Study Region Major river basins in India. Study Focus Floods pose major risks to human lives and infrastructure, requiring short to long-term mitigation measures. Modeling floods under changing conditions needs a better understanding of hydro-meteorological drivers beyond extreme rainfall. The long-assumed link between annual maximum rainfall (AMR) and annual maximum floods (AMF) is not well understood, as AMF events are not always associated with the highest rainfall. To address this, we examine the relative role of rainfall, soil moisture, evapotranspiration, and baseflow in generating AMF across Indian river basins. We define three rainfall-streamflow cases: (1) AMR leading to AMF, (2) AMR producing flow lower than AMF, and (3) heavy rainfall (less than AMR) triggering AMF. Temporal composite analysis, Bayesian multi-linear regression, and Moran’s I spatial autocorrelation are employed to disentangle these dynamics. New Hydrological Insights for the Region Our results demonstrate that AMF is frequently caused by rainfall events smaller than AMR, underscoring the inadequacy of relying solely on maximum rainfall for flood estimation. Temporal analysis reveals that rainfall and baseflow are dominant drivers of high-magnitude AMF events, while spatial clustering reveals that floods are more likely when heavy rainfall coincides with low evapotranspiration and high soil moisture. These findings provide a novel synthesis of temporal and spatial drivers of AMF, emphasizing the need for flood risk assessments that incorporate multiple hydro-meteorological factors, especially under climate change.

Abstract Effective water resource management under climate change is dependent on reliable modeling of future water availability, yet significant uncertainties remain due to shifting precipitation patterns and limitation in climate modeling. This study investigates a multiple‐lines‐of‐evidence approach aimed at reducing the uncertainty in projections of streamflow, specifically focusing on extreme flood events as proxies for total annual flows. Using subdaily rainfall‐runoff models calibrated for four climatically diverse Australian catchments, we compare projections from regional climate model (RCM) downscaling with a continuous precipitation generation approach conditioned on stable climatic covariates such as temperature. Our findings suggest that in wetter regions with pronounced extreme precipitation events, using flood events as proxies for total annual flows can reduce variance and bias, potentially offering more reliable inputs for water resource planning. Both continuous simulation and RCM downscaling demonstrated similar biases for rainfall when evaluated against historical observations. However, continuous simulation typically produced lower biases in modeled streamflow, and performed marginally better in representing low‐frequency climate variability; whilst also offering reduced computational demands, presenting a parsimonious alternative for climate impact assessment. Despite the improved precipitation representation by RCMs, compared to General Circulation Models (GCMs), epistemic uncertainty and sampling biases persist, limiting the confidence in projections of extreme streamflow events for water supply modeling. These results highlight the need for improved modeling of precipitation extremes under warming climates to refine and enhance the robustness of future water resource projections.

Atmospheric Rivers (ARs) are key drivers of precipitation variability and extremes, yet their contribution to long-term precipitation trends remains poorly constrained. Using multiple AR detection algorithms combined with high-resolution precipitation data, we assess spatiotemporal variations in AR frequency and associated precipitation. We find that ARs account for up to 70–90% of interannual precipitation variability in the mid-latitudes. We also identify statistically significant increases in the annual frequency of ARs and the precipitation they produce. These trends are particularly strong in the eastern United States, central Brazil, western Africa, and southern Australia, and often more pronounced than the trends in total annual precipitation. Crucially, we find an intensification of AR-driven extreme precipitation events, indicating that the most impactful storms are becoming more frequent and severe due to ARs. Our findings underscore an increasing risk of AR-related hydrometeorological hazards, with implications for water resource management and climate adaptation strategies.

