Gauge-independent, multi-source precipitation data merging is a well-established approach for improving precipitation estimation across regions and scales. However, existing merging techniques often assume precipitation follows a unique, state-independent statistical distribution, overlooking the inherently mixed nature of rain and no-rain occurrence states. This assumption leads to two major limitations: (i) uncertainty in rain/no-rain classification and (ii) biased merging weights, resulting in suboptimal precipitation magnitude estimates. Here we introduce RainMerge, a novel two-stage merging framework designed to address these limitations. In Stage 1 (Binary Merging), we employ the Categorical Triple Collocation Merging (CTC-M) to enhance rain/no-rain detection by optimally combining three input datasets. In Stage 2 (Conditional Merging), we use the binarymerged product to conditionally group samples in time, and merge rain-flagged groups with at least two positive detections using Signal-to-Noise Ratio Optimisation (SNR-opt) to refine precipitation magnitudes. Using gauge observations from diverse hydroclimatic settings, we show that RainMerge improves daily rain/no-rain classification by 25-60 % in False Alarm Ratio (FAR) and Heidke Skill Score (HSS) relative to the parent products (ERA5, SM2RAIN, and IMERG-final). In terms of overall magnitude error (RMSE), RainMerge reduces the parent-product RMSE by 2-6 mm/day and outperforms the state independent (termed One-stage) framework, by 1.5-2.5 mm/day. Additionally, RainMerge reduces underestimation bias by 2-8 mm/day for extreme precipitation events (75th-99th percentiles) compared to One-stage, demonstrating its superior performance in capturing heavy events. These results highlight the potential of RainMerge for hydrological and climate applications, offering a globally applicable approach to improve detection and magnitude accuracy in multi-source precipitation products.

Choosing effective models for accurately estimating potential evapotranspiration (PET) is essential in various fields, including climatology, ecology, hydrology, and agronomy. However, many currently dominant PET products rely on multiple models with default parameters, which can introduce uncertainties into PET-related research. To address this issue, we derived the parameters for five widely used PET models using observations from 124 eddy covariance sites worldwide. By comparing their performances across various biomes, we identified the calibrated Priestley-Taylor and Milly-Dunne models as the optimal choices, capable of being applied effectively beyond their original observation sites. Using these optimal models, along with four widely-used meteorological datasets and annual land use and land cover data, we generated a monthly PET dataset at a 0.1° resolution during 1992-2022 across global vegetation zones. Finally, we compared the new dataset with the Global Land Evaporation Amsterdam Model V4.2a PET, including daily comparisons for each biome and annual trend comparisons during 1992-2022. This new PET dataset serves as an alternative resource for conducting PET-related research.

The Budyko hypothesis is commonly used for attributing changes in evapotranspiration (ET) and runoff (R), but it has limitations in distinguishing the specific impacts of vegetation-such as leaf area index (LAI) and the physiological effects of CO2-and land use/cover change (LUCC) on ET and R. This study aims to fill this gap by combining the Budyko hypothesis with the Shuttleworth-Wallace (SW) potential evapotranspiration (PET) model, which allows for the explicit incorporation of vegetation and LUCC. Using this new framework, we quantified the contributions of climate, vegetation, LUCC, and other land properties to changes in ET and R during the 2000s and 2010s, relative to the period from 1991 to 2000, across 732 catchments in the continental U.S. The results indicate that ET increased in over half of the catchments during both decades, while R decreased in 69 % of the catchments in the 2000s and in 46 % in the 2010s. Attribution analysis revealed that precipitation (P) was the dominant factor influencing changes in ET in over 45 % of the catchments during both periods. NonPCV (non-precipitation climatic variables) affected more than 15 % of the catchments, while the Budyko parameter n was the dominant factor in approximately 35 % of them. Regarding changes in R, P, non-PCV, and parameter n dominated in about 60 %, 10 %, and 30 % of the catchments, respectively. From the 2000s to the 2010s, around 45 % of the catchments experienced shifts in the dominant factor affecting ET or R changes, with more than 20 % transitioning primarily between P and parameter n. A comparison of the attribution of changes in ET or R, using the Budyko-type equation combined with the SW model and the Food and Agriculture Organization of the United Nations-56 Penman-Monteith model, revealed over 7 % of the catchments with different dominant factors. This indicates that overlooking the effects of CO2, LAI, and LUCC can significantly influence the attribution of changes in ET or R. This integrated framework offers a valuable reference for other regions seeking to comprehensively attribute changes in ET and R by incorporating the effects of vegetation and LUCC within the Budyko hypothesis.

