Climate elasticity of streamflow represents a nondimensional measure of the sensitivity of streamflow to climatic factors. Estimation of such elasticities from observational records has become an important alternative to scenario-based methods of evaluating streamflow sensitivity to climate. Nearly all previous elasticity studies have used a definition of elasticity known as arc elasticity, which measures changes in streamflow about mean values of streamflow and climate. Using observational records in western U.S., our findings reveal that elasticity definitions based on power law models lead to both regional and basin specific estimates of elasticity which are physically more realistic than estimates based on arc elasticity. Evaluating the ability of arc and power law elasticity estimators in reproducing Dooge’s complementary relationship (DCR) between potential evapotranspiration and precipitation elasticities reveal that power law elasticities estimated from at-site, panel and hierarchical statistical models reproduce DCR, whereas corresponding estimators based on arc elasticity cannot reproduce DCR. Importantly, our regional elasticity formulations using either panel and/or hierarchical formulations led to estimates of both regional and basin specific estimates of elasticities, enabling and contrasting streamflow sensitivity to climate across both basins and regions.

Despite advances in hydrological Deep Learning (DL) models using Single Task Learning (STL), the intricate relationships among multiple hydrological components and model inputs might not be comprehensively encapsulated. This study employed a Long Short-Term Memory (LSTM) neural network and the CAMELS dataset to develop a Multi-Task Learning (MTL) model, predicting streamflow and evapotranspiration across multiple basins. An optimal multi-task loss weight ratio was determined manually during the validation phase for all 591 selected basins with streamflow data-gaps under 5%. During test period, MTL showed median Nash-Sutcliffe Efficiency predictions for streamflow and evapotranspiration at 0.69 and 0.92, consistent with two STL models. The MTL’s strength appeared when predicting the non-target variable, surface soil moisture, using probes derived from LSTM cell states—representative of the internal DL model workings. This prediction showed a median correlation coefficient of 0.90, surpassing the 0.88 and 0.89 achieved by the streamflow and evapotranspiration STL models, respectively. This outcome suggests that MTL models could reveal additional rules aligned with hydrological processes through the inherent correlations among multiple hydrological variables, thereby enhancing their reliability. We termed this as “variable synergy,” where MTL can simultaneously predict varied targets with comparable STL performance, augmented by its robust internal representation. Harnessing this, MTL promises enhanced predictions for high-cost observational variables and a comprehensive hydrological model.

In mountain rivers, sediment from landslides or debris flows can alluviate portions or even full reaches of bedrock channel beds, influencing bedrock river incision rates. Various landscape evolution models have been developed to account for the coevolution of alluvial cover and sediment-flux-dependent bedrock incision. Despite the commonality of their aims, one major difference between these models is the way they account for and conserve sediment. We combine two of the most widely used sediment conservation schemes, an Exner-type scheme and an erosion-deposition scheme, with the saltation-abrasion model for bedrock incision to simulate the coevolution of sediment transport and bedrock incision in a mixed bedrock-alluvial river. We compare models incorporating each of these schemes and perform numerical simulations to explore the transient evolution of bedrock incision rates in response to changes in sediment input. Our results show that the time required for bedrock incision rates to reach a time-invariant value in response to changes in sediment supply is over an order of magnitude faster using the Exner-type scheme than the erosion-deposition scheme. These different response times lead to significantly different time-averaged bedrock incision rates, particularly when the sediment supply is periodic. We explore the implications of different model predictions for modeling mixed bedrock-alluvial rivers where sediment is inevitably delivered to rivers episodically during specific tectonic and climatic events.

Climate projections suggest an increase in drought frequency and intensity in various places over the globe, one of them being Southern Europe, expected to become a hotspot. However, 2018 presented an anomaly with the emergence of a rare “water seesaw” phenomenon, leading to severe drought in Central and Northern Europe while Southern Europe experienced high humidity. This unexpected event resulted in significant agricultural disparities, emphasizing the influence of interannual variability. The commentary underscores the danger of overlooking short-term climate variability, vital for accurate adaptation planning, especially for vulnerable regions, when focusing solely on long-term trends. This case serves as a motivation for exploration of global atmospheric circulation changes, emphasizing the need for nuanced modeling approaches to grasp subtle complexities in climate predictions and considering short-term climate variability alongside long-term trends.

