Many efforts have been devoted to extracting impervious surfaces based on different methods from multiple spatial resolution images. Differences in extraction methods and spatial resolutions are significant and have led to discrepant performances in terms of the impervious surface extraction accuracy. However, which extraction method is more suitable for which kind of spatial resolution image in practice is poorly understood. This study systematically compared the performances of 12 methods of impervious surface extraction for four spatial resolution images (i.e. Landsat 8 [30 m], Sentinel-2A [20 m], Sentinel-2A [10 m], and Gaofen-2 [4 m]) in three testing areas. The results indicated that for the medium-spatial resolutions of 30 and 20 m, the support vector machine (SVM) method was considered as the optimal classification method with the highest accuracy of impervious surface extraction. For the high-spatial resolutions of 10 and 4 m, the object based image analysis (OBIA) method obtained the highest accuracy of the impervious surface distribution. Furthermore, the perpendicular impervious surface index (PISI) outperformed the other indices in obtaining the impervious surface distribution, with the highest accuracy for four spatial resolution images. These comprehensive assessments can provide a valuable guidance for future impervious surface extraction from different spatial resolutions.

The widespread presence of mixed pixels in remotely sensed images is a pressing challenge for accurate target detection and classification. Linear spectral mixture analysis (LSMA) is commonly used to address this problem by deriving remotely sensed information at the subpixel level. In the implementation of LSMA, the effects of mixed-pixel spectral interference need to be taken into account; mixed spectra would exhibit as a pure spectral characteristic when the abundance of one endmember in a mixed pixel exceeds a specific threshold. However, the thresholds of endmember abundance resulting in mixed-pixel spectral interference remain unclear. Thus, this study designed an experiment to analyze the effect of the spectral interference of mixed pixels and to identify the thresholds causing such interference by spectral similarity measures (spectral angle and spectral distance). Four types of pure endmember spectra (vegetation, high-albedo impervious surface (HIS), low-albedo impervious surface (LIS), soil) and corresponding representative mixed spectra with endmember abundances of 95%–5% at intervals of 5% were collected from Earth Observing-1 Hyperion imagery. Spectral similarity measures among the pure endmember spectra and representative mixed spectra were used to determine the thresholds of endmember abundance that cause spectral interference. The results verified the effect of the spectral interference of mixed pixels. The thresholds of abundance causing mixed-pixel spectral interference in vegetation, HIS, LIS, and soil endmembers were 70%, 75%, 80%, and 70%, respectively. Therefore, when the endmember abundance within mixed pixels exceeds the abovementioned thresholds, these mixed spectra are interfered and would exhibit as a pure spectral characteristic. Accordingly, interfered mixed pixels have to be removed before applying LSMA or other unmixing methods to avoid the effect of spectral interference.

Ecological changes affected by increasing human activities have highlighted the importance of ecological quality assessments. An appropriate and efficient selection of ecological parameters is fundamental for ecological quality assessments. On the basis of remote sensing data and methods, this study developed an enhanced ecological evaluation index (EEEI) with five integrated ecological parameters by containing pixel and sub-pixel information: normalized difference vegetation index, impervious surface coverage, soil coverage, land surface temperature, and wetness component of tasseled cap transformation. Significantly, the EEEI simultaneously considered the five aspects of land surface ecological conditions (i.e., greenness, human activities, dryness, heat, and moisture), which provided an effective guide for the systematic selection of ecological parameters. The EEEI has a clear theoretical framework, and all the parameters can be obtained quickly on the basis of the remote sensing datasets and methods, which is suitable for the promotion and application of ecological quality assessments to various areas and scales. Furthermore, the EEEI was applied to assess and detect the ecological quality of the Guangdong–Hong Kong–Macau Greater Bay Area (GBA) of China. Assessment results indicated that the ecological quality of the GBA is currently facing great challenges with a degradation trend from 2000 to 2020, which emphasizes the significance and urgency for eco-environmental protection of the GBA. This provided evidence that the EEEI can be used as an effective index for scientific, objective, quantitative, and comprehensive ecological quality assessment, which can also aid regional environmental management and ecological protection.