A solely human-centered approach to drought assessment is inadequate. Monitoring drought from an ecological perspective is essential for developing scientifically grounded response measures, particularly in drought-prone countries such as Australia. However, the Vegetation Health Index (VHI) may be limited in accurately representing vegetation conditions because of shortcomings in the Normalized Difference Vegetation Index (NDVI). To address this, we developed a Vegetation Optical Depth (VOD)-based VHI (VHIVOD) by integrating long-term harmonized VOD data sets into the conventional VHI framework. Vegetated Australia was used as a case study to evaluate the performance of VHIVOD during 1990-2024. Based on Spearman correlations with Gross Primary Production and soil moisture, VHIVOD outperformed the NDVI-based VHI across over 70% of vegetated Australia at both national and biome levels. Although no significant long-term trend in annual VHIVOD was detected, a significant breakpoint occurred in 2002. During 1990-2001, annual VHIVOD increased across more than 78% of vegetated Australia, with 44% showing significant increases, whereas only 19% showed significant increases during 2002-2024. Regarding vegetation drought indicators, including severity, intensity, duration, and area percentage, both vegetated Australia and most biomes experienced weakening drought conditions during both sub-periods. Over 70% of vegetated areas experienced declines in severity, intensity, and duration, with significant weakening observed in over 39% of vegetated Australia during 1990-2001 and over 20% during 2002-2024. These findings highlighted the improved capability of VHIVOD in representing vegetation conditions and demonstrated its potential for vegetation drought monitoring under changing hydroclimatic conditions.

Southwest China (SWC), a pivotal carbon sink region, has experienced increasingly frequent and intense droughts under climate change. Yet, the relative roles of soil moisture (SM) and vapor pressure deficit (VPD) in regulating ecosystem productivity remain unclear. This study combines Solar-Induced Chlorophyll Fluorescence (SIF) with a two-dimensional copula framework to quantify SM and VPD dominance and critical thresholds across the SWC from 2001 to 2024, using the 40th percentile of SIF values as the threshold for productivity reduction. Results show that SM predominantly drives SIF reduction across 90.8

Merging multiple uncertain datasets cancels random errors, resulting in a merged product that is equivalent to or better than the best individual parent product used. Merging proceeds via a weighted average of individual products, with weights derived using second-order error statistics (variance/covariance), which cannot fully capture skewed or heavy-tailed error structures, and often restrictive assumptions (e.g., zero error cross-correlation, ECC), which are frequently violated when products share retrieval algorithms or calibration data. We propose here a novel alternative for merging geophysical data, Constrained Negentropy Optimisation (CoNE-opt), which assumes the true dataset exhibits the greatest departure from Gaussianity, an assumption commonly invoked in performing Independent Component Analysis (ICA) to derive source signals. CoNE-opt maximises negentropy, which is a measure of non-Gaussianity, and uses it as the objective function while incorporating a linear error model constraint. This formulation allows joint estimation of the full error spectrum and the merging weights. The method is validated through synthetic experiments followed by merger of three global satellite-derived surface soil moisture products, SMAP, SMOS, and SMOS-IC. Comparison against reference datasets demonstrate that CoNE-opt outperforms existing merging alternatives that rely on second order statistics instead. Notably, CoNE-opt improves error magnitudes in the presence of high ECC and outliers, where most Ordinary Least Squares (OLS)-based merging methods struggle. Superior performance was observed in soil moisture merging, where the overall normalised RMSE equals 0.15 (vs. 0.34 for the best OLS-based alternative), consistently surpassing input products and existing methods. These findings support the potential of CoNE-opt as a robust framework for generating global soil moisture datasets, thereby enhancing land surface and hydrological modelling in data-sparse regions.