Tectonic and/or climatic perturbations can drive drainage adjustment. The capture events, significantly changing the river network topology, are the major events in river network evolution. While they could be identified through field observations and provenance analysis, reconstructing this evolution process and pinpointing the capture time remain challenging. Following a capture event, the steady-state elevation of the captor river will be much lower than that of the beheaded river. Then, the newly-formed drainage divide will migrate towards the beheaded river, a process also known as river-channel reversal. The migration of the newly-formed drainage divide provides a new perspective for identifying the reorganization of the river network. Here, we employ numerical modeling to reproduce the characteristic phenomena of drainage-divide migration following capture events and analyze the effects of different parameters on the migration rate. We find that (1) the migration of newly-formed drainage divides can last for tens of millions of years, with the migration rate decreasing exponentially over time; (2) larger captured area, higher uplift rate, and lower erosional coefficient, all of which cause a higher cross-divide difference in steady-state elevation, will cause higher migration rate of the newly-formed drainage divide. This insight was further applied to the Dadu-Anning and Yarlung-Yigong capture events. We predict the present Dadu-Anning drainage divide would further migrate ~65–92 km southward to reach a steady state in tens of millions of years. The Yarlung-Yigong capture event occurred in the early-middle Cenozoic, which implies that the late-Cenozoic increased exhumation rate is not related to the capture event.

A document by Yi Yang. Click on the document to view its contents.

Reliable nutrient load estimation of a reservoir is challenging due to inconsistent spatial extent and temporal frequency of water quality and quantity. This study aims to collect consistent spatial extent and temporal frequency of water depths and nitrate concentrations of a reservoir in South Korea using uncrewed surface vehicle (USV). In this study, reservoir nitrate loads were estimated using four methods to examine how spatial variation in water depth and nitrate concentrations affected load estimates. Based on dual measurements of water depth and nitrate concentration, reservoir nitrate loads across 30 sampling dates (0.7 million tons of fresh water on average) ranged from one to four tons. Results showed that a point measurement of water depths and nitrate concentrations can cause up to -17% of underestimation of nitrate loads, particularly after intense rainfall events. This study highlights potential opportunities and  challenges of the USV-based dual monitoring systems for water quality and quantity.

The emission potential, which represents the total leachable mass in landfill waste body, is hard to measure directly. Therefore we propose to quantify it by assimilating available measurements. The leachate production rate is influenced by the total water storage in the waste body, while both total chloride mass and total water storage in the waste body influence the chloride concentration in the leachate. Thus assimilating leachate volume and chloride concentration simultaneously will help quantify the uncertainties in emission potential. This study investigated the feasibility of using particle filter in a concentration-volume coupled travel time distribution model to estimate the emission potential. Leachate production rates and chloride concentrations were assimilated simultaneously by a weakly coupled data assimilation(WCDA) method. The time lag issue in the travel time distribution model was solved by adding a daily model error to cover layer states. The proposed method was tested in synthetic experiments firstly to investigate the performance. The results show that the uncertainties in chloride mass and waste body total water storage were quantified and reduced. The predictions of chloride concentrations were also improved.

Increasing wildfire frequency and severity in high-elevation seasonal snow zones presents a considerable water resource management challenge across the western U.S. Wildfires can affect snowpack accumulation and melt patterns, altering the quantity and timing of runoff. While prior research has shown that wildfire generally increases snow melt rates and advances snow disappearance dates, uncertainties remain regarding variations across complex terrain and the energy balance between burned and unburned areas. Utilizing multiple paired in-situ data sources within the 2020 Cameron Peak burn area during the 2021–2022 winter, we found no significant difference in peak snow water equivalent (SWE) magnitude between burned and unburned areas. However, the burned south aspect reached peak SWE 22 days earlier than burned north. During the ablation period, burned south melt rates were 71% greater than unburned south melt rates, whereas burned north melt rates were 94% greater than unburned north aspects. Snow disappeared 7 to 11 days earlier in burned areas than unburned areas. Net energy differences at the burned and unburned AWS sites were seasonally variable, with the burned area losing more energy during the winter but gaining significantly more energy during the spring. Net shortwave radiation was 56% greater at the burned area during the winter and 137% greater during the spring driving a ~60% greater cumulative net energy at the burned site during May. These findings emphasize the need for post-wildfire water resource planning that accounts for aspect-dependent differences in energy and mass balance to accurately predict snowpack storage and runoff timing.

