A spatiotemporal dataset for NDVI prediction on Sentinel-2 imagery
The dataset was designed for prediction of vegetation health on Sentinel-2 imagery and utilized in the AI4Copernicus service: “Long Short-Term Memory Neural Network for NDVI prediction" in which an LSTM neural network was trained in order to create an AI model for NDVI (Normalized Difference Vegetation Index) prediction with time-horizon of one month.
The dataset and the LSTM model was created for the SandMap platform, a tangible GIS educational platform which aims to inform and educate students and citizens about the advanced topics of AI and EO including forecasting using a spatio-temporal dataset. Forecasting in precision agriculture is extremely important, since farmers and related organizations are able to refine significant operations during the production process, increasing the quality and quantity of agricultural products.
The spatiotemporal dataset includes NDVI values for three regions representing data for training, validation and test respectively in several dates from January to September of 2023, aiming to predict the vegetation health of September or October NDVI values.
The dataset needs to be enriched with more data in time dimension but currently it can be used for initial experimentation with an LSTM neural network for NDVI prediction.