Evapotranspiration (ET) estimation was driven by application of machine learning (ML) techniques, utilizing Penman-Monteith-Leuning's second variety (PML-V2). The evaluation was centered on two ecologically diverse environments: China's Duolun grassland and Japan's Seto mixed forest. The primary advancement for PML-V2 is incorporation of canopy conductance and ability to account carbon dioxide effect on transpiration through integration of gross primary production. To predict PML-V2 ET, ML models such as shallow learning-based feedforward neural network (FFNN), support vector regression (SVR), adaptive neuro-fuzzy inference system (ANFIS), and deep learning based long short-term memory (LSTM) network as well as ensemble shallow learning models applied. While LSTM excelled in capturing sequence relationships, ensemble technique achieved overall better performance so that, the prediction accuracy increased by up to 23% and 17% over the validation in Seto mixed forest in Japan and the Duolun grassland region in China, respectively. Comparative analyses demonstrated that shallow learning and PML-V2 consistently performed better than the PM model, particularly over the growing seasons, improving dynamic time warping (DTW) metric by up to 23% and 19% in Duolun and Seto sites, respectively. Then, the proposed ensemble technique was applied to forecast ET using Model for Interdisciplinary Research on Climate, version 6 (MIROC6) Global Climate Models (GCM) data under Shared Socioeconomic Pathways (SSP2-4.5 and SSP5-8.5) scenarios. Projections for 2050 revealed a marginally increasing trend under SSP2-4.5 in Duolun, while SSP5-8.5 climate change indicated a slight decline. In contrast, the Seto site exhibited a declining trend under SSP2-4.5, with SSP5-8.5 showing relative stability.

Increasing drought frequency threatens ecosystem health, necessitating comprehensive resilience assessments to understand ecological responses. However, current frameworks often overlook adaptability (the capacity of systems to learn from disturbances and self-adjust). This omission can lead to overestimating resilience loss and misjudging ecosystem collapse thresholds. To address this, we developed an integrated framework that incorporates adaptability as a third core dimension alongside resistance and recovery, and constructed a comparable composite resilience indicator using the entropy weight method. Applying this framework to the Yellow River Basin (YRB, 1982-2017) yielded several key insights. Resistance and recovery showed opposite trends and a clear trade-off across ecosystems. This reflects the regulatory role of adaptability in dynamically balancing drought tolerance and post-drought regeneration. Among ecosystems, forests exhibited the lowest resistance but the highest recovery, grasslands displayed the opposite pattern, and rain-fed croplands and shrublands were intermediate. Temporally, resistance, adaptability, and overall resilience increased (p < 0.001) across ecosystems, whereas recovery declined (p < 0.001). Analysis of the driving factors revealed that seasonal temperature variability, soil-topography conditions (available water capacity and elevation), and drought characteristics (severity and frequency) were key drivers of ecosystem resilience metrics. Compared with resilience estimates based on the first-order lagged autocorrelation coefficient (AR(1)), the proposed framework produced consistent results and detected early-phase resilience decline. Overall, this study provides a novel, three-dimensional perspective on ecosystem resilience and informs targeted ecosystem management strategies in the YRB and similar dryland regions.

The alteration of lakes modifies the catchment landscapes, and thus the associated hydrological processes (e.g., baseflow). However, the role of lakes in the baseflow dynamics of the Tibetan Plateau is unclear under the effects of climate change. Here, the revised baseflow separation is considered for a quasi-paired catchment. A no-lake scenario is then constructed using machine learning (ML) to quantify the influence of lakes on baseflow signatures. Our results indicate that: (i) The winter baseflow duration curve is in good agreement with the summer baseflow duration curve (R2 = 0.982, %BiasV = 15.38%) within the lake catchment; (ii) Lakes show bidirectional (positive and negative) contributions to the baseflow index, and exhibit a strong reduction effect on the frequency of high baseflow days (-46.82%); (iii) Lakes amplify the influence of temperature on the frequency of both high baseflow days and low baseflow days, while attenuating the impact of soil moisture on rising limb density. These results reveal that lakes act as fundamental modulators, altering how physiographic and climatic variabilities propagate into baseflow behavior in headwater systems. This study enhances our understanding of how lakes regulate catchment baseflow dynamics, supporting the design of water management strategies under future changing conditions.

