A novel index for robust and large-scale mapping of plastic greenhouse from Sentinel-2 images

Abstract

As an efficient mode of modern agriculture, plastic greenhouse (PG) has significantly increased crop yields, but it is also criticized for changing the agriculture landscape and posing a threat to the environment. Accurate and timely information on PG distribution is essential for the strategic planning of modern agriculture as well as the projection of the environmental impacts. However, PG mapping over a large area has been a long-term challenge. Compared with classifier-based methods, index-based methods have the advantages of fast speed and convenience, which are very suitable for rapid large-scale mapping. The existing PG indices face the diversity of PG types and background environments, and the seasonal variation of PG spectra. To solve these problems, this study proposes a novel spectral index using Sentinel-2 images, namely the Advanced Plastic Greenhouse Index (APGI), to map PGs at a large scale. Four typical PG planting regions in the world, including Almería (Spain), Anamur (Turkey), Weifang (China), and Nantong (China), were selected as study areas. Based on the spectral analysis, some common spectral characteristics of PGs (i.e., high reflectance in NIR wavelengths and strong absorption in red and SWIR2 wavelengths) were observed and used in the APGI for highlighting PG areas. Besides, the coastal aerosol band and the red band were selected as optimal indicators to distinguish PG from other land covers which share similar spectral characteristics with PG. The experimental results indicate that the APGI has obvious advantages in enhancing PG information and suppressing non-PG backgrounds in various scenes compared with the existing indices. The APGI achieved the PG mapping accuracy with an OA of 90.63% ~ 97.50% and an F1 score of 80.56% ~ 96.20% in all study cases. Furthermore, the APGI also showed its robustness in seasonal variations and different datasets.