<p>Land cover and crop classification play a crucial role in agricultural monitoring and land development. However, existing models often struggle to fully exploit the spatial and temporal patterns in large-scale multi-temporal, multispectral datasets. In this paper, we propose a Temporal Attention U-Net model for multi-temporal land cover and crop classification. By integrating the attention mechanism, the model effectively captures complex spatiotemporal dynamics, focusing on relevant temporal features across time steps. Experimental results demonstrate that the model achieves high average classification accuracy and mean Intersection over Union (IoU) scores, outperforming existing methods with a relative gain of ~ 4.7% in accuracy and ~ 5.5% in IoU. Additionally, it shows significant performance improvements in key land use and crop classes, validating its effectiveness in tackling the challenges of multi-temporal classification tasks.</p>