Utilizing multi-source meteorological data for lightning identification is crucial for monitoring severe convective weather. However, several key challenges persist in this field: dimensional imbalance and modal competition among multi-source heterogeneous data, model training bias caused by the extreme sparsity of lightning samples, and an imbalance between false alarms and missed detections resulting from complex background noise. To address these challenges, this paper proposes a lightning identification network guided by physical priors and constrained by supervision. First, to tackle the issue of modal competition in fusing satellite (high-dimensional) and radar (low-dimensional) data, a physical prior-guided asymmetric radar information enhancement mechanism is introduced. This mechanism uses radar physical features as contextual guidance to selectively enhance the latent weak radar signatures. Second, at the architectural level, a multi-source multi-scale feature fusion module and a weighted sliding window–multilayer perceptron (MLP) enhanced decoding unit are constructed. The former achieves the coupling of multi-scale physical features at a 2 km grid scale through cross-level semantic alignment, building a highly consistent feature field that effectively improves the model’s ability to detect lightning signals. The latter leverages adaptive receptive fields and the nonlinear modeling capability of MLPs to effectively smooth spatially discrete noise, ensuring spatial continuity in the reconstructed results. Finally, to address the model bias caused by severe class imbalance between positive and negative samples—resulting from the extreme sparsity of lightning events—an asymmetrically weighted BCE-DICE loss function is designed. Its “asymmetric” characteristic is implemented by assigning different penalty weights to false-positive and false-negative predictions. This loss function balances pixel-level accuracy and inter-class equilibrium while imposing high-weight penalties on false-positive predictions, achieving synergistic optimization of feature enhancement and directional suppression. Experimental results show that the proposed method effectively increases the hit rate while substantially reducing the false alarm rate, enabling efficient utilization of multi-source data and high-precision identification of lightning strike areas.

Radar echo extrapolation based on real-world data is a fundamental problem in meteorological forecasting. Existing extrapolation models typically assume complete radar echo sequences, but in practice, data loss frequently occurs due to equipment failures and communication disruptions. Although traditional solutions can handle missing values through a interpolation-then-prediction pipeline, they suffer from a major limitation: interpolating the missing data and then extrapolating will introduce a cumulative error. To address these issues, we propose MissPred, a radar echo extrapolation model specifically designed for missing data patterns. MissPred employs a dual encoder–decoder architecture. Specifically, the training process involves the sequential execution of interpolation and extrapolation as two distinct serial tasks. In order to circumvent the occurrence of cumulative errors, interpolation and extrapolation are required to share encoder parameters. Furthermore, a missing spatiotemporal feature fusion module (MSTF) that is absent has been designed for the purpose of extracting fine-grained complete spatiotemporal features. Finally, the incorporation of adversarial training is introduced to enhance the authenticity of the prediction results. In order to evaluate the proposed model, case studies are conducted on real radar datasets. Our dataset covers missing rates ranging from 10% to 50%. The experimental results show that the model outperforms the baseline model with the prior interpolation of missing data in the missing mode with stable robustness.

The Gravity Recovery and Climate Experiment (GRACE) mission has revolutionized our understanding of Earth’s mass redistribution. However, enhancing the utility of GRACE products remains challenging due to inherent trade-offs between the temporal and spatial resolution, constrained by the mission design. High autocorrelation at the first lag in geoid coefficient time series is a key feature that enables the development of improved forecasting frameworks. Building on this temporal persistence of gravity coefficients, we propose a predictive framework to characterize the relationship between monthly and finer temporal scale-uncertainties of geoid coefficients. The proposed method enables reducing the finer temporal scale-uncertainties to levels comparable with known monthly uncertainties within the ranges of higher spectral degrees. With reduced uncertainties of finer temporal scales for higher spherical harmonic degrees, this predictive framework sets the groundwork for enhancing the analyses of rapid mass variations in the context of regional hydrology.

High Mountainous Areas (HMAs) with extensive snow cover presents considerable modelling challenges due to their intricate topography and data scarcity. However, these regions are particularly of concern as they are experiencing rapid warming, resulting in accelerated snowmelt along with more intense rainfall events, complicating the notion of flood typology that has existed for the region since long. Here, we present a comprehensive framework to evaluate projected changes in flood typology in HMAs for future relative to the Historical period. The results suggest that there will be a notable increase in rainfall-induced floods, particularly of the short-duration variety, coupled with a decrease in snowmelt-induced floods as the future periods advance. Additionally, there is a noticeable shift in the mean timing of floods, suggesting a delay in their occurrence. Although these results are specific to the three regions studied, similar changes will likely occur in other snow-dominated basins across HMAs. These insights could empower policymakers to make informed decisions and enhance regional risk assessment and management strategies.

