Effects of Climate Change and Application of Causal Inference in Environmental Science

Souvick Bera Chair
Colorado School of Mines
 
Tuesday, Aug 6: 8:30 AM - 10:20 AM
5090 
Contributed Papers 
Oregon Convention Center 
Room: CC-B111 

Main Sponsor

Section on Statistics and the Environment

Presentations

A Bayesian Model of Citizen Science Data for Monitoring Environments Stressed by Climate Change

We propose a new method to adjust for the bias that occurs when citizen scientists monitor a fixed location and report whether an event of interest has occurred or not, such as whether a plant has bloomed. The bias arises as monitors note whether the event has happened upon arrival, lacking the precise day of occurrence. Adjustment is important because differences in monitoring patterns can make local environments appear more or less anomalous than they actually are, and the bias may persist when the data are aggregated across space or time. To correct for this bias, we propose a nonparametric Bayesian model that uses monotonic splines to estimate the distribution of bloom dates at different sites. We then use our model to determine whether the lilac monitored by citizen scientists in the northeast US bloomed anomalously early or late, preliminary evidence of environmental stress caused by climate change. Our analysis suggests that failing to correct for monitoring bias would underestimate the peak bloom date by 32 days on average. In addition, after adjusting for monitoring bias, several locations have anomalously early bloom dates that did not appear anomalous before adjustment. 

Keywords

nonparametric Bayes

monotonic splines

monitoring bias

bias correction

crowdsourcing

climate change 

View Abstract 2815

Co-Author(s)

Theresa M. Crimmins, University of Arizona
David Kepplinger, George Mason University
Ruishan Lin
E.M. Wolkovich, University of British Columbia

First Author

Jonathan Auerbach, George Mason University

Presenting Author

Ruishan Lin

Bayesian spatially-varying functional regression for crop yield modeling

Accurate crop modeling is essential to maintain food security in the United States, especially in the face of ongoing climate change. As climate conditions change regionally, it will be essential to understand both marginal and interaction effects of climate variables on crop yield.

We present a Bayesian Spatially-Varying Functional Regression model for crop yield. This model looks at the combined effect of two climate variables on crop yield. Previous models have either modeled the combined effect as a scalar term, or have used functional regression but neglected the interaction. The estimated weight functions from the functional regression are used to identify periods of vulnerability during different crop growth stages. We also account for spatial variability in modeling climate variable effects on yield.

We apply our model to recent crop yield data in the midwestern United States using the climate variables of vapor pressure deficit and soil moisture. We demonstrate the performance and interpretability of the model and compare to other crop yield models. 

Keywords

spatial statistics

functional data analysis

Bayesian statistics

crop yield modeling 

Abstracts


Co-Author

Veronica Berrocal, University of California, Irvine

First Author

Lynsie Warr

Presenting Author

Lynsie Warr

Spatial causal inference in the presence of sampling bias

Environmental data are often observational and exhibit spatial dependence, making causal effects of treatments or policies difficult to estimate. Unmeasured spatial confounders, i.e., spatial processes correlated with both the treatment assignment mechanism and the outcome, can introduce bias when estimating causal effects of interest since important assumptions in causal inference are violated. Spatial data can also be subject to preferential sampling, where sampling of locations are related to unmeasured confounders or the response variable, which introduces additional bias to the estimation of model parameters. We propose a spatial causal inference method that simultaneously accounts for unmeasured spatial confounders in both sampling locations and treatment allocation. We prove the identifiability of key parameters in the model and the consistency of the posterior distributions of those parameters. We also show via simulation studies that the causal effect of interest can be reliably estimated under the proposed model. The proposed method is applied to assess the effect of policies that govern marine protected areas on fish biodiversity. 

Keywords

Poisson process

Preferential sampling

Spatial confounding

Potential outcomes 

View Abstract 2745

Co-Author(s)

Brian Reich, North Carolina State University
Erin Schliep, North Carolina State University

First Author

Dongjae Son

Presenting Author

Dongjae Son

Estimating population change as heterogenous treatment effects with citizen science data

The increasing volumes of species observation data being collected by citizen-science projects around the world have great potential for monitoring populations and helping to identify the drivers of population change. However, to realize this potential requires methods that can 1) estimate heterogenous patterns of population change that arise when multiple drivers (e.g. change in land use and climate) affect species populations simultaneously, and 2) control for confounding sources of inter-annual variation common in citizen science datasets. In this presentation we investigate the use of machine learning-based estimators designed for Conditional Average Treatment Effect (CATE) estimation (including Causal Forests and meta-learners) to address these challenges. Using a simulation study and data from the citizen-science project eBird, we assess performance estimating spatially varying trends in population size and identifying drivers of population change in the face of real-world confounding. We discuss results showing how this approach can recover heterogenous trends and discuss outstanding challenges. 

Keywords

Biodiversity monitoring, Conservation, Ecology, Causal machine learning, Double machine learning, Spatiotemporal, Species distribution modelling 

View Abstract 3164

First Author

Daniel Fink, Cornell Lab of Ornithology

Presenting Author

Daniel Fink, Cornell Lab of Ornithology

Exploring spatiotemporal trends in air pollutants with Quantile Regression

Concern regarding climate change and its influential impact on humanity is the talk of the hour. Air pollutant levels in air are constantly monitored, and we use the United States Environmental Protection Agency's available resources to access the distribution of particular pollutants for a given number of sites, over the years. Various spatial locations have their spatially dependent pollutant′s quantile functions which varies with time. Using an approach of simultaneously modelling the quantiles, our aim is to reduce the computational complexity than the existing methodologies. We use a quantile regression method that uses functional principal components to reduce the dimensions over space and quantile levels while testing for trends in air pollution data over the last 20 years. Extensive comparison among the existing methods in literature is demonstrated. 

Keywords

spatial

quantile

regression

functional

computation

pollutants 

View Abstract 3357

Co-Author(s)

Ana-Maria Staicu, North Carolina State University
Brian Reich, North Carolina State University

First Author

Sukanya Bhattacharyya, North Carolina State University

Presenting Author

Sukanya Bhattacharyya, North Carolina State University

Inferring causal relationships between spatio-temporal processes using tail-descriptive estimands

We propose a latent spatio-temporal causal model for a class of causal estimands that go beyond the conditional expectation. In particular, we focus on estimands for contemporaneous and lagged effects that serve as descriptors of the tail behaviour of the predictive distribution of the underlying spatio-temporal process. Under mild sufficient conditions, we theoretically validate the correctness of causal interpretation and further prove: i) the identifiability of causal effects using the full observational distribution; and ii) the consistency of our model estimator. We provide a simulation study to illustrate the correctness of our asymptotic consistency theorem and showcase the advantages of using a causal estimand, that focuses on the tails, over the traditional conditional expectation. Finally, we apply our framework to quantify causal spatio-temporal structures in U.S. wildfire and air quality data. 

Keywords

air quality data

causal inference

extreme event

spatio-temporal process

tail-descriptive estimand

wildfire data 

Abstracts


Co-Author(s)

Jordan Richards, King Abdullah University of Science and Technology
Raphael Huser, KAUST
Marc Genton, King Abdullah University of Science and Technology

First Author

Zipei Geng

Presenting Author

Zipei Geng