Global prediction of extreme floods in ungauged watersheds.

Floods are one of the most common natural disasters, with a disproportionate impact in developing countries that often lack dense streamflow gauge networks1. Accurate and timely warnings are critical for mitigating flood risks2, but hydrological simulation models typically must be calibrated to long data records in each watershed. Here we show that artificial intelligence-based forecasting achieves reliability in predicting extreme riverine events in ungauged watersheds at up to a five-day lead time that is similar to or better than the reliability of nowcasts (zero-day lead time) from a current state-of-the-art global modelling system (the Copernicus Emergency Management Service Global Flood Awareness System). In addition, we achieve accuracies over five-year return period events that are similar to or better than current accuracies over one-year return period events. This means that artificial intelligence can provide flood warnings earlier and over larger and more impactful events in ungauged basins. The model developed here was incorporated into an operational early warning system that produces publicly available (free and open) forecasts in real time in over 80 countries. This work highlights a need for increasing the availability of hydrological data to continue to improve global access to reliable flood warnings.

Different roads, same destination: The shared future of plant ecophysiology and ecohydrology.

Terrestrial water fluxes are substantially mediated by vegetation, while the distribution, growth, health, and mortality of plants are strongly influenced by the availability of water. These interactions, playing out across multiple spatial and temporal scales, link the disciplines of plant ecophysiology and ecohydrology. Despite this connection, the disciplines have provided complementary, but largely independent, perspectives on the soil-plant-atmosphere continuum since their crystallization as modern scientific disciplines in the late 20th century. This review traces the development of the two disciplines, from their respective origins in engineering and ecology, their largely independent growth and maturation, and the eventual development of common conceptual and quantitative frameworks. This common ground has allowed explicit coupling of the disciplines to better understand plant function. Case studies both illuminate the limitations of the disciplines working in isolation, and reveal the exciting possibilities created by consilience between the disciplines. The histories of the two disciplines suggest opportunities for new advances will arise from sharing methodologies, working across multiple levels of complexity, and leveraging new observational technologies. Practically, these exchanges can be supported by creating shared scientific spaces. This review argues that consilience and collaboration are essential for robust and evidence-based predictions and policy responses under global change.

Long-term climate and hydrologic regimes shape stream invertebrate community responses to a hurricane disturbance.

Disturbances can produce a spectrum of short- and long-term ecological consequences that depend on complex interactions of the characteristics of the event, antecedent environmental conditions, and the intrinsic properties of resistance and resilience of the affected biological system. We used Hurricane Harvey's impact on coastal rivers of Texas to examine the roles of storm-related changes in hydrology and long-term precipitation regime on the response of stream invertebrate communities to hurricane disturbance. We detected declines in richness, diversity and total abundance following the storm, but responses were strongly tied to direct and indirect effects of long-term aridity and short-term changes in stream hydrology. The amount of rainfall a site received drove both flood duration and flood magnitude across sites, but lower annual rainfall amounts (i.e. aridity) increased flood magnitude and decreased flood duration. Across all sites, flood duration was positively related to the time it took for invertebrate communities to return to a long-term baseline and flood magnitude drove larger invertebrate community responses (i.e. changes in diversity and total abundance). However, invertebrate response per unit flood magnitude was lower in sub-humid sites, potentially because of differences in refuge availability or ecological-evolutionary interactions. Interestingly, sub-humid streams had temporary large peaks in invertebrate total abundance and diversity following recovery period that may be indicative of the larger organic matter pulses expected in these systems because of their comparatively well-developed riparian vegetation. Our findings show that hydrology and long-term precipitation regime predictably affected invertebrate community responses and, thus, our work underscores the important influence of local climate to ecosystem sensitivity to disturbances.

Organic Matter Accumulation and Hydrology as Drivers of Greenhouse Gas Dynamics in Newly Developed Artificial Channels.

