Satellite soil moisture for yield prediction in water limited regions

Abstract

Many parametric or index-based drought risk financing instruments are based on satellite-derived rainfall, temperature and/or vegetation health data. However, an underlying issue is that indices often do not perfectly correlate to the actual losses experienced by the policy holders. The resulting increased basis risk can diminish demand for parametric drought risk insurance. Remotely sensed soil moisture (SM) can help decrease basis risk in parametric drought insurance through 1) complementary and/or improved parameters and variables used in existing models such as the Water Requirement Satisfaction Index (WRSI), 2) shadow-models to cross-check, test or validate payouts triggered through other indicators and models or 3) potentially through development of a stand-alone product. Here, we will demonstrate the use of a combined Sentinel-1 and Metop ASCAT high resolution soil moisture dataset to predict yield and develop an early-warning yield deficiency indicator for Senegal.

Soil moisture is retrieved from the Advanced SCATterometers (ASCAT) on-board the Metop satellite series, which has an original spatial sampling of 12.5 km. Sentinel-1 backscatter data at 500 m spatial sampling is used to downscale the Metop ASCAT surface soil moisture data to 500 m. The underlying concept is the temporal stability of surface soil moisture: In the temporal domain surface soil moisture measured at specific locations is correlated to the surface soil moisture content of neighbouring areas, where neighbours with similar physical properties (like soil texture, land cover and terrain) show a higher coherence to the local surface soil moisture than others. In addition to soil moisture, freely available rainfall from Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) and Copernicus Global Land Service NDVI were used. All datasets were spatially resampled to a 500 m grid, temporally aggregated to monthly anomalies and finally detrended and standardized. Data on yields was obtained from the Food and Agriculture Organization of the United Nations (FAO). Data on crop growth areas is based on FAO Global Agro-Ecological Zones (GAEZ) information and Livelihood zones (2015).

First, regression analysis with yearly yield data was performed per EO dataset for single months. The EO datasets were aggregated over areas where the specific crop was grown. Secondly, based on these results multiple linear regression was performed using the months and variables with the highest explanatory power. The multiple linear regression was used to provide spatially varying yield predictions by trading time for space. The spatial predictions were validated using sub-national yield data from Senegal and reports from the African Risk Capacity (ARC).

The analysis demonstrates the added-value of satellite soil moisture for early yield prediction. Soil moisture showed a high predictive skill early in the growing season: negative early season soil moisture anomalies often lead to lower yields. NDVI showed more predictive power later in the growing season. Combining anomalies of the optimal months based on the different variables in multiple linear regression improved yield prediction. Especially at the start of the season soil moisture improves predictions, with the ability to explain 60% (groundnut), 63% (millet), 76% (sorghum) and 67% (maize) of yield variability. These findings are particularly relevant for parametric drought insurance, because an earlier detection of drought conditions enables earlier payouts, which then help to mitigate the development of shocks into serious crises with often long-lasting socioeconomic effects.

Based on the analysis a yield deficiency indicator can be developed, which can provide spatial information on yield deficiencies. Yield deficiencies were compared to sub-national yield information and WRSI information as reported by the African Risk Capacity end of season reports. Strong spatial correspondence was found between the yield deficiency indicator and WRSI. For example, for millet in Senegal for the drought 2019 strong yield deficiencies in the provinces of Ziguinchor, Fattick, Kaolack and Kaffrine and moderate deficiencies in Thies, Louga and Tambacounda were found. This corresponded to low WRSI as reported by the African Risk Capacity in its end of season report of 2019. The analysis shows very clearly that soil moisture can be a valuable tool for anticipatory drought risk financing and early warning systems.

This analysis was performed in collaboration with the World Bank Disaster Risk and Financing Program and Global Risk Financing Facility.