Diagnosis of GLDAS LSM based aridity index and dryland identification

The identification of dryland areas is crucial for guiding policy aimed at intervening in water-stressed areas and addressing the perennial livelihood or food insecurity of these areas. However, the prevailing aridity indices (such as UNEP aridity index) have methodological limitations that restrict their use in delineating drylands and may be insufficient for decision-making frameworks. In this study, we propose a new aridity index based on based on 3 decades of soil moisture time series by accounting for site-specific soil and vegetation that partitions precipitation into the competing demands of evaporation and runoff. Our proposed aridity index is the frequency at which the dominant soil moisture value at a location is not exceeded by the dominant soil moisture values in all of the other locations. To represent the dominant spatial template of the soil moisture conditions, we extract the first eigenfunction from the empirical orthogonal function (EOF) analysis from 3 GLDAS land surface models (LSMs): VIC, MOSAIC and NOAH at 1 × 1 degree spatial resolution. The EOF analysis reveals that the first eigenfunction explains 33%, 43% and 47% of the VIC, NOAH and MOSAIC models, respectively. We compare each LSM aridity indices with the UNEP aridity index, which is created based on LSM data forcings. The VIC aridity index displays a pattern most closely resembling that of UNEP, although all of the LSM-based indices accurately isolate the dominant dryland areas. The UNEP classification identifies portions of south-central Africa, southeastern United States and eastern India as drier than predicted by all of the LSMs. The NOAH and MOSAIC LSMs categorize portions of southwestern Africa as drier than the other two classifications, while all of the LSMs classify portions of central India as wetter than the UNEP classification. We compare all aridity maps with the long-term average NDVI values. Results show that vegetation cover in areas that the UNEP index classifies as drier than the other three LSMs (NDVI values are mostly greater than 0). Finally, the unsupervised clustering of global land surface based on long-term mean temperature and precipitation, soil texture and land slope reveals that areas classified as dry by the UNEP index but not by the LSMs do not have dry region characteristics. The dominant cluster for these areas has high water holding capacity. We conclude that the LSM-based aridity index may identify dryland areas more effectively than the UNEP aridity index because the former incorporates the role of vegetation and soil in the partitioning of precipitation into evaporation, runoff and infiltration.