Skip to main content

Master of Environmental Data Science
(2022)

An Open Source Pipeline for Remote Sensing of Crop Yields Under Environmental Change in Sub-Saharan Africa

Fruits and avocados on market table

Group Members: Steven Cognac, Juliet Cohen, Grace Lewin, Cullen Molitor

Faculty Advisors: Tamma Carleton

Client: UCSB Bren School

Description

Remotely-sensed and large-scale agricultural data is widely available globally for many countries, but agricultural data specific to sub-Saharan Africa is limited. This leaves policymakers, businesses, and researchers unable to accurately forecast short-run food security risks posed by weather events or generate long-run climate change projections. Machine learning approaches paired with high-resolution satellite imagery have recently improved predictions, though they come at a high computational and financial cost. Reducing these costs will enable the study of how temperature, rainfall, and in turn climate change affect predictions of agricultural productivity across sub-Saharan Africa. The novel machine learning approach, MOSAIKS, dramatically decreases the computational and financial cost of using satellite imagery in machine learning. Reducing these barriers democratizes access to environmental monitoring and empowers data-poor countries to inform their own climate and social policies.

The MOSAIKS machine learning approach will be used to make fine-resolution agricultural crop yield predictions over time across all of sub-Saharan Africa through the following steps:

  1. First, the satellite imagery over Nigeria, Tanzania, and Zambia will be summarized over annual growing seasons and merged with the respective administrative crop yield data.
  2. Subsequently, the resulting dataset will be used to train a machine learning model on the relationship between the MOSAIKS satellite imagery features and the administrative crop yield data.
  3. Finally, the trained model will then be applied to the remaining countries in sub-Saharan Africa to make annual crop yield predictions from 2000-2020.

This overall process will improve the accuracy of the featurized MOSAIKS machine learning approach, which can then be applied to other dynamic environmental and socioeconomic factors impacted by climate change. 
 

Client contact: Tamma Carleton, Bren School

arrow up icon