It is well established that China’s rapid urban expansion has led to a substantial loss of cropland. However, few studies have examined how different urban expansion patterns contribute to cropland consumption, which has hindered the formulation of sustainable urban development and cropland protection policies. To fill this gap, we analyzed the occupation of cropland under three urban expansion patterns (leap-frogging, edge-spreading, and interior filling) in China from 1990 to 2020, using long-term land use data. The dominant driving forces of cropland loss were then explored using the XGBoost model and SHAP values. Our findings indicate that urban expansion in China from 1990 to 2020 resulted in a 6.3% reduction in cropland, with edge-spreading (4.0%) contributing the most, followed by leap-frogging (2.1%) and interior filling (0.2%). Change in urban intensity (CUI) proved to be the most critical driver of cropland loss, with SHAP values of 0.38, 0.28, and 0.37 for edge-spreading, leap-frogging, and interior filling, respectively. Over time, the driving forces evolved from a single demographic-economic dominance to a more diversified and integrated set of drivers. Based on these findings, we propose tailored planning and policies for different urban expansion patterns; for regions dominated by edge-spreading, stricter controls on urban boundaries and stronger land use planning constraints are required. For regions with prominent interior filling expansion, efforts should be made to improve internal land use efficiency while preserving existing cropland spaces. In regions characterized by leap-frogging expansion, further optimization of construction land allocation is needed to reduce the occupation of productive suburban cropland. These findings not only offer new empirical evidence for understanding the interplay between urban expansion and cropland conservation but also provide transferable insights that can inform sustainable land-use planning and cropland protection strategies in other rapidly urbanizing regions facing similar challenges.

Pest infestations significantly impact rice production and threaten food security. Remote sensing offers a vital tool for the non-destructive, rapid detection of rice pests. Existing studies often focus on laboratory conditions at the leaf level, limiting their applicability for precise pesticide application. Therefore, this study aimed to develop a method for detecting rice pests (rice leaf folders) in paddy fields based on unmanned aerial vehicle (UAV) hyperspectral data. Firstly, a UAV imaging system collected hyperspectral images of rice plants in both the jointing and heading stages. A total of 222 field plots for investigating rice leaf folders was established during these two periods. Secondly, 23 vegetation indices were calculated as candidates for identifying rice pests. Then, hyperspectral data and field investigation data from the jointing stage were used to construct a machine learning (extreme gradient boosting, XGBoost) algorithm for detecting rice pests. The results showed that the XGBoost model exhibited the best performance when eight vegetation indices were considered as the selected input features for model construction: the Red-edge Normalized Difference Vegetation Index (red-edge NDVI), Structure Insensitive Pigment Index (SIPI), Enhanced Vegetation Index (EVI), Atmospherically Resistant Vegetation Index (ARVI), Soil-Adjusted Vegetation Index (SAVI), Red-edge Chlorophyll Index (CIred-edge), Pigment-Specific Simple Ratio₆₈₀ (PSSR₆₈₀), and Carotenoid Reflectance Index₇₀₀ (CPI₇₀₀). The training and testing accuracies reached 87.46% and 86%, respectively. Furthermore, the heading stage application confirmed the model’s feasibility. Thus, the XGBoost model with input features of eight vegetation indices provides an effective and reliable method for detecting rice leaf folders, supporting real-time, precise pesticide use in rice cultivation.

The accurate extraction and temporal monitoring of abandoned croplands are essential for the effective scientific management of such abandonments. Currently, analyzing time-series normalized difference vegetation index (NDVI) variations serves as a widely-used method for abandoned cropland extraction. However, acquiring complete cloud-free images to establish time-series NDVI data across the entire crop growth cycle is usually challenging. To enhance the accuracy of abandoned cropland extraction, the method of annual maximum of NDVI value was proposed to extract abandoned cropland using Gaofen-2 (GF-2) and Sentinel-2 imagery in Zengcheng District, Guangdong Province, China. The method involves the following steps. Firstly, the GF-2 images were used to generate land use map by object-based image analysis (OBIA) method. Subsequently, the images of annual maximum of NDVI value from 2018 to 2022 of Zengcheng District were calculated based on Sentinel-2 data on the Google Earth Engine platform. Crucially, the optimal threshold distinguishing planted cropland and unplanted cropland was determined. After conducting repeated comparisons from Google Earth images and field survey data, it was determined that an object with an annual maximum NDVI value below 0.4 should be regarded as an unplanted object. With this threshold (NDVI = 0.4), the spatial distribution of unplanted cropland within year was identified. Finally, croplands that remained unplanted for two or more consecutive years were extracted as abandonment. The extraction results and accuracy assessment showed that our method achieved an overall accuracy ranging from 0.80 to 0.85. In summary, this study presents a novel approach for accurate abandonment extraction based on available NDVI data.