Quantifying the contribution of agricultural non-point sources to watershed export of nitrogen (N) depends on the integrated modeling of soil biogeochemistry and watershed hydrology. However, scale mismatch between biogeochemical and hydrological models prevents the reliable explicit tracking of N fate and transport in watersheds. To address this issue, we developed a spatially-explicit watershed model-INtegrated SImulator for bioGeochemical and Hydrological Trackings (INSIGHT)-by embedding a soil biogeochemical model at fine-resolution grids while simultaneously coupling surface and subsurface flowpath models fed by the outputs of water and N fluxes in each grid and layer. An agricultural-forested watershed in the lower Yangtze River basin was used to test if the INSIGHT model is capable of capturing N loss, removal, retention, and exports in different pathways as well as quantifying the contribution of agricultural non-point sources to N exports at high spatiotemporal scales. The model explained the variations in watershed TN export by 79% and enabled the tracking of contributions from different land uses and transport pathways. Surface and subsurface N generation were 26.0 and 206.9 Mg N yr⁻¹, respectively, but 89.6% of surface-derived and 28.3% of subsurface-derived N were only exported to the watershed outlet, because 53.5% of N generation was retained as soil organic N and 11.9% was removed by denitrification during horizontal transport. Tea cropland showed the highest export intensity (31.4 kg N ha⁻¹ yr⁻¹), followed by mixed forest (20.4 kg N ha⁻¹ yr⁻¹) and rice (18.5 kg N ha⁻¹ yr⁻¹). Tea cultivation on steep slopes, with short surface lag time and high fertilization rates, drives high N export intensity. The results suggest that subsurface transport as the dominant mechanism for N retention and export in a watershed, calling for targeted mitigation strategies. These findings highlight that INSIGHT provides a useful tool for tracking N fate and transport in watersheds and supporting targeted agricultural non-point pollution mitigations.

Study region A typical urban residential compound in China. Equipped with an extensive underground parking. With considerable green space area providing potential benefits of runoff mitigation via imperviousness disconnection. Study focus This study focuses on the impacts of the underground parking structure on the hydrological processes and the green space functions. Using a fully-distributed and coupled surface-subsurface model, we analyzed the changes in soil saturation, runoff, and evapotranspiration caused by the underground structure. New hydrological insights for the region Results show that the underground parking structure causes soil water accumulation on its top in long term, leading to higher runoff and evapotranspiration. The magnitude of these effects relies on water supply to the green space. Lower water supply (e.g., when the climate is dryer or soil permeability is lower) would result in less soil water accumulation, thereby weakening the hydrological effects. In contrast, these effects are amplified when imperviousness disconnection is implemented, as it increases water supply to the green space through focused infiltration. Moreover, for the condition of imperviousness disconnection, these effects are not confined to the disconnection point, as the water accumulation zone can develop via subsurface lateral preferential flow along the underground parking slab, potentially influencing runoff generation patterns over a larger area. These findings highlight the importance of considering urban underground structure in stormwater management, especially for regions with wet climate.

Reliable high-resolution precipitation is essential for monitoring hydrologic extremes and informing climate-risk decisions, yet satellite precipitation products often show biases and remain too coarse (5–25 km) to resolve localized processes. Conventional downscaling also tends to overlook dynamic moisture–cloud mechanisms that drive precipitation variability. We develop a Physics-Informed Geospatial Machine-Learning Downscaling rainfall model (PIGMLD) to produce 1-km daily precipitation over China (2000–2020) by combining the 10-km Global Precipitation Measurement (GPM) Integrated Multi-satellitE Retrievals for GPM (IMERG) (GPM IMERG) product with rare ground observations, ERA5-Land precipitation, and physically traceable covariates describing moisture, clouds, and land–atmosphere coupling. Evaluation against independent gauges across China and nine major river basins shows broad skill gains: 84.1 % of stations achieve KGE > 0.60, with improved event detection and reduced systematic bias. Gains are constrained by complex topography and sparse in-situ stations in the Northwest and Southwest basins (median RMSE = 1.18 mm; KGE = 0.45). For heavy and torrential rainfall, RMSE decreases by 36.1 % and 28.3 %, respectively. Relative threshold assessments indicate robust corrections under dry and wet extremes: for events below the 10th percentile, BIAS is reduced by ∼66.7 % at >90 % of stations; for events ≥90th percentile, underestimation is substantially alleviated, with BIAS typically reduced by ∼50.1 %. XGBoost–SHAP attribution reveals scale-dependent controls: 10-km estimates are dominated by cloud and column moisture, whereas 1-km estimates are more sensitive to near-surface humidity and land-surface states, and heavy rainfall reflects coupled moisture–dynamics–thermodynamics interactions. Overall, PIGMLD provides a mechanism-aware pathway for producing and interpreting 1-km precipitation fields and clarifies when finer-scale information improves extreme-event characterization.

