Uncovering Causal Pathways of Agricultural Droughts using Climate and Vegetation Signals

The impacts of climate change are directly visible in the intensification and increasing frequency of extreme climate events, such as floods and droughts. Since droughts result from complex, multivariate, and non-linear land-atmosphere interactions, understanding these relationships is crucial for developing impactful future measures to reduce or mitigate drought impacts. Many studies have performed correlation analysis among these variables, but (1) correlation does not fully resolve causality and complexity of drought occurrence, due to which (2) the nonlinear behavior of drought propagation remains poorly understood. This study applies conditional independence tests, such as the Peter and Clark Momentary Conditional Independence algorithm (PCMCI+), to identify and analyze the causal drivers of drought at different lag periods using multivariate time series data. We investigated the influence of seven important variables for drought incidences on drought-induced vegetation responses in the drought-prone Bundelkhand region of Central India as well as its seven districts (Banda, Chitrakoot, Hamirpur, Jalaun, Jhansi, Lalitpur, Mahoba) separately. We used ERA-5 land monthly data at 0.1˚ spatial resolution for climatic variables, including precipitation, temperature, evaporation, relative humidity, soil moisture, and biophysical variables as Leaf Area Index (LAI) and the Normalized Difference Vegetation Index (NDVI) which measures vegetation health via greenness used as a proxy for drought-induced vegetative stress, were taken from Peking University’s Global Inventory Modelling and Mapping Studies version 1.2 (PKU GIMMS) at 0.0833˚ spatial resolution. The analysis spans 32 years temporally from 1990 to 2021 and is carried out at a monthly scale by temporally aggregating the data through monthly averages.

For each variable, PCMCI+ measures partial correlation  as a function of the maximum time delay and the significance threshold applied.  Results here are presented for a maximum time lag of 3 months and a significance value of 0.05. At the investigated spatiotemporal scales, precipitation is the primary driver of soil moisture in Bundelkhand given a 3-month lag. Temperature primarily affects LAI with a 1-month lag, while accumulated warmth supports vegetation on longer timescales (3-month lag). Among atmospheric factors, relative humidity emerges as the strongest control on vegetation greenness and canopy development, influencing both NDVI, and LAI. The results also reveal important land-atmosphere feedback. The negative feedback between soil moisture, NDVI, and LAI indicates self-limiting plant growth under water stress 2-3-month lag. Vegetation contributes to surface cooling as expected, reflected in the inverse relationship between LAI and temperature. Furthermore, vegetation regulates evaporation, with NDVI affecting evaporation at a 2-month lag and LAI at a 3-month lag. Spatially, district-level patterns generally mirror the regional findings, except for Lalitpur, where fewer and different causal links were identified. Overall, the study shows that humidity-driven vegetation dynamics and multi-lag feedback between the land surface and atmosphere are central to drought evolution, highlighting the importance of explicitly representing these coupled processes in ecohydrological assessments. Future work should translate these identified causal pathways into next-generation drought monitoring and forecasting systems that incorporate lag-aware vegetation-climate interactions to improve drought early-warning capabilities and anticipatory mitigation planning.