Obtaining timely, accurate, and automated data on the spatial distribution and planting area of cotton is crucial for production management and informed trade decision-making. In this regard, remote sensing technologies are important and effective means. Methods based on machine learning, and deep learning, rely on a large number of training samples, which is time-consuming and laborious. The feature index method is limited in that the determination of the classification threshold is an obstacle to realizing automatic classification. This study proposed an automatic mapping method for cotton based on cumulative spectral, phenological characteristics, and spatial thresholds. First, an index was designed, which combined the cumulative spectral and phenological characteristics (CSP) within the cotton-growing season to effectively distinguish cotton from other features and contemporaneous crops. Second, using the maximum between-class variance method (OTSU) and Sauvola algorithms, a new local adaptive threshold method (Otsu–Sauvola) was developed for the automatic determination of the classification threshold. In this study, Xinjiang Province, with a planted area of 25,000 km2 and 84.94% of China's total production, was selected as the study area, and fully automated mapping experiments were performed using Sentinel-2 time-series images at four experimental sites with different planting structures. The overall accuracies of cotton classification at the four sites were 91.20%, 90.45%, 93.00%, and 91.80%, and the F1-scores were 90.85%, 90.33%, 92.62%, and 92.26%, respectively. In the absence of samples, the accuracy of the CSP method was comparable to that of SVM and RF-supervised classification results, which could be realized 60–70 days before cotton harvest. The CSP method was applied to 10 major cotton-producing counties in Xinjiang, and the MRE between the CSP-detected area and the statistical area was 14.1%. Further analysis revealed that the CSP index can accurately and effectively distinguish cotton from other features and contemporaneous crops and that the Otsu–Sauvola automatic threshold method has robustness and regional consistency, thus providing an automatic and effective method for large-scale mapping of cotton in the early growing season.

Grassland fires represent a significant natural disaster affecting global grassland ecosystems, posing serious threats to ecological safety, livestock production, and the lives and property of herders. This study focuses on Qinghai Province, utilizing historical grassland fire data to pre-screen the factors influencing both natural and anthropogenic grassland fires. By applying the maximum entropy model and analyzing data from 150 fire incidents, this research predicts the spatial distribution of grassland fire susceptibility within the study area. The outcomes were as follows. (1) A maximum entropy–Kvamme gain statistical model was developed and validated for reliability. (2) The natural grassland fire-prone areas are predominantly located in southern Qinghai, covering a vast portion of the Guoluo Tibetan Autonomous Prefecture, as well as Zeku County, Henan Mongolian Autonomous County, Yushu City, and Mado County in the Yushu Tibetan Autonomous Prefecture. These regions are distinguished by their proximity to rivers and the presence of abundant vegetation. In contrast, grassland areas prone to anthropogenic fires are mainly concentrated in eastern Qinghai Province, where human activities are more intensive and population centers are located nearby. (3) The prediction results identify dominant grassland fire factors and their thresholds. (4) Natural grassland fires in Qinghai Province primarily result from spontaneous combustion, while anthropogenic grassland fires are primarily caused by electrical faults from high-voltage power lines, heating activities related to grazing, and the use of firecrackers. This study presents a disaster prediction model to support grassland management and fire prevention in Qinghai Province, providing a scientific basis for effective fire control strategies.