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.

Accurate soil moisture (SM) estimation is essential for applications such as flood forecasting, irrigation management, and drought monitoring. Data assimilation (DA) provides a powerful means to integrate multi-source SM observations, yet conventional approaches-such as the ensemble Kalman filter (EnKF) and its variants-are constrained by linear update formulations and Gaussian assumptions. To overcome these limitations, we introduce the ensemble deep learning filter (EnDLF), a novel DA framework that replaces the Kalman-based formula with a deep learning (DL)-based nonlinear mapping capable of capturing complex relationships and non-Gaussian patterns. To reduce the computational burden of retraining the DL model at each assimilation step, EnDLF employs a transfer learning strategy that fine-tunes model parameters across assimilation cycles, substantially shortening training time. Applied to an SM assimilation case with pronounced non-Gaussianity, EnDLF outperforms EnKF, reducing root-mean-square error (RMSE) by 10.53 % and 9.09 % and improving Kling-Gupta efficiency (KGE) by 5.13 % and 3.90 % for surface (0-10 cm) and subsurface (10-50 cm) SM, respectively. Enhanced SM estimates further translate into improved streamflow predictions, increasing Nash-Sutcliffe efficiency (NSE) from 0.848 with EnKF to 0.864 with EnDLF. These results demonstrate the effectiveness of integrating DL into DA to capture nonlinear dependencies and highlight EnDLF's potential for assimilating other hydrological variables with non-Gaussian characteristics, such as precipitation and streamflow.

Achieving water budget closure using remote sensing datasets remains a major challenge for global hydrological monitoring due to uncertainties in key water budget components, including precipitation (P), evapotranspiration (ET), streamflow (Q), and terrestrial water storage change (TWSC). While various water budget closure correction (BCC) methods have been developed to address imbalances, their effectiveness is often compromised by the neglect of water surface evaporation (WSE) and the substitution of runoff (R) for observed Q. These practices introduce considerable uncertainties. This issue remains insufficiently explored, particularly the impact of different ratios of water surface area to total basin size on total uncertainty, as well as the contribution of substituting R for Q to total uncertainty, under different climates (e.g., in humid and arid watersheds). In this study, we introduce a research framework to assess the impact of substituting R for Q, neglecting WSE, and their combined effects on water budget closure across 62 river basins worldwide. Multiple datasets were used for each budget component, and four widely used BCC methods with varying levels of complexity were employed to enhance the robustness of results: Proportional Redistribution (PR), Constrained Kalman Filter (CKF), Multiple Collocation (MCL), and Minimized Series Deviation (MSD). Our findings reveal that the effects of neglecting WSE are mainly influenced by the ratio of water surface area to land area within each basin. When this ratio exceeded 9.92%, neglecting WSE introduces non-negligible errors. Substituting R for Q has more pronounced effects in humid basins (approximately 11.53%) than in arid basins (around 4.13%). Furthermore, combined uncertainties from both sources are higher than that by a single factor alone in basins with a high proportion of water surface area. These findings provide critical insights into the uncertainties associated with satellite-based water budgeting and offer valuable guidance for improving the reliability of hydrological datasets, particularly in basins lacking direct Q and WSE observations.

