Comparison of spectral indices extracted from Sentinel-2 images to map plastic covered greenhouses through an object-based approach

Abstract

One of the most important challenges of agriculture today is increasing its productivity gains, while controlling its environmental footprint. Because of that plastic covered greenhouses (PCG) mapping via remote sensing is receiving a great attention throughout this century. In this study, a fair comparison was carried out in four PCG study areas around the world to test 14 spectral indices mainly focused on the detection of plastic. To the best knowledge of the authors, this is the first research that fairly compares all these spectral indices in such variable number of study sites. The applied OBIA approach was based on the combined use of very high-resolution satellite data (Deimos-2 pansharpened images) to address the segmentation process and Sentinel-2 time series to compute the spectral indices. When dealing with Sentinel-2 single images, the Plastic GreenHouse Index (PGHI) stood out among all the indices tested in the study areas dedicated to the cultivation of vegetables, such as the cases of Almería (Spain), Agadir (Morocco) and Antalya (Turkey). Better Overall Accuracy (OA) values of 94.09%, 92.27%, 92.77% and 92.17% were achieved for Almería, Agadir, Bari and Antalya study sites, respectively, when using statistical seasonal spectral indices based on Sentinel-2 time series, being the maximum and mean values of PGHI (MAX (PGHI) and MEAN (PGHI)) the best ranked. Meanwhile, the PCG area of Bari (Italy), with a monoculture in vineyards, presented the worst and most irregular results.

Keywords

• plastic greenhouse
• spectral indices
• obia
• segmentation
• sentinel-2
• decision tree