Ya Li

Land-cover studies often compare feature configurations at one pixel size and report aggregate accuracy, leaving their behavior across aggregation supports, classes, and landscape strata incompletely characterized. We compared five pixel sizes (2, 5, 10, 15, and 30 m) derived from one pan-sharpened Gaofen-2 image in Huajiang Town, Guizhou Province, China. Random forests with eight combinations of spectral, index, texture, and terrain features were evaluated for seven classes using the same 1583 reference polygons and five repeated spatial-block partitions. The reported performance compares predictions for these polygons, not accuracy over all map pixels. Macro-F1 for spectral decreased from 70.59 ± 0.62% at 2 m to 62.47 ± 0.38% at 30 m; the standard deviations describe sensitivity to the repeated partitions. Averaged across the five pixel sizes, the index-, texture-, and terrain-augmented pipelines exceeded their separately screened spectral baselines by 1.68, 2.35, and 5.39 percentage points, respectively. These contrasts include changes in feature screening and do not isolate individual feature effects. Performance also varied across classes and terrain-zone sample groups. Within one image, one sampled region, and the specified aggregation and validation design, feature-pipeline contrasts varied across scale, class, and terrain-zone strata; external map accuracy and transferability remain to be tested with independent, support-matched reference data.

Long-term Earth observation has reliably documented vegetation greening, biomass increases, and the mitigation of rocky desertification across the South China karst. However, “greening” alone neither demonstrates an incremental effect of ecological restoration interventions nor substitutes for the verification of functional recovery in hydrology, soils, carbon, biodiversity, and ecosystem resilience. Based on structured searches completed on 10 August 2026 in the Web of Science Core Collection, Scopus, and China National Knowledge Infrastructure (CNKI), which yielded 837 records in total, this review is organized around three evidence chains: the identification of state changes, the attribution of intervention effects, and the verification of ecological functions. The available evidence indicates that (1) optical, microwave, LiDAR, thermal infrared, and UAV observations have clear advantages for monitoring vegetation cover, canopy structure, aboveground biomass, evapotranspiration, and landscape patterns; (2) key functional variables, including soil thickness, soil physicochemical properties, groundwater processes, soil and dissolved carbon, and species composition, can generally only be constrained indirectly and must be jointly validated using field plots, sensor observations, laboratory analyses, and process models; and (3) existing quasi-experimental studies in karst regions show that, where project boundaries, implementation timing, and comparable controls are available, methods such as difference-in-differences can strengthen the credibility of intervention-effect identification, although such evidence remains limited. Future research should prioritize function-oriented multi-source Earth observation products, independent validation across years and catchments, the explicit representation of uncertainty, and testable designs such as BACI and DID, thereby advancing remote sensing from merely documenting surface greening to providing verifiable evidence for judging ecosystem functional recovery.

In karst mountainous tobacco fields, white plastic mulch, bare rock, and weeds are spatially interspersed and can affect semantic segmentation accuracy in UAV RGB imagery. This study constructed seven background-composition scenarios from UAV RGB orthomosaics acquired in Zhenfeng County, Guizhou Province, China, and evaluated DeepLabV3+, U-Net, U-Net++, and U-Net Former in terms of overall performance, scene-specific accuracy, and cross-scene stability. DeepLabV3+ achieved the highest validation mIoU (0.91) during hyperparameter selection, whereas U-Net performed best on the spatially independent test set, with an mIoU of 0.86, a tobacco-class IoU of 0.74, and an OA of 0.99. Under single-background conditions, bare rock yielded the highest IoU for all four models, while the relative effects of white plastic mulch and weeds were model-dependent. Composite-background effects were non-monotonic: adding weeds to bare-rock scenes reduced IoU by 16.89–28.98%, whereas the three-background scenario did not yield the lowest IoU. U-Net achieved the highest scene-wise IoU in all seven scenarios and the lowest cross-scene coefficient of variation (8.06%). This study provides a reference for sample optimization, model selection, and error diagnosis in complex mountainous tobacco plant identification.