The Penman–Monteith–Leuning (PML) model is a widely recognized diagnostic framework for estimating coupled terrestrial evapotranspiration (ET) and gross primary production (GPP). To address the critical need for high-fidelity, long-term, and near-present eco-hydrological records, we developed the PML-V2.2 dataset, spanning from 1982 to 2025. Driven by observation-constrained Multi-Source Weighted-Ensemble Precipitation (MSWEP) and Multi-Source Weather (MSWX) meteorological variables, the dataset comprises three complementary products: (1) PML-V2.2a, an 8 d 500 m MODIS/VIIRS satellite-based product (2000–2024 and 2012–2025) optimized for near-present monitoring (updated annually); (2) PML-V2.2b, a half-month 0.1° AVHRR-based product (1982–2020) anchoring long-term climate attribution; and (3) PML-V2.2c, a consolidated half-month 0.1° record integrating the above products for seamless 44-year continuity (1982–2025). Our methodological framework features an expanded bottom-up calibration using 208 flux sites (∼ 1400 site-years) across various plant functional types (PFTs) and a refined parameterization that explicitly distinguishes between irrigated and rainfed croplands. This distinction effectively mitigated systematic biases in agricultural regions, reducing ET and GPP estimation errors by 8.7 % and 16.2 %, respectively. Performance evaluation reveals high accuracy across PFTs (cross-validation Nash-Sutcliffe Efficiency, NSE > 0.60, absolute bias < 5 %), while top-down water-balance validation across 56 large river basins during 1982–2016 and 152 basins during 2003–2020 confirms high reliability (NSE: 0.89–0.91) as compared with other products. The MODIS- and VIIRS-based PML-V2.2a datasets are internally consistent, and exhibit high agreement with PML-V2.2b during their overlapping period (NSE = 0.90 and 0.79 for annual ET and GPP anomalies), ensuring a seamless transition across satellite epochs. Based on the consolidated PML-V2.2c dataset, global terrestrial ET and GPP during 1982–2025 are estimated at 65.8 × 103 km3 yr−1 (with 58.2 % from transpiration) and 143.4 PgC yr−1, respectively. Long-term analysis reveals significant (p < 0.05) increasing trends in GPP (0.343 PgC yr−2) and ET (0.019 × 103 km3 yr−2) during 1982–2025, where vegetation greening impact on ET is partially offset by physiological water saving under rising atmospheric CO2, consequently enhancing water use efficiency. By bridging the gap between satellite epochs, PML-V2.2 provides an internally consistent long-term global dataset for hydrology, ecology, and other Earth science studies. The dataset is freely accessible, with the 500 m resolution PML-V2.2a product hosted on Google Earth Engine, and all 0.1° PML-V2.2a/b/c versions archived at the National Tibetan Plateau Data Center under https://doi.org/10.11888/Terre.tpdc.303314 (Xu et al., 2026).

Satellite observations from the Gravity Recovery and Climate Experiment and its follow-on mission (GRACE/-FO) enable an integrated monitoring of terrestrial water storage (TWS) dynamics, offering a novel perspective for hydrological drought assessment. Existing studies utilizing TWS data often rely on simplistic indices, limiting their diagnostic capability. To address this gap, we develop a Standardized Terrestrial Water Storage Index (STWSI) by optimizing probability distribution fitting across five parametric models (Beta, Johnson-SB, Gamma, Weibull, and Pearson III) for monthly TWS changes in 40 globally distributed major river basins. Results indicate that the Johnson-SB distribution provides the optimal fit for STWSI construction in 34 basins, outperforming traditional Gamma/Pearson III distributions used in meteorological drought indices (SPI/SPEI). Multi-scale comparisons (1- to 12-month) reveal significantly stronger correlations between STWSI and SPI/SPEI at longer timescales (R = 0.40-0.80 at 12 months) over global major basins. Crucially, STWSI detects hydrological droughts with lower frequency but longer duration and higher intensity than meteorological droughts across most basins, except for tropical basins (e.g., Amazon). At the moderate drought category and different time scales, the total basin area proportion detected by STWSI (44-54 %) is approximately twice that identified by meteorological indices (17-28 %). This divergence underscores the dominant control of long-term TWS depletion on hydrological drought genesis. Our findings establish STWSI as a transformative tool for monitoring composite hydrological droughts in a warming and changing climate.

