A robust integrated agricultural drought index under climate and land use variations at the local scale in Pakistan

This study aims to develop an integrated agricultural drought index (IADI) that incorporates multiple remote sensing-based agricultural drought indices, considering variations in climate and land use at a local scale in ten different agro-ecological zones (AEZs) of Pakistan from 2000 to 2020. The IADI focuses on soil moisture stress to vegetation, utilizing the soil moisture condition index (SMCI) derived from various soil moisture datasets. The correlation between SMCI and the Standardized Precipitation Evapotranspiration Index (SPEI) is assessed at different time scales (1-, 3-, 6-, and 12-month) using Pearson correlation coefficient (r). Results indicate that SMCIERA5-Land is superior to others and is selected to develop the IADI. To account for local variations in climate and land use, agricultural drought indices (ADIs) are calculated on a 0.25°×0.25° grids. These ADIs are then averaged over a 3×3 window and analyzed using structural equation modeling (SEM) to understand the causal relationship between SMCIERA5-Land and ADIs. Temperature condition index (TCI) and vegetation health index (VHI) demonstrate a significant causal relationship with SMCIERA5-Land in most AEZs, while the relationship between temperature-vegetation dryness index (TVDI) and vegetation condition index (VCI) with SMCIERA5-Land is insignificant across the AEZs except in northern Pakistan. The dominance of land surface temperature (LST)-derived indices in most AEZs highlights the role of temperature in drought onset and propagation. To develop the IADI, Bayesian principal component analysis (BPCA) is employed to calculate dynamic weights between the selected ADIs and SMCIERA5-Land. The distribution of BPCA weights exhibits extreme variability across different AEZs, with some windows being sensitive to LST-derived ADIs and others to NDVI-derived ADIs. The average BPCA weights for VCI, TCI, VHI, and TVDI range from 9.42%–23.05%, 23.20–49.83%, 13.00–31.08%, and 16.38–25.50%, respectively. The IADI demonstrates the best fit with TCI and VHI, while showing good agreement with SPEI, VCI, and TVDI. The correlation between IADI and TCI/VHI ranges from 0.62/0.72–0.94/0.97. Furthermore, ordinary least square regression (OLSR) is used to assess the accuracy of IADI, ADIs, and SPEI in analyzing the impact of drought on wheat/maize yield at the district level in Punjab province, Pakistan. OLSR analysis reveals the dominance of IADI, followed by VHI, in analyzing the role of drought on crop yield. For example, in Rajanpur district, one unit increase in drought severity results in an 11.94 t/ha decrease in maize yield, with IADI explaining 59% of the variations in maize yield. Similarly, one unit increase in drought severity leads to a 7.22 t/ha decrease in wheat yield, with IADI explaining 49% of the variations in wheat yield. Overall, results indicate that the developed IADI effectively captures agricultural drought with high spatial and temporal resolutions across different agro-ecological zones, providing valuable insights for monitoring agricultural drought in data-scarce regions.