km, 1 day spatiotemporal resolution. Unlike many of the models based on coarser data, the gold standard results in whale locations spread over the coast with continuous population foraging behavior. In model iterations parametrized by the more extreme resolutions, like 12 km in space or averaged over 8 days in time, using the coarse data resulted in perceiving the whale population to act completely differently.
When tracking movement based on data coarsened in space (Figure 1), the whales tended to move closer to each other as the simulation progressed, resulting in them clustering unrealistically at the coast instead of being spread over the domain. Adding temporal coarsening to the data only exacerbated this clumping effect.
In addition to causing the whales to cluster over time, temporal coarsening alone also altered
Figure 7: Zoomed in utilization distribution maps of spatially coarse data vs. the gold standard. Curves show the 95% contour of the UDs. The full domain is shown on the right, with the regions on the left indicated by the orange box.
whale behavior. Abrupt jumps in the percentage of whales foraging became apparent at the end of each time-slice, making the whales’ feeding behavior appear unrealistically stepwise for extremely coarse time (Figure2).
Figure 8: Percentage of population foraging over time for temporally coarse data vs. the gold standard.
These deviations from the gold standard output required that we alter the IBM or its inputs such that the IBM would continue to produce realistic results, even with poor data.
2.5 Model Adaptations: Spatial
In an effort to have the IBM running on coarse data reproduce the same movement and behavioral patterns as the gold standard IBM running on the gold standard ROMS data, we modified parts of its algorithm; in particular, we scaled the time-steps of the IBM.
To mitigate the issues that arose specifically from spatially coarse data, we adjusted the rate at which the whales’ locations and behavioral states were updated. In the gold standard model, the whales take 4 steps per day (every 6 hours), with step length selected from a Gamma distribution with parameters µ0 and σ0. In order to recover realistic whale movements, i.e. to have the whales move the same distance over the same amount of time, we scaled this distribution by a parameter h, defined in equation 1, to accommodate for the increased size of grid cells caused by coarse spatial data.
From the gold standard model’s rate of 4 time-steps, we changed our rates to 1 and 2, corresponding, respectively, to time-steps of 24 and 12 hours. This way, the whales updated locations less often, but took larger steps that were matched to the larger grid cells [5], preventing them from getting trapped in grid cells due to limited movement ability, allowing them to move freely as they did when basing their movements on the ROMS data.
2.6 Model Adaptations: Temporal
To mitigate the disruption to whale behavior caused by temporally coarse data in the IBM, we introduced temporally finer data (when available) to the coarsened data, resulting in a hybrid coarse-fine environmental dataset that we could pass to the IBM. Our intention was to develop a methodology for handling temporally coarse input data while remaining consistent with the available satellite data. Temporally finer satellite data (e.g. 1 day, or daily) is often prohibitively sparse for IBM use due to missing data from clouds or the location of the satellite, but 8-day averages are sufficiently dense. We used this structure to our advantage and augmented the coarse data with available finer resolutions.
We regard the coarse data as our primary input, and the fine data as our secondary input. First, we created gaps in the coarse data by using Matlab’s randi function to select 30% of the indices in the data matrix uniformly at random. Then, to fill in the gaps in the coarse data, we replaced the randomly selected indices of the coarse data with the fine. In short, given temporally coarse data, we backed it up, or combined it with data of similarly coarse spatial, but finer temporal (1 day) resolution. For example, we might modify the 6 km, 8 day coarse data by replacing 30% of it with the finer 6 km, 1 day data.
Note that combining data of different resolutions in this way required some interpolation of the coarser data to the larger dimensions of the fine.
Before selecting 70% as the ratio of temporally coarse data, we analyzed real GOES (Geostationary Operational Environmental Satellites) satellite data recorded from the same area of the ocean as our ROMS data. Measurements of the GOES data are gathered by the GOES Imager, a multi-channel radiometer carried aboard the satellite. This satellite data is available in 6 km spatial resolution (about 0.05 degree latitude-longitudinal resolution) and in 1, 3, 8, 14, and 30 day composites averaged in time. Satellite SST data was extracted from NOAA GOES Imager Western Hemisphere satellite and accessed via [6]. Examples of data with few holes and many holes can respectively be found in Figure 9.
(a) A day with very few holes
(b) A day with many holes
Figure 9: SST GOES satellite data on days with varying amounts of holes. Dark blue indicates missing data.
Plotting the percentage of the data that was missing (Figure 10) showed that the highest percentage of clouds at any time, for the first temporal resolution (1 day), was 80%, with an average of 40%. That is, on any given day, there is a wide range of data that might be missing. We also tested a range of coarse to fine ratios (i.e. percentages of missing data) on our own data, and the lowest percentage of fine data that returned reasonably improved results from no fine data at all was 30%. Thus, in what follows, we fix 30% as the amount of fine data to introduce, and 70% as the amount of ‘missing’ or coarse data. The improvements will be further discussed in the Results section; see table 3.
2.7 Metrics
The behaviors that we are trying to replicate in the gold standard are the foraging behaviors over time and over the whales’ habitat area. We used two established metrics to compare model outputs between the adjustments, as well as a third that we designed to be a combination of the two.
The L2 norm, or the Mean Square Error, was used to measure how close the whales’ foraging behavior resulting from any of the models was to that of the gold standard model output. We extracted the vector of the percentage of whales foraging over time for both models and, after applying a moving mean, compared them using the L2 norm of the data. This metric provided an understanding of the foraging behaviors of the whale population through time. We define
Note that XGS represents the output produced by the gold standard model. Lower L2 norms indicate a more accurate reproduction of the gold standard outcomes, and an L2 norm of zero indicates that the models have identical foraging behavior.
Figure 10: Percentage of the missing data for real, daily satellite GOES data. Dashed vertical lines indicate the beginning of each month.
Utilization distributions are probability distributions built from the data points of individual positions and are commonly used to understand and compare animal habitats. To quantify the accuracy of the utilization distribution produced by a given model, we used the volume intersection (‘VI’ in the adehabitathr R package) between the utilization distribution of the final positions of the whales from the given model versus the gold standard model. The positions were taken from September 1, 2008, a date by which the whales had had sufficient time to explore and interact with the domain. This gives a measure of how close the model output is to the gold standard in terms of whale positions. We refer to this volume as the ‘VI’. The equation for the VI integral is
for model run X compared to the gold standard model at all points (x, y) in our domain.
Because we are measuring overlap of the utilization distributions, a higher value indicates better model performance, with the maximum being 1. Due to the stochastic, meaning somewhat random, nature of the model, a VI value of 1 was not achieved even by comparing two of the same model runs from the gold standard, which returned an overlap value of around 99%; thus, for our purposes, VI values near 0.9 are considered good.