Microplastics are pervasive in inland waters, yet large-scale association patterns of abundance and traits remain unclear. Here we compiled a global lake database integrating surface-water and sediment observations with 56 hydroclimatic and anthropogenic indicators. Most hydroclimatic variables show positive associations with abundance in surface waters but act oppositely in sediments. Leading contribution patterns differ: leaf area index and 2-m dew-point temperature rank highest in surface waters, whereas solar radiation and wind speed lead in sediments. To enable finer-scale evaluation, we conducted dense sampling in China, which hosts one of the world's largest and most intensively studied lake systems. Hydroclimatic and anthropogenic fingerprints show clear cross-compartment contrasts: surface-water traits associate strongly with aquaculture production and rural income, while sediment traits more link to population aged 0-14 and chemical oxygen demand. Our findings imply a hydroclimate-human game across compartments, highlighting fingerprints structuring microplastic abundance and traits.

Accurate estimation of weather variables is essential for climate science and real-world applications, yet sparse official weather station networks often limit data reliability in many regions. This study highlights the transformative potential of integrating third-party automatic weather station (TPAWS) data to improve gridded climate data products. Daily rainfall, one of the most important yet challenging weather variables to estimate, is used as a case study in southwestern Western Australia. By incorporating quality-controlled TPAWS observations, we reduce the root mean square error (RMSE) of rainfall estimates by over 15% and false no-rain rates by 30%, with notable improvements during extreme events. These results illustrate how TPAWS data can augment official networks, offering a scalable, cost-effective approach to improve the accuracy of diverse weather variables beyond rainfall alone. Our findings provide compelling evidence of the scientific and practical value of leveraging non-traditional datasets to address data sparsity, opening new avenues for research and development in climate data integration worldwide.

The rainy season characteristics are directly modulated by atmospheric circulation and moisture transport dynamics. Focusing on the characteristics of the rainy season onset date (RSOD), this study aims to advance the understanding and prediction of climate change impacts on agricultural production and disaster mitigation strategies. Based on rainfall data from 66 meteorological stations in northeast China (NEC) from 1961 to 2020, this study determined the patterns of the RSOD in the region and established its mechanistic linkages with atmospheric circulation and water vapor transport mechanisms. This study identifies a climatic regime shift around 2000, with the RSOD transitioning from low to high interannual variability in NEC. Further analysis reveals a strong correlation between the RSOD and atmospheric circulation characteristics: cyclonic vorticity amplifies before the RSOD and dissipates afterward. Innovatively, this study reveals a significant transition in the water vapor transport paths during the early rainy season in NEC around 2000, shifting from eastern Mongolia–Sea of Japan to the northwestern Pacific region. Moreover, the advance or delay of the RSOD directly influences the water vapor transport intensity—an early (delayed) RSOD is associated with enhanced (weakened) water vapor transport. These findings provide a new perspective for predicting the RSOD in the context of climate change while providing critical theoretical underpinnings for optimizing agricultural strategies and enhancing disaster prevention protocols.