Integrated hydrologic models can simulate coupled surface and subsurface processes but are computationally expensive to run at high resolutions over large domains. Here we develop a novel deep learning model to emulate continental-scale subsurface flows simulated by the integrated ParFlow-CLM model. We compare convolutional neural networks like ResNet and UNet run autoregressively against our novel architecture called the Forced SpatioTemporal RNN (FSTR). The FSTR model incorporates separate encoding of initial conditions, static parameters, and meteorological forcings, which are fused in a recurrent loop to produce spatiotemporal predictions of groundwater. We evaluate the model architectures on their ability to reproduce 4D pressure heads, water table depths, and surface soil moisture over the contiguous US at 1km resolution and daily time steps over the course of a full water year. The FSTR model shows superior performance to the baseline models, producing stable simulations that capture both seasonal and event-scale dynamics across a wide array of hydroclimatic regimes. The emulators provide over 1000x speedup compared to the original physical model, which will enable new capabilities like uncertainty quantification and data assimilation for integrated hydrologic modeling that were not previously possible. Our results demonstrate the promise of using specialized deep learning architectures like FSTR for emulating complex process-based models without sacrificing fidelity.

The permeability coefficient of seepage is an important factor that is widely used in various engineering fields. There are numerous issues that influence the permeability coefficient, among which the porosity, particle size, particle size distribution and Reynolds number are of great importance. In this paper, a C++ code based on the three-dimensional lattice Boltzmann method (LBM) was developed and used to investigate the effects of the abovementioned factors on the permeability coefficient. A multiple relaxation time (MRT) collision scheme of the LB equations was used in the simulation. Porous media were prepared using the random packing method. Laminar flow and turbulent flow were simulated separately for particle Reynolds numbers in a range from 0.001 to 3,000. It was proven that in addition to the influence of porosity and particle size distribution on the permeability, the influence of the Reynolds number was obvious and could not be ignored. As the Reynolds number increased, the permeability of porous media decreased gradually. Based on the numerical simulation results, a modified formula for the permeability coefficient is proposed.

Inverse problems are ubiquitous in hydrological modelling for parameter estimation, system understanding, sustainable water resources management, and the operation of digital twins. While statistical inversion is especially popular, its sampling-based nature often inhibits the inversion of computationally costly models, which has compromised the use of the Generalized Likelihood Uncertainty Estimation (GLUE) methodology, e.g., for spatially distributed (partial) differential equation based models. In this study we introduce multilevel GLUE (MLGLUE), which alleviates the computational burden of statistical inversion by utilizing a hierarchy of model resolutions. Inspired by multilevel Monte Carlo, most parameter samples are evaluated on lower levels with computationally cheap low-resolution models and only samples associated with a likelihood above a certain threshold are subsequently passed to higher levels with costly high-resolution models for evaluation. Inferences are made at the level of the highest-resolution model but substantial computational savings are achieved by discarding samples with low likelihood already on levels with low resolution and low computational cost. Two test problems demonstrate the similarity of inferred parameter posteriors and uncertainty estimates of MLGLUE and GLUE as well as increased computational efficiency. Findings are furthermore compared to inversion results from Markov-chain Monte Carlo (MCMC) and from multilevel delayed acceptance MCMC. The computation time of inversion of a groundwater flow model was decreases by ≈45% and ≈57% when using MLGLUE instead of conventional formulations of GLUE and MCMC, respectively.

The Horn of Africa region has frequently been affected by severe droughts and food crises over the last several decades, and this will increase under projected global-warming and socio-economic pathways. Therefore, exploring novel methods of increasing early warning capabilities is of vital importance to reducing food-insecurity risk. In this study, we present the XGBoost machine-learning model to predict food-security crises up to 12 months in advance. We used >20 datasets and the FEWS IPC current-situation estimates to train the machine-learning model. Food-security dynamics were captured effectively by the model up to three months in advance (R2 > 0.6). Specifically, we predicted 20% of crisis onsets in pastoral regions (n = 84) and 40% of crisis onsets in agro-pastoral regions (n = 23) with a 3-month lead time. We also compared our 8-month model predictions to the 8-month food-security outlooks produced by FEWS NET. Over a relatively short test period (2020–2022), results suggest the performance of our predictions is similar to FEWS NET for agro-pastoral and pastoral regions. However, our model is clearly less skilled in predicting food security for crop-farming regions than FEWS NET. With the well-established FEWS NET outlooks as a basis, this study highlights the potential for integrating machine-learning methods into operational systems like FEWS NET.