The UDs and VI values were computed using R. We formatted the final whale positions of each model, labeled by the resolution of its input data, into a dataframe, then used the kerneloverlap and kernelUD functions from the R (version 4.0.2) package adehabitatHR to return the VI overlap value [7]. Our choice of setting the method argument to ’VI’ gave us the computed volume of intersection between the gold standard and coarse data model outputs.
Finally, our own ∆ metric is the combination of the L2 and VI norms: for a model outcome X, we defined
Since all efforts to adapt the model were done in order to produce equivalent results to the gold standard when fed into the simulator, our goal was to minimize the ∆ of the model; i.e. minimize the L2 and maximize the VI values. We measured the effectiveness of a model adjustment by quantifying model improvement using the fold decrease in ∆ values before and after the adjustment was applied.
3 Results
3.1 Changing Rate Counteracts Spatially Coarsened Data
We focus first on purely spatially coarsened data, which has the same 1 day temporal resolution as the data used in the gold standard. The model run on spatially coarsened data resulted in unrealistic behaviors from the whales: resolutions lower than the gold standard of 3 km prevented the whales from exploring the entirety of the domain and caused them to move closer together as the IBM progressed. In particular, the agents ended up in clusters near the coast instead of spread over the domain. In an attempt to recreate the spread-out behavior of the gold standard IBM, we decreased the rate at which the whales updated their locations and behavioral states.
Without any temporal coarsening, adjusting the rates significantly mitigated the clustering effects of spatial coarsening on the locations of the whales. Changing the rate in the IBM caused the whales to forage in generally the same areas with coarse data as they did with the gold standard data. The utilization distributions of the whales’ positions on the date September 1, which we noted was chosen as a sufficiently advanced time in the simulation to show model- representative whale behaviors (Figure 7), became visibly more similar to that of the gold standard.
Table 1: Original and corrected VI for spatially coarsened data, with fold decrease in ∆ values.
The original VI values for the coarse data for 9 km, and 12 km were 0.705, 0.480, and 0.256 respectively. Recall that higher VI values, closer to the maximum of 1, are more desirable. After adjusting the rate, these values became 0.710, 0.654, and 0.549; all improvements, even to the 6 km, which show less dramatic improvement due to already being relatively close to the gold standard. In addition , before changing the rate, the ∆ values were about 1.9, 2.8, and 3.0; all very high and
Figure 11: Utilization distribution maps of spatially coarsened data with and without altered rates.
indicative of undesirable model results that did not align with the gold standard. After changing the rate, these values reduced to 1.2, 1.4, and 1.7; at minimum a 1.6-fold decrease each, with the 9 and 12 km resolutions seeing the greatest favorable impacts.
3.2 Introducing Finer Data Counteracts Temporally Coarsened Data
We now consider purely temporally coarsened data, keeping the spatial resolution constant at 3 km. Recall that coarsening the data temporally caused issues with population-level whale foraging behavior. For lower temporal resolutions than the gold standard’s 1 day, the whales experienced abrupt, en masse changes in foraging state, causing the percentage of whales foraging over time to look unrealistically segmented. To mitigate this effect, we introduced finer data as a replacement for 30% of the coarse data as described in the Methods section. With the addition of the finer data, the foraging percentages were visually less step-like and numerically closer to the gold standard results.
Table 2: L2 norms for temporally coarsened data before and after the addition of 30% 1 day data, with fold decrease in ∆ values.
When naively fed temporally coarse data, i.e. when no changes are made to the data or the model to counter the effects of the temporal coarseness, the IBM results had ∆ values of 1.4 and 2.0 respectively for 3 day and 8 day. These high ∆ values are largely due to jump-like behaviors in the percentage of whales foraging. After replacing 30% of the coarse data with available fine data, we found that the ∆ values decreased over 1.6-fold and 2-old, to a value near 1. Similarly, the L2 norms decreased, as desired, from 1.2 and 1.7 to 0.77 and 0.93 for the coarse temporal resolutions of 3 day and 8 day.
Figure 12: Proportion of the population foraging over time for temporally coarsened data with and without added finer data.
3.3 Combined Spatial and Temporal Fixes
Negative influences on whale foraging behaviors were amplified when the environmental data had both poor spatial and temporal resolutions: with no fixes, the spatiotemporally coarsened data had a minimum ∆ of 3 across all resolutions. Combining the fixes by substituting available fine data, then changing model rate was successful in improving the behaviors in the combined coarseness scenario. By reducing the jumps in the number of whales in each behavioral state, and increasing the overlap between the utilization distributions of the corrected and gold standard models, we were able to decrease the ∆ value to be capped at 1.6. This was at least a 2.6-fold decrease in ∆s for each of the spatiotemporally coarsened datasets, greater than for any of the adjustments applied to data that was coarse in only one dimension.
Table 3: Fold decrease in ∆ for all spatiotemporal coarsening combinations.
Figure 13: Bar chart of ∆ values for spatiotemporally coarse data before and after algorithm modifications.
4 Discussion
Our long-term goal was to run the IBM on satellite data to predict real-time locations of blue whales. However, the IBM had only been shown to produce realistic predictions on fine-scale input data. Specifically, gaps and poor resolution of SST and krill data would lead to inaccurate simulations and predictions of whale movements. To the end of mimicking this ‘bad’ data, we coarsened our data spatially and temporally. By decreasing the rate at which the whales updated their locations and introducing available fine data to the coarse data, we successfully improved the results of the IBM when running on coarse data, as measured by the VI, L2, and ∆ metrics,.
Of our two directions of coarsening, the first we dealt with was spatial. The lower spatial resolution (e.g. 12 km instead of the gold standard 3 km), or larger grid cells, caused the whales to clump together over time instead of exploring the entire domain. We anticipated that this was because the increased size of grid cells caused the whales to be trapped in areas of high foraging; the whales’ step lengths were too short to exit areas with high foraging probability. Since the step lengths are a function of the step rate, with larger average lengths corresponding to lower rates, by decreasing the rate at which the whales updated locations, the step length became better aligned with the spatial size of the grid cells. Larger step lengths prevented the whales from getting stuck in rich foraging locations and allowed agents to step out of a grid cell and continue to explore the domain. For example, changing the update from 6 to 24 hours (rate 4 to rate 1) shifted the average step length of foraging from 6.3 km to 25 km. The larger average step length then allowed individuals to exit a 12 km grid cell and continue to explore the domain.