Abstract. Rainfall-runoff events are widely used in hydrological applications, from flood estimation and flood forecasting, to understanding catchment responses in a climate/anthropogenic affected world. The majority of methods used to identify rainfall-runoff events are statistical in nature, relying on subjective, user-defined ‘rules’ (i.e., parameters) that define a rainfall-runoff event. Since no ground-truth information is available to confirm the exact beginning and end of rainfall-runoff events, there is noticeable inconsistency (i.e. uncertainty) within the results. In this study, we propose the Robust Variance-based Event Identification Method (RVEIM), a new parsimonious rainfall-runoff event identification method which uses fewer parameters and better mimics the natural runoff generation process; decreasing the uncertainty in rainfall-runoff event identification. RVEIM detects runoff events by focusing on changes of streamflow variance and pairs to the corresponding rainfall event(s) simultaneously. RVEIM was compared to a benchmarking event identification method in 8 representative catchments in Australia. A sensitivity analysis was performed using a comprehensive set of plausible event identification parameter values for both methods. Results revealed that the variation of rainfall-runoff events characteristics – including annual number, length of runoff events, and the mean volume of runoff events – showed limited uncertainty from the RVEIM (standard deviation within ±15 % of the mean across 8 representative Australian catchments). This demonstrates robustness of RVEIM, while the benchmarking method exhibited considerably greater uncertainty in the event characteristics, with coefficients of variation exceeding 23 % and more than an order of magnitude difference across catchments. Using RVEIM, we present the first comprehensive summary of rainfall-runoff event characteristics across 467 Australian catchments. The distribution of event-scale runoff coefficients for individual catchments shows a strong climate gradient. Such systematic shifts in the distribution of runoff coefficient across climate regions indicate that climate variables play an important role in catchment response and point to potential contrasts in dominating runoff generation mechanisms across climate zones.

Calibrating conceptual rainfall-runoff models to predict a catchment's response to extreme rainfall events is a significant challenge, especially when historical data is limited and the event exceeds recorded extremes. This challenge is amplified in non-stationary environments where the frequency and intensity of extreme rainfall and floods are increasing. Traditional parameter estimation methods thus may not adequately capture the behavior of extreme events. To address this, we propose an Adaptive Markov Chain Monte Carlo (MCMC) algorithm that incorporates a model robustness-likelihood function; a novel approach in the context of hydrologic parameter sampling. Our two-stage sampling method first uses an automated scheme to identify the posterior distribution of parameter sets, then in the second stage, we sample robust parameter sets from this posterior using a likelihood function based on hypothetical (design) storm events. We demonstrate this approach using the GR4H model applied to a catchment in Victoria, Australia. The robust parameter sets selected through this process exhibit reduced variability across simulations, providing more reliable predictions for extreme events not observed in historical data. This methodology increases confidence in simulating hydrologic responses to unprecedented events. While use is made of probabilistic design storms in defining the robust parameter space, this rationale can be extended to incorporate other alternatives including site analogs, future climates or paleo reconstructions across the broad realm of hydrological applications.

Streamflow events are critical in providing water and ensuring the environmental health of our ecosystems. Despite intensifying rainfall events under climate change, streamflow events do not show similar increases, reflecting the complex interactions between rainfall, temperature, soil moisture, and catchment conditions. While it is well known that antecedent soil moisture modulates the streamflow response, recently it has been proposed that surrogate variables for antecedent moisture may also explain streamflow dynamics. Here, we investigate the drivers of large streamflow event volumes by building a Bayesian Hierarchical Model to predict large streamflow event volumes as a function of their drivers across Australia. Compared to the baseline model of only using event rainfall as a predictor, adding antecedent dry spells yielded little to no improvement in model performance. Adding soil moisture as a predictor, however, led to a significant improvement compared to the baseline, particularly across southeast and southwestern Australia. By assessing the relative importance of model predictors, we found rainfall to be the most important predictor overall, followed by soil moisture and antecedent dry spells. Our results show that while antecedent dry spells are not impactful in Australia for predicting large streamflow event volumes, accurate representation of antecedent soil moisture conditions in hydrological modelling is crucial for understanding future changes in streamflow event dynamics in Australia.

2025 featured geographically concentrated flood anomalies shaped by shifting large-scale atmospheric circulation and ocean–atmosphere variability, with antecedent hydrologic conditions and topography amplifying impacts in several regions. The year ranked in the lower tier of the past two decades for flood exposure, totalling over US$ 28 billion in damages and 4,200 flood-related fatalities globally.