Climate extremes exert detrimental influences on the water retention capacity and carbon sequestration functions of forest ecosystems. However, the response mechanisms of carbon-water cycles and their coupling relationships to climate extremes remain unclear. To address these issues, we investigated in an evergreen forest ecosystem located in Southern China, with comprehensive datasets and machine learning (ML) algorithms, where water-use efficiency (WUE) is defined as the ratio of gross primary production (GPP) to evapotranspiration (ET). The spatial-temporal distribution characteristics of GPP, ET, and WUE, as well as their responses to observed extreme weather events (torrential rains, drought, heat wave and cold wave) and climate extreme indices (CEIs) during 1979-2017 were investigated. We evaluated the accuracy of 8 types of ML algorithms in modelling observed GPP and ET, and the extreme gradient boosting model showed the best performance (GPP: R-2 = 0.91-0.98, ET: R-2 = 0.84-0.98). Regionally averaged annual values indicated significant increasing (p < 0.01) trends for GPP (3.28 gC m(-2) a(-2)), ET (0.62 mm a(-2)), and WUE (0.0023 gC m(-2) mm(-1) a(-1)), with mean annual values exhibiting a gradient increase from the northwest to southeast. Across the 21 CEIs, the precipitation-related indices showed positive correlations with GPP and ET, and the warm (cold)-related temperature indices showed positive (negative) correlations with these fluxes. These fluxes were more sensitive to torrential rains and cold wave, during which the response of ET was more immediate, while the negative impact of cold wave on GPP gradually intensified over time. As a result, WUE initially remained stable but then declined sharply during cold period. Overall, the carbon-water fluxes for evergreen forests on Southern China were most likely to exhibit high risk during cold events. Our findings provide valuable references for the response of evergreen forests to climate extremes.

Uncertainty estimation in integrated dynamic models, especially for water quality simulations, is challenging due to its complexity and limited observations. This study employs variance decomposition analysis (VDA) to estimate multi-source input uncertainties (precipitation, temperature, point source pollution, and social economy) and their propagation for integrated water quantity and quality simulations in the HEQM model. The Zhugan River watershed, a typical agricultural area in southern Huai River, China, was chosen for the case study. Results showed that precipitation is the main source of uncertainty for both runoff and TN load simulations, with contributions of 64.1% and 61.3%, respectively. Temperature is the main direct uncertainty source for TN load simulation (80.3%), along with contributions from point source pollution (15.1%) and social economy (4.6%). Furthermore, 70.7% of TN load simulation uncertainty propagates from upstream runoff simulation. Estimating multi-source input uncertainties and their propagation is essential for improving integrated water system models.

It is widely recognised that air pollutants including sulphur dioxide (SO2), respirable suspended particulates (PM10), nitrogen oxides (NOx), nitrogen dioxide (NO2), and ozone (O3), as well as weather conditions such as temperature (Temp) and relative humidity (RH), are major causes of respiratory illnesses. To quantify the unknown and highly nonlinear relationships between these factors and respiratory illness, and the cumulative effect from exposure to symptoms, in this paper, we propose a semiparametric index model with constraints to capture the cumulative effect additively and the nonlinearity nonparametrically. As a case study, the model is applied to a dataset from the Hong Kong SAR. As the data period includes the SARS (severe acute respiratory syndrome) epidemic in 2003, we further construct a growth curve model to account for the extra impact of public health measures. The results show that the effects of SO2, NO2, and PM10 decay quickly, while the other pollutants have a period of stable accumulation (18–38 days for O3, 2–30 days for NOx, 1–13 days for RH, and 4–12 days for temperature). The results also show that the proposed model has a better fitting performance than previous models and hence has potential applications in health monitoring programs.

Achieving water budget closure improves the consistency of water budget component datasets, including precipitation (P), evapotranspiration (ET), streamflow (Q) and terrestrial water storage change (TWSC), thereby advancing our understanding of basin-scale water cycle dynamics. Existing water budget closure correction (BCC) methods typically aim to eliminate the entire water imbalance error (Delta Res) by fully redistributing it across budget components. However, this often overlooks the trade-off between achieving perfect closure and the errors introduced into the corrected components through this redistribution. Moreover, inaccurate estimation of redistribution weights can lead to contradictory outcomes, such as negative values in P, ET, or Q. In this study, we quantify the uncertainties introduced by four existing BCC methods (CKF, MCL, MSD, and PR) at the monthly scale across 84 basins spanning diverse climate zones. We then propose a novel method, IWE-Res, which identifies an optimal redistributing strategy by minimizing the combined error from both the errors introduced to individual budget components and the remaining Delta Res error. This method also reduces the occurrence of negative values in the corrected datasets. Our results show: (1) Existing BCC methods can result in negative values in 0 %-10 % of the time series for each corrected budget component (typically <5 %); (2) The proposed IWE-Res method improves the accuracy of corrected components compared to existing methods, reducing RMSE by 29.5 % for P, 24.7 % for ET, 69.0 % for Q, and 6.8 % for TWSC; and (3) For most basins, excluding those in cold regions, the optimal redistribution is achieved when 40 %-90 % of Delta Res is redistributed. By offering a more balanced approach to water budget closure, this study improves the accuracy and reliability of corrected budget component datasets.