Irrigation is a major human intervention in the global land-atmosphere system. However, increasing climate variability and associated regional water scarcity may lead to abrupt reductions in irrigation. In this study, we use the Community Earth System Model to investigate the global near-surface air temperature response to irrigation cessation over Northwest India. Our results indicate that evaporative cooling is the dominant local mechanism regulating temperature in response to irrigation in Northwest India. Following irrigation cessation, average near-surface temperatures over Northwest India increased by similar to 0.24 degrees-0.38 degrees C during November-February. Notably, irrigation cessation over Northwest India also induces temperature changes in remote regions. These remote responses are primarily associated with the advection of warm, moist air, enhanced atmospheric humidity, and increased downward longwave radiation. Our findings highlight that the temperature impacts of irrigation practices can extend beyond local regions and should be considered in assessments of climate and water management.

Estimating root-zone soil moisture (RZSM) under sparse in-situ observations is closely related to the broader Prediction in Ungauged Basins (PUB) challenge. This study introduces a novel application of the Tabular Prior-data Fitted Network (TabPFN) for predicting RZSM using small sample sizes (fewer than 10,000) and evaluates its robustness under extreme data sparsity conditions. Using five-fold spatial cross-validation conducted at the grid-cell level, TabPFN was trained on sampled data, while the baseline models (RF, XGBoost, CatBoost and Tabular ResNet) were trained on both sampled and full datasets. All models were evaluated on full test sets to ensure a consistent benchmark. Our results indicate that under extremely sparse training conditions (one sample per grid), TabPFN achieves the highest prediction accuracy (R = 0.726 ± 0.002; RMSE = 0.069 ± 0.0004; MAE = 0.053 ± 0.0003), improving correlation by 2.5–9.2% and reducing prediction errors by 2.6–9.8% compared to benchmark models. In time-specific modeling, where separate models were constructed for each day, week, and month, its performance advantage was broadly maintained across changing hydro-climatic states and under consistent temporal aggregation. Its performance also remained relatively stable as training grid density decreased, including when fewer than 40% of the available training grids were used. These results demonstrate that TabPFN enables robust RZSM prediction using small samples and provides a practical tool for drought monitoring and water assessment in data-scarce or poorly monitored regions.

Abstract Understanding large‐scale hydrological responses to climate and vegetation change is critical, yet the propagation effects through the hydrological process on streamflow signatures in large scale are not quantified yet. Here, we analyzed 2,252 unregulated catchments worldwide to disentangle climate and vegetation change effects by employing a well‐performed HBV‐PML hydrological model, which features enhanced representations of evapotranspiration processes. Our analysis reveals consistent propagation patterns across most climate zones: (a) Increasing Leaf Area Index (LAI) enhances interception and transpiration but suppresses soil evaporation, decreases streamflow magnitude and dynamics, shortens high‐flow durations, and increases low‐flow frequency/duration, indicating reduced water availability, and vice versa. (b) The effects of precipitation changes are opposite to those of LAI. (c) Temperature modulates hydrological processes via rain‐snow partitioning, soil evaporation, and evaporation from canopy interception. The impacts of climate and vegetation change are magnified as they propagate through total actual evapotranspiration to the remaining hydrological processes. Notably, the relative changes of key hydrological variables grow with increasing temporal scales. Contribution analysis identifies precipitation as the dominant driver of changes in streamflow signatures (73.6 ± 10.5% of catchments). However, contributions vary by climate zones and streamflow signatures, with median precipitation contributions exceed 60% in equatorial zones, while temperature dominates (61.4%–73.6%) low‐flow changes in polar zones. This study provides global insights into hydrological responses under environmental changes and offers an essential process‐based understanding of impact propagation mechanisms for water resources management.