Over the past two decades, the Amazon has experienced four severe large-scale droughts (i.e. 2005, 2010, 2015/16 and 2023), leading to drastically reduced water availability, slowed vegetation growth and higher forest mortality. As future droughts are expected to become more frequent and severe, accurately predicting the unprecedentedly low water storage levels and water shortages in advance is crucial. Herein, we developed a new approach to predict terrestrial water storage (TWS) during droughts, based on monthly changes in TWS (ΔTWS) and meteorological variables from 2003 to 2023. The model was trained during non-drought months and assessed during the four droughts when TWS values are well below the range of training data. The ΔTWS-based model excels in predicting drought-month TWS even only using precipitation and incoming solar radiation, with average correlation ( R ) over 0.9 and RMSE below 50 mm. The model also showed superior skills for predicting drought TWS months lead-time, with the 3-month prediction achieved high performance ( R > 0.8, RMSE < 80 mm). We further examined TWS predictions during the large-scale 2023 drought and found that the predicted TWS showed high spatial agreement with observed TWS, with all 1-, 2-, and 3-month lead-times reaching average R values over 0.9. Then we evaluated water deficits in the driest months (September—December) in 2023. The model predicted the affected regions with reasonable accuracy, achieving an average of 72% even at 3-month lead-time. We also analyzed how uncertainty in meteorological inputs affects model performance, revealing higher input uncertainty reduced the model performance. This study presents a reliable approach for estimating and predicting low water storage during severe large-scale droughts, enabling early warnings of water deficits across the Amazon. This study could be generalized to other regions, supporting proactive water resource management, water security policies, ecosystem protection and climate adaptation strategies.

As climate change exacerbates, droughts have become more intense, causing significant, widespread, and enduring adverse effects on carbon dynamics within terrestrial ecosystems. Therefore, it is crucial to quantitatively assess how drought influences carbon fluxes in these ecosystems. This understanding is essential for improving predictions of ecosystem responses to water stress and providing critical information to mitigate the effects of drought. In this study, we collected daily observations from 2004 to 2020 at two eddy covariance sites: one in an evergreen needle-leaved forest (abbreviated as ENFTha) and the other in grassland (abbreviated as GRAGri), both located in the North Temperate Zone. We first quantified the impacts of growing season (GS) droughts (GSD) on gross primary productivity (GPP) and ecosystem respiration (RECO). Our results indicated contrasting impacts of GSD on the two ecosystems: GPP and RECO increased in ENFTha, while they decreased in GRAGri. To explain these contrasting effects, we developed XGBoost models for GPP and RECO in ENFTha and GRAGri during the GS using eight environmental variables. We then applied the TreeExplainer-based SHapley Additive exPlanations framework to assess the significance of these variables in regulating GS carbon fluxes and to analyze their contributions to changes in GSD GPP and RECO. During the GS, four environmental variables—downwelling shortwave radiation (SW), vapor pressure deficit (VPD), longwave radiation (LW), and soil water content (near-surface air temperature (TA), SWC, SW, and LW)—significantly influenced GPP (RECO) in ENFTha and GRAGri, although their importance varied between the two ecosystems. These variables and their interactions also had nonlinear effects on GS GPP and RECO, with distinct threshold effects observed. In ENFTha, the enhanced SW demonstrated greater interaction effects with VPD, leading to increased GPP during GSD. In contrast, the rise in RECO could be attributed to the direct effects of TA. On the other hand, the declined SWC determined the decreases in carbon fluxes in GRAGri.

To improve the prediction of Typhoon Doksuri (2023), this paper explores how variations in horizontal scale factors used in assimilating radar-derived wind velocities influence the performance of numerical simulations and forecasts. Using the WRF-ARW model in conjunction with the WRF-3DVAR data assimilation system, two assimilation configurations were tested with horizontal length scale factors of 1.0 and 0.25. Results show that a reduced length scale facilitates a more detailed reconstruction of mesoscale features, including the typhoon’s eye and inner-core circulation, leading to improved accuracy in short-term intensity and structure forecasts. The experiment utilizing the 0.25 length scale exhibited a tighter warm core, stronger cyclonic wind bands, and a better representation of the vortex’s three-dimensional structure. However, this configuration also led to growing forecast deviations in the latter stages, likely due to imbalances introduced by excessive localization. In contrast, the 1.0-scale experiment produced smoother but less accurate structures and demonstrated larger track deviations. These findings highlight a key trade-off between localized observational influence and long-term forecast stability. The study underscores the importance of optimizing horizontal scale parameterization in variational assimilation to enhance the forecasting accuracy of high-impact tropical cyclones and offers practical insights for operational forecasting systems in regions frequently affected by typhoon activity.