Canopy temperature is an important indicator of plants’ water status. The so-called PeriodiCT model was developed to forecast canopy temperature using ambient weather variables, providing a powerful tool for planning crop irrigation scheduling. As this model requires observed data in its parameter training before implementing the forecast, it is important to understand the data requirements in the model training such that accurate forecasts are attained. In this work, we conduct a comprehensive assessment of the PeriodiCT model in terms of sample size requirement and predictabilities across sensors in a field and across seasons for the full model and sub-models. The results show that (1) 5 days’ observations are sufficient for the full model and sub-models to achieve very high predictability, with a minimum coefficient of efficiency of 0.844 for the full model and 0.840 for the sub-model using only air temperature. The predictability decreases in the following order: full model, sub-model without radiation S, with air temperature Ta and vapor pressure VP, and with only Ta. The predictions perform reasonably well even when only one day’s observations are used. (2) The predictability into the future is very stable as the prediction steps increase. (3) The predictabilities of the full and sub-models when using a trained model from one sensor for another sensor perform comparatively well, with a minimum coefficient of efficiency of 0.719 for the full model and 0.635 for the sub-model using only air temperature. (4) The predictabilities of the sub-models without solar radiation when using trained models from one season for another season perform comparatively well, with a minimum coefficient of efficiency of 0.866 for the full model and 0.764 for the sub-model using only air temperature, although the cross-season performances are not as good as the cross-sensor performances. The importance of the predictors is in the order of air temperature, vapor pressure, wind speed, and solar radiation, while vapor pressure and wind speed have similar contributions, and solar radiation has only a marginal contribution.

Remote sensing data primarily provide insights into surface soil moisture (SM). By integrating auxiliary data such as topography and meteorology, deep learning (DL) can enable the estimation of SM beneath the soil surface. However, DL models often overlook underlying physical mechanisms and lack interpretability, which may compromise their reliability in predicting SM. In this study, we introduce a physics-guided DL (PGDL) approach for estimating multi-layer SM. Our method utilizes convolutional neural networks (CNN) to analyze spatial features and long short-term memory networks (LSTM) to capture temporal dynamics, effectively modeling the spatiotemporal characteristics of SM. Crucially, we integrate the Richardson-Richards equation, which describes the spatiotemporal dynamics of SM, as a constraint within the CNN-LSTM model's loss function. Additionally, we constrain the LSTM activation function to keep estimated values within the residual and saturated SM range. This integration guides the model to learn physical characteristics, thereby enhancing accuracy and interpretability. Our method (named PGDL-CNN-LSTM) outperforms other machine learning methods (CNN-LSTM, extreme gradient boosting, and random forest) across multi-layer (10 cm, 20 cm, 30 cm, 40 cm, and 50 cm). Specifically, for 10 cm SM, our method achieves a 9.6 % reduction in root mean square error and a 92.0 % reduction in physical inconsistency compared to CNN-LSTM alone. Furthermore, we extend surface SM to estimate root-zone SM at 10 km resolution. Our method accurately captures dynamic changes in SM and remains effective even with a 50 % reduction in data. In conclusion, integrating physical constraints significantly enhances the depth and interpretability of SM estimation from remote sensing data. These findings highlight the potential of our method for various hydrological applications.