Soil moisture (SM) regulates water availability in land-atmosphere interactions and constrains vegetation recovery in water-limited ecosystems. The Loess Plateau (LP) is one of China's most ecologically fragile regions, and the large-scale vertical SM structure plays a vital role in drought propagation and vegetation response. However, this structure has been poorly investigated under anthropogenically induced ecological restoration. In this study, we derived the standardized soil moisture index (SSMI) from multi-depth (0-200 cm) SM data to characterize the spatiotemporal evolution of soil drought across the LP during 2001-2022. Using SSMI, the enhanced vegetation index, solar-induced chlorophyll fluorescence, and Sen's slope, we quantified the cumulative and lagged effects of vegetation response to soil drought. The findings are (1) Drought dynamics exhibit distinct vertical stratification. In shallow layers, SM fluctuations were rapid and frequent (33 events). In the deepest layers, SM fluctuations were damped and deficits were more persistent (mean duration of 4.1 months). (2) Vegetation recovery was sensitive to shallow SM variability, while cumulative and lagged effects of vegetation recovery were primarily governed by drought in the 10-100 cm layer. (3) Deep-layer SM had an asynchronous effect on vegetation growth in cold and high-elevation regions. Vegetation recovery was more strongly linked to ecophysiological drivers than to climate or multilayer soil drought. Plant transpiration (r = 0.44) and water use efficiency (0.41) had strong indirect effects on vegetation recovery that were mediated through soil drought. The findings highlight the role of soil drought in vegetation restoration and contribute towards drought detection and early warning in dryland regions.

Accurate prediction of flood events is important for flood control and risk management. Machine learning techniques contributed greatly to advances in flood predictions, and existing studies mainly focused on predicting flood resource variables using single or hybrid machine learning techniques. However, class-based flood predictions have rarely been investigated, which can aid in quickly diagnosing comprehensive flood characteristics and proposing targeted management strategies. This study proposed a prediction approach of flood regime metrics and event classes coupling machine learning algorithms with clustering-deduced membership degrees. Five algorithms were adopted for this exploration. Results showed that the class membership degrees accurately determined event classes with class hit rates up to 100%, compared with the four classes clustered from nine regime metrics. The nonlinear algorithms (Multiple Linear Regression, Random Forest, and least squares-Support Vector Machine) outperformed the linear techniques (Multiple Linear Regression and Stepwise Regression) in predicting flood regime metrics. The proposed approach well predicted flood event classes with average class hit rates of 66.0%-85.4% and 47.2%-76.0% in calibration and validation periods, respectively, particularly for the slow and late flood events. The predictive capability of the proposed prediction approach for flood regime metrics and classes was considerably stronger than that of hydrological modeling approach.