Typhoons are extreme weather events that pose significant threats to human life and property. Sequential satellite imagery of typhoon clouds, which is rich in spatiotemporal information, plays a critical role in understanding their formation, development, and evolutionary dynamics. Recently, the rapid advancement of deep learning that combines physical insights with data-driven has opened new avenues for research in Earth Science. In this study, for high-quality prediction of sequential typhoon cloud images, we propose a physics-enhanced deep learning model termed C$^{2}$PhyNet. Specifically, we introduce a novel disentangling spatiotemporal block integrated with a criss-cross physics-enhanced unit. To further improve the fine structural details in the predicted typhoon cloud images, a concurrent spatial and channel squeeze-and-excitation attention mechanism is incorporated into both the encoder and decoder modules. Our quantitative analysis demonstrates the superiority of the proposed approach over existing sequential image prediction models on the publicly available Digital Typhoon dataset. The experimental results show that our method achieves superior performance, with a structural similarity index measure of 0.8200 and a peak signal-to-noise ratio of 23.26. C$^{2}$PhyNet is capable of generating high-quality sequential typhoon cloud images, which can significantly enhance the ability of meteorologists to forecast typhoon-related details with greater accuracy. Furthermore, our research contributes to improved risk mitigation and more effective disaster warning and management strategies.

End-to-end disparity estimation algorithms based on cost volume deployed in edge-end neural network accelerators have the problem of structural adaptation and need to ensure accuracy under the condition of adaptation operator. Therefore, this paper proposes a novel disparity calculation algorithm that uses low-rank approximation to approximately replace 3D convolution and transposed 3D convolution, WReLU to reduce data compression caused by the activation function, and unimodal cost volume filtering and a confidence estimation network to regularize cost volume. It alleviates the problem of disparity-matching cost distribution being far away from the true distribution and greatly reduces the computational complexity and number of parameters of the algorithm while improving accuracy. Experimental results show that compared with a typical disparity estimation network, the absolute error of the proposed algorithm is reduced by 38.3%, the three-pixel error is reduced to 1.41%, and the number of parameters is reduced by 67.3%. The calculation accuracy is better than that of other algorithms, it is easier to deploy, and it has strong structural adaptability and better practicability.

Abstract In mixed‐phase cumulus clouds, droplets and ice crystals are inhomogeneously distributed, such spatial inhomogeneity can be enhanced by inhomogeneous entrainment as it can strengthen the particle clustering, which may further influence the phase partitioning and interactions among hydrometeors. However, the scale of particle clustering induced by inhomogeneous entrainment is not well known. Utilizing high‐resolution in‐situ aircraft measurements in mixed‐phase cumulus clouds, this study shows due to inhomogeneous entrainment‐mixing, the cluster scales of droplets and ice crystals decrease by approximately 10 m from the cloud center to the edge. Changes in the clustering are correlated with the intensity of entrainment‐mixing. Clouds that are significantly affected by entrainment exhibit stronger enhancement of particle clustering and a more noticeable reduction in cluster scale. The findings from this study improve our understanding of the scale of particle clustering induced by entrainment‐mixing, and are potentially helpful in evaluating models.

