Master Recession Curves (MRCs) exhibiting apparent semi-logarithmic linearity are commonly interpreted as evidence of homogeneous linear-reservoir behaviour. However, physically based groundwater theory predicts that recession dynamics may become nonlinear as storage, transmissivity, and hydraulic connectivity evolve during catchment depletion. Whether apparent MRC linearity represents a structurally invariant catchment property or an emergent signature of aggregated recession processes remains unresolved, particularly in tropical island catchments characterised by steep terrain, rapid runoff generation, and heterogeneous subsurface flow pathways. This study addresses this question using 217 streamflow recession events from five tropical monsoonal catchments on Ambon Island, eastern Indonesia. A hierarchical multi-diagnostic framework integrated global exponential recession modelling, hyperbolic recession analysis, phase-specific recession-factor estimation, residual diagnostics, continuous segmented regression, information-theoretic model comparison, and uncertainty-aware interpretation of the Recession Steepness Index RSI=−lnk. Although global exponential models consistently achieved excellent descriptive performance (R2 = 0.979–0.987), all catchments exhibited systematic phase dependence kearly<klate, temporally organised residual curvature, and substantially stronger support for segmented models. Statistically supported breakpoints clustered between approximately 7 and 10 days, indicating reproducible transitions in effective recession behaviour. These findings demonstrate that high global semi-logarithmic goodness of fit may conceal internally evolving depletion dynamics and that a single recession constant does not necessarily provide a structurally complete representation of catchment recession. Because groundwater observations, transmissivity measurements, tracer evidence, evapotranspiration estimates, and coastal boundary effects were not independently quantified, the inferred mechanisms remain process-consistent hypotheses rather than direct hydrogeological verification. Overall, this study provides a reproducible, uncertainty-aware framework that explicitly distinguishes descriptive model performance, structural recession behaviour, and physical process interpretation, thereby strengthening the interpretation of low-flow persistence and storage–discharge dynamics in tropical island catchments.
Incorporating human influence, particularly reservoirs, into Land Surface Models (LSMs) is critical for improving hydrologic predictions. As incorporating basin-specific reservoir models – typically simulation-optimization models – into LSMs is challenging, efforts have focused on deriving data-driven generalizing reservoir operational policies over a larger domain. This study derives reservoir operational groups/policies from the Piece-wise Linear Regression Trees (PLRT) model classification obtained using observed inflows, storage and releases from 76 reservoirs from four major river basins – Colorado, Columbia, Missouri and Tennessee – over the coterminous US (CONUS). Using this dataset, PLRT classifies the 76 reservoirs into two broader groups – Run-of-the-River systems (ROR) and storage dam systems – into five different groups. ROR having residence time less than 10 days are further classified into two groups based on reservoir storage considering 124,900 acre-ft small ROR and large ROR. Storage dam groups are further classified into operations based on small, medium and large storage systems to develop piece-wise regression based on inflows, past releases and past storages for each group. Simulated storage and release from the PLRT trees also match well with the observed counterparts. Given the PLRT rules mimic the actual operation over four major basins into five groups based on reservoir characteristics and data, this provides a simple, but generalized and effective approach to incorporate the reservoir operation into LSMs to estimate downstream releases.
Precipitation extremes, fundamental in hydrologic design and risk assessment, are typically estimated through purely statistical methods. Knowledge and integration of the physical characteristics of the generating storms are limited, although they strongly influence hydro-geomorphological response to extremes.A storm-based framework is proposed to explore whether and how the spatial variability of extreme precipitation statistics reflects the spatial variability of storm characteristics. The study combines a non-asymptotic extreme value approach with storm characterization based on 22 physical descriptors, leveraging high-resolution precipitation and temperature data from about 400 stations in the Greater Alpine Region.The spatial organization of extreme precipitation statistics is found to be duration-dependent and structured by physical storm characteristics: their relationships strongly reorganize from sub-hourly to daily durations and reveals partial decoupling between scale and tail heaviness of the distribution. Intermediate durations emerge as a transitional regime, likely linked to the shift from convective- to stratiform-dominated extremes. Although the spatial variability of the statistical parameters requires multiple physical descriptors, for extreme intensities (return levels) it's captured by a simpler underlying structure.This framework is easily transferable in other regions and provides a promising pathway to identify, interpret and apply process-based proxies for improving traditional approaches and modelling of extremes.