Highlights It accurately predicts regional groundwater levels with an R2 of 0.91 +/- 0.009 in cross-validation mode. Farming activities and high soil permeability can be important factors contributing to a large error in groundwater prediction. Groundwater level in YRB is strongly declined in fall and winter, particularly in the middle and lower reaches. What are the main findings? The model accurately predicts regional groundwater levels, achieving an R2 of 0.91 +/- 0.009 in cross-validation. Groundwater levels in the YRB show a strong decline in fall and winter, particularly in the middle and lower reaches. What are the implications of the main findings? Farming activities and high soil permeability are key factors contributing to large errors in groundwater prediction. The identified seasonal and spatial patterns provide critical insights for sustainable groundwater management in the YRB.Highlights It accurately predicts regional groundwater levels with an R2 of 0.91 +/- 0.009 in cross-validation mode. Farming activities and high soil permeability can be important factors contributing to a large error in groundwater prediction. Groundwater level in YRB is strongly declined in fall and winter, particularly in the middle and lower reaches. What are the main findings? The model accurately predicts regional groundwater levels, achieving an R2 of 0.91 +/- 0.009 in cross-validation. Groundwater levels in the YRB show a strong decline in fall and winter, particularly in the middle and lower reaches. What are the implications of the main findings? Farming activities and high soil permeability are key factors contributing to large errors in groundwater prediction. The identified seasonal and spatial patterns provide critical insights for sustainable groundwater management in the YRB.Highlights It accurately predicts regional groundwater levels with an R2 of 0.91 +/- 0.009 in cross-validation mode. Farming activities and high soil permeability can be important factors contributing to a large error in groundwater prediction. Groundwater level in YRB is strongly declined in fall and winter, particularly in the middle and lower reaches. What are the main findings? The model accurately predicts regional groundwater levels, achieving an R2 of 0.91 +/- 0.009 in cross-validation. Groundwater levels in the YRB show a strong decline in fall and winter, particularly in the middle and lower reaches. What are the implications of the main findings? Farming activities and high soil permeability are key factors contributing to large errors in groundwater prediction. The identified seasonal and spatial patterns provide critical insights for sustainable groundwater management in the YRB.Abstract Groundwater storage is vital for managing water resources, especially as global water scarcity intensifies. Estimating groundwater levels regionally is challenging due to natural heterogeneity. We employed a large groundwater observation sample, along with Global Land Data Assimilation System (GLDAS) and Gravity Recovery and Climate Experiments (GRACE) datasets, to develop a random forest model for predicting groundwater levels in China's Yellow River Basin. The model showed robustness, achieving an R2 of 0.95 in calibration and an R2 of 0.91 +/- 0.009 in 10-fold cross-validation with 100 repetitions. Temporal predictability was lower, with an R2 of 0.72 for April-May 2023; however, the temporal prediction is preliminary and limited by the short validation period (April-May 2023), which should be interpreted with caution. Spatial maps revealed significant seasonal declines in fall and winter, particularly in the middle and lower reaches. This study highlights the potential of machine learning with extensive observations to estimate regional groundwater levels and supports groundwater analysis with robust data.

Climate change and rising human water demand are intensifying water scarcity across arid Asia, but their relative contributions to terrestrial water storage (TWS) decline in endorheic basins remain poorly quantified. Using satellite gravity observations from GRACE and GRACE‑FO, which measure changes in TWS, we show that TWS anomalies across Asian endorheic basins decreased by 64.26 ± 2.43 Gt yr −1 during 2002–2023. Multi-dataset attribution for 2002-2021 reveals that while climate-driven changes account for roughly half of the total TWS decline (−29.41 Gt yr⁻¹), increasing human water demand severely exacerbates both the rate and magnitude of this depletion. Human water use accounts for 38.0 ± 4.55% (−21.30 Gt yr⁻¹) of the total TWS loss, primarily through irrigation and groundwater abstraction. Depletion is spatially heterogeneous, with faster losses near human activity centres and in drier areas with greater irrigation demand. Current global hydrological models substantially underestimate the observed magnitude of TWS depletion and fail to reproduce these spatial hotspots. Our results demonstrate that localized human water demand is driving severe basin-scale water depletion, underscoring the urgent need for a policy shift from supply-side infrastructure to transboundary demand management to avert irreversible ecological and societal collapse across Earth's fragile drylands.

Most existing attribution assessments of runoff changes mainly focused on direct effects of climate change and human activities on runoff changes, and their indirect effect assessments were not widely reported, i.e., impacts on underlying surface conditions and their consequences to runoff changes. In this study, we adopted an improved Budyko framework, employing the elasticity coefficient method and principal component regression, to comprehensively assess both the direct and indirect impacts of climate changes and human activities on runoff changes. Thirty-one source catchments were selected in the upper and middle reaches of the Yellow River Basin, comprising the Yellow River Source Region (YRSB), Wei River Basin (WRB) and Yiluo River Basin (YRB). Results showed that during 1995-2022, the mean annual runoff changes in the post-change period ranged from -24.0 to 129.9 mm in 31 source catchments compared to the pre-change period. The underlying surface parameter n in the Budyko model ranged from 0.4 to 3.8, with lower values mainly in the YRSB. Sensitivity analysis revealed that parameter n was most sensitive to potential evapotranspiration, particularly in the YRB, and runoff changes exhibited the greatest sensitivity to precipitation, with the highest sensitivity observed in the WRB. Changes in parameter n were primarily attributed to human activities (77.4%) rather than climatic factors (22.6%), with irrigation water (22.6%) and temperature (8.4%) exhibiting the largest impacts, respectively. Contributions to runoff changes were 54.7% from the direct impact of climate change, highest in the YRSB at 59.8%, while 12.2% from its indirect impact, and 33.1% from human activities, with the greatest influence both in the YRB at 16.2% and 42.4%, respectively.