China's Losses Plateau (LP) is one of the ecologically vulnerable and the most severe soil erosion regions. Thus, knowing spatiotemporal changes in evapotranspiration (ET) and its components (soil evaporation, E; transpiration, T; and vegetation interception evaporation, EI) and revealing the underlying mechanisms are vital for ecosystem and water resources sustainability for this region. Here, we investigate the spatiotemporal changes in ET and its components and then quantify the impacts of climate variables (i.e., precipitation, radiation, temperature, and relative humidity) and vegetation dynamics (e.g., land use/cover changes [LUCC] and changes in leaf area index [LAI]) on their annual trends, by using a process-based terrestrial ecosystem model and a joint-solution method with multiple sensitivity numerical experiments. Results show that over 67% of the study region experienced significant (p < 0.05) increases in annual ET, T, and EI, with regional average rises of 4.05, 3.67, and 0.74 mmyear-1, respectively. However, there are significant (p < 0.05) decreases in regional mean E of 0.38 mmyear-1, and the negative trend covers 35.8% of the study area. E, T, and EI changes dominate the annual ET trends over 11.8%, 87.3%, and 0.9% of the study area, respectively. Attribution analyses highlight the increased LAI as the critical factor governing these trends across most of the LP (>58%). At the same time, precipitation and LUCC play a more dominant role in the remaining areas. This study emphasizes the spatial heterogeneity in the drivers of changes in ET and its components and highlights the critical role of vegetation dynamics. These findings provide valuable insights for understanding the ET processes and guiding sustainable water resource management in the LP.

Three severe droughts impacted the Amazon in 2005, 2010, and 2015, leading to widespread above-average land surface temperature (LST) (i.e., positive thermal anomalies) over the southern Amazon in the dry season (Aug-Sep) of these years. Below-average dry-season incoming solar radiation (SW down arrow) and terrestrial water storage (TWSA) were simultaneously observed in 2005 and 2010, whereas the opposite was observed in 2015. We found that anomalies in precipitation (P), SW down arrow, and TWSA combined can well explain dry-season thermal anomalies during these droughts (average R (2) similar to 0.51). We investigated the causes for opposing anomalies in dry-season SW down arrow and TWSA, and found different hydro-meteorological conditions preceding the drought-year dry seasons. In 2005 and 2010, P was considerably below average during the wet-to-dry transition season (May-Jul), causing below-average TWSA in dry season that was favourable for fires. Increased atmospheric aerosols resulting from fires reduced solar radiation reaching the ground. In 2015, although below-average dry-season P was observed, it was above the average during the wet-to-dry transition season, leading to reduced fires and aerosols, and increased dry-season SW down arrow. To further examine the impact of opposite hydro-meteorological processes on the drought severity, we compared dry-season LST during droughts with the maximum LST during non-drought years (i.e., LST max ) for all grid cells, and a similar analysis was conducted for TWSA with the minimum TWSA (i.e., TWSA min ). Accordingly, the regions that suffered from concurrent thermal and water stress (i.e., LST > LST max and TWSA < TWSA min ) were identified. These regions are mainly observed over the southeast in 2005 and southern Amazon in 2010. In 2015, large-scale dry-season thermal stress was found over central and southeast Amazon with little water stress. This study underlines the complex interactions of different hydrological components and the importance of understanding the evolution of droughts to better predict their possible impacts on the Amazon rainforest.

These datasets are results from merging three FengYun passive microwave soil moisture observations at a 15kmx15km spatial resolution from 2011 to 2020 with continuous extension as data becomes available. Here, we rely on a merging technique that minimizes mean square error (MSE) using the signal-to-noise ratio (SNRopt) of the input parent products to first merge subdaily soil moisture products into dail averages (FY3_Merged_*). From these, these are gap-filled using a Data INterpolating Convolutional Auto-Encoder, DINCAE (FY3_Reoconstructed_*). The advantage of this method is that it comes with error variances(FY3_ErVar_*) for each pixel and time step which are useful for sevral applications.

Recent satellite and in-situ measurements show that forests can influence regional and global cloud cover through biophysical processes. However, forest’s possible local and non-local impacts on clouds remain unclear. By analyzing the model simulations from the Coupled Model Intercomparison Project Phase 6, here we show that deforestation-induced cloud cover changes have a strong latitudinal dependence, with decreased cloudiness in the tropics but increased cloudiness in the temperate and boreal regions. We further disentangle the local and non-local effects in influencing the cloudiness changes in model simulations. Results show that deforestation leads to a local cloud reduction in the tropics and a non-local cloud enhancement in the temperate and boreal regions. We demonstrate that the relationship between changes in cloud cover and deforestation would be misinterpreted without considering the non-local signals. Furthermore, our modeling results are inconsistent with recent observational studies, with enhanced clouds in model simulations but reduced clouds in observations in the temperate and boreal regions. Further efforts to explore the non-local effect and to reduce the model uncertainty could help advance our understanding of the biophysical effects of deforestation.