Artificial channels, common features of inland waters, have been suggested as significant contributors to methane (CH4) and carbon dioxide (CO2) dynamics and emissions; however, the magnitude and drivers of their CH4 and CO2 emissions (diffusive and ebullitive) remain unclear. They are characterized by reduced flow compared to the donor river, which results in suspended organic matter (OM) accumulation. We propose that in such systems hydrological controls will be reduced and OM accumulation will control emissions by promoting methane production and outgassing. Here, we monitored summertime CH4 and CO2 concentrations and emissions on six newly constructed river-fed artificial channels, from bare riparian mineral soil to lotic channels, under two distinct flow regimes. Chamber-based fluxes were complemented with hydrology, total fluxes (diffusion + ebullition), and suspended OM accumulation assessments. During the first 6 weeks after the flooding, inflowing riverine water dominated the emissions over in-channel contributions. Afterwards, a substantial accumulation of riverine suspended OM (≥50% of the channel's volume) boosted in-channel methane production and led to widespread ebullition 10× higher than diffusive fluxes, regardless of the flow regime. Our finding suggests ebullition as a dominant pathway in these anthropogenic systems, and thus, their impact on regional methane emissions might have been largely underestimated.

Hydrologic Connectivity Regulates Riverine N2O Sources and Dynamics.

Indirect nitrous oxide (N2O) emissions from streams and rivers are a poorly constrained term in the global N2O budget. Current models of riverine N2O emissions place a strong focus on denitrification in groundwater and riverine environments as a dominant source of riverine N2O, but do not explicitly consider direct N2O input from terrestrial ecosystems. Here, we combine N2O isotope measurements and spatial stream network modeling to show that terrestrial-aquatic interactions, driven by changing hydrologic connectivity, control the sources and dynamics of riverine N2O in a mesoscale river network within the U.S. Corn Belt. We find that N2O produced from nitrification constituted a substantial fraction (i.e., >30%) of riverine N2O across the entire river network. The delivery of soil-produced N2O to streams was identified as a key mechanism for the high nitrification contribution and potentially accounted for more than 40% of the total riverine emission. This revealed large terrestrial N2O input implies an important climate-N2O feedback mechanism that may enhance riverine N2O emissions under a wetter and warmer climate. Inadequate representation of hydrologic connectivity in observations and modeling of riverine N2O emissions may result in significant underestimations.

Integrating conceptual and machine learning models to enhance daily-Scale streamflow simulation and assessing climate change impact in the watersheds of the Godavari basin, India.

This study examined and addressed climate change's effects on hydrological patterns, particularly in critical places like the Godavari River basin. This study used daily gridded rainfall and temperature datasets from the Indian Meteorological Department (IMD) for model training and testing, 70% and 30%, respectively. To anticipate future hydrological shifts, the study harnessed the EC-Earth3 data, presenting an innovative methodology tailored to the unique hydrological dynamics of the Godavari River basin. The Sacramento model provided initial streamflow estimates for Kanhargaon, Nowrangpur, and Wairagarh. This approach melded traditional hydrological modeling with advanced multi-layer perceptron (MLP) capabilities. When combined with parameters like lagged rainfall, lagged streamflow, potential evapotranspiration (PET), and temperature variations, these initial outputs were further refined using the Sac-MLP model. A comparison with Sacramento revealed the superior performance of the Sac-MLP model. For instance, during training, the Nash Sutcliffe efficiency (NSE) values for the Sac-MLP witnessed an improvement from 0.610 to 0.810 in Kanhargaon, 0.580 to 0.692 in Nowrangpur, and 0.675 to 0.849 in Wairagarh. The results of the testing further corroborated these findings, as evidenced by the increase in the NSE for Kanhargaon from 0.890 to 0.910. Additionally, Nowrangpur and Wairagarh experienced notable improvements, with their NSE values rising from 0.629 to 0.785 and 0.725 to 0.902, respectively. Projections based on EC-Earth3 data across various scenarios highlighted significant shifts in rainfall and temperature patterns, especially in the far future (2071-2100). Regarding the relative change in annual streamflow, Kanhargaon projections under SSP370 and SSP585 for the far future indicate increases of 584.38% and 662.74%. Similarly, Nowrangpur and Wairagarh are projected to see increases of 98.27% and 114.98%, and 81.68% and 108.08%, respectively. This study uses EC-Earth3 estimates to demonstrate the Sac-MLP model's accuracy and importance in climate change water resource planning. The unique method for region-specific hydrological analysis provides vital insights for sustainable water resource management. This research provides a deeper understanding of climate-induced hydrological changes and a robust modeling approach for accurate predictions in changing environmental conditions.