As the global climate warms, sea level rise (SLR) is expected to accelerate, leading to significant changes in tidal characteristics. This study investigates the dynamic relationships between tidal characteristics and SLR in the Yangtze River Estuary (YRE). The analysis draws on 33 numerical simulations performed with the MIKE21 model, focusing on tidal asymmetry intensity A (based on amplitudes of the M2 and M4 tidal constituents), the average durations of flood and ebb tides, as well as average flow velocity and average flow direction. Results show a linear relationship between SLR and tidal asymmetry intensity A, average flow velocity, and average flow direction is relatively strong, whereas its linear relationship with the average duration of flood and ebb tides is comparatively weaker, and this pattern varies by station. Specifically, the average duration of ebb tide shows a decreasing trend, while the average duration of flood tide increases, with these nonlinear changes being particularly pronounced in the upper reaches of the YRE. Among the five Shared Socioeconomic Pathway (SSP) scenarios, SSP5-8.5 exhibits the strongest amplification effects on average flow velocity, average flow direction, and tidal asymmetry intensity A, with values generally exceeding 2.0. The modeling framework and analytical methods presented here are broadly applicable, and can be adapted to investigate tidal responses induced by SLR in other estuarine systems with similar physical and hydrodynamic characteristics.

Providing high-resolution, skilful weather sequences for months across extensive regions is crucial for climate-sensitive sectors. Achieving this goal is particularly challenging for precipitation due to limitations in General Circulation Models (GCMs) that provide ensemble Seasonal Climate Forecasts (SCFs) with discrepancies from daily observations. Traditional downscaling techniques like Quantile Mapping (QM) rely on pre-defined relationships. Deep learning methods have shown promise in short-range forecasts or fortnightly precipitation totals, but often assume alignment between low-resolution forecasts and high-resolution observations. This paper introduces a novel Generative Adversarial Network (GAN)-based approach, called Downscaling with Enhanced Super-Resolution GAN (DESRGAN), for addressing discrepancies in long-range precipitation forecasts. DESRGAN employs cascaded Residual-in-Residual Dense Blocks (RRDB) in its generator to produce realistic higher-resolution images, a U-Net discriminator to assess mismatches at both pixel and image levels, and an augmented loss function to account for losses at different scales. It is finalised according to the Continuous Ranked Probability Score (CRPS), an overall skill metric for ensemble forecasts that assesses both accuracy and reliability. Tested on downscaling both versions 1 and 2 of the Australian Community Climate and Earth-System Simulator - Seasonal model (ACCESS-S) from a 60 km resolution to 5 km with forecast lead times up to 216 days, DESRGAN outperforms bicubic interpolation, QM, Very Deep Super-Resolution (VDSR), and a long-standing benchmark Climatology in terms of CRPS skill score, Mean Absolute Error (MAE), and relative bias. Compared to its downscaling counterparts, DESRGAN achieves substantial improvements, reducing CRPS by at least 3.27% and MAE by 4.16%. Concerning reliability alone, a critical factor for operational applications, preliminary findings from the current ACCESS-S2 operational system show that DESRGAN performs less well than QM and Climatology. Consequently, further work is required to improve deep learning methodologies for downscaling SCFs.

Third-party automatic weather stations (TPAWS) provide a compelling data source for scientists and practitioners to observe and estimate more accurate fine-scale atmospheric conditions, including daily maximum and minimum temperature (denoted as Tmax and Tmin, respectively), than the current primary weather observation network can offer. Several uncertainties and errors arise in data from TPAWS as the quality control applied to these stations may be inadequate or ad hoc. In this study, we develop a statistical approach to evaluate the quality of daily Tmax and Tmin observations collected from TPAWS in Australia. Our approach compares a target observation with multiple types of reliable reference data, including neighbouring primary weather observations from the official Bureau of Meteorology of Australia stations, Australian Gridded Climate Data, and numerical weather prediction data. Guided by the operational requirements in terms of automation, interpretability, and simplicity as well as expandability, a separate test is formed for each type of reference data and then all the individual tests are merged to generate a single result based on a Gaussian mixture model that is used to provide the final overall assessment for each TPAWS observation. The overall assessment is made in the form of a p-value-based confidence score that measures the difference between the target observation and trusted reference data. Our method is validated by synthetic datasets based on high-quality observations and is also applied to daily Tmax and Tmin observations from 184 TPAWS owned by the Department of Primary Industries and Regional Development of Western Australia. The framework can be readily applied to different regions with different reliable or trusted data sources. We present a statistical method assessing daily Tmax and Tmin data from third-party automatic weather stations (TPAWS). Our approach employs p-value-based confidence scores, detecting disparities between TPAWS observations and trusted references. The figure depicts 2019 daily Tmax data from a TPAWS, highlighting eight potentially erroneous readings (labelled in red; Section 5 elaborates). Some anomalies may not be immediately evident in time series trends alone, but comparing them with surrounding reference observations highlights the significance of our method's contribution. image

