Smoothed-LSTM: Advancing spatio-temporal NDVI prediction in semi-automated dataset for rice crop

Abstract

Accurate Normalized Difference Vegetation Index (NDVI) forecasting is crucial for effective agricultural planning. However, a good prediction of the same requires sufficient data, but structured data is not available in the public domain or open-source community. Also, most of the existing methods do not consider spatial information. This study presents a novel semi-automated dataset generation framework that utilizes Sentinel-2, POWER Data Access Viewer, and Google Earth Engine to create a comprehensive time-series dataset. We propose a smoothed long short-term memory (LSTM) model considering time series, historical meteorological, and spatial information. The proposed Smoothed-LSTM-based model outperforms Traditional-LSTM models, demonstrating its effectiveness in NDVI prediction for agricultural applications.