Abstract The groundwater consumption for agriculture has increased since the green revolution, and its depletion severely threatens food security, especially in major rice-growing areas of Southeast Asia. This research investigated the spatiotemporal distribution of land use land cover (LULC) from 2000 to 2018 in a rice-dominated canal command area. The study compared the classification performance of two machine learning algorithms, i.e., Support Vector Machines (SVM) and Random Forest (RF). The time-varying response of LULC transition on groundwater dynamics was investigated using a 3-D numerical groundwater flow model (MODFLOW-NWT). The MODFLOW-NWT model was calibrated and validated with the observed hydraulic heads. The results indicated that RF outperformed SVM in overall classification during the testing period. The LULC of the command area revealed a seven-fold increase in built-up area from 19.12 km2 in 2000 to 133.72 km2 in 2018. Further, the Boro rice cultivated area has increased from 39.2 to 56.4 of the command area during the study period. The results of transient state calibration (R2?=?0.987, NSE?=?0.987) and validation (R2?=?0.978, NSE?=?0.974) of MODFLOW-NWT indicated satisfactory match between simulated hydraulic heads and observed hydraulic heads. The area under the hydraulic head of -32?m to -5?m was consistently increasing, which requires contemplation on the future sustainability of groundwater. The methodology and results of this study can be used for LULC classification in a heterogeneous landscape and accurate groundwater flow simulation in data inadequacy scenarios in major rice-growing areas of Southeast Asia.
展开▼
