Skip to main content

Master of Environmental Data Science
(2023)

Measuring Agricultural Adaptation to Climate Change in Zambia Using Satellite Imagery and Machine Learning

Group Members: Andrew Bartnik, Carl (Carlo) Broderick, Gabrielle Smith, Hailey Veirs

Faculty Advisors: Tamma Carleton

Client: UCSB Bren School

Deliverables:

Proposal

Technical Documentation

Final Presentation

Description

Climate change is anticipated to greatly impact agriculture in Sub-Saharan Africa, potentially exacerbating the existing food insecurity issues in the region. Agricultural data has been used to guide policy decisions and address food insecurity in other parts of the world. However, high-quality historical data is sparse in Sub-Saharan Africa due to resource constraints. Satellite imagery and machine learning approaches have successfully modeled agricultural and environmental variables in other countries, yet traditional machine learning approaches are computationally expensive and inaccessible to this region. In this project, we conduct a case study of Zambia to demonstrate a recently developed machine learning pipeline. We use the “Multi-task Observations using Satellite Imagery & Kitchen Sinks” (MOSAIKS) machine learning approach. In this approach, we use processed numerical data from satellite images and agricultural survey data to develop machine learning models that predict various agricultural variables over time. Our models use ridge regression to predict agricultural variables from the following groups: crop loss mechanisms, crop yields, harvested area and production, and tillage practices. The predictions from these models increased the spatial and temporal resolution of existing agricultural survey data in Zambia. We produced processed satellite images, regression models, and high spatial resolution agricultural predictions across Zambia. These products and our documented code serve the research community, machine learning users, and those who are interested in studying Zambia’s agriculture. The produced code and pipeline lowers the barriers to entry for machine learning and allows more groups and individuals to implement the MOSAIKS approach into their analyses.

Acknowledgements

Bren School: Tamma Carleton, Assistant Professor; Ruth Oliver, Assistant Professor; Naomi Tague, Professor

2022 Master of Environmental Data Science Alumni: Steven Cognac, Juliet Cohen, Grace Lewin, Cullen Molitor

Baylis Research Group, University of Illinois: Kathy Baylis, Professor; Protensia Hadunka, PhD Student

Sitian Xiong, PhD Student, Clark University Graduate School of Geography

Zambia Statistics Agency
 

arrow up icon