Geodaysit 2023

Emma Izquierdo


Double Crop Mapping using Sentinel-2 Data in Support to Implementation and Monitoring of the 2023-2027 Common Agricultural Policy within Rural Development Interventions
Enrico Borgogno-Mondino, Filippo Sarvia, Emma Izquierdo, Francesco Vuolo

Sustainable agriculture is one of the main focus of the 2023 – 2027 Common Agricultural Policy (CAP). For this reason, the new CAP strategic plan presents greater ambitions on climate and environment action in comparison of the previous programming period and stronger incentives that promote climate- and environment-friendly farming practices (i.e. minimizing soil disturbance, organic and carbon farming, maintaining permanent ground cover and adopting combined rotations) are provided. Among the several options, avoiding bare soil conditions and consequently promoting cover crops, or even to cultivate two main crops in a year, can provide excellent benefits. In particular, soil erosion and nitrate percolation are limited and soil structure, fertility, organic carbon sequestration and adaptability to climate change are supported. Consequently, an estimation of how much cultivated area is currently managed in this way should be estimated. Within the farmer CAP application, single (i.e. winter or summer) and a double crop could be included even if more crops can indeed be cultivated afterwards. Accordingly, the scope of this research is to design and validate an approach to classify and map the fields where a crop cover maintenance is promoted rather than the single crop based on Copernicus Sentinel-2 (S2) data. The study area is located in Austria, where a representative sample of the main crop types cultivated in the region was derived from the declarations to the Integrated Administration and Control System (IACS) for the year 2021. The approach relies on the classification of reflectance data from S2 time series including nine vegetation indices that were used to identify single or double crop systems. For this purpose, two supervised classifiers were applied namely One-Class Support Vector Machine (OneClassSVM) and Random Forest (RF). Statistical measures such as Overall Accuracy and Cohen's kappa coefficient were derived from the confusion matrices and the differences between field data and mapping results were analysed. A new map showing single vs double-crop systems was generated for further spatial analysis and interpretation.

AIT Contribution
Sala Biblioteca @ PoliBa