Karst regions are characterized by fragmented topography and significant micro-relief mosaics, leading to prominent spectral aliasing of land features, which can result in insufficient spatial representativeness of remote sensing samples for Land Use and Land Cover (LULC). The accuracy of LULC data directly affects the scientific basis of decision-making for rocky desertification control and ecological conservation. This study selected the Beipanjiang River Basin in Guizhou Province, a typical karst region, as the study area. The study selected the SOS, LOS, OM, and EOS indices from the 2001–2020 MODIS MCD12Q2 phenological dataset, combined with topographic zoning data. This study developed a sample spatial optimization scheme for complex karst terrain by integrating Spearman’s correlation analysis, SKATER spatially constrained clustering, statistical tests, adaptive stratified sampling, and Random Forest classification. The scheme was designed to test a phenology–landform joint stratification strategy for spatial sample allocation. The results indicate that (1) the study area was divided into six phenological pattern subregions, with significant spatial differentiation observed among them; (2) the “phenology–landform joint stratification + dual-weighted sample allocation” method was associated with improved sample representativeness and greater internal homogeneity within sample strata under the current experimental setting; and (3) compared to simple random sampling, the remote sensing phenological pattern-driven spatial optimization scheme improved overall accuracy from 71.33% to 77.55% and increased the Kappa coefficient from 0.43 to 0.62. These results suggest that, under the current study-area, sample-size, and validation settings, the phenology–landform joint stratification and dual-weighted allocation scheme can improve the spatial organization of training samples and classification performance over complex karst terrain, although weakly vegetated or bare classes remain difficult to separate.

Karst regions are commonly characterized by highly interwoven bare rock–bare soil–vegetation mosaics, strong coupling between surface and subsurface processes, and pronounced geomorphic fragmentation. Conventional land cover classification systems, which are primarily organized around land use patterns or generic ecological types, are often unable to accurately represent these key surface components and their roles in ecological processes. From the perspective of reconstructing classification semantics, this study proposes a Natural-Attribute-Based Karst Land Cover Classification framework (NALCC). The framework takes bare rock, bare soil, vegetation, water bodies, and impervious surfaces as primary classes, and further develops a hierarchical system consisting of subclasses, attribute labels, hierarchical coding, multi-scale organization, and parameter mapping with ecosystem service models. Compared with conventional land cover classification systems, the innovation of this framework lies not in increasing the number of categories, but in reconstructing the semantic organization of classification units, so that land cover classification can move beyond surface-type description toward the expression of process-sensitive information. The classification objective of NALCC is not to develop a universal land cover classification system, but to establish a process-oriented classification framework for ecosystem service monitoring, rocky desertification diagnosis, and governance zoning in karst regions, which can directly represent key surface components and their ecological-process significance. However, its regional transferability and mapping performance still need to be further validated through case studies in representative areas.

Land use/land cover (LULC) data serve as a critical information source for understanding the complex interactions between human activities and global environmental change. The subtropical karst region, characterized by fragmented terrain, spectral confusion, topographic shadowing, and frequent cloud cover, represents one of the most challenging natural scenes for remote sensing classification. This study reviews the evolution of multi-source data acquisition (optical, SAR, LiDAR, UAV) and preprocessing strategies tailored for subtropical regions. It evaluates the applicability and limitations of various methodological frameworks, ranging from traditional approaches and GEOBIA to machine learning and deep learning. The importance of uncertainty modeling and robust accuracy assessment systems is emphasized. The study identifies four major bottlenecks: scarcity of high-quality samples, lack of scale awareness, poor model generalization, and insufficient integration of geoscientific knowledge. It suggests that future breakthroughs lie in developing remote sensing intelligent models that are driven by few samples, integrate multi-modal data, and possess strong geoscientific interpretability. The findings provide a theoretical reference for LULC information extraction and ecological monitoring in heterogeneous geomorphic regions.