Our second direction of coarsening was temporal; lower temporal resolutions were apparent in sudden changes in the percentage of whales foraging. The jump-like behaviors were due to the whales operating on old information: the coarser the temporal resolution, the more outdated the whales’ knowledge of their surroundings. Our model correction was to provide the agents with the most up to date information when possible. Adding finer data into the coarse data removed the unrealistic jumps in the whales’ behavior and yielded population-level foraging activity that better resembled the gold standard.
Our solution for temporal coarsening is particularly powerful because of its scale: we only required 30% of the coarse data to be randomly replaced with the fine to see great improvement. Analysis of GOES satellite data showed that the average missing data during the summer and fall, commonly due to cloud cover, is usually less than 40% and rarely exceeds 70%. Thus, utilizing 30% of 1 day temporal resolution data with 70% of temporally coarse data is a realistic quantity, and actually uses the upper bound of 70% for the amount of missing satellite data. We introduced the fine data to the coarse data uniformly at random as the natural first step, since it did not require any assumptions about the locations or shapes of the missing data. One next step would be to target localized, perhaps even moving, regions for more ‘cloud-like’ replacement.
The methods we develop here are not at all unique to modeling blue whale populations; they can be applied to any individual based model in which the agents make decisions based on environmental input data. These solutions are particularly useful for models that, like ours, have inputs that are dynamic in space and time; nearly any IBM for animal movement could be similarly adapted. By incorporating finer temporal data when possible and aligning the step lengths to be of the same order as the spatial grid resolution, we believe other individual based models can be used with coarse environmental data to make predictions that are considerably more similar to those they might make with finer, more accurate data.
The problem of predicting animal behavior based on environmental conditions is an important one, especially with continually changing impacts from humans and climate change. With these results and further exploration in our suggested direction, it is our hope that it will be possible to make accurate, real-time predictions of whale positions based on satellite data.
References
- Scales KL, Hazel EL, Jacox Mg, Edwards CA, Boustany AM, Oliver MJ, Bograd SJ. 2017. Scale of inference: on the sensitivity of habitat models for wide-ranging marine predators to the resolution of environmental data. Ecography [Internet]. 40:210-220. doi:10.1111/ecog.02272.
- Dodson S, Abrahms B, Bograd SJ, Fiechter J, Hazen EL. 2020. Disentangling the biotic and abiotic drivers of emergent migratory behavior using individual-based models. International Journal on Ecological Modelling and Systems Ecology [Internet]. 432. doi:10.1016/j.ecolmodel.2020.109225.
- Rose KA, Fiechter J, Curchitser EN, Hedstrom K, Bernal M, Creekmore S, Haynie A, Ito S, Lluch-Cota S, Megrey BA, Edwards CS, Checkley D, Koslow T, McClatchie S, Werner F, MacCall A, Agostini V. 2015. Demonstration of a fully-coupled end-to-end model for small pelagic fish using sardine and anchovy in the California Current. Progress in Oceanography [Internet]. 138(B): 348-380. doi:10.1016/j.pocean.2015.01.012.
- Fiechter J, Huckstadt LA, Rose KA, Costa DP. 2016. A fully coupled ecosystem model to predict the foraging ecology of apex predators in the California Current. Marine Ecology Progress Series [Internet]. 556:273-285. doi:10.3354/meps11849.
- Bailey H, Mate BR, Palacios DM, Irvine L, Bograd SJ, Costa DP. 2009. Behavioural estimation of blue whale movements in the Northeast Pacific from state-space model analysis of satellite tracks. Endangered Species Research [Internet]. 10:93-106. doi:10.3354/esr00239.
- 2019. SST, GOES Imager, Day and Night, Western Hemisphere. https://coastwatch. pfeg.noaa.gov/erddap/index.html.
- Calenge C. 2006. The package adehabitat for the R software: tool for the analysis of space and habitat use by animals. Ecological Modelling 197:516-519. doi:10.1016/j.ecolmodel.2006.03.017.
The Neck Raising Behavior of Branta canadensis
By Cristina Angelica Bilbao, Biological Sciences ‘22
Author’s Note: I performed this ethological research study for my Zoology class at Las Positas College. I love animals and was excited to have the opportunity to conduct research on an animal of my choice. I chose to research Canada Geese because I grew up around them and was initially scared of them. I wanted this project to be something that would help me understand the complex behavior of geese and provide knowledge to my community. I chose to go to Shadow Cliffs Lake in Pleasanton California because that is where a large population of Canada Geese live year round.
Above all, I hope this paper can convey that Canada Geese are not as aggressive as we may think. I hope that this paper can provide the knowledge that geese are actually a lot humans. With an understanding of their emotional displays, we can begin to understand them in ways similar to how we understand and interact with one another.
ABSTRACT
For centuries, biologists have tried to understand the mechanics of animal behavior. In 1963, Niko Tinbergen published four questions that allowed zoologists to focus on animal behavior in a scientifically rigorous manner. Tinbergen’s Four Questions center around four different concepts that could be related to animal behavior: Causation, Development, Adaptation, and Evolution.
In this ethological study, I aimed to explain if the neck raising behavior of Canada Geese was due to causation or adaptation. An adaptive behavior is a behavior that has been shaped by an environment over a long period of time, while causative behaviors can be linked to physiological responses to stimuli. I hypothesized that the neck raising behavior done by Canada Geese is an adaptation and not a causative behavior in response to a negative stimulus.
In order to understand this behavior, I observed a population of Canada Geese residing at Shadow Cliffs Lake in Pleasanton, California. Previous studies proposed that Canada Geese raise their necks due to an adaptation and not solely as a reaction to a negative stimulus. This study was done over a period of ten days with a close study and a distant observation study. If the distance was minimal, the geese were indifferent and only raised their necks out of curiosity. If the distance was further away with no external stimulus, the geese would still raise their necks at the same frequency. The results of this experiment was confirmed by concluding that the neck raising behavior of Canada Geese is an adaptation rather than just a reaction to a stimulus. I was also able to conclude that my observations were able to support the null hypothesis because the geese raised their necks at the same frequency during the close and distant studies.
INTRODUCTION
In the book Geese, Swans and Ducks [1], Canada Geese are described to be sociable and family oriented birds that show their emotions through a variety of displays.
Emotional responses can be the key to understanding Canada Geese behavior. It has been observed that dominance and aggression tend to be expressed simultaneously among Canada Geese. Bernd Heinrich, a professor of biology at University of Vermont, conducted an observational study on a breeding pair of Canada Geese. He made many personal accounts of how the breeding pair had a high intensity of aggression towards him while guarding their eggs. When Heinrich came back once their young had hatched, the pair was not aggressive and seemed to adjust to the researchers presence [2].
