HOME ABOUT RESEARCH CV CONTACT
ANNA BOSER

ABOUT ME

Environmental science, remote sensing, and machine learning.

I am a PhD student at the Bren School for Environmental Science and Management at UC Santa Barbara. Ashley Larson and Kelly Caylor co-advise me, and my work is supported by the NSF GRFP and the Eugene Cota-Robles Fellowship. I hold a BA in Statistics from UC Berkeley, and am the 2020 recipient of UC Berkeley's University Medal.

Photo of Me

I am a statistician, remote sensor, and environmnetal scientist. My primary interests lie in using data-driven approaches, especially remote sensing and machine learning, to understand earth systems that humans rely on for health and wellbeing. I am most interested in food and water security. However, my previous work spans West Nile virus risk mapping, public health effects of prescribed burns, fine particulate matter gridding, HIV in French Guiana, early marriage in Niger, and the effect of school closures on the COVID-19 epidemic.

RESEARCH

MY WORK

Here are some of the projects I've been involved in.

agriculture

Saving water in California Agriculture

California is one of the most agriculturally productive and water stressed regions in the world. This project uses satellites and machine learning techniques to estimate crop water consumption in the California Central Valley. These estimates can be used to quantify the water saving potential of different water management strategies, such as crop switching, fallowing, and improvements in irrigation efficiency. This work has been accepted for publication in Nature Communications. See the preprint here and the Supplementary Information here.

stagg:: A data pre-processing R package for climate impacts analysis

stagg is an R package that transforms raw gridded climate data into tabular administrative-level variables intended for use in climate econometrics analyses. Flexible options let users control specifications like nonlinear transformations, weighted spatial aggregation, and temporal aggregation with a few lines of code. Our package GitHub is here, and a helpful cheatsheet can be downloaded here. It's a work in progress -- please send us feedback!

stagg
ML validation

Validating spatio-temporal environmental machine learning models

Machine learning has revolutionized environmental sciences by estimating scarce environmental data. However, methods for validating the models used to generate these data often ignore the spatial or temporal structure commonly found in environmental data. In this work, I invesitgate the impact such pitfalls can have on the interpretation of the validity of environmental machine learning models. I find that models with poor performance can appear to perform well when evaluated on the wrong data splits, and that this effect is exacerbated when not accounting for Simpson's paradox. The manuscript is in review.

West Nile Virus Risk Maps

West Nile virus (WNV) is the most prevaluent mosquito-borne disease in California, and mitigation hinges on fine grained understanding of the risk landscape. We use high resolution land surface temperature measurements from ECOSTRESS in conjuction with mechanistic understandings of WNV risk to map mosquito biting rates and transmission rates at unprecedented spatial and temporal scales. The manuscript is published in Environmental Research Letters, and a follow-on study further validating this method is in review.

WNV risk maps
Super-resolution

Downscaling remote sensing data using deep learning

High resolution remote sensing data has revolutionized the environmental sciences, but some key data types are only available at coarse resolutions. This project uses deep learning to downscale coarse resolution data to fine resolution data. We use a U-Net architecture to predict fine resolution land surface temperature from coarse resolution land surface temperature and high resolution red, green, and blue imagery. We additionally devise a strategy to minimize the amount of ground truth data needed to train the model.

COVID-19 in Schools

In an effort to curb the COVID-19 epidemic, government officials have halted the functioning of many institutions, including schools. However, many questions remain unanswered regarding the effectiveness of this intervention, and if and when to reopen schools. This project, conducted by the Remais Group at UC Berkeley, uses an agent based SEIR model to assess the effect of school closures on the COVID-19 epidemic in the Bay Area. The manuscript is published in the Journal of the Joyal Society Interface and can be seen here.

COVID modeling
Photo of French Guiana

HIV in French Guiana

According to the World Health Organization, French Guiana is the site of a generalized HIV epidemic. On the border of French Guiana and Brazil, the prevalence and severity of HIV cases is significantly more severe. In partnership with IDsante, a local NGO, this project aims to characterize the epidemic on the border and investigate the contibuting social, economic, and political factors. Read more about this research here, here, and here.

CONTACT

GET IN TOUCH

Santa Barbara, CA
Email: annaboser at ucsb.edu