This study developed machine learning models to predict the Vegetation Health Index (VHI) using predictors derived from the Standardized Streamflow Index (SSI), Standardized Precipitation Index (SPI), and lagged VHI values. Three machine learning approaches were evaluated: Stepwise Quadratic Regression (SQR), Random Forest (RF), and Artificial Neural Network (ANN). Correlation analysis revealed that SPI at 4-5 month timescales and SSI at 1–3 month timescales exhibited the strongest relationships with VHI, particularly during January-July when vegetation is most sensitive to drought conditions. The models demonstrated reliable forecasting capability for 1–3 month lead times. ANN showed superior performance during the dry season (February-April), achieving correlation coefficients (R) of 0.79-0.87, Willmott's index of agreement (d) of 0.86-0.93, and RMSE of 11.5-16.4 for one-month lead time forecasts. In contrast, the SQR model yielded comparable accuracy to ANN with lower RMSE values and demonstrated better performance for longer lead times and during the rainy season, while RF yielded the lowest performance among the three models. The relative contribution of predictor variables varied by lead time, with VHI dominating one-month forecasts while SPI and SSI played primary roles in 2–3 month forecasts. These findings provide a scientific foundation for developing early warning systems to monitor meteorological and hydrological drought impacts on vegetation.