Given the increasing demand for food production, accurately monitoring the spatiotemporal distribution of plastic-covered greenhouses (PCGs) is crucial for managing agricultural expansion and its environmental impacts. This study employs a random forest (RF) algorithm to create multi-temporal maps of PCGs in three intensive agricultural areas in Morocco (Loukkos, Gharb, and Souss-Mass). The classification is done using Landsat-8 (L8) and Sentinel-2 (S2) imagery through the Google Earth Engine (GEE) platform, with both object-based (OB) and pixel-based (PB) classification approaches applied.
The study compares RF performance across five feature scenarios, combining spectral bands, spectral indices, and texture information. The analysis covers PCGs’ spatiotemporal dynamics from 2018 to 2023. Results show that the Blue and SWIR bands, Normalized Difference Tillage Index (NDTI), Plastic-Mulched Landcover Index (PMLI), variance, and sum average texture are the most significant variables contributing to classification accuracy.
The study achieves an average overall accuracy of over 93.10% in mapping PCGs. The combination of all features (bands, indices, and texture) produces the highest overall accuracy (>94.9%) and F1-score (>91.8%). However, combining only spectral indices with texture also yields high accuracy (>94.6%). The OB and PB approaches demonstrate similar performance, with slight differences.
The analysis of PCGs’ estimated area reveals that the Souss-Mass area has the largest PCG coverage, approximately three times larger than the other areas, with an overall growth of +8.45% between 2018 and 2023. This research highlights the RF algorithm and OB-PB approaches’ capability to accurately identify PCGs in GEE. The findings can provide valuable insights for policymakers and agricultural planners to monitor and manage PCGs expansion efficiently in intensive agricultural areas.