An extended time-varying Budyko framework for quantifying the hydrological effect of vegetation restoration under climate variations at watershed scale.

The Budyko framework, widely used to quantify the watershed hydrological response to the watershed characteristics and climate variabilities, is continuously refined to overcome the disadvantages of steady state assumption. However, dynamic variations in vegetations and climate variables are not fully integrated including coverages and precipitation regimes of intensity, frequency, and duration. To address this, we developed an innovative approach for determining the parameter ω in the Budyko framework to quantify the hydrological effects of vegetation restoration in a mesoscale watershed located in northern China. We found that fractional vegetation coverage (FVC), heavy precipitation amount (95pTOT), and the number of precipitation days (R01mm) are significant variables for estimating ω to improve the predictive capability of the watershed response. This extended time-varying Budyko framework can rigorously capture the temporal variations and underlying mechanisms of interactions between vegetation dynamic and precipitation regime partitioning precipitation (P) to R. Under the Budyko-Fu framework, compared to constant ω (ω‾) or ω that only considers FVC (ωP) or precipitation regimes (ωFVC) for simulating R, using ω that integrated FVC and precipitation regimes (ωP-FVC) can improve Nash-Sutcliffe efficiency coefficient (NSE) by 24.81%, while reduced the root mean squared error (RMSE) and relative error (RE) by 64.08% and 65.77%, respectively. Although the increase in climatic dryness (PET/P) resulted in decreased R, the increase in FVC has also a significant contribution to this decrease due to vegetation restoration. We highlight that decrease precipitation intensity (95pTOT) and frequency (R01mm) amplified the hydrological effects of vegetation restoration, causing a 79.09∼100.31% increase in R compared to the independent impact of changes in FVC. We conclude that the extended time-varying Budyko framework by precipitation regime is more rigorous for quantifying the hydrological effects of ecological restoration under climate change and providing more reliable approach for adaptive watershed management.

Seasonal hydrological dynamics affected the diversity and assembly process of the antibiotic resistome in a canal network.

The significant threat of antibiotic resistance genes (ARGs) to aquatic environments health has been widely acknowledged. To date, several studies have focused on the distribution and diversity of ARGs in a single river while their profiles in complex river networks are largely known. Here, the spatiotemporal dynamics of ARG profiles in a canal network were examined using high-throughput quantitative PCR, and the underlying assembly processes and its main environmental influencing factors were elucidated using multiple statistical analyses. The results demonstrated significant seasonal dynamics with greater richness and relative abundance of ARGs observed during the dry season compared to the wet season. ARG profiles exhibited a pronounced distance-decay pattern in the dry season, whereas no such pattern was evident in the wet season. Null model analysis indicated that deterministic processes, in contrast to stochastic processes, had a significant impact on shaping the ARG profiles. Furthermore, it was found that Firmicutes and pH emerged as the foremost factors influencing these profiles. This study enhanced our comprehension of the variations in ARG profiles within canal networks, which may contribute to the design of efficient management approaches aimed at restraining the propagation of ARGs.

Unravelling the influence of cattle stocking rate on the macroinvertebrate community of freshwater wetlands subjected to hydrological modifications in three hydroclimatic periods.

