Evaluating Crop Yield Sensitivity to Climate Change Using a Deep‐Learning Framework Incorporating Vegetation and Groundwater Memory

Abstract

Abstract Crop yields integrate contemporaneous weather and lagged vegetation and subsurface water states that reflect eco‐hydrological memory. However, these controls are observed only during the satellite era, creating a temporal mismatch between multi‐decadal crop‐yield records and limiting data‐driven models ability to learn antecedent vegetation and groundwater effects. We developed an interpretable spatiotemporal framework to reconstruct historical crop yields responses and assess future climate‐scenario impacts in Northeast China (NEC). Central to this framework is the reconstruction module (ReGen‐Net), which extends vegetation Normalized Difference Vegetation Index and groundwater storage‐anomaly memory back to 1981, bridging satellite period observations with four decades of yield records. Model validation used a temporal hold‐out design to prevent leakage and emulate forecasting conditions. The final GATAE‐TFT model achieved test R 2 values of 0.857, 0.777, 0.798, and 0.837 for maize, rice, soybean, and wheat, respectively, and was used for spatially informed yield assessment and future projections. YieldTransformer was used as a temporal baseline, whereas GATAE‐TFT served as the final spatially informed model for scenario‐based projections. Bias‐adjusted Inter‐Sectoral Impact Model Intercomparison Project projections from 2017 to 2060, indicate gradual declines in maize, rice, and wheat yields, while soybean yields remain stable under both SSP1‐2.6 and SSP5‐8.5. SHAP‐based attribution identifies vegetation greenness and hydro‐climatic memory as the most influential predictors across the crops, with crop‐specific sensitivities to drought indices and groundwater anomalies. These climate‐conditioned yield trajectories provide a physically grounded and interpretable benchmark for assessing water‐stress risks to crop production in NEC, China's primary grain‐producing region and a critical node in national food security under climate change.