Causal loop diagrams (CLDs) based on expert and/or stakeholder inputs inform the quantitative structure of socio-hydrological models (SHMs). However, a systematic exploration of the sensitivity of CLDs and SHMs to different levels of stakeholder inputs is lacking. For a large multi-purpose reservoir in southern India, we explore this sensitivity by developing three CLDs that integrate reservoir water balance, groundwater pumping, and consumer water use patterns. CLD1 is a conventional water balance-based reservoir model, while CLD2 additionally incorporates the reservoir operatorâ\euro™s judgment and groundwater pumping. CLD3 further incorporates the adaptive behavior of water users by adjusting demands in response to long-term (5-year) droughts. The correlation between observed and simulated monthly reservoir storage (2000-2013) for SHM1, SHM2, and SHM3 is 0.57, 0.85, and 0.87, respectively. SHM3 also outperforms SHM1 and SHM2 in simulating the relative use of surface and groundwater for irrigation purposes in the command area of the reservoir. Simulated demand deficits, command area groundwater levels, and minimum environmental flow satisfaction downstream of the reservoir for 1968-2013 using the three models exhibit substantial differences. SHM1 and SHM2 simulate deteriorating groundwater levels under the multi-year drought of 2001-2003 while SHM3 does not due to the consideration of adaptive farmer behavior. Thus, our understanding of water and food security during a multi-year drought can be significantly affected by the level of stakeholder inputs incorporated in the models. We highlight the importance of testing different SHMs structures to better understand human-water interactions under extreme conditions.

A document by G. Aaron Alexander. Click on the document to view its contents.

Channel networks across fluvial landscapes are believed to have evolved to minimize energy expenditure[1–3], as evidenced by the similarities between computer-generated optimal channel networks (OCNs) and real networks[4,5]. However, the specific mechanisms driving energy minimization in fluvial landscapes remain largely elusive[6]. Here we propose that randomness has a profound role in landscape evolution[7] and that efficient channel networks emerge when the probability of a channel pixel changing its flow direction decreases with drainage area. The proposed probabilistic growth model then employs a power function to simulate channel-network evolution, with positive exponent (𝜂) values leading to asymptotic decrease of energy expenditure. An interpretation of this result is energy minimization tendency of river networks is a result of landscape evolution following specific adaptive rules rather than being the cause of landscape evolution itself. A greater 𝜂 ensures a greater restriction on the role of randomness and thus results in a more stable channel network configuration, and vice versa. Interestingly, the most efficient networks are observed to emerge always at 𝜂 =0.5, suggesting that randomness plays an important but limited role in the emergence of efficient channel networks. The proposed framework holds promise for explaining the evolution of other tree-like networks in nature and for developing more efficient optimization methods for practical applications.

A reliable understanding of linkages between meteorological, hydrological and agricultural droughts (MD, HD and AD respectively) is crucial to building resilience and planning for future climate changes. Despite Australia being prone to severe droughts, lagtimes of propagation (and recovery) from meteorological to hydrological and agricultural droughts across its large hydroclimatic regions are not well understood. Therefore, we investigate the characteristics of drought propagation and recovery time lags for droughts of four timescales and a combination of drought onset and cessation criteria in 407 unregulated catchments within six major precipitation zones across Australia. We find that the propagation and recovery lags are dependent on climatic conditions, drought criteria and timescales. The average propagation times from MD to HD across Australia varied from 0.8 to 1.7 months for monthly timescales, increasing to 2 to 4.5 months for 12-monthly timescales. The corresponding recovery lagtimes were 1.3 to 3.7 and 1.7 to 7.5 months respectively. Similarly, the average propagation times from MD to AD ranged from 0.9 to 1.9 months for monthly timescales, increasing to 0.8 to 5 months for 12-monthly timescales. The corresponding recovery lagtimes were 0.7 to 2.8 and 0.3 to 9.4 months respectively. For droughts of smaller timescales, propagation and recovery lags are linearly correlated with recovery lagtimes consistently greater than the propagation times. As the timescale increases, these relationships weaken suggesting effects of other catchment attributes (e.g. groundwater contributions) on lag relationships. Notably, recovery lagtimes are generally longer for the high-yielding catchments in eastern Australia compared to the other regions