Extreme climate occurred frequently in subtropical region, which seriously affects carbon and water fluxes such as evapotranspiration (ET) and gross primary productivity (GPP) of terrestrial ecosystems. The process-based biome biogeochemical cycles (Biome-BGC) model is widely used for simulating carbon and water fluxes of forest ecosystems. However, the lack of the interaction information of climate, vegetation and soil, such as the hysteresis effect, canopy stratification on photosynthesis, impedes better simulations of the ecohydrological processes. Here, we tended to improve the simulation accuracy of Biome-BGC model at a subtropical forest on the Xin’an River in Southeastern China by reconstructing the precipitation series, modifying the ET and canopy multilayers modules, and optimizing the parameters. The spatiotemporal patterns of GPP, ET, water use efficiency (WUE) and their response to environmental factors across the Xin’an River Basin from 1982 to 2018 were further explored. The results showed that the improved model performed well, with the determination coefficient, root means square error and mean absolute error being 0.730, 1.522 gC/m2/d and 1.218 gC/m2/d for GPP, 0.857, 1.082 mm/d and 0.838 mm/d for ET, respectively. Basin-averaged GPP, ET and WUE increased during 1982-2018 and these increasing trends were more pronounced during 1999-2018. Significant positive trends of WUE occurred in the northeast corner. The increasing air temperature and precipitation respectively dominated the increase in GPP and ET, the increasing CO2 concentration and NDVI mitigated the negative effect of extreme precipitation on WUE. Given that human activities such as afforestation have effectively reduced the extent of damage to forest ecosystems from extreme precipitation, we highlight an urgent need to formulate adaptation strategies aimed at reducing the risk of extreme climate in humid regions.

Compound drought and heatwaves (CDH) have garnered increasing attention because concurrent extreme events can exacerbate the harmful impacts caused by univariate extremes. However, various severities in CDH events and their relationships with sea surface temperature (SST) variations in China remain little understood. Here, we accurately identify CDH events and multi-aspect of characteristics using the standardized precipitation evapotranspiration index (SPEI) and the excess heat factor (EHF) during the extended summer (May–September) of 1961–2017. The evolution of multifaceted characteristics of CDH and their association with SST variation are further explored. The results suggest that the number, frequency, duration and intensity of regional CDH events show heterogeneous spatial patterns, with a significant increasing trend. A consistent abrupt transition in CDH characteristics averaged over China occurred in the period of 1993–1996. Mild and moderate CDHs occur more commonly in Northwest and North China, whereas severe CDHs are mainly found in central and eastern regions. Mild and moderate CDHs are more susceptible to SST modes than severe CDH, and there are strong positive correlations between mild and moderate CDH characteristics and SST variations in the northwest and northern regions. Compared to El Niño-Southern Oscillation (ENSO), Indian Ocean Dipole (IOD) plays a dominant role in the intensifications of mild and moderate CDH events. Regionally, the northwest and north have experienced longer, more frequent and severe CDH events during the positive phase of IOD. These findings reveal the divergent evolutions in CDH characteristics with various severities and inconsistent impacts of different SST modes on the compound events.