The prominent neck of the Canada Goose is an important indicator of emotional responses. Emotional displays characterized by head pumping, withdrawn necks, and vocalizations have been designated as situational and related to attacking or fleeing action [3]. The threat postures were categorized as having a variety of neck movements accompanied by vocalized hissing noises [4]. Another study was able to observe specific members of Canada Geese families taking up ‘guard positions’, which involved various sets of alert and alarm postures [3]. The alert and alarm postures were characterized with similar behaviors relating to threat postures. If the geese were in an alert posture, they would raise their necks high and freeze. If they were in alarm postures, the geese would more than likely begin to display similar threat behaviors of wing flapping and vocalization to warn the rest of the group. This behavior was seen equally between the male and female geese, especially at the time when they are protecting their nests from predators [3].
While it seemed that aggression was the only explanation for this reaction, I began to question if that was the sole conclusion that could be made. The Shadow Cliffs lake population usually could be seen on the beach, in the lake, or on the grass with minimal worry about threats. I observed that the geese lived mutually with the other waterfowl in the area and they seemed to have grown comfortable around humans due to constant contact. This could be one of the reasons that the group chose to stay in this area for prolonged periods of time.
Before I performed my observational study, I conducted two preliminary observations over two days and was able to notice an unusual behavior. At specific times of the day when the geese fed, members of the group would raise their necks at random. The neck extension would occur for about a minute before they went back to eating again. I believe that this may have been a form of communication, until I noticed another behavior that proved my initial hypothesis to be incorrect. While feeding, the geese would form a tactical perimeter around those who were eating in the center. The geese acting as guards on the outside would raise their necks for the longest periods of time and remain alert.
As I performed my preliminary research, I aimed to learn more about the population of Canada Geese residing at the lake. I interviewed Mark Berser, one of the Shadow Cliffs Park Rangers who has been working at the park for over 10 years. He confirmed the geese were usually around the lake throughout the year. He stated the tall grass by the water edge close to the picnic benches was a place the geese went to both eat and sleep. He had not observed the frequency of the geese’s neck raising behavior, but he was aware of the fact that they all seemed watchful in their close groups regardless of the threat.
I expanded my research to online scholarly journals, and I found more information regarding threat postures instead of the specific behavior I aimed to observe. These studies provided a generalized explanation on threat postures, which included neck raising [4]. I was able to find foraging studies that mentioned neck raising behavior as well, but it did not provide details about why the behavior occurred [4]. I was able to deduce that most researchers thought of this behavior as a biological reaction to a stimuli, also known as the attack-flee response [4]. The attack-flee response is a neurological response to a threat that prepares an animal to make the decision fight or to flee. I aim to challenge this idea by hypothesizing that the neck raising behavior done by Canada Geese is an adaptation and not a causative behavior in response to a negative stimulus.
MATERIALS AND METHODS
Before I began my study, I gathered binoculars for observations and a field notebook to keep records of my observations. Next, I developed a detailed schedule relating to how and when I would observe the geese. I planned to conduct a ten day study on the population of Canada Geese in their habitat at Shadow Cliffs Lake. My observation days were on Mondays and Fridays before noon, in order to reduce the possibility of outside interference. If there was a situation that prevented me from not attending on Monday or Friday, I was able to make observations on another day at the same time in order to prevent discrepancies in the results.
During the first five days of the experiment, I positioned myself ten feet away from the geese. I casually sat at one of the tables in order not to disturb them in their normal routines. I then focused my observations on the frequency of the neck raising behavior over the course of one hour. Through these five days, I made sure to make note of the frequency of the behavior in my field notebook. Once the first five observation days were complete, the second portion of the experiment began. I made sure to arrive unnoticed as I positioned myself at a further distance away from the population. By keeping distance, I eliminated the possibility of the geese raising their necks as a reaction to my presence. During this time, I watched the geese with binoculars and took note of their neck raising behavior for one hour. I continued to take notes on the frequency of the behavior and I began to transfer the data to a table in order to properly visualize the results.
RESULTS
The target population of Canada Geese typically remained in a large group. On random occasions, smaller groups would split off from the rest of the family to either get an early start on bathing in the lake or to find food elsewhere. While they did split, they would eventually come back together in one location.
In large and small groups, there was a known presence of ‘guard’ geese. These geese would take position around the perimeter of the large or smaller groups. When the guard geese noticed me, they raised their necks more frequently and held the position for the longest period of time. The geese observed me before leaning down to eat and switch roles with another member.
During the first five days of observation, the geese were observed in close proximity. The geese were aware of my presence but they did not spread their wings or pump their necks in a threatened manner. Instead, they appeared to be indifferent towards me, raising their necks at a very low frequency as they waited to see if I had food. However, the geese would raise their necks more frequently if they felt that I was too close.
The second half of the observational study was performed at a further distance. The geese were unbothered as they continued with their usual daily routine. The guards and the others in the population raised their necks at low frequency, similar to what was observed in the first part of the study.
CONCLUSION AND DISCUSSION
In this ethological study, I aimed to observe the neck raising behavior of Canada Geese over the course of 10 days. During this observation period, I intended to explain whether or not the behavior was due to causation or an adaptation, as it relates to Tinbergen’s Four questions. Causation would describe the behavior as physiological while adaptation would describe the behavior as a reaction to stimuli developed over time. Through this study I was able to compare whether or not this behavior was more physiological or adaptive by taking certain factors such as visual stimuli and general environment into account. Initially, I hypothesized that the neck raising behavior done by Canada Geese is an adaptation and not solely a physiological response to a negative stimulus.
To successfully observe this behavior in the chosen target population, a variety of variables were taken into account. The Canada Geese were set to be observed without an external stimuli, which meant that time, presence of food and distance were all important variables. There was no food present and the population was observed at a time with minimal external stimuli exposure. The geese were observed from a far and close proximity in order to prove that neck raising behavior was an adaptation.
To support my hypothesis, observational data was collected over the course of ten days. Chart 1 presents the behavior observations from a short distance. The geese were aware of my presence because they raised their necks frequently, but they seemed indifferent towards me. More specifically, the perimeter guard geese seemed to be raising their necks the most and for the longest period of time. When the designated guards vocalized, the group of geese raised their necks collectively. I kept note of these results as I moved into the next phase of the study. Chart 2 depicts the behavioral observations made from a distance. Similarly, the geese raised their necks at the same frequency to look at their surroundings as they went about their daily routine.
