Assessment of the performance of drought indices for explaining crop yield variability at the national scale: Methodological framework and application to Mozambique

Abstract

[Abstract:] Droughts are one of the most damaging and complex natural disasters in the world, and they frequently affect agricultural production. Drought monitoring is essential for decision-makers seeking to minimize the socio-economic impacts related to drought events. In this study, we propose a methodology to identify the most suitable drought indices and data sources for monitoring the impact of drought on crops. Mozambique is used as a case study, as it represents a challenging example because of its poor hydroclimatic monitoring network and a lack of disaggregated data for agricultural production. A total of seven standardized drought indicators (SPI, SPEI, SSI, SVCI, STCI, SVHI, and STWS) at different scales (1, 3, 6, and 12 months) were obtained from global databases and evaluated as possible predictors of the annual variability of agricultural yields at the national level. A statistical model of crop yields based on time series was used to measure the explanatory capacity of each index. SPEI and SSI were the most effective at detecting the country's historical drought records regardless of whether nationally averaged values or the percentages of area affected by drought (PAA) were used. However, PAA was found to be a more accurate predictor of variability in crop yields. The variability of most cereals (maize, millet and sorghum) was adequately explained by the PAA of SPEI-3, with that of other crops (cashew nuts, cassava, potatoes, tea, tobacco and vegetables) being explained by the PAA of SSI-12. Specific indicators were proposed for monitoring wheat and sugar cane. These results can directly support managers and decision makers in developing drought contingency plans in Mozambique. To further demonstrate the potential of this methodology, it should be tested in other regions with a greater availability of agricultural data, including spatial disaggregation.