Comparison of hydrological and vegetation remote sensing datasets as proxies for rainfed maize yield in Malawi

Abstract

Weather Index-based Insurances (WIIs) have emerged as a promising risk coping mechanism to compensate for weather-induced damage to rainfed agriculture. Remote sensing may provide cost-effective information capable of discriminating the weather spatial variability thus reducing the spatial basis risk, i.e., the mismatch between the weather-based index triggering the insurance payout and the actual damage experienced by the farmers, which is often one of the causes hindering the wide implementation of WIIs. In this work we assess which indices based on remote sensing datasets are the best proxy indicators for rainfed maize yield in Malawi. We analyse the spatial (district scale) and temporal (monthly) correlations of historical maize yield data and several remote sensing datasets including the Climate Hazards group Infrared Precipitation with Stations (CHIRPS) dataset, the ESA CCI Soil Moisture combined dataset (version 4.2), the Evaporative Stress Index (ESI) from the Atmosphere-Land Exchange Inversion model (ALEXI), the MOD13Q1 Normalized Difference Vegetation Index (NDVI) and Enhanced Vegetation Index (EVI). With respect to the previous literature, this work exploits a historical crop yield dataset at the sub-national level which allows us to analyse the correlation of the hydro-meteorological and vegetation variables at a higher spatial resolution than what is commonly done (i.e., at the national level using FAO national yield statistics) and ultimately explore the issues related to WII spatial basis risk. Results show that the correlations between crop yield and satellite datasets show high spatial and temporal variability, making it difficult to identify a unique WII index that is at the same time simple and effective for the entire country. Precipitation, particularly the standardized March precipitation anomaly, has the highest correlations with maize yield (with Pearson correlation values higher than 0.55), in Central and South Malawi. Soil moisture and NDVI do not add much value to precipitation in anticipating historical maize yield at the district scale. From a methodological perspective, our work shows that WII indexes are best identified by: i) considering datasets with fine spatial resolution, whenever possible; ii) accounting for the vulnerability of the different crop growing stages to water-stress; iii) distinguishing between water scarce and water abundant events.