An Artificial Intelligence-Based Framework for Modeling Vegetation Photosynthesis Anomalies in India Corresponding to...

Climatic oscillations substantially influence the interannual variability of the land carbon cycle and global food anomalies through teleconnections, underscoring the need to understand and predict terrestrial vegetation variability in response to large-scale oceanic perturbations. This study scrutinized the temporal variability of vegetation photosynthesis in India using solar-induced chlorophyll fluorescence (SIF) anomalies and evaluated the response of Indian vegetation activity to sea surface temperature (SST) vacillations in the tropical Pacific, the strongest interannual climate variability. SIF variability over the Indian mainland demonstrated a robust inverse association with Niño 3 SST anomaly (R = -0.68) during June-August (JJA) over the period 2001-2020. Precipitation and temperature variations over India mirror interannual Niño 3 SST and SIF fluctuations, validating their mediating role in ocean-vegetation interactions. This study advances the representation of oceanic influences on vegetation photosynthesis by modeling SIF anomalies over India through a non-linear framework that integrates dominant tropical SSTs with local hydroclimate anomalies. A specifically parameterized multi-layer perceptron (MLP) architecture was configured to capture the dynamic SST teleconnections on broad-scale interannual vegetation variability in India. From the fine-tuned model iterations, only the good-fit models were screened out based on the loss function, followed by a systematic ensemble approach to obtain a stable and generalized predictive model for accounting SIF anomalies based on SST variability. The top ten good-fit models performed satisfactorily on the test data (R² = 0.55, RMSE = 0.006 Wm^-2 μm^-1 sr^-1). Data augmentation incorporating the seasonal lags (1-4) of predictors elevated the ensemble model performance (R² > 0.7). The best-performing ensemble at lag 1-3 (Ensemble_JJA,1-3) effectively captured SIF variability in JJA (R² = 0.9, RMSE = 0.004 Wm^-2 μm^-1 sr^-1). Explainable AI revealed the substantial contribution of Niño 3 and western Indian Ocean SST anomalies (~45%) to vegetation photosynthetic variability in the developed MLPRegressors. This AI-enabled non-linear SST-SIF advances the capacity for early warnings of vegetation stress in India.