This study presents a novel approach to improving the accuracy of flood forecast models with limited training data. Flood forecast information is crucial for early evacuation planning. However, the probability of flooding caused by continuous heavy rainfall is increasing, even in areas where we have not installed flood forecasts. New methods exist to provide flood forecasts, but they require long-term observations and regular updating of extensive data on the basin. Existing methods of providing new flood forecast information require long-term observations and regular updates of extensive data on the watershed. These requirements are related to the construction time and cost of providing flood forecasts. To address this issue, we propose Informed Neural Networks (INN) that integrate existing domain knowledge of river engineering to enhance the performance of flood forecasts with a limited amount of training data. We evaluated the performance of our proposed method with Japanese real-world river water levels and compared it to conventional methods such as artificial neural networks (ANN). Our results demonstrate that the INN can significantly improve forecasting accuracy with only a small amount of training data, comparable to conventional methods trained with eight times the amount of flood data. This study highlights the potential of INN as a novel approach for accurate and efficient flood forecasting with limited training data.

Streambed grain sizes and hydro-biogeochemistry (HBGC) control river functions. However, measuring their quantities, distributions, and uncertainties is challenging due to the diversity and heterogeneity of natural streams. This work presents a photo-driven, artificial intelligence (AI)-enabled, and theory-based workflow for extracting the quantities, distributions, and uncertainties of streambed grain sizes and HBGC parameters from photos. Specifically, we first trained You Only Look Once (YOLO), an object detection AI, using 11,977 grain labels from 36 photos collected from 9 different stream environments. We demonstrated its accuracy with a coefficient of determination of 0.98, a Nash–Sutcliffe efficiency of 0.98, and a mean absolute relative error of 6.65% in predicting the median grain size of 20 testing photos. The AI is then used to extract the grain size distributions and determine their characteristic grain sizes, including the 5th, 50th, and 84th percentiles, for 1,999 photos taken at 66 sites. With these percentiles, the quantities, distributions, and uncertainties of HBGC parameters are further derived using existing empirical formulas and our new uncertainty equations. From the data, the median grain size and HBGC parameters, including Manning’s coefficient, Darcy-Weisbach friction factor, interstitial velocity magnitude, and nitrate uptake velocity, are found to follow log-normal, normal, positively skewed, near log-normal, and negatively skewed distributions, respectively. Their most likely values are 6.63 cm, 0.0339 s·m-1/3, 0.18, 0.07 m/day, and 1.2 m/day, respectively. While their average uncertainty is 7.33%, 1.85%, 15.65%, 24.06%, and 13.88%, respectively. Major uncertainty sources in grain sizes and their subsequent impact on HBGC are further studied.

Water resources planning and management requires the estimation of extreme design events. Anticipated climate change is playing an increasingly prominent role in the planning and design of long-lived infrastructure, as changes to climate forcing are expected to alter the distribution of extremes in ways and to extents that are difficult to predict. One approach is to use climate projects to force hydrologic models, but this raises two challenges. First, global climate models general focus on much larger scales than are relevant to hydrologic design, and regional climate models that better capture small scale dynamics are too computationally expensive for large ensemble analyses. Second, hydrologic models systematically misrepresent the variance and higher moments of streamflow response to climate, resulting in a mischaracterization of the extreme flows of most interest. To address both issues, we propose a new framework for non- stationary risk-based hydrologic design that combines a stochastic weather generator (SWG) that accurately replicates basin-scale weather and a stochastic watershed model (SWM) that accurately represents the distribution of extreme flows. The joint SWG-SWM framework can generate large ensembles of future hydrologic simulations under varying climate conditions, from which design statistics and their uncertainty can be estimated. The SWG-SWM framework is demonstrated for the Squannacook River in the Northeast United States. Standard approaches to design flows, like the T-year flood, are difficult to interpret under non-stationarity, but the SWG-SWM simulations can readily be adapted to risk and reliability metrics which bare the same interpretation under stationary and non-stationary conditions. As an example, we provide an analysis comparing the use of risk and more traditional T-year design events, and conclude that risk-based metrics have the potential to reduce regret of over-and under-design compared to traditional return-period based analyses.

