High-frequency GPS tracking provides fine-scale insight into animal space use, but spatial thinning—routinely used to meet independence assumptions—discards substantial information and can underestimate core habitat quality. Using 65,548 high-frequency GPS fixes from 42 wintering black-necked cranes (Grus nigricollis), we developed a framework that converts these fixes into continuous kernel density estimation (KDE) surfaces of space-use intensity and then models these surfaces using random forest (RF) regression. We compared this KDE + RF approach with two thinning-based alternatives: conventional thinning + Maximum Entropy (MaxEnt) and thinning + RF, across nine environmental variable combinations. The 100 m KDE bandwidth, corresponding to the median step length of the GPS-tracked black-necked cranes, generally outperformed the 500 m bandwidth, particularly for models incorporating remote-sensing variables, and was therefore used in subsequent analyses. For the same predictor combinations, KDE + RF models consistently achieved higher predictive performance than their thinning + RF, with higher area under the curve (AUC) and continuous Boyce index (CBI) across all nine combinations; the CBI gain is particularly relevant because it concerns the ranking of habitat quality rather than the discrimination of presence from absence. For example, under the Clim + LC combination, KDE + RF achieved a higher CBI than thinning + RF (0.989 ± 0.007 vs. 0.947 ± 0.042), and their AUC values were comparable (0.989 ± 0.006 vs. 0.984 ± 0.007). KDE + RF also generally outperformed the conventional thinning + MaxEnt framework across environmental predictor combinations. Independent field validation supported closer correspondence between KDE + RF predictions and observed flock distributions at both wintering sites, with stronger enrichment in the High-quality habitat and a more consistent increase in flock occurrence ratios along predicted habitat-suitability gradients. By preserving space-use intensity information and improving the representation of habitat-suitability gradients, the framework may support more effective conservation prioritisation for black-necked cranes. More broadly, it offers a transferable methodological concept for other species monitored with high-frequency GPS tracking.
Vegetation dynamics in the Mu Us Sandy Land are influenced by the complex interplay of climate change and human activities, with response processes often characterized by temporal lag. Studies have insufficiently addressed these lag effects, which limits the accuracy of assessments of ecological restoration outcomes and hinders the formulation of sustainable management strategies. On the basis of long-term Landsat NDVI data, meteorological records, and land use data spanning 1989–2024, this study integrated Theil–Sen trend analysis, the Mann–Kendall test, partial correlation analysis, and residual analysis to identify the effects of various factors on the vegetation NDVI and quantitatively decomposed the relative contributions of climate and human activities while explicitly incorporating lag effects. The results reveal a distinct stagewise pattern of vegetation coverage in the Mu Us Sandy Land: stable fluctuation (1989–2001), slow recovery (2001−2012), accelerated improvement (2012–2017), and maintenance of stability (2017–2024). Temperature increase exerted a predominantly promoting effect in the early period (1989–1999), with significant positive correlations between temperature and the NDVI accounting for 79.09% of the area; this area expanded to 84.14% during 2000–2024, indicating a continuously strengthening positive impact of warming on vegetation growth. The positive effect of increased precipitation intensified over time, and the vegetation response to precipitation exhibited a pronounced annual lag. Precipitation with lags of one and two years provided dominant explanatory power for vegetation growth in a given year, with the affected area (68.07%) far exceeding that of current-year precipitation (lag-0 accounting for only 21.49%). The relative contribution of human activities increased steadily from 36.78% during 1989–1999 to more than 50% after 2000 and was revised to 58.91% for 2000–2024 when lag effects were considered, indicating that human activities were the dominant driver of vegetation greening. However, this contribution was 3.79 percentage points lower than that estimated without considering lags. Within the human activity contribution, 39.00% of the area (with consideration of the lag) exhibited negative effects during 1989–1999; notably, in extremely dry years, the positive regulatory capacity of human activities significantly weakened. The benefits of large-scale ecological projects, represented by the Grain for Green program, entered a period of significant release after 2010, reached a peak of accelerated NDVI increase during 2012–2017, and subsequently decelerated. After the regional average NDVI exceeded approximately 0.40, the combined benefits of climate and human activities on vegetation restoration became increasingly difficult to enhance, indicating the existence of an indicative threshold. These findings provide a direct scientific basis for understanding climate–human interaction mechanisms in the vegetation restoration of the Mu Us Sandy Land, for objectively evaluating the long-term effectiveness of ecological projects, and for optimizing adaptive management strategies.
