Comparative Analysis of Machine Learning Methods for Imputing Missing Daily Rainfall Data in Complex Himalayan Terrain

Rainfall data plays a pivotal role in modelling complex and nonlinear hydro‐meteorological systems. In this process, the availability and fidelity of rainfall data at different spatial and temporal scales are of utmost importance. Gaps in the rainfall data are inevitable, resulting from human errors and instrumental/sensor malfunctions. The sparse availability of station data in the complex terrains of the Himalayas compounds the problem. The imputation of these gaps requires robust methods with higher accuracy and precision to capture the observed rainfall skewness in frequency and magnitude. This study has considered gap‐filling of missing data at different elevations under three agro‐climatic zones in Himachal Pradesh, India. For that, seven machine learning methods were used to estimate the missing daily rainfall data, including multiple linear regression (MLR), support vector regression with radial basis function kernel (SVR‐RBF), K‐nearest neighbours (KNN), random forest (RF), multilayer perceptron (MLP), extreme gradient boosting (XGBoost) and Lasso regression. For a comprehensive evaluation of the estimations, the analysis was structured into two tiers: overall (the entire time series) and RIC (rainfall‐intensity‐classes) assessments, ensuring a more robust comparison of the imputation methods. The methods were evaluated using the coefficient of determination ( R 2 ), correlation coefficient ( r ), root mean square error (RMSE) and mean absolute error (MAE). The results showed that, in both overall and RIC analyses, MLP consistently demonstrated higher accuracy and lower estimation errors, making it the most suitable method for imputing missing daily rainfall data across different elevation zones in the complex Himalayan terrain considered in this study.