The land-lake interface is a unique zone where terrestrial and aquatic ecosystems meet, forming part of the Earth’s most geochemically and biologically active zones. The unique characteristics of this interface are yet to be properly understood due to the inherently high spatiotemporal variability of subsurface properties, which are difficult to capture with the traditional soil sampling methods. Geophysical methods offer non-invasive techniques to capture variabilities in soil properties at a high resolution across various spatiotemporal scales. We combined electromagnetic induction (EMI), electrical resistivity tomography (ERT), and ground penetrating radar (GPR) with data from soil cores and in-situ sensors to investigate hydrostratigraphic heterogeneities across land-lake interfaces along the western basin of Lake Erie. Our Apparent electrical conductivity (ECa) maps matched soil maps from a public database with the hydric soil units delineated as high conductivity zones (ECa > 40 mS/m) and also detected additional soil units that were missed in the traditional soil maps. This implies that electromagnetic induction (EMI) could be relied upon for non-invasive characterization of soils in sampling-restricted sites where only non-invasive measurements are feasible. Results from ERT and GPR are consistent with the surficial geology of the study area and revealed variation in the vertical silty-clay and till sequence down to 3.5 m depth. These results indicate that multiple geophysical methods can be used to extrapolate soil properties and map stratigraphic structures at land-lake interfaces, thereby providing the missing information required to improve the earth system model (ESM) of coastal interfaces.

Functional relationships capture how variables co-vary across specific spatial or temporal domains. However, these relationships often take complex forms beyond linear, and they may only hold for sub-sets of the domain. More problematically, it is often a priori unknown how such sub-domains are defined. Here we present a new method called SONAR (diScovery Of fuNctionaAl Relationships) that enables the automated discovery of functional relationships in large datasets. SONAR operates on existing unstructured data and is designed to be an explorative tool for large datasets where manual search for functional relationships would be impossible. We test the method on groundwater recharge outputs of several global hydrological models to explore its usefulness and limitations. Further, we compare SONAR to the established CART (Classification and Regression Trees) and CIT (Conditional Inference Trees) methods. SONAR results in smaller trees with functional relationships in the leaf nodes instead of specific classes or numbers. SONAR provides a robust and automated method for the exploration of functional relationships.

The Gotvand dam was built on the most important Iranian river to support a number of populated cities with freshwater, provide irrigation water for million hectares of fertile farmlands, and meet water demand for the country’s hub industrial zones. This dam is known as one of the worst engineering failures in Iran’s history because its impoundment submerged the enormous salty unit of Gachsaran evaporite formation (GEF) outcropped in the reservoir, leading to reservoir water salinization in deep layers up to several times greater than that of in the high-seas. Given the failed practical application of direct intervention strategies to control the salinity crisis, we suggested a low-cost salinity management strategy based on the reservoir operation to mitigate the dam outlet salinity and preserve the downstream environment from the salinity hazards. The three-dimensional MIKE3 model, was run to calculate the GEF dissolution rate, accumulated salt in the reservoir, and the dam outlet salinity. Then, we ran the model considering different outlet salinity levels to explore the best reservoir operation strategy to prohibit the accumulated salt in the reservoir and keep the safe salinity for downstream irrigation-use. Simulation results suggested that the GEF dissolution rate varied from 0.5 to 7 cm/hr, mainly due to incremental submergence of the GEF during multi-stage impoundment of the reservoir. Considering the final dissolution rate of 0.5 cm/hr and inlet salinity from the upstreams, salt accumulation inside the reservoir can be gradually prevented by setting the outlet salinity to its maximum historical downstream level, i.e., 1400 µmhos/cm.

Global flood mapping has developed rapidly over the past decade, but previous approaches have limited scope, function, and accuracy. These limitations restrict the applicability and fundamental science questions that can be answered with existing model frameworks. Harnessing recently available data and modelling methods, this paper presents a new global ~30 m resolution Global Flood Map (GFM) with complete coverage of fluvial, pluvial, and coastal perils, for any return period or climate scenario, including accounting for uncertainty. With an extensive compilation of global benchmark case studies – ranging from locally collected event water levels, to national inventories of engineering flood maps – we execute a comprehensive validation of the new GFM. For flood extent comparisons, we demonstrate that the GFM achieves a critical success index of ~0.75. In the more discriminatory tests of flood water levels, the GFM deviates from observations by ~0.6 m on average. Results indicating this level of global model fidelity are unprecedented in the literature. With an optimistic scenario of future warming (SSP1-2.6), we show end-of-century global flood hazard increases are limited to 9% (likely range -6–29%); this is within the likely climatological uncertainty of -8–12% in the current hazard estimate. In contrast, pessimistic scenario (SSP5-8.5) hazard changes emerge from the background noise in the 2040s, rising to a 49% (likely range of 7–109%) increase by 2100. This work verifies the fitness-for-purpose of this new-generation GFM for impact analyses with a variety of beneficial applications across policymaking, planning, and commercial risk assessment.