Heatwaves in particularly tropical and sub-tropical countries impact public health, resilience, and sustainable development; hence, accurate forecasting of heatwaves is of great importance. Sub-seasonal to seasonal (S2S) forecasts offer vital insights into weather patterns weeks in advance, supporting proactive mitigation, especially in heatwave-prone countries like India. Indian heatwaves are associated with extended dry spells during the pre- and post-monsoon periods and are influenced by large-scale climate variability modes. While S2S temperature forecasts generally outperform precipitation forecasts in short lead times, their predictive skill remains limited for heatwave warnings. To overcome these limitations, this study introduces an Adaptive & Aggregated Intelligence (AAI) framework to enhance the operational S2S-based heatwave prediction across India. The AAI framework post-processes raw S2S (real-time sets) outputs, significantly improving temperature anomaly predictions associated with heatwaves. It achieves a probability of detection exceeding 50%, compared to near-zero in unprocessed forecasts, and reduces the false alarm ratio from 1.0 to 0.3, corresponding to approximately 70% confidence in event occurrence. Furthermore, the framework's capability to extend lead times up to six weeks further enhances its operational value. These results demonstrate the effectiveness of AI-driven post-processing in improving S2S forecast's reliability. This framework is particularly useful for sectors like agriculture, water management, and civic awareness.
Tropical cyclones over land do not simply decay; they reorganize and interact with orography, producing extreme rainfall and flooding far from the center of the storm. In the Appalachian region, orographic enhancement results in areas with extreme rainfall amounts, leading to precipitation and flood peak distributions that do not exhibit evidence of an upper bound based on available observations and statistical analyses, although the existence of a physical upper limit cannot be ruled out. Here we use an empirical approach revolving around Hurricane Helene and long-term records to support the notion that design practices anchored in upper bounds (e.g., Probable Maximum Precipitation) are increasingly misaligned with observed extremes and projected risks. We argue for probabilistic frameworks that explicitly account for extremely low annual exceedance probabilities, mechanism-dependent tail behavior, and uncertainty beyond historical records. By integrating statistics reflective of the lack of an upper bound, high-resolution modeling, and by accounting for different processes in flood frequency analyses, the meteorological and hydrologic community can better inform resilient infrastructure and risk communication in an era of accelerating climate extremes.
Water scarcity is a problem that is likely to become more prevalent in many parts of the world in the near future. Changing water pricing schemes to reduce households’ water demand could be one policy to address the problem. Given households’ price-inelastic water demand, such an approach would likely require substantial price increases to reduce water demand. We propose an alternative pricing scheme that issues water-use rights (permits) to households at prices that cover a water supplier’s costs. These permits can be traded at the end of the year: households with excess permits are redeemed, while households in need of additional permits must purchase them. We show that, in such a setting, households’ opportunity cost considerations can curtail water consumption and yield lower water bills than a policy that increases volumetric rates.
The Budyko framework is widely used to infer long-term water-energy partitioning. Its original uniform parameters limited accuracy, while locally calibrated formulations are difficult to apply in ungauged basins. This study develops a machine-learning (ML)-based parameterization linking Budyko parameters to catchment attributes, enabling annual estimates of evapotranspiration (ET) and streamflow in gauged and ungauged basins. Using 671 CAMELS catchments, we optimized parameters for six Budyko-type equations with the Shuffled Complex Evolution (SCE) algorithm, then adjusted them to catchment attributes using a ML model. The resulting attribute-based parameterization improved evaporative-index estimates relative to both SCE-based calibration and a calibrated SAC-SMA model. Applied to test catchments, it achieved a mean KGE of 0.67 for annual streamflow, outperforming calibrated SAC-SMA (0.48) and site-specific Budyko predictions (0.61). Results highlight a scalable approach for annual water-budget estimation in data-limited regions.