Agricultural water consumption accounts for more than 90 IWD_net , mm) and actual irrigation water demand ( IWD_actual , m3) for major crops, as well as the underlying drivers of these changes. These uncertainties pose significant challenges for sustainable water management practices in the region. This paper investigates future changes in IWD_net and IWD_actual for four primary crops: corn, cotton, rice, and winter wheat. Using the partial derivative method, we analyze the factors contributing to these changes. The results indicate that IWD_net for corn, cotton, and rice is expected to increase by 20–40 IWD_net for winter wheat is projected to decrease, particularly under the SSP126 scenario, due to an increase in effective precipitation ( P_eff ). In terms of IWD_actual , the dominant influencing factor is the crop-planted area, with winter wheat showing a significant decrease in IWD_actual . Understanding these dynamics is crucial for effective agricultural water resources management in CA. This knowledge will support the development of strategies to ensure sustainable water use in the region.

The Lunar Neutron and Gamma-ray Spectrometer (LNGS) is one of the scientific payloads of the Chang'E-7 mission. This paper describes the process of calibration of LNGS on the Back-streaming white neutron beamline (Back-n) of the China Spallation Neutron Source (CSNS). The neutron detection efficiencies in the energy range of 0.4 eV-1000 eV were measured by the time-of-flight method on the beamline produced by 1.6 GeV protons bombarding a tungsten target, and the discrepancy between the experimental and simulated results is less than 6%. The angular response curves of the detector rotating in two directions with respect to the neutron beamline were measured, and the experimental and simulated results are also in very good agreement.

Recently, climate change and human activities such as deforestation are putting drastic pressure on water security in the Upper Blue Nile River Basin, Ethiopia. Despite this, the spatial and temporal variability in streamflow responses to these drivers remains insufficiently understood. In this study, we used the modified Mann-Kendall test and Sen's slope estimator to quantify streamflow trends from 1981 to 2018 in 25 catchments, and we applied a Budyko-based elasticity framework, considering the impact of potential evapotranspiration (E0), precipitation (P), and catchment characteristics (ω) to attribute these trends. Results showed that six catchments- Gumara, Azuari, Koga, Megech, Main Beles, and Rib-exhibited statistically significant rise in annual streamflow, with rates ranging from 3.5 to 10.5 mm year-1. In contrast, catchments such as Dura and Upper Rib showed decreasing streamflow trend despite rising precipitation, indicating complex hydroclimatic interactions. Elasticity of streamflow to P (εp) ranged from 0.60 to 0.83. In contrast, elasticity to Eo(εEo), and ω(εω) ranged from -0.40 to -0.17, and -0.48 to -0.11, respectively. Upstream catchments-such as Gelgel Abay and Gumara-exhibited higher sensitivity to climatic variability, as reflected in their coefficients. Generally, 67% of the observed streamflow variability was attributed to climate change, while catchment characteristics contributed to the remaining 33% of the streamflow variability, with distinct spatial differences. This conclusion shows the need for customized management approaches to address both shifts in climatic variables and changes in vegetation.

