Detecting abandoned citrus crops using Sentinel-2 time series. A case study in the Comunitat Valenciana region (Spain)

• DOI: 10.1016/j.isprsjprs.2023.05.003
• Publication Date: 2023-05-24T16:33:13Z
• Authors: Sergio Morell-Monzó; María-Teresa Seba...; Javier Estornell; Enrique Moltó
• Keywords: Abandonment (legal); Agricultural land

Agricultural land abandonment (ALA) is becoming a growing phenomenon around the world that needs to be monitored and quantified. A massive of citrus orchards has been experienced in last years Comunitat Valenciana (CV) region (Spain) driven by different socio-economic factors. Therefore, developing time cost-efficient methods for monitoring ALA priority. Citrus are perennial crop trees which make have low spectral variation during year. In CV region, they planted relatively small parcels, thus creating highly fragmented heterogeneous landscape. This study proposes machine learning-based classification framework uses annual series indices extracted from Sentinel-2 images identify status at parcel level. The method based on features reconstructed OSAVI NDMI used train Random Forest classifier. Then, parcel-based performed using boundaries probabilities belonging each category full pixels inside boundaries. research assessed potential three statuses crops (non-productive, productive, abandoned). Results temporal spatial datasets provided an overall accuracy ranging 89 92 %, demonstrating importance multi-temporal data crops. Furthermore, we studied ability model spatially temporally transferred. Limitations recall abandoned parcels when models trained other areas or periods exposed, opening way improvements.