Climate change and anthropogenic pressures unpredictably alter ecological niches, exacerbating the extinction risk for threatened species. However, spatial projections often suffer from limited transferability and inadequate uncertainty quantification, hindering their application in dynamic conservation planning. Here, utilizing nine bioclimatic, topographic, and land cover variables, we developed an ensemble modeling framework integrating ten machine learning algorithms (AUC: 0.9099 ± 0.0130, TSS: 0.6432 ± 0.0428, and BI: 0.9370 ± 0.0352) to evaluate habitat suitability and climate-driven range shifts for the endangered large flying-fox (Pteropus vampyrus), a keystone umbrella species in Southeast Asia. Model interpretability revealed that slope aspect is strongly associated with predicted suitability, reflecting its role in shaping microclimatic exposure, moisture dynamics, and wind shelter, whereas temperature seasonality and precipitation of the driest month support population persistence across fragmented landscapes. Projecting spatial trajectories and mapping explicit uncertainties across three Shared Socioeconomic Pathways (SSP1–2.6, SSP3–7.0, and SSP5–8.5) spanning three future time horizons (2011–2040, 2041–2070, and 2071–2100) indicated severe contraction and fragmentation of suitable ecological niches, with the area classified as General Decrease (29.13%) being more than threefold larger than the area classified as General Increase (9.29%) under the highest emission scenario SSP5–8.5. Beyond emphasizing the critical necessity of uncertainty quantification in transferability across climate scenarios and rendering ecological interactions transparent through model interpretability techniques, monitoring the current status and changes of the Pteropus vampyrus ecological niche provides valuable scientific evidence, serving as a critical foundation for formulating robust conservation strategies.
Rapid urban expansion in the lake domain urban agglomeration (LRUA) is profoundly reshaping regional land use patterns and ecological networks while also driving declines in ecological resilience. Yet where and how land use regulation and ecological restoration should be spatially prioritized remains insufficiently understood. Focusing on the urban agglomeration around Poyang Lake (UAPL), this study develops an integrated framework of “resilience assessment–multi-scenario simulation–zoning regulation” that couples the PLUS, Zonation, and MCR models. The framework is applied to analyze the co-evolution of land use change and ecological resilience (ER) in UAPL from 1980 to 2020 and to simulate future ER trends and optimal spatial optimization plans under five development scenarios from 2030 to 2060. Results indicate that the average ER of UAPL declined by 2.05% over the past 40 years, with a spatial pattern of “higher in the periphery and lower in the center” and a negative correlation with construction land expansion. In future simulations of UAPL, this negative correlation can be alleviated through ecological restoration, and ER levels in major cities of the region can be maintained between 0.4508 and 0.5222. Further scenario comparisons show that the sustainable development scenario (SDS) most effectively enhances ecological resilience. Under the optimal SDS scenario, the Zonation model is therefore used to delineate four levels of ecological resilience restoration zones, and the Minimum Cumulative Resistance Model (MCR) is used to construct an ecological restoration network comprising 33 first-level ecological corridors, 46 s-level ecological corridors, 59 third-level ecological corridors, and 141 key nodes, forming a “source–corridor–node” structure. This study can provide scientific support for resilience-oriented land governance and sustainable development in lake domain urban agglomerations.
Successful identification and characterization of drought are vital steps in drought risk assessment and management, requiring improved characterization approaches and suitable indices that capture different drought features. Although many drought indices have been developed, their broader suitability is often limited because local conditions frequently play a dominant role. Furthermore, multiple aggregation timescales for standardized indices make selecting the most appropriate timescale challenging. This study aims to identify suitable drought indices for the Horn of Africa (HoA) by evaluating 15 configurations of 6 drought indices with varying complexity. Employing a recently proposed combined cluster filtering mechanism, the performance of the indices in capturing spatial and temporal characteristics of droughts was evaluated against past drought records, specifically the 2010/11 and 2016/17 droughts. Results show that indices incorporating potential evapotranspiration (PET), particularly the Standardized Precipitation Evapotranspiration Index (SPEI) and Multivariate Standardized Drought Index (MSDI), performed better. SPEI9 and MSDI9 showed strong performance in identifying drought-affected areas, with Matthews Correlation Coefficient (MCC) values of 0.66 and 0.60 for the 2010/11 drought and 0.64 and 0.65 for the 2016/17 drought, respectively, while showing moderate temporal performance. In contrast, SPEI6 and MSDI6 effectively captured temporal characteristics while showing moderate spatial performance, with MCC values of 0.56 and 0.56 for 2010/11 and 0.58 and 0.53 for 2016/17, respectively. No individual index consistently outperformed across all aspects. Considering performance and computational complexity, SPEI6 and SPEI9 are recommended in parallel as complementary timescales for drought monitoring and to support related planning and management actions in the HoA, where droughts are frequent, prolonged, and widespread.