The similarity in the results prove that the behavior of neck raising is not just linked to the presence of a stimulus. In relation to the work done by Blurton-Jones, neck raising and held erect posture could be perceived as a threat posture to drive away predators [8]. At a closer distance, the geese would raise their necks. While this could be seen as a threatened reaction, the behavior seemed to be more of an awareness of my presence rather than aggression. This was confirmed when the geese did not flex their wings or alert the group of a threat. The geese raised their necks and did not give any further reaction unless I got significantly closer. This allowed me to conclude that neck raising behavior is linked to threatening external stimuli.
With these results in mind, my hypothesis was confirmed. This experiment proves that neck raising behavior seems to be an adaptation developed through time as a method of protection rather than just a reaction to a stimulus. Through the act of raising their necks, the Canada geese make it known that they are aware of threats and are watching out for members of their family.
In the future, I would hope to perform more in-depth studies. Within my study, I was limited by the population of Canada Geese that I studied. The population of geese was the only significantly large population in Pleasanton, which ultimately limited my results. There was also a limiting factor relating to human contact; the group of geese that I observed had grown used to human contact instead of perceiving them as threats. While the group of geese displayed the behavior I aimed to observe, the results could be different in situations with limited human contact.
In future research, populations of Canada geese that have not experienced human influence should be observed. This would be able to prove whether or not the environment is a factor contributing to the neck raising behavior. Further research should include sexing the geese in order to rule out sex being a possible factor contributing to the frequency of neck raising behavior.
References
- Kear J. 2005. Canada goose (Branta canadensis).Ducks, geese and swans: general chapters, species accounts (Anhima to Salvadorina). New York (NY): Oxford University Press. P.306-316.
- Blurton-Jones NG.1960. Experiments on the causation of the threat postures of Canada geese. Wildfowl. 11(11): 46-52.
- Klopman RB. 1968. The agonistic behavior of the Canada goose (Branta canadensis canadensis): I. attack behavior. Behaviour. 30 (4): 287-319.
- Raveling DG. 1970. Dominance relationships and agonistic behavior of Canada geese in winter. Behaviour. 37(3/4):291-319.
- Hanson HC. 1953. Inter-family dominance in Canada geese. The Auk. 70(1):11-16.
- Herrmann D. 2016. Canada geese. Avian cognition: exploring the intelligence, behavior, and individuality of birds. Boca Raton (FL):CRC Press. P.72-143.
- Akesson TR, Raveling DG. 1982. Behaviors associated with season reproduction and long-term monogamy in Canada geese. The Condor. 84(2): 188-196.
- Heinrich B. 2010. Parenting in pairs. The nesting season: cuckoos, cuckolds, and the invention of monogamy. Cambridge (MA): Harvard University Press.p. 210-213.
Appendix
Chart 1: Field Notes Observation Table Results for Close Study
Date | Day/Description | Summary of Behavior |
9/30/19 | Day 1- 1:30-2:30 p.m.
The weather at the time was comfortably warm. The sun was out and there was a light breeze. While it did seem like the perfect day to go out to the lake, there actually weren’t a lot of people in the area. This made it easier to reduce any interference. |
–In the large family of geese and even in the smaller sub groups, there were designated ‘guards’. These rotating guards seemed to have a strategy of holding the perimeter. They were the members of the group that actually raised their necks the longest and most frequently. They would hold their necks up for about 10-20 seconds.
-When it comes to noticing me or if I was too close, the guard geese signaled a call to the rest of the group. The call would get the rest of the group to raise their necks. Overall though, they didn’t seem threatened but instead either somewhere in between indifferent or curious about if I had food. |
10/4/19 | Day 2- 11:30-12:32 p.m.
The weather at the time was a little colder than the first observational day. The sun was out but it was all around a little colder with a slight breeze. There were not many people at the lake. |
–The entire family of geese would raise their necks at random points. It became clear that the ‘guards’ were in a position where the ones that had the most power to alert the group of any potential problems.
-The guards raised their necks the most frequently. -In an area with no remote threats, the geese still seemed to be very clearly aware of their surroundings. |
10/8/19 | Day 3-12:30-1:30 p.m.
The weather today easily symbolized that fall was near. The sun was out but it was cloudy and there was a strong wind. This ultimately made it a cold day. There were people but not many near the geese. |
–Similarly to the other two observation days, the geese were still mostly indifferent to my presence.
-If I got too close, the guard geese would make sure others knew about me, but they weren’t even close to being threatened. -The neck raising behavior occurred frequently when I was in the presence of the population. |
10/11/19 | Day 4-11:27-12:30 p.m.
The weather today was cold. I could still see the sun but there were more clouds than most of the other days. There also was a slight breeze. Once again there were not many people at the park. |
–The geese retained the same indifferent behavior towards me.
-Most of the time I could justify that they raised their necks simply because they wanted to be aware of me and maybe were hoping that I had food. -The guards at the same time still were the ones that raised their necks the most frequently. |
10/16/19 | Day 5-6:40-7:40 p.m.
Mainly because it was the evening, the sun was actually setting. The weather itself started to get windy and cold as the darkness approached. There were still a few people at the lake but they weren’t providing interference to the geese. |
–The geese were raising their necks every few minutes like usual and holding the position before they went back to eating.
-It was interesting to see that there were actually more guards around vocalizing and raising their necks quite frequently. -The guards were also still raising their necks quite frequently when it came to watching out for the small groups flying to the water to sleep. |
Chart 2: Field Notes Observation Table Results for Distance Study:
Date | Day and Description | Summary of Behavior |
10/19/19 | Day 1- 1:20-2:20 p.m.
The weather at the time was sunny and warm. There weren’t too many people at the lake at this time, which minimized experimental interference. |
–The geese still seemed to display the same alert behavior through the action of neck raising.
-The perimeter guard geese still seemed to be the particular members of the family that raised their necks the most often and longest. The longest neck raised posture that the geese held was about 10 seconds. -The neck raising behavior would occur at random points of time to be alert of the surroundings and what the other members of the population were doing. |
10/22/19 | Day 2- 2:13-3:30 p.m.
The weather was sunny and warm, but there was a stronger breeze than the previous observation day. There were a bit more people at the lake today, but they weren’t anywhere near the geese population |
–The guard geese were still on alert both on the beach and in the park area.
– It is important to note that while the guards were the ones still raising their necks the most and the longest, the rest of the family was also doing the action. -When the guards seemed to vocalize, they had the power to get the whole group to stop eating and raise their necks to attention. The behavior also seemed to occur if two groups were calling to each other from two different locations. |
10/25/19 | Day 3- 1:13:2-13 p.m.
The sun was out but it was a bit colder, hinting at the arrival of the fall season. There were people fishing but since the geese took particular interest in feeding within an enclosed area, observation was not affected. |
–The geese in the parking lot and picnic area actually seemed to have similar frequencies of neck raising behavior.