The Paraná River Delta in South America, a large wetlands macromosaic, faces threats from climate change, human activities like livestock intensification, and hydrological modifications driven by the construction of water management infrastructure to prevent flooding in productive lands. Macroinvertebrates, essential for wetland health, are affected by cattle-induced changes in water quality, nutrient enrichment, and trampling, posing challenges to the ecosystem's ecological balance and long-term survival of these organisms. In this study, we analyzed the impact of two categories of cattle stocking rates (low and high) on the taxonomic and functional structure of the aquatic macroinvertebrate community in freshwater marshes. In addition, we compare the influence of cattle stocking rate on macroinvertebrates in natural and modified freshwater marshes, and, finally, the effect of cattle stocking rate in three contrasting hydrometeorological periods: a drier, a humid, and an extreme drought period. Samplings were conducted in 16 freshwater habitats of the Lower Paraná River Delta, examining variables such as temperature, pH, dissolved oxygen, coliforms, and nutrient concentrations. Macroinvertebrates were collected and functional and taxonomic metrics were estimated. Statistical analyses, including ANOVA and Kruskal-Wallis tests, were conducted to evaluate the effects of cattle stocking rates, hydrological modifications, and hydrometeorological periods on macroinvertebrate metrics and environmental variables. RDA, PERMANOVA, and SIMPER analyses explored the relationships between assemblage composition and environmental factors. High stocking rate altered the community structure, modifying its composition and decreasing the density, taxonomic and functional richness. Moreover, hydrological alterations exacerbated these negative impacts of cattle overstocking in macroinvertebrates. Under severe drought conditions, only tolerant species can survive cattle overstocking conditions. Our findings provide relevant insight into the ecological risks associated with cattle overstocking in natural and modified freshwater marshes and underscore the need to control cattle stocking rates in extreme drought to avoid loss of ecological functions.

Comparison of the performance of SWAT and hybrid M5P tree models in rainfall-runoff simulation.

Stream flow forecasting is a crucial aspect of hydrology and water resource management. This study explores stream flow forecasting using two distinct models: the Soil and Water Assessment Tool (SWAT) and a hybrid M5P model tree. The research specifically targets the daily stream flow predictions at the MH Halli gauge stations, located along the Hemvati River in Karnataka, India. A 14-year dataset spanning from 2003 to 2017 is divided into two subsets for model calibration and validation. The SWAT model's performance is evaluated by comparing its predictions to observed stream flow data. Residual time series values resulting from this comparison are then resolved using the M5P model tree. The findings reveal that the hybrid M5P tree model surpasses the SWAT model in terms of various evaluation metrics, including root-mean-square error, coefficient of determination (R2), Nash-Sutcliffe efficiency, and degree of agreement (d) for the MH Halli stations. In conclusion, this study shows the effectiveness of the hybrid M5P tree model in stream flow forecasting. The research contributes valuable insights into improved water resource management and underscores the importance of selecting appropriate models based on their performance and suitability for specific hydrological forecasting tasks.

Parameter Estimation Using the Inverse Problem Method for Simulating Lateral Inflow and Runoff Depth in a small catchment of Amazon.

The inverse problem method can be applied to determine the properties of hydrological phenomena and estimate the parameters, which cannot be measured directly. This type of inverse focus can facilitate the implementation of the kinematic wave model (direct model-DM), to fill gaps for lateral inflow rate and runoff depth in watersheds. Thus, the goal of the study was the application of the inverse problem method (IP). The lateral inflow rate was generally obtained as a Fourier transform to represent any watersheds. The study was developed using a small catchment in the Amazon where intense rainfall events occur, producing runoff and sediments, which affect rural populations. Lateral inflow rate and runoff depth were derived using precipitation data and parameters estimated through the KINEROS2 (K2)/direct model (DM) model and the ensuing solution methods with MCMC (Markov chains Monte Carlo)/Fourier transform. The developed method was applied to four rainfall-runoff events, leading to a good fit between the observed and predicted data (Nash-Sutcliffe coefficients between 0.76 and 0.85 and RMSE values between 1.80 mm and 6.72 mm).

Assessment of monthly runoff simulations based on a physics-informed machine learning framework: The effect of intermediate variables in its construction.

