AI-Enabled Climate–Phenology Coupling and Future Productivity Assessment for Semi-Arid Bundelkhand under CMIP6 Forcings

Phenological development in semi-arid landscapes is susceptible to climatic variability, making it a powerful early indicator of climate stress. The Bundelkhand region of Central India—characterised by erratic monsoon rainfall, frequent droughts, high atmospheric dryness, and fragile agroecosystems—offers a critical natural laboratory for assessing climate-driven phenological changes. This study develops an integrated framework that combines multi-sensor remote sensing, phenology–climate machine learning models, and bias-corrected CMIP6 projections to quantify how vegetation dynamics and crop productivity in Bundelkhand will evolve under future climate scenarios. A multi-year phenology record was constructed using Sentinel-2 NDVI/EVI (10 m) and MODIS MCD12Q2 phenometrics to derive start-of-season (SOS), end-of-season (EOS), length-of-season (LOS), and peak greenness. Temporal smoothing using harmonic regression and double-logistic models enabled robust extraction of phenological markers for croplands (rice, wheat, pulses), natural vegetation, and fallow systems. Historical climate variables—temperature extremes, monsoon onset variability, vapour pressure deficit (VPD), heatwave duration, and solar radiation—were obtained from IMD and NASA POWER datasets. Climate–phenology linkages were quantified using generalized additive models, Random Forest, LightGBM, and LSTM networks to capture nonlinear responses and climate lag effects.

Future projections were developed using five CMIP6 GCMs (ACCESS-CM2, MPI-ESM1-2-HR, MIROC6, NorESM2-LM, and FGOALS-g3) under SSP2-4.5 and SSP5-8.5 scenarios. Bias correction followed the ISIMIP3BAS protocol. ML-derived phenology models were then forced with CMIP6 futures to simulate phenological trajectories for the 2030s, 2050s, and 2080s. SHAP sensitivity analysis identified VPD, Tmax anomalies, and pre-monsoon rainfall deficits as dominant drivers controlling phenological timing in Bundelkhand’s water-limited environment. Results reveal region-wide advancement of SOS by 10–25 days, shortening of LOS by 6–20 days, and reductions in peak greenness due to compounded heat and moisture stress—particularly under SSP5-8.5. Projected declines in vegetation productivity range from 12–30%, with drought-prone districts (Tikamgarh, Chhatarpur, Mahoba, Hamirpur) emerging as phenological stress hotspots. These shifts threaten major rabi crops (wheat, gram, mustard) and already stressed natural vegetation. By integrating phenology–climate modelling, remote-sensing dynamics, and CMIP6 climate trajectories, this study provides a first-of-its-kind, high-resolution assessment of how Bundelkhand’s vegetation will respond to future climate change. The framework supports climate-smart agricultural planning, phenology-based early-warning systems, and long-term drought adaptation strategies in one of India’s most climate-vulnerable regions.
Keywords: Vegetation Phenology; Climate Change; Bundelkhand; CMIP6 Projections; SHAP sensitivity analysis