The integration of machine and deep learning models into Earth science applications has become increasingly crucial for analyzing time-series satellite data, particularly in monitoring environmental changes and land cover classification. This report presents the development and implementation of the sits_lstm_fcn function within the SITS package, an open-source R framework for satellite image time series analysis. Based on the LSTM Fully Convolutional Networks (LSTM-FCN) model, this function combines temporal convolutions and LSTM blocks to effectively capture both local temporal patterns and long-term dependencies in the data. The report details the theoretical background, model architecture, and training methodology, along with the challenges and solutions encountered during implementation. Results from experiments on multi-band satellite data demonstrate the model’s strong classification performance, achieving 85% accuracy on the valida tion dataset. By providing a user-friendly interface, this contribution expands the SITS package’s capabilities, enabling researchers to utilize advanced deep learning models with minimal programming effort.