Event-based rainfall-runoff models are commonly used to derive the magnitude of floods with a specified exceedance probability for infrastructure planning and design and flood management. However, the natural variability and joint interaction between key flood producing factors means it is difficult to reproduce flood frequency curves. This difficulty is in part due to the challenge of characterising the antecedent catchment wetness. Here, for the first time, we quantify the distributions of antecedent catchment wetness parameters prior to flood events. We examine the empirical distributions of initial and continuing loss values for 349 catchments across Australia. We find that the probability of how wet or dry a catchment is prior to a flood is governed by a non-dimensional distribution that is largely invariant across a wide range of hydroclimatic regimes. This is despite the catchments spanning a wide range of sizes and climatic regimes that include tropical, temperate and semi-arid climates. These findings are of considerable practical benefit as they demonstrate one needs to only specify a single median loss parameter to fully describe the distribution of losses that moderate flood responses, and this can be used to explicitly assess the joint probability of flood generating processes. We characterise the aleatory and epistemic sources of uncertainty that govern the loss distributions and demonstrate their incorporation in a Monte Carlo scheme deriving flood frequency curves. Representing losses in a stochastic manner is a conceptually attractive improvement over deterministic simulation schemes and is easily implemented with traditional event-based models commonly used by practitioners.

Atmospheric rivers (ARs) are narrow corridors of concentrated moisture known to distribute moisture globally and influence extreme precipitation events. However, the contribution of ARs to rare flood risk, the resultant hazard from extreme precipitation events, remains unquantified. Here, using 2686 largely regulation-free catchments across the world, we find that in some regions ARs are linked to over 70% of the largest precipitation and streamflow events. AR related precipitation significantly shortens the recurrence intervals of large flood events, making them 2–8 times more likely. In parts of Northern America, Europe and Australia, rare flood events are up to 12 times more likely when an AR is present. We conclude that ARs play a crucial role in increasing the frequency of extreme hydrological events on a global scale.

Much of the world's population faces significant threats to water security that are likely to be exacerbated under a warmer climate. To understand the impact of climate change on water security, we characterise the relationship between total annual streamflow, a surrogate for water supply, and low frequency flood events. We first calculate the proportion of annual streamflow attributable to flood events at approximately 5000 stations globally, then characterise this relationship as a function of climate and catchment characteristics. To explore the impact of climate changes we investigate trends in these relationships. We find that across the world on average 25% of annual streamflow comes from a single flood event and this proportion is well described by the variability in streamflow and precipitation, which is a function of catchment aridity. In arid to semi-arid catchments, where on average over 40% of annual streamflow comes from a single flood event, we conclude that water availability may be reasonably approximated by focusing on a selection of low frequency events, simplifying the projection of water supply changes under a future climate. Flood generating mechanism is crucial in determining trends of annual streamflow volumes from low frequency events. Where snowmelt is a significant flood process, increasing temperatures are causing a reduction in the proportion of total annual streamflows accounted for by flood events. Where snow is not a significant flood driver, trends are dominated by hydroclimatic variability and aridity, which is expected to be exacerbated with a shift to rainfall driven flooding under climate change.

When undertaking hydrologic modelling for design of infrastructure, storms of varying rainfall depths and durations are generated for event-based models by combining univariate distributions of rainfall depths and rainfall temporal patterns. Flood characteristics, especially peak flows, are sensitive to the temporal pattern of the causative rainfall, however there is limited research on how the temporal uniformity of these patterns might change with event severity. Here we investigate how the uniformity of rainfall temporal patterns varies with rainfall magnitude. This analysis was conducted for Australia, which spans a wide variety of meteorological conditions. Three country-wide datasets were used: sub-daily point rainfall observations, daily gridded rainfall observations, and sub-daily gridded rainfalls from modelled reanalyses. Using each of these datasets, storms of different durations were identified, and regression relationships were established between rainfall depths over the entire duration of the storm and rainfall depths that occurred within fixed subset periods embedded in the storm. Regression relationships were derived for each pixel in the two gridded datasets, and for each gauge in the sub-daily point observation dataset. It was found that rainfall bursts become temporally more uniform with increasing storm depth. This finding was consistently observed for different storm durations across each of the three datasets, though the strength of the relationship varied spatially across Australia. This relationship was also observed when considering the regional maximum rainfall bursts across meteorologically homogeneous zones, implying the relationship also holds for extreme storms. A numerical experiment was undertaken using synthetic data for a range of distributional characteristics, and the results obtained were found to be consistent with those based on observed data. This increase in temporal uniformity with storm depth highlights the need to vary design inputs with the storm magnitudes of interest.

