A statistical evaluation of Earth-observation-based composite drought indices for a localized assessment of agricultural drought in Pakistan

Abstract

Drought is a complex phenomenon that impacts a multitude of sectors globally and is difficult to characterize due to the variation in defining conditions. With over 2.5 billion people dependent on agriculture for their livelihoods, agricultural drought impacts are particularly acute. Specific indicators based on different aspects of the hydrologic cycle can be used to better characterize drought and mitigate its impacts. The overall objective of this study is to develop a monthly Composite Drought Index (CDI) using Earth observation datasets to provide an assessment of droughts in Pakistan. Further, district level wheat production data was used to optimize variables to create a customized composite drought index (CCDI) specifically for agriculture and evaluate the two indices. Pakistans economy and communities rely heavily on the agricultural sector and many areas are at high risk of crop failure due to the dependence on precipitation and other environmental conditions. A total of 10 environmental variables are considered that account for supply and demand of water, soil moisture, and vegetative conditions. Statistically important variables are chosen for each district with respect to wheat production, creating a subset of the original inputs. Weights for each dataset are calculated using a Principal Component Analysis (PCA), identifying what variables contribute the most to the index. The CDI and CCDI are evaluated with nationally reported, district level, wheat production data, for the harvest years 2005–2017, carried out specifically for the Rabi season, October - March. The CCDI, on average used 5 variables, as compared to the full 10 in the CDI. Overall, the two composite indices were highly correlated and both captured well known climatological events. When compared to production trends, the CDI has the ability to identify agricultural droughts with a true negative rate of 0.742 in rainfed districts, while irrigated districts have a true negative rate of 0.568. For the CCDI the true negative rates in rainfed and irrigated districts were 0.667 and 0.602, respectively. The results distinguished the importance of each variable in the contribution to the CDI and CCDI. The findings from this study demonstrate the potential of using the methodology for the CDI to enhance pre-existing drought monitoring and forecasting systems in the region.