With the ever-ongoing debate over the death of stationarity in flood time series, developing methods for flood frequency analysis that can address both stationarity and non-stationarity simultaneously has become increasingly important for design flood estimation. Existing non-stationary flood frequency analysis (NFFQ) methods are either limited to providing time-varying conditional flood quantile estimates or lack closed-form expression to estimate the design flood over the planning period. We propose a novel framework, MM-NFFQ, that introduces marginal moments (MM) estimation techniques to provide a closed-form expression to estimate design flood under non-stationarity. We demonstrate MM-NFFQ using the LP3 distribution for estimating conditional moments, but in principle, it can work for any 3-parameter distribution. We first show the proposed MM-NFFQ collapses to stationary flood frequency analysis analytically using synthetic data and then demonstrate the MM-NFFQ approach for two basins exhibiting non-stationarity in their flood time series. We further extend the analysis to selected 40 basins across CONUS and find that arid basins exhibit higher deviation from stationarity. Thus, the proposed MM-NFFQ framework can estimate traditional flood frequency curves for both, stationary and non-stationary flood processes, and can also be utilized to analyze the changes in conditional moments and marginal moments over different planning horizons.
The development of AI models is increasing at a rapid rate. However, when are they ready to be deployed in real-world operational settings? In this paper, we introduce a framework to support such assessments and apply it to Google’s recently released AI-based flood prediction system, which is claimed to achieve “reliability in predicting extreme riverine events” and provide “accurate and timely warnings” that are available “earlier and over larger and more impactful events in ungauged basins”. The system has been integrated into an operational early-warning platform producing open, real-time forecasts in more than 80 countries. While this development promises to usher in a new and exciting age in global flood forecasting, the supporting evidence relies heavily on several subjective choices, the implications of which have not been acknowledged or assessed. Here, we evaluate the consequences of these choices on claims of operational deployment readiness across four dimensions: predictive accuracy, forecast timeliness, the characterization of extreme events, and benchmarking against state-of-the-art models. Our assessment reveals that the system’s actual predictive accuracy is likely to be substantially lower than reported—particularly for extreme events—raising concerns about responsible practices across modelling and publicity in high-stakes applications. The deployment of the Google AI model therefore risks misinforming those who depend on its outputs for evacuation and preparedness decisions, particularly in less-developed countries such as those targeted by the enterprise, given its alarmingly high (>90%) rates of false positives and false negatives. Beyond the immediate operational consequences, if left unaddressed, these outcomes may erode public trust in AI within hydrological sciences. We conclude by calling for greater transparency, accountability, and methodological rigor in the integration of AI into flood forecasting.
Understanding the effects of land use changes on hydrological cycling is essential for managing water resources. While deforestation is a known driver of hydrological regime shifts, few studies have assessed its impacts in the steep landscapes of the Brazilian Amazon-Cerrado transition, a region experiencing rapid deforestation and agricultural expansion. Here, we evaluate streamflow responses to land use and land cover changes using three years of field monitoring across eight watersheds (2.5 – 39 km2) in eastern Mato Grosso, Brazil. Our study watersheds spanned a gradient of slopes (2% – 8.6%) and native vegetation cover (10% – 80%). Annual and daily flows were consistently higher in deforested watersheds, and these areas also exhibited greater flow seasonality and stormflow peaks. Notably, deforested watersheds experienced reduced baseflow during the dry season, increasing water scarcity at a critical period of the year. These findings highlight the importance of considering terrain slope and seasonal dynamics, as analyses at annual scales may mask the dual impacts of deforestation – increased flood risk during the rainy season and exacerbated water scarcity during the dry season. Integrating slope, land cover, and multiyear data is therefore essential to support more effective water resource management in rapidly changing tropical landscapes.