-In the picnic area, the geese seemed extremely indifferent to the people in the distance and actually laid down. It was interesting to see the guards lay down as well, yet still raise their necks in the same frequency. -In regards to the group in the parking lot, it seemed that the neck raising behavior still occurred at the same frequency. The difference was more observable in how long the posture was held, which came out to be about 30 seconds. |
11/8/19 | Day 4- 11:30 a.m. -12:30 p.m.
There were a bit more clouds and a stronger breeze, but I could still clearly see the sun. The observations were carried out with no interference. |
– The guard geese prominently were still the usual members of the population raising their necks the longest and most frequently compared to the rest of the population.
-The geese that took up the roles as the guards happened to be the most alert and aware of the surroundings and other members of the family. – The neck raising behavior almost seemed summed as both a reaction to family calls and a general adaptation to watch the surroundings for any threats to the population. |
11/15/19 | Day 5- 12:40-1:40 p.m.
The sun was out but the overall weather was cold, mainly due to the strong breeze. Once again there weren’t many people around, which allowed for the geese to not get distracted. |
–On the final day of observation, the geese were in a much larger group feeding in the park area. There was also a small group on the beach.
-It was clear that there were designated guards, which told me that these roles were not just temporary. -The whole family would raise their necks, but it was the guards that raised their necks most frequently. -It seemed that the neck raising behavior was occuring quite randomly. Realistically this could mean that the geese raising their necks could be an adaptation to be aware of what the family is doing. It could also be an adaptation to avoid predators. |
Field Note Photos and Supplementary Images:
- The first set of images are entries from my field journal. Though they may be difficult to see, the data has been summarized in the tables above.
- The second set of images depicts two instances where the geese were displaying neck raising action. They also show the guard geese in place.
Image 1: This image was taken at the time the close distance observations were being conducted. In this image, a group of geese can be seen eating on the grass of the park hill. On the far corners of the image, two guard geese can be seen taking up perimeter positions around the rest of the group.
Image 2: This image was taken at the time close observations were being conducted for the experiment. In this image, two Canada Geese, possibly a mating pair, were foraging in the grass. The current acting guard can be seen on the left.
The Parable of the Passenger Pigeon: How Colonizers’ Words Killed the World’s Largest Bird Population
By Jenna Turpin, Wildlife, Fish, and Conservation Biology ‘22
Author’s Note: I started this piece as an assignment for my undergraduate expository writing class under the guidance of my supportive professor Hillary Cheramie. Hillary urged me to take my writing beyond her course. In May, I had the wonderful opportunity to share this research at the 2020 UC Davis Annual Undergraduate Research Conference. I want to continue to share my work through publication. I wrote this piece with the intention of inspiring both students and teachers. From this paper, students can learn the parable of the passenger pigeon and teachers can come to understand why teaching about the passenger pigeon matters.
I learned of the passenger pigeon during my first week of college at UC Davis. One of my professors, Dr. Kelt, explained a brief history of the passenger pigeon to my first-year wildlife ecology and conservation class. The lesson was about wildlife-human interactions and the destruction humans can execute on the environment. The passenger pigeon’s story shook me to my core. It was a disturbing portrayal of how people sometimes negatively shape ecosystems. For me, it reinforced all of the reasons I decided to study wildlife conservation. I want people who read this piece to feel the emotions I felt when I first took in the parable of the passenger pigeon and come to the belief that humans have a responsibility to conserve species through management, policy, and education. The more people who hear this parable, the more people who hold sympathy for our wildlife. It should be built into schools’ science and history curriculums. A greater understanding of the passenger pigeon will save future species from extinction.
Abstract
Genre is the literary process through which people collectively communicate about a topic. Applied to a species, genre helps us understand how society communicates about that animal. Species’ genres change over time as different people interact with them. This influences human-wildlife interactions and thus plays a critical role in determining the fate of that species. In the passenger pigeon’s (Ectopistes migratorius) prime, it was the most abundant bird species in existence but went extinct. The dynamics of human-wildlife interactions over time defined the progression of the passenger pigeon’s recorded history. These interactions varied based on how the dominant people in North America thought about the bird and the genre surrounding its existence. The parable of the passenger pigeon is a poignant example of why genre matters in preserving species and how this can go wrong. The analysis of the historical evolution of the passenger pigeon’s genre showed that the European colonization of North America is why these birds went extinct. I conducted a survey that showed that the passenger pigeon’s genre is fading among young people. Failing to spread the parable of the passenger pigeon is a threat to every currently endangered species and their respective genres.
Introduction
The passenger pigeon (Ectopistes migratorius) lived in North America and was described as having a “small head and neck, long tail, and beautiful plumage” [1]. In its prime, it had the largest population size of any bird species at the time but went extinct due to overexploitation and habitat loss caused by European settlers [1]. The dynamics of human-wildlife interactions over time defined the progression of the passenger pigeon’s recorded history.
These interactions varied based on how the dominant people in North America thought about the bird and the genre surrounding its existence. Genre refers to “repeating rhetorical situations” to aid human interaction. In other words, it is the collection of how people refer to a specific topic. The definition of genre can be applied broadly. Genres are dynamic and develop over time, as people face new situations to apply them to. Every species has its own genre surrounding its existence. People participate in many genres on a daily basis, even if they do not know it. Genre is a “social action,” people shape genres and genres shape people [2]. The way groups of people collectively feel about anything is communicated through language. Thus, looking at the way people talked about passenger pigeons explains the processes that led to their downfall. The passenger pigeon is an effective ambassador for teaching youth about conservation because of the population’ rapid decline.
Historical Evolution
Indigenous People
The passenger pigeon’s parable begins with Indigenous people who lived within the range of the bird, mostly covering only the Eastern half of America [1]. These Indigenous people were the first humans to interact with the passenger pigeon and create its genre. Simon Pokagon, a Potawatomi tribe member was interviewed about seeing them in flight: “When a young man I have stood for hours admiring the movements of these birds. I have seen them fly in unbroken lines from the horizon…” [3]. Under Indigenous peoples’ care, the passenger pigeon numbers rose beginning in the years 100 to 900 C.E. [4]. This was because Indigenous people and passenger pigeons had a well-balanced relationship that allowed both populations to thrive.
