Prediksi Hasil Panen Padi di Provinsi Bali Menggunakan Metode CNN-LSTM Berdasarkan Data Citra Satelit

Abstract

Rice production in Bali Province continues to face pressure due to land conversion and variations in agroecological conditions across districts, necessitating an accurate harvest prediction system as a basis for food planning. This study aims to develop a rice yield prediction model using a CNN-LSTM architecture that integrates Landsat 8 satellite imagery, CHIRPS rainfall data, and BPS Bali Province production data for the period 2018–2025. The data were processed through a series of preprocessing stages and modeled using three CNN-LSTM architecture scenarios evaluated with RMSE and MAE metrics. The results show that Scenario 1 achieved the best performance with an RMSE of 2,567.99, MAE of 1,853.56, and an error percentage of 33.96%. The model was generally able to follow the production trend patterns of rice across all nine districts/cities in Bali Province, although deviations remained under extreme fluctuation conditions. This study demonstrates the potential of integrating remote sensing data and CNN-LSTM architecture in supporting a more adaptive and accurate agricultural prediction system.