The proof-of-concept study presents, for the first time, the results of integrated numerical hydrological simulations at kilometer resolution on a global scale. Using available datasets and applying significant simplifications to the real terrestrial system, the model was informed with hydrofacies, soil texture and topographic slopes, and effective recharge at the upper boundary. A steady-state spin-up was performed, resulting in a 3D pressure head distribution of the water continuum from 60 m deep variably saturated groundwater to surface water. Relative saturation and diagnostic water table depth were examined, resolving variability over several orders of magnitude. In our opinion, the added value of the partial differential equation (PDE) based simulations outweighs the computational resources required, which are considerable. These simulations are possible, because of the advent of massively parallel, accelerator based supercomputer architectures and performance portable scientific software. While the current simulation results may not be reliable from the perspective of stakeholders at this stage of model development, the study demonstrates the feasibility of prognostic groundwater simulation at the global scale, and will stimulate future model improvements, including the quantification of uncertainties. Simultaneously, the study opens new avenues for future research in the context of hyper-resolution global Earth system modelling.
Streamflow shifts threaten water security under climate change. While gauged basins have been extensively studied, ungauged regions remain poorly characterized. Using high- and low-flow trends from 2,213 unregulated catchments, we predict four types of streamflow changes globally. From 1982 to 2018, 81.1% of studied regions exhibited coherent trends: 49.3% showed wetting (increased high and low flows), and 31.8% showed drying (decreased flows). The remaining areas experienced diverging (6.2%) or equalizing (12.6%) changes. Wetting dominated arid, snow, and polar climate regions, whereas drying prevailed in equatorial and temperate zones. Streamflow trends aligned with precipitation changes in 58.3% of areas, while 39.6% were also driven by non-precipitation factors, highlighting complex hydroclimatic interactions. Ongoing shifts demand enhanced water management and hazard adaptation.
ChatGPT, a generative AI, is applied and compared to the PCSWMM hydrological model for modelling peak flow in a small watershed in the runoff period of April to September. A new approach for fuzzy mathematical representation of rainfall and peak-flow errors was developed to lead to a fuzzy based GPT model and fuzzy based PCSWMM model. This led to fuzzy output for both models and a more appropriate application of both models given data errors and large language model structure. Training and validation were conducted with an approximately 25/75 split of the data and again using a 75/25 data split. Evaluation metrics were used to compare model performance under the different data-split scenarios. Calibrated and validated PCSWMM outperformed GPT in the 25/75 data split but ChatGPT 4o mini's generation outperformed PCSWMM in the 75/25 split and with comparable validation metrics and an application that was less onerous than when using PCSWMM. The fuzzy-based error analysis showed that for both models, a fuzzy-based approach produced more interpretable and reasonable results than either original model. Moreover, the trade-off between coverage (uncertainty range) and precision for GPT-4o mini model's fuzzy output at high membership levels (x-cut) demonstrated enhanced predictive performance under data-scarce conditions.
Optimal management of water resources is challenging due to uncertainty in future conditions. One promising approach is to directly incorporate future uncertainty into objective function formulations of optimization problems, enabling system performance evaluation across multiple potential conditions. However, this creates additional uncertainties as both the choice of objective function formulation and the plausible future conditions included in optimization are subjective. Given the inherent uncertainty in plausible future conditions, it is highly unlikely that future conditions included in optimization can cover all conditions that might occur. Therefore, identifying objective function formulations that perform well regardless of future uncertainties is crucial; however, it has not been formally explored. In this study, the performance of different objective function formulations under both expected (i.e., similar to conditions used in optimization) and unexpected (i.e., vastly different from conditions used in optimization) future conditions is investigated using a real-world case study. Results reveal that percentile and expected-value-based formulations generally perform consistently under both expected and unexpected conditions, whereas extreme-case-based formulations can lead to highly variable results depending on the actual conditions that will be realized in the future. Finally, variance-based formulations offer the greatest consistency across all conditions but may lead to compromised performance under favorable conditions.