Hydrological forecasting is of great importance for water resources management and planning, especially given the increasing occurrence of extreme events such as floods and droughts. The physics-informed machine learning (PIML) models effectively integrate conceptual hydrologic models with machine learning (ML) models. In this process, the intermediate variables of PIML models serve as bridges between inputs and outputs, while the impact of intermediate variables on the performance of PIML models remains unclear. To fill this knowledge gap, this study aims to encompass the construction of PIML models based on various hydrologic models, conduct comparative analyses of different intermediate variables based on a case study of 205 CAMELS basins, and further explore the relationship between the performance of PIML models and catchment characteristics. The optimal ML model for constructing PIML is first selected among four ML models within the 205 basins. The PIML models are then developed based on five monthly water balance models, namely TM, XM, MEP, SLM, and TVGM. To quantify the potential impact of difference in intermediate variables, two sets of experiments are further designed and performed, namely S1 with actual evapotranspiration as the intermediate variable and S2 with soil moisture as the intermediate variable. Results show that five PIML models generally outperformed the optimal standalone ML models, i.e., the Lasso model. Specifically, regardless of the choice of intermediate variables, the PIML-XM model consistently outperformed the other models within the same basins. Almost all constructed PIML models are affected by the intermediate variables in monthly runoff simulations. Typically, S1 exhibited better performance compared to S2. A greater impact of aridity index, forest fraction, and catchment area on model performance is observed in S2. These findings improve our understanding of constructing PIML models in hydrology by emphasizing their excellent performance in runoff simulations and highlighting the importance of intermediate variables.

Spatiotemporal convolutional long short-term memory for regional streamflow predictions.

Rainfall-runoff (RR) modelling is a challenging task in hydrology, especially at the regional scale. This work presents an approach to simultaneously predict daily streamflow in 86 catchments across the US using a sequential CNN-LSTM deep learning architecture. The model effectively incorporates both spatial and temporal information, leveraging the CNN to encode spatial patterns and the LSTM to learn their temporal relations. For training, a year-long spatially distributed input with precipitation, maximum temperature, and minimum temperature for each day was used to predict one-day streamflow. The trained CNN-LSTM model was further fine-tuned for three local sub-clusters of the 86 stations, assessing the significance of fine-tuning in model performance. The CNN-LSTM model, post fine-tuning, exhibited strong predictive capabilities with a median Nash-Sutcliffe efficiency (NSE) of 0.62 over the test period. Remarkably, 65% of the 86 stations achieved NSE values greater than 0.6. The performance of the model was also compared to different deep learning models trained using a similar setup (CNN, LSTM, ANN). An LSTM model was also developed and trained individually to predict for each of the stations using local data. The CNN-LSTM model outperformed all the models which was trained regionally, and achieved a comparable performance to the local LSTM model. Fine-tuning improved the performance of all models during the test period. The results highlight the potential of the CNN-LSTM approach for regional RR modelling by effectively capturing complex spatiotemporal patterns inherent in the RR process.

Assessment of land use/ land cover change derived catchment hydrologic response: An integrated parsimonious hydrological modeling and alteration analysis based approach.

Land use/land cover (LULC) change, often a consequence of natural or anthropogenic drivers, plays a decisive role in governing global catchment dynamics, and subsequent impact on regional hydrology. Insight into the complex relationship between the drivers of LULC change and catchment hydrology is of utmost importance to decision makers. Contemplating the dynamic rainfall-runoff response of the Indian catchments, this study proposes an integrated modeling-based approach to identify the drivers and relative contribution to catchment hydrology. The proposed approach was evaluated in the tropical climate Nagavali River Basin (NRB) (9512 km2) of India. The Soil and Water Assessment Tool (SWAT) hydrological model, which uses daily-scale rainfall, temperature, wind speed, relative humidity, solar radiation, and streamflow information was integrated with the Indicators of Hydrologic Alteration (IHA) technique to characterize the plausible changes in the flow regime of the NRB. Subsequently, the Partial Least Squares Regression (PLSR) based modeling analysis was performed to quantify the relative contribution of individual LULC components on the catchment water balance. The outcomes of the study revealed that forest land has been significantly converted to agricultural land (45-59%) across the NRB resulting in mean annual streamflow increase of 3.57 m3/s during the monsoon season. The affinity between land use class and streamflow revealed that barren land (CN = 83-87) exhibits the maximum positive response to streamflow followed by the built-up land (CN = 89-91) and fallow land (CN = 88-93). The period 1985-1995 experienced an increased ET scenario (911-1050 mm), while the recent period (2005-2020) experienced reduced ET scenario owing to conversion of forest to agricultural land. Certainly, the study endorses adopting the developed methodology for understanding the complex land use and catchment-scale hydrologic interactions across global-scales for early watershed management planning.

