Temporal change in land cover (LC) patterns have profound implications on the natural ecosystem, biodiversity, climatic cycle and human well-being. As such, study of LC change is essential for proper land management and planning. Recently, deep learning-based methods applied on Synthetic Aperture Radar (SAR) data have become increasingly popular to cater to a number of geospatial applications including LC classification. In this work, a novel supervised algorithm - Spatial Attention-Enhanced Dual-Path Skip Connection Siamese Neural Network (SAE-DP-SCSNN) is proposed that combines Siamese architectures, attention modules and multi-path channel networks for LC change detection. Performance evaluation of SAE-DP-SCSNN with other state-of-the-art models using percentage of correct classification (PCC), false positive (FP), false negative (FN), overall error (OE) and kappa coefficient (KC) revealed superior and competitive performance of proposed model over popular benchmark datasets. Through this study, we also propose a novel benchmark dataset comprising of bi-temporal SAR image pairs of Kelavarapalli reservoir in Tamil Nadu, India to act as a new standardized resource for evaluating and advancing deep learning models in SAR data based change detection studies.