Indigenous people carefully interacted with the passenger pigeon because it was an important game bird to them, second only to the wild turkey [1]. The passenger pigeon was a staple food of the Seneca, who named the bird jah’gowa, meaning “big bread” [4]. Tribes followed specific procedures for hunting the birds. Almost all tribes had a strict policy—based in both religion and biology—against taking nesting adult passenger pigeons. This strategic wildlife management policy promoted chick survival by allowing parents to care for their young. The Sioux and the Iroquois League were among those known to enforce their rules on other hunters. Instead of hunting the birds during this time, they often used nests as an opportunity to closely observe the bird. Individual tribes also had additional policies. For the Ho Chunks, hunting of the passenger pigeon could only happen if the chief held a feast. When the birds returned in spring, they offered much needed seasonal food. Before the Seneca began the hunt, they monitored the nests until the chicks were two to three weeks old. The Seneca even went as far as managing the habitat of the passenger pigeon, for instance they did not allow the cutting of any tree a “chief” pigeon nested in [4]. Chief Pokagon of the Potawatomi tribe credits strategies such as this for not only allowing the pigeon to maintain its numbers but actually increasing them [1]. By thinking about the needs of the pigeons and adjusting behaviors to accommodate those needs rather than freely hunting them, the population was able to continue on as a reliable food resource for the tribes that used them.
Furthermore, the connection between the two species went beyond the typical predator and prey relationship. To many Indigenous people, the pigeons were not just food, they were a being. Passenger pigeons were included in the religion of some tribes through stories, song, and dance [1]. The Seneca believed that the pigeon gave its body to create their children. The passenger pigeon was so important to the Seneca that they termed albino ones “chief of all pigeons” and strictly forbade hunting them. The Cherokee and the Neutrals told similar stories of the bird as a guide to avoid starvation. The Seneca and the Iroquois opened their Maple Festival every year with a dance song about the bird. The Cherokee Green Corn Festival featured a dance mimicking a pigeon hawk in pursuit of a pigeon [4]. The pigeons held value in the lives of the people who benefited from them.
European Arrival
The Europeans recorded their first passenger pigeon on July 1, 1534 [1]. Right away, colonizers of every walk of life made note of the massive number of pigeons Indigenous people had maintained. The average European enjoyed the sight, “…I was perfectly amazed to behold the air filled and the sun obscured by millions of pigeons…” [1]. Many accounts told the narrative of an undiminishable population. Schorger, a professional ornithologist confirmed this notion, stating that “no other species of bird, to the best of our knowledge, ever approached the passenger pigeon in numbers” [1]. More ornithologists like Alexander Wilson took records, “In the autumn of 1813…I observed the pigeons flying from northeast to southwest, in greater numbers than I thought I had ever seen them before…The light of the noon day was obscured as by an eclipse” [5]. Even Leopold described them as a “biological storm” that used the resources of the land to their advantage [6].
While everyone knew the birds to be copious, not everyone understood the science behind it. Ornithologists knew that the pigeons could thrive because they had ample food and habitat when the Europeans arrived. However, for the vast majority of Europeans who were not trained in biology, the flock of birds blocking out the sky was frightening and unexplainable. This is where the genre began to separate itself from Indigenous peoples’ understanding of the bird. Europeans constructed urban legends in an effort to explain what was unknown to them. When only one acorn was found in pigeons’ crop (food storage pouch), Europeans predicted death and sickness. The evidence they saw supported their beliefs, “It is a common observation in some parts of this state, that when the Pigeons continue with us all the winter, we shall have a sickly summer and autumn” [1].
Diminishing
As colonizers made themselves more at home in North America, they encroached on passenger pigeon habitat and depleted their numbers. The colonizers did not take wildlife management into consideration while hunting. Instead, they killed far more than they took and failed to leave young and nesting birds alone [1]. Extinction seemed entirely impossible, they did not see a need to ensure the next generation of pigeons could continue, it was understood as a given. To compound this, the birds were generally not thought of highly in European cultures. The passenger pigeon was merely a thing to exploit rather than a being to feel for. They began to disappear from the places humans occupied, retreating into what wilderness remained [3].
People began to notice that the passenger pigeon populations were fading. Some states, like Ohio, actively avoided policies to protect the species, claiming “the passenger pigeon needs no protection. Wonderfully prolific, having the vast forests of the North as its breeding grounds, travelling hundreds of miles in search of food, it is here to-day, and elsewhere to-morrow, and no ordinary destruction can lessen them or be missed from the myriad that are yearly produced” [1]. Other states, particularly Wisconsin, wrote laws to protect the species: “It shall be unlawful for any person or persons to use any gun or guns or firearms, or in any manner to main, kill, destroy, or disturb any wild pigeon or pigeons at or within three miles of the place or places where they are gathered for the purpose of brooding their young, known as pigeon nestings”. Laws along these lines were enacted in several states but no efforts were made to actually enforce them. Much of this was due to pushback by settlers to the laws. Farmers in particular protested any enforcement, worrying that allowing the pigeons to thrive would mean crop destruction [1]. To them, the pigeons were pests to get rid of, not preserve.
Gone
The last passenger pigeon, named Martha, was a resident of the Cincinnati Zoo up until her passing on September 1, 1914. She was named after First Lady Martha Washington and was housed with her companion named George. The pair never produced fertile eggs, the zoo’s captive breeding effort was too late to save the population [1]. The end of the species “removed more individual birds than did all the other 129 [previously recorded bird] extinctions put together”. They went extinct because of introduced species, chains of extinction, overexploitation, and habitat loss—all four of these were human-driven factors. Captive breeding, regulating hunting, and habitat protection could have saved them. However, these efforts were seldom made and not done early enough in the population’s decline [3]. The passenger pigeon was lost because of the genre Europeans created for it during the time it was still around.
The majority of non-indigenous Americans only appreciated the passenger pigeon and shifted their genre once they were no longer around. People now found a soft spot for the birds in their memories, “Alas, the pigeons and the frosty morning hunts and the delectable pigeon-pie are gone, no more return”. Artists incorporated these fond memories into their paintings, poems, and music. Monuments were erected around the United States inscribed with laments such as “this species became extinct through the avarice and thoughtlessness of Man” and “the conservationist’s voice was heard too late” [3]. People regretted the fact that future generations would not get to see the bird in the sky so they attempted to etch the passenger pigeon into everyone’s minds [6].
Making an effort to remember the passenger pigeon is important because the species’ story functions as a lesson and a guide for the future. However, in the past decade, the passenger pigeon is being forgotten. Many high school students are not taught about the population’s time on Earth and why they are now gone, as shown by a case study in Pennsylvania [7]. If people are no longer talking about it then the same mistake will be made again. At the same time, there are also organizations, like the Project Passenger Pigeon founded in 2014, working to tell the parable “through a documentary film, a new book, their website, social media, curricula, and a wide range of exhibits and programming for people of all ages” [8]. However, a small group of thoughtful individuals will not be enough to save the next species from human destruction if the story of the passenger pigeon does not make it into enough of the right hands.