Identifying rainfall-runoff events is routinely performed in many hydrologic applications. Absence of a groundbased truth makes rainfall-runoff event identification largely subjective. As a result, current algorithms often disagree on the start and end of events, leading to events within a given set of rainfall and runoff time-series with inconsistent properties - referred to hereafter as 'uncertainty in rainfall-runoff event identification'. In this study, the uncertainty associated with identifying rainfall-runoff events is assessed across Australia. A considerable uncertainty exists in the characteristics of identified rainfall-runoff events, including in their Runoff Coefficients (RCs). We propose a new objective metric to narrow the plausible set of parameters for identifying rainfall-runoff events. The metric demonstrates a substantial reduction in the uncertainty in rainfall-runoff event identification while improving the plausibility of the rainfall-runoff events chosen (up to a 25 % reduction in RCs >1) making the metric applicable for large-sample analyses of rainfall-runoff events.

Hydroclimatic variability at the catchment scale modulates spatiotemporal patterns of water availability, potentially inducing hydrological extremes such as flooding and drought. These events alter streamflow and pose significant challenges for water resources management, ultimately impacting local ecosystems and communities. To understand the changes in hydroclimatic variability we examine the patterns of rainfall intermittency using aggregated catchment average rainfall. 467 Hydrological Reference Stations (HRS) catchments are used with data spanning from 1950 to 2022 across the Australian continent. We investigate changes in intermittency characteristics such as spell duration, frequency and intensity at the annual and seasonal scale. There is a clear trend towards an increase in rainfall intermittency with an increase in the number of both wet and dry spells per year across Australia. Wet spells are becoming shorter across 80 % of catchments, with an increase in the number of dry days per year. Despite this increase in dry days, there are no robust trends for changes in dry spell length. Catchments with drying trends are typically in southern and eastern Australia, whilst the catchments in northern and northwestern Australia exhibit wetting trends. This wetting trend comes from fewer dry days and increases in both annual rainfall totals and rainfall intensity during wet spells. We find that the trends in the seasonal scale are regionally dependent and align with changes in the large-scale drivers of regional rainfall dynamics. In the south, winter rainfall and wet spells are the most impacted, whereas in the north, it is the summer monsoon that is most impacted by these trends. Our results show rainfall intermittency has increased in recent decades, suggesting that intermittency could potentially continue to change into the future. These results also highlight the need to investigate wet and dry spells concurrently to form a foundational understanding of how rainfall intermittency dynamics are changing. We conclude that changes in rainfall intermittency across Australian catchments have the potential to impact water resources management and need to be considered in future planning.

It is now well understood that anthropogenic induced global warming is increasing extreme rainfalls, with the more extreme the rainfall, the greater the intensification. This in turn is increasing the magnitude of rare floods. For floods with annual exceedance probabilities rarer than 1 in 20, the intensification of rainfall offsets any decreases in soil moisture. Less well understood, however, is the changing impact of spatial and temporal patterns of extreme rainfall on flooding under a warming climate. The spatial and temporal distribution of rainfalls during storm events has a significant influence on runoff volumes (and hence water availability) and on flood peaks. Hence, robust datasets are required to model hydrologic risk with changes in storm spatial and temporal patterns.To this end, we have developed Australia-Wide Extreme Storms Database (or AWESD) which characterises storm patterns with high spatial (12km) and temporal (hourly) resolution for hydrologic risk assessments. Whilst the record length for such high-resolution data is currently 30 years, the availability of information at a high resolution over large homogeneous regions allows the trading of space for time, which has the potential to provide equivalent independent record lengths that are much longer than 30 years. To develop such a database, we first identify and track storms using two data sets: a low resolution (daily) gridded dataset based on observations, and a higher resolution (hourly) reanalysis dataset. Identified storms are then filtered to ensure they are independent in both space and time. Storms identified in the high-resolution reanalysis dataset are checked for consistency with the observation-based data set to ensure a grounding in reality.Having developed the database, we then created a software for storm selection. Storms are selected based on an input catchment location plus a prespecified buffer region and within a range of prespecified ratios of catchment size. Storms are then transposed to the catchment centre. The final storms selection can then be formatted to facilitate input to event-based flood modelling software.The development of the extreme storms database and storm selection software facilitates the undertaking of hydrologic risk assessments as storms may be sampled on depth/rarity, spatial homogeneity and temporal homogeneity. For example, it can be used to investigate how spatