Selecting reliable negative samples (NS) is a crucial step in enhancing the robustness of landslide susceptibility assessments (LSA). Existing studies frequently utilize InSAR-derived deformation data to identify stable areas, thereby refining the negative samples; however, InSAR is susceptible to residual errors due to decorrelation and geometric distortions, particularly in regions with significant topographic relief and dense vegetation cover. Additionally, the processing workflow for InSAR can be complex and costly.To address these issues, we examine the Lijiang River Basin in Guangxi, China, as a case study. We propose a novel negative sampling strategy constrained by the temporal stability of two SAR-based indices: Long-term Distillation and Identification (LDI). First, we delineate temporally stable areas (S_SAR) by selecting pixels that exhibit minimal long-term change rates in the Radar Vegetation Index (RVI) and the Radar Forest Degradation Index (RFDI). We then apply Positive-Unlabeled Learning (PU-learning) to refine S_SAR further, resulting in a high-confidence NS set (Nopt). Next, we evaluate stability differences between Nopt and various NS sets generated by conventional sampling strategies, using cumulative deformation and deformation-rate metrics obtained from SBAS-InSAR. Finally, we built Landslide Susceptibility Assessment (LSA) models utilizing Random Forest (RF), Extreme Gradient Boosting (Xgboost), and Categorical Boosting (Catboost). We assess model performance using the Area Under the Curve (AUC) and confusion-matrix-based metrics. Additionally, we analyze spatial patterns in LSA, area proportions across susceptibility classes, and their relationship with the multi-year means and long-term change rates of RVI and RFDI.The results indicate the following: (1) Deformation values in S_SAR are primarily clustered around 0 mm, confirming the consistency between “stable long-term vegetation change” and “stable ground deformation.” After refining with PU-learning, Nopt shows more minor fluctuations in deformation and exhibits the highest internal consistency. (2) LSA models based on LDI perform the best, with the Xgboost-based LSA achieving the highest AUC (0.843). Additionally, feature contributions quantified by Shapley Additive Explanations (SHAP) are more concentrated and stable, demonstrating that LDI effectively reduces noise. (3) Although various NS sampling strategies result in significant differences in LSA spatial patterns, the Very High Susceptibility (VH) class consistently displays a typical pattern of “higher RFDI and lower RVI, with a weaker RFDI trend and a stronger RVI trend”. This suggests that areas classified as VH have lower vegetation cover, greater inter-annual variability, and weaker disturbance resistance. Overall, LDI provides a cost-effective approach to obtaining reliable NS data in complex terrains, serving as a valuable reference for LSA modeling in the Lijiang River Basin and similar regions.

Significant radiometric discrepancies commonly exist among multisource optical satellite images due to differences in sensor characteristics, acquisition geometry, atmospheric conditions, and imaging time, which severely limit their application accuracy in land-cover classification, change detection, and geophysical parameter retrieval. Therefore, relative radiometric normalization (RRN) has become an essential preprocessing step for the integrated application of multisource optical satellite imagery. However, existing RRN methods often suffer from low automation in radiometric reference sample selection and are highly sensitive to anomalous samples, resulting in limited normalization accuracy and robustness. To address these issues, this study proposes an automated cross-sensor RRN framework based on automatic construction and intelligent refinement of radiometric reference sample set (RRSS). An unsupervised deep-learning-based change detection approach is employed to automatically identify radiometrically stable samples and construct an initial RRSS. Subsequently, a dual-stage sample refinement strategy integrating statistical outlier detection and topographic prior knowledge is developed to progressively remove anomalous samples and improve sample reliability. Linear regression is then performed using the refined samples to establish the radiometric normalization model. Experiments conducted on multisource optical satellite imagery acquired under different terrain conditions demonstrate that the proposed framework consistently outperforms five mainstream RRN methods, achieving lower Root Mean Square Error (RMSE) and higher Spectral Angle Cosine (SAC) and Structural Similarity Index Measure (SSIM) values. The results indicate that automatic construction and intelligent refinement of RRSS effectively improve inter-image radiometric consistency, particularly in topographically complex regions. Root Mean Square Error (RMSE), Spectral Angle Cosine (SAC), and Structural Similarity Index Measure (SSIM).

Changes in vegetation phenology affect ecosystem carbon uptake and water use, thereby regulating water use efficiency (WUE). However, in alpine ecosystems of the Qinghai-Tibetan Plateau (QTP), uncertainty remains regarding the phenological information characterized by different remote-sensing data sources and its associations with WUE. Using solar-induced chlorophyll fluorescence (SIF) and MODIS normalized difference vegetation index (NDVI) data from 2001 to 2018, we derived the start of growth (SOG) and end of growth (EOG) using multiple phenology extraction methods. WUE was calculated using gross primary productivity (GPP) and evapotranspiration (ET) data. We then employed trend analysis, statistical modeling, and a machine learning interpretive framework to systematically evaluate spatiotemporal differences in phenology derived from SIF and NDVI and their associations with WUE. The results showed that: (1) WUE generally increased across the QTP at approximately 0.15 g C m⁻² mm⁻¹ decade⁻¹, with significant increases mainly in the central-eastern and southeastern regions. Both NDVI- and SIF-derived SOG advanced at rates of −1.08 and −1.14 doy decade⁻¹, respectively. In contrast, EOG showed clear data source divergence: EOG_NDVI was delayed by 0.62 doy decade⁻¹, whereas EOG_SIF advanced by −0.48 doy decade⁻¹. SOG_SIF occurred on average 6.6 days later than SOG_NDVI, EOG differences were larger, with EOG_SIF occurring 17.2 days earlier than EOG_NDVI on average. Trend consistency was also higher for SOG than for EOG, whereas opposite EOG trends accounted for 25.3%. (2) After accounting for climatic covariates, SIF- and NDVI-derived phenological indicators showed distinct model-based associations with WUE, but their explanatory contributions were generally weaker than those of key climatic variables. (3) GAM results further showed that SOG was generally negatively associated with standardized WUE in both phenological datasets, whereas the EOG–WUE partial association differed between SIF and NDVI, with positive associations for EOG_SIF and negative associations for EOG_NDVI. This study highlights the differences between SIF- and NDVI-derived phenological indicators and their model-based associations with WUE, providing complementary remote-sensing information for interpreting vegetation phenology and WUE dynamics on the QTP.