Enhanced SWAT calibration through intelligent range-based parameter optimization.

Hydrological models are vital tools in environmental management. Weaknesses in model robustness for hydrological parameters transfer uncertainties to the model outputs. For streamflow, the optimized parameters are the primary source of uncertainty. A reliable calibration approach that reduces prediction uncertainty in model simulations is crucial for enhancing model robustness and reliability. The optimization of parameter ranges is a key aspect of parameter calibration, yet there is a lack of literature addressing the optimization of parameter ranges in hydrological models. In this paper, we introduce a parameter calibration strategy that applies a clustering technique, specifically the Self-Organizing Map (SM), to intelligently navigate the parameter space during the calibration of the Soil and Water Assessment Tool (SWAT) model for monthly streamflow simulation in the Baishan Basin, Jilin Province, China. We selected the representative algorithm, the Sequential Uncertainty Fitting version 2 (SUFI-2), from the commonly used SWAT Calibration and Uncertainty Programs for comparison. We developed three schemes: SUFI-2, SUFI-2-Narrowing Down (SUFI-2-ND), and SM. Multiple diagnostic error metrics were used to compare simulation accuracy and prediction uncertainty. Among all schemes, SM outperformed the others in describing watershed streamflow, particularly excelling in the simulation of spring snowmelt runoff (baseflow period). Additionally, the prediction uncertainty was effectively controlled, demonstrating the SM's adaptability and reliability in the interval optimization process. This provides managers with more credible prediction results, highlighting its potential as a valuable calibration tool in hydrological modeling.

A novel longitudinal connectivity index to evaluate reticular river networks based on the combination of network maximum flow and resistance distance.

Reticular river networks, essential for ecosystems and hydrology, pose challenges in assessing longitudinal connectivity due to complex multi-path structures and variable flows, exacerbated by human-made infrastructures like sluices. Existing tools inadequately track water flow's spatiotemporal changes, highlighting the need for targeted methods to gauge connectivity within complex river network systems. The Hydraulic Capacity Connectivity Index (HCCI) was developed adopting complex network theory. This involves river networks mapping, nodes and edges construstion, weight factor definition, maximum flow and resistance distance calculation. The connectivity between nodes is represented by the product of the maximum flow and the inverse of the resistance distance. The mean connectivity of each node with all other nodes, denoted as the node connectivity capacity Ci, and the HCCI of the whole river network is defined as the mean of the Ci for all nodes. The HCCI was firstly applied to a symmetrical virtual river network to investigate the factors influencing the HCCI. The results revealed that Ci showed a radial decreasing pattern from the obstructed river reach outwards, and the boundary rivers play the most significant role in regulating the flow dynamics. Subsequently, the HCCI was applied to a real river network in the Yandu district, followed by spatiotemporal statistical analysis comparing with 1D hydraulic model's simulated river discharge. Results showed a high correlation (Pearson coefficient of 0.89) between the HCCI and monthly average river discharge at the global scale. At the local scale, the geographically weighted regression model demonstrated the strong explanatory power of Ci in predicting the distribution of river reach discharge. This suggests that the HCCI addresses multi-path connectivity assessment challenge in reticular river networks, precisely characterizing spatiotemporal flow dynamics. Furthermore, since HCCI is based on a complex network model that can calculate the connectivity between all river node pairs, it is theoretically applicable to other types of river networks, such as dendritic river networks. By identifying low-connectivity areas, HCCI can guide managers in developing scientifically sound and effective strategies for restoring river network hydrodynamics. This can help prevent water stagnation and degradation of water quality, which is beneficial for environmental protection and water resource management.