Experiment
Over the period of one month during March 2019, I surveyed teenagers regarding their passenger pigeon knowledge. At the time of the survey, the teenagers were high school or college students in the United States. The overall purpose of my survey was to investigate if young people are talking about the passenger pigeon in contemporary society. Of the fifty-one responses, three (6%) subjects spoke about the passenger pigeon accurately. Furthermore, eight (16%) subjects believed to know the true story of the passenger pigeon, all of those eight falsely stated that the passenger pigeon was used to carry messages. Eight (16%) subjects even claimed to have seen a live passenger pigeon since 2000.
My survey found that very few teenagers have heard the parable of the passenger pigeon’s extinction. This group of people has gone through a large amount of schooling in their lives so far without being taught about the passenger pigeon despite its intertwining with significant historical events. The convenient fact about passenger pigeons is that they feel familiar to people since most have seen today’s common pigeon, the rock dove. It is easy for the uninitiated to imagine what a passenger pigeon was like based off of what they know about rock doves. The parable of the passenger pigeons can be taught in any classroom—science, history, art, and more.
The experiment shows that the genre is not being passed on. This is exactly why the way contemporary society talks about this species and its genre matters. Education that advocates for proper wildlife management and policy is the key to saving species from extinction.
Conclusion
The passenger pigeon species went from the world’s largest bird population to complete extinction, due to mistreatment from European colonizers. My survey of high school teenagers shows that people are not learning from this parable. This species makes itself an ideal candidate because of the rapid severity of its decline. It is increasingly important that we care about our environment before it is too late to take action. The clock is ticking, the passenger pigeon told us so. If we can learn to mourn a bird we never met, we will not have the opportunity to mourn the birds we know.
Works Cited
- Schorger, A.W. The Passenger Pigeon: Its Natural History and Extinction. Wisconsin, University of Wisconsin Press, 1955.
- Dirk, Kerry. “Navigating Genres”. Writing Spaces: Readings on Writing, edited by Charles Lowe and Pavel Zemliansky, vol. 1, Parlor Press, 2010.
- Avery, Mark. A Message from Martha. London, Bloomsbury Publishing, 2014.
- Greenberg, Joel. A Feathered River Across the Sky: The Passenger Pigeon’s Flight to Extinction. New York, Bloomsbury USA, 2014.
- Wilson, Alexander. Wilson’s American Ornithology. Boston, Otis Broader and Company, 1853.
- Leopold, Aldo. A Sand County Almanac. New York, Oxford University Press, 1954.
- Soll, David. “Resurrecting the Story of the Passenger Pigeon in Pennsylvania.” Pennsylvania History: A Journal of Mid-Atlantic Studies, vol. 79, no. 4, 2012, pp. 507–519. JSTOR, www.jstor.org/stable/10.5325/pennhistory.79.4.0507.
- Project Passenger Pigeon. The Chicago Academy of Sciences and its Peggy Notebaert Nature Museum, 2012, passengerpigeon.org. Accessed 19 February 2019.
Abundance Estimates And Vegetation Preferences Of The Suisun Song Sparrow In The Interior And Along Edges Of Impounded Wetlands
By Debi Fanucchi, Oscar Garzon, Julia F. Herring, and Kevin M. Ringelman
ABSTRACT
The Suisun Song Sparrow (Melospiza melodia maxillaris) is a subspecies of Song Sparrow that is endemic to the Suisun Marsh of California. It is listed as a state species of special concern by the California Department of Fish and Wildlife due to its restricted range, small breeding population, and susceptibility to encroaching human development. The Suisun Marsh ecosystem is highly altered, and is comprised of both natural tidal wetlands and impounded wetlands that are cut off from the natural tidal cycle. Suisun Song Sparrows are believed to prefer natural tidal wetlands, but there is a lack of information on sparrow densities and habitat associations in impounded wetlands. To address this knowledge gap, we examined the abundance and habitat preferences of Suisun Song Sparrows in an impounded and heavily managed wetland complex. We conducted surveys at seven sites within interior managed (impounded) wetlands, and seven sites along edges between tidal and impounded wetlands. We found significantly more birds in interior sites than at edge sites, but within edge habitat, abundances were higher on the tidal side of the levee. We found that Song Sparrows used tall vegetation in greater proportion than its abundance, and specifically preferred bulrush and common reed as calling perches. Interior sites contained relatively less of this preferred vegetation than edge sites, suggesting that beneficial habitat heterogeneity in interior sites, and/or deleterious edge effects along the dikes may be important drivers of abundance. In the face of sea-level rise and shifting conservation priorities, many managed wetlands are expected to be converted back into fully tidal systems, and our results provide an important baseline for future research on the effects of tidal restoration.
Fecundity Decrease in Drosophila simulans in the absence of Wolbachia
By: Jack Taylor, Biochemistry and Molecular Biology ’15
Sponsor: Michael Turelli, Ph.D.
Ecology and Evolution
Many species of arthropods host Wolbachia, maternally transmitted bacteria that often influence host reproduction. This manipulation of host reproduction has contributed to Wolbachia becoming a normally-present infection of many Drosophila simulans. The Y36 isofemale line, a population of Drosophila simulans created from a single female collected in 2010 in Yolo County, produces flies which have an unusual phenotype when reciprocally crossed with uninfected simulans populations denoted U. The cross between Y36 male and U female produces female offspring with significantly lower fecundity than the reciprocal cross (Y36 female with U male). It is possible that this effect is a result of a paternal defect gene that is masked by the Wolbachia infection. Ten separate sublines were created from the Y36 isofemale line. The goal of my experiment was to determine whether this phenotype is still present among each of the ten sublines and if there is any variation in the phenotype among the sublines. Analysis of the data indicates that the qualitative effect of the phenotype has been preserved across the sublines but suggests that there is quantitative variation amongst them.
Dissecting the signaling pathway regulating early stages in parasitic plant, host plant interactions
By Lee Nguyen, Biotechnology ’14
Parasitic plants pose a serious threat to the world’s agriculture and environment. Understanding the parasitism signaling pathway will help identify methods of pest control as well as pest resistance. One gene that enters the parasitic signaling pathway early is TvQR1, a gene that encodes an enzyme that catalyzes an oxidation-reduction reaction crucial for development of a root like outgrowth called a haustorium. In parasitic plants, TvQR1 is transcriptionally activated upon host contact and my project is to study the promoter of this gene, pQR1, in a nonparasitic plant. (more…)