Cropland mapping using Earth observation derived phenological metrics

Satellite Earth observation provides timely and spatially explicit information on crop phenology that can support sustainable agricultural land management, in terms of natural resource protection and adaptation to climate and other environmental changes. Accurate classification and mapping of croplands is a primary information for environmental impact assessment in the agricultural sector. This research study presents a digital agriculture approach that integrates Earth observation big data analytics based on machine learning technologies to classify and map main crops types. Two supervised machine learning models were calibrated using Random Forest algorithm from phenological metrics, estimated from NDVI and LAI vegetation indices time series calculated using Copernicus Sentinel-2 MSI satellite acquisitions. Models were calibrated for the Toscana region in Italy, where reliable spatially explicit agricultural data were available. Results show that the proposed method can achieve a satisfactory overall accuracy (~78%) in croplands classification, and that model calibrated using phenological metrics estimated from LAI time series performs slightly better than the model calibrated using NDVI derived phenological metrics. The proposed approach offers a potential to accurately map crop types, offering a simple yet reliable way to derive important agricultural land management information and to support agricultural monitoring systems for large areas at field level over time.