Reservoir storage flash droughts (RFDs), characterized by the rapid decline in reservoir storage, and conventional (long-term) reservoir storage droughts (RDs) impact water availability, hydropower generation, and agricultural activities. However, the mechanism and drivers of flash and conventional reservoir storage droughts in India remain unexplored. Using daily observations of reservoir storage, we identify RFDs and RDs in 81 major reservoirs in India during the 2000-2023 period. 46 out of 81 reservoirs are dominated by upstream climate as reservoir storage trends are driven by changes and variability in upstream precipitation, while the remaining 35 reservoirs are identified as human-dominating reservoirs. RFDs occur more frequently in human-dominating reservoirs than climate-dominating, especially in small reservoirs. About 70 % of RFDs in climate and humandominating reservoirs are caused by sudden release to meet increased water demands in the downstream regions. Additionally, upstream precipitation deficit and downstream water demand control RDs, while downstream water demands can solely drive RFDs. Unlike reservoir storage trends, reservoir storage droughts are mostly linked with downstream water demands. We highlight the role of climate and human interventions in reservoir storage/droughts in India.
The interdependence of crucial resources and the imperative for ensuring sustainability through integrated management approaches is underscored by the Water-Energy-Food (WEF) Nexus. The current study focuses on Alabama, Arkansas, Louisiana, Mississippi, and Tennessee in the Deep South USA to analyze the trade-offs and synergies in WEF Nexus. We propose an Integrated WEF Sustainability Index (IWSI) to provide a quantitative assessment of sustainability across these states. The IWSI is constructed by integrating standardized indicators across the water, energy, and food sectors, with weights derived from inter-sectoral economic interactions, to capture both trade-offs and synergies in a single composite score to provide an aggregated sustainability assessment. USA has an IWSI value of 1.62. Tennessee has an IWSI value of 2.34, characterized by efficient water utilization, substantial contributions from renewable sources, and robust agricultural productivity. Conversely, Louisiana and Arkansas encounter notable sustainability challenges, respectively, primarily attributable to low energy and water efficiency, reliance on fossil fuels, high emissions, and large water footprints. Arkansas demonstrates a significant water footprint in agriculture, well above the national average, highlighting its heavy reliance on irrigation. There is variation in hydropower conditions across states, with Tennessee leading in renewable energy use. The study underscores regional disparities in sustainability and emphasizes the need for tailored strategies to enhance resource efficiency and renewable energy adoption. A global assessment using datasets from the World Bank and Our World in Data highlights disparities across regions, providing insights into region-specific opportunities and challenges.
Worldwide, stormwater systems are increasingly stressed due to increased rainfall and runoff caused by climate change and urbanization. Traditional static strategies for addressing these challenges, including increasing infrastructure capacity, are often inadequate as they are not suited to dealing with large uncertainties. In contrast, adaptive strategies, such as smart real-time control (RTC), are suited to dealing with such uncertainties, as they are able to respond to future changes as they occur. However, existing RTC approaches are not truly adaptive, as they require information on future rainfall. In this paper, we modify an existing RTC approach that does not require such information so that it is able to match desired outflow hydrographs in the face of changing inflow hydrographs. The utility of the proposed Target Flow Control for Hydrographs (TFC-H) approach is demonstrated by simulating its ability to achieve desired target flow hydrographs for multiple future worlds of a simplified lot-scale system, in which peak flows increase from 7 % to 95 % and storm volumes increase from 25 % to 57 %. The results show that use of the TFC-H approach effectively maintains the desired target outflow hydrograph with less than 5 % error for this wide range of "future worlds". Importantly, unlike other RTC approaches, the TFC-H approach is able to adapt without any knowledge/predictions of future rainfall/inflow hydrographs. This clearly demonstrates the potential of the TFC-H approach to enable existing stormwater systems to adapt to future changes.
