Detailed national-scale land use/land cover (LULC) maps are critical for managing natural resources, urban planning, and addressing food security in disaster-prone countries like Bangladesh, yet high-resolution time-series maps have historically been lacking. To address this gap, this study generated 10-meter high-resolution LULC maps for Bangladesh for 2020 and 2023 across 14 distinct categories—including locally critical classes such as jhum, brickfields, and salt pans—to evaluate spatio-temporal land dynamics. The methodology integrated multitemporal Sentinel-2 optical data, Sentinel-1 C-band SAR, PALSAR-2 L-band SAR, and ancillary topographical and OpenStreetMap data, utilizing JAXA's SACLASS2 3D Convolutional Neural Network (CNN) deep learning model across six annual seasonal composites. Reference data for 2020 were acquired from 2023 baseline points using an automated spectral angle and a Euclidean distance migration technique. The data were compiled through a combination of field observations, random sampling and label checking from existing high-quality LULC maps, and manual interpretation of satellite imagery. A total of 67,034 reference data were used for the year 2023 and 45,515 for the year 2020. The resulting maps achieved high overall accuracies of 94.55 ± 0.41% (Kappa: 0.93) for 2020 and 94.32 ± 0.42% (Kappa: 0.93) for 2023, revealing substantial land transformations between the two years. Specifically, brickfields expanded by 56.11 ± 25.33%, aquaculture by 20.17 ± 4.94% , and single cropland by 17.23 ± 4.08% , whereas salt pans decreased by 47.38 ± 16.32%, bare land by 23.77 ± 4.27%, built-up areas by 13.83 ± 2.56%, and natural mangroves by 11.36 ± 2.44%. Ultimately, the fusion of multisensory imagery with deep learning enables more accurate and spatially detailed land cover classification than existing global datasets, providing vital spatial information for sustainable urban, agricultural, and environmental policy in Bangladesh.