A Comprehensive Benchmarking of the Available Spectral Indices Based on Sentinel-2 for Large-Scale Mapping of Plastic-Covered Greenhouses

Abstract

Plastic-covered greenhouses (PCG) have been extensively used in agricultural practices around the world. Remote sensing based on spectral indices is a key asset tomonitor the spatial distribution of these structures on a large scale. The primary objective of this research was to conduct a comprehensive benchmarking of the available spectral indices based on Sentinel-2 data for largescale PCG mapping. For that, eight PCG indices were thoroughly analyzed by systematically investigating their optimal thresholds in five study sites located in Almería (Spain), Antalya (Turkey), Agadir (Morocco), Weifang (China), and Nantong (China), including also different growing seasons. The experimental results demonstrated that the Plastic GreenHouse Index (PGHI) achieved the best PCG mapping accuracy in almost all study sites and growing seasons tested. Fromthe visual analysis carried out on thePGHI mapping results, it was made out that the main misclassification between PCG and background classes took place in water bodies and industrial building land covers, particularly in the Weifang and Nantong study areas. Based on this fact, the original version of PGHI was modified by adding two processes aimed at masking water bodies and industrial buildings. This new composite index, called Improved PGHI (IPGHI), attained better accuracy results in all study sites, especially in Chinese PCG areas. The average F1 score calculated for all the study cases improved from86.05% using PGHI to 90.51% applying IPGHI. The new approach provided a significant and robust improvement in PCG large-scale mapping for several types of PCG sites, even considering different growing seasons.