The Tibetan Plateau (TP), one of the most climate-sensitive and hydrologically crucial regions on Earth, poses significant challenges for the accurate estimation of surface air temperature. In this study, we evaluate five widely used global reanalysis datasets (ERA-Interim, JRA-55, MERRA-2, ERA5, ERA5-Land) against observations from 132 meteorological stations during 1980-2018. For daily temperature accuracy, the overall ranking is JRA-55 > ERA-Interim > ERA5 > MERRA-2 > ERA5-Land. JRA-55 and ERA5 provide the most reliable overall representation of warming trends over the TP. The observed median annual warming trend is about 0.45 degrees C/decade, with the strongest warming in winter (0.47 degrees C/decade). All five reanalysis datasets show negative trend biases, indicating systematic underestimation of observed warming. Errors are largest in winter and are consistently concentrated in the topographically complex and monsoon-influenced southeastern TP, a hydrologically important headwater region. The post-2000 warming dipole, characterized by accelerated warming in the southeast and relative slowdown toward the northeastern TP, is reproduced best by JRA-55 and ERA-Interim and is weakest in MERRA-2. Overall, JRA-55 provides the highest overall accuracy, while JRA-55 and ERA5 provide the most reliable representation of warming trends over the TP.

The integration of machine learning (ML) into geohazard assessment has successfully instigated a paradigm shift, leading to the production of models that possess a level of predictive accuracy previously considered unattainable. However, the black-box nature of these systems presents a significant barrier, hindering their operational adoption, regulatory approval, and full scientific validation. This paper provides a systematic review and synthesis of the emerging field of explainable artificial intelligence (XAI) as applied to geohazard science (GeoXAI), a domain that aims to resolve the long-standing trade-off between model performance and interpretability. A rigorous synthesis of 87 foundational studies is used to map the intellectual and methodological contours of this rapidly expanding field. The analysis reveals that current research efforts are concentrated predominantly on landslide and flood assessment. Methodologically, tree-based ensembles and deep learning models dominate the literature, with SHapley Additive exPlanations (SHAP) frequently adopted as the principal post-hoc explanation technique. More importantly, the review further documents how the role of XAI has shifted: rather than being used solely as a tool for interpreting models after training, it is increasingly integrated into the modeling cycle itself. Recent applications include its use in feature selection, adaptive sampling strategies, and model evaluation. The evidence also shows that GeoXAI extends beyond producing feature rankings. It reveals nonlinear thresholds and interaction effects that generate deeper mechanistic insights into hazard processes and mechanisms. Nevertheless, several key challenges remain unresolved within the field. These persistent issues are especially pronounced when considering the crucial necessity for interpretation stability, the demanding scholarly task of reliably distinguishing correlation from causation, and the development of appropriate methods for the treatment of complex spatio-temporal dynamics.

Aerosol-cloud interactions exert substantial influences on atmospheric chemistry and regional climate, yet process-resolved characterizations of chemical and microphysical evolution within cloud droplets remain limited. Here, two intensive campaigns were conducted at the high-altitude Shanghuang station in southeastern China during autumn 2023 and spring 2024, capturing nocturnal orographic and long-persistence stratiform cloud events. Using two complementary cloud-droplet sampling systems, the ground-based counterflow virtual impactor and a custom-designed aerosol-cloud sampling inlet system, along with aerosol chemical speciation and cloud microphysical measurements, we resolved the composition of interstitial (INT), residual (RES) and ambient (AMB) particles. Organic aerosols (OA) dominated particle mass across both seasons, while inorganic species (nitrate, sulfate, ammonium) exhibited high scavenging efficiencies (>= 65%-70%) and strong enrichment in RES particles. Organic components showed seasonally contrasting partitioning patterns, with differences between INT and RES particles indicating variability in aqueous-phase processing. Air-mass analysis further revealed pronounced source-dependent variability, with polluted westerly inflow leading to the highest particle loadings and most aged organic signatures. Linking chemistry with microphysics, we found that secondary organic aerosol (SOA) formation is favored in smaller droplets, whereas primary organic aerosol is preferentially incorporated into larger droplets through collision-coalescence. The contrasting evolution of oxygen-to-carbon ratio in RES and INT particles as a function of OA/Delta CO indicates distinct oxidation pathways for activated and non-activated aerosols. These results demonstrate that droplet size and cloud dynamics jointly regulate aerosol processing and that in-cloud oxidation pathways differ between particle types. This study provides process-level constraints for improving the representation of aqueous-phase SOA formation and aerosol-cloud interactions in atmospheric models, particularly in regions influenced by complex terrain and variable cloud regimes.