Sugarcane in Telangana is highly sensitive to climatic variability due to its long growth cycle and water demand. This study develops a unified diagnostic framework integrating district-level yield data (1997-2022) with daily climate records to assess resilience and risk. Yield statistics were standardized across 33 districts, while climate predictors, including rainfall, temperature, solar radiation, and drought indices, were aggregated seasonally. Derived metrics such as growing degree days (GDD) and entropy captured thermal accumulation and variability. Five machine learning models were tested, with robustness evaluated under 5% Gaussian noise. Ensemble methods outperformed linear regression, achieving R 2 = 0.72-0.81 and RMSE = 480-620 kg/ha. SHAP analysis revealed monsoon rainfall and GDD as dominant drivers, explaining over 40% of yield variance, while entropy of radiation highlighted localized anomalies. Prediction uncertainty was greater in semi-arid districts, reflecting heightened vulnerability. Resilience indices ranged from 0.42 in high-variability districts to 0.78 in more stable zones. Composite risk scores classified 11 districts as high risk, 14 as medium, and 8 as low risk. The framework's novelty lies in combining robustness testing, SHAP-based interpretability, entropy diagnostics, and resilience-risk mapping into a single pipeline. This approach provides fine-scale, policy-relevant insights for climate-smart sugarcane management and establishes transferable standards for agrometeorological research. By situating Telangana as a case study, the findings demonstrate that variability and extremes, rather than mean climate conditions, are the dominant stressors for long-cycle, water-intensive crops, offering lessons for adaptive crop planning in similar agro-climatic regions worldwide.