Crop Yield Estimation at Field Scale in Sub-Saharan Africa Using Multisource Earth Observation Data

Abstract:

Accurate assessment of local crop yield and its spatial variation is a crucial need for policymaking, understanding determinants of yield gaps, and improving food security in Sub-Saharan Africa. Challenges of reliable crop yield estimation using earth observation data in small-holder farming systems include cloud-contamination of optical satellite data and coarse spatial resolutions compared to the small field size and heterogeneous landscape. We aim to tackle these challenges by combing statistical algorithms, in-situ data, and multisource satellite data at different spatial scales (1m, 3 m, 10 m, and 20 m), focusing on Malawi as a case study. Specifically, we aim to

1) Identify crop area and crop type using the Sentinel-1 and Sentinel-2 dataset.

2) Estimate yield at field level using high-resolution PlanetScope data and in-situ data on yield and leaf area index.

3) Test scaling effects on yield estimation accuracy and feasibility of using freely available satellite data (Landsat-8, Sentinel-1/Sentinel-2) for yield estimation in small-holder farming systems.

We hypothesized that either the usage of high-resolution satellite data (i.e. PlanetScope) or a combination of time series of optical (Sentinel-2) and radar imagery (Sentinel-1) can improve accuracy of crop area/type identification and yield estimation compared to using single-source coarse satellite data. We expect relevant results to be delivered and presented by December 2020.