A unified deep learning framework for water quality prediction based on time-frequency feature extraction and data feature enhancement.

Deep learning methods exhibited significant advantages in mapping highly nonlinear relationships with acceptable computational speed, and have been widely used to predict water quality. However, various model selection and construction methods resulted in differences in prediction accuracy and performance. Hence, a unified deep learning framework for water quality prediction was established in the paper, including data processing module, feature enhancement module, and data prediction module. In the established model, the data processing module based on wavelet transform method was applied to decomposing complex nonlinear meteorology, hydrology, and water quality data into multiple frequency domain signals for extracting self characteristics of data cyclic and fluctuations. The feature enhancement module based on Informer Encoder was used to enhance feature encoding of time series data in different frequency domains to discover global time dependent features of variables. Finally, the data prediction module based on the stacked bidirectional long and short term memory network (SBiLSTM) method was employed to strengthen the local correlation of feature sequences and predict the water quality. The established model framework was applied in Lijiang River in Guilin, China. The maximum relative errors between the predicted and observed values for dissolved oxygen (DO), chemical oxygen demand (CODMn) were 12.4% and 20.7%, suggesting a satisfactory prediction performance of the established model. The validation results showed that the established model was superior to all other models in terms of prediction accuracy with RMSE values 0.329, 0.121, MAE values 0.217, 0.057, SMAPE values 0.022, 0.063 for DO and CODMn, respectively. Ablation tests confirmed the necessity and rationality of each module for the established model framework. The established method provided a unified deep learning framework for water quality prediction.

Evolution of a tidal channel network in the Yellow River Delta, China, and simulation of optimization scenarios.

Tidal channel networks, which characterize all river deltas, control the exchange of water and nutrients (hydrological connectivity) between the ocean and the delta area. Therefore, a tidal channel network in optimal conditions ensures the maintenance of the diversity and stability of the deltaic ecosystem. However, the developmental status of channel networks in the Yellow River Delta, China, has not been clearly determined. Here, we selected a typical tidal channel network in this delta that showed different spatial patterns (e.g., connectivity attributes) in the past three decades and explored its evolution using entropy as an index of connectivity. Seven scenarios were set up to determine the optimal status of the tidal channel network by optimizing its structure. The optimization effect was evaluated by comparing the connectivity attributes of the channel network before and after optimization. The results showed that the network experienced two obviously different developmental phases: an evolution before 2005 and a regression after 2005. Mann-Kendall analysis indicated that the channel network achieved dynamic stability before 2014 and became unstable thereafter. The simulations conducted to optimize the system showed that adding outlets changed the current patterns of the network' structural and functional connectivity. As the optimization proceeded, structural connectivity increased while functional connectivity decreased, and the tidal channel network tended to be dynamically stable. Our study elucidated the quantitative relationship between outlet number and stability within tidal channel networks, providing reference information that could be incorporated into future projects for the restoration and management of river deltas.

Coupling machine learning and physical modelling for predicting runoff at catchment scale.

In this paper, we present an approach that combines data-driven and physical modelling for predicting the runoff occurrence and volume at catchment scale. With that aim, we first estimated the runoff volume from recorded storms aided by the Green-Ampt infiltration model. Then, we used machine learning algorithms, namely LightGBM (LGBM) and Deep Neural Network (DNN), to predict the outputs of the physical model fed on a set of atmospheric variables (relative humidity, temperature, atmospheric pressure, and wind velocity) collected before or immediately after the beginning of the storm. Results for a small urban catchment in Madrid show DNN performed better in predicting the runoff occurrence and volume. Moreover, enriching the input primary atmospheric variables with auxiliary variables (e.g., storm intensity data recorded during the first hour, or rain volume and intensity estimates obtained from auxiliary regression methods) largely increased the model performance. We show in this manuscript data-driven algorithms shaped by physical criteria can be successfully generated by allowing the data-driven algorithm learn from the output of physical models. It represents a novel approach for physics-informed data-driven algorithms shifting from common practices in hydrological modelling through machine learning.