Object-Based Mapping of Plastic Greenhouses with Scattered Distribution in Complex Land Cover Using Landsat 8 OLI Images: A Case Study in Xuzhou, China

The extraction of plastic greenhouses (PGH) has been addressed by remote sensing mainly in areas with densely distributed PGH and simple land cover. However, there still remain substantial challenges in extracting the PGH with scattered distribution in some developing agricultural regions. This paper proposed a threshold model to extract the PGH with scattered distribution from Landsat 8 OLI (L8 OLI) images. The threshold model was created by seven discriminative features with the help of an object-based image analysis and spectral analysis. Then, the proposed model was successfully applied to extract the PGH in Xuzhou city of Jiangsu Province, China, a large area of 11,258 km(2) with scattered PGH distribution and complex land cover, in 2014 and 2018. A total of 18,000 random points were generated to evaluate the extraction accuracy. The evaluation results show that the overall accuracy was higher than 98%, the producer's accuracy was over 85%, and the user's accuracy was over 95%. A visual interpretation with the high-spatial-resolution Google Earth Images also manifests the effectiveness of our results. Meanwhile, the applications in different years demonstrate the temporal consistency of the results. The study indicates that our presented method has large potential to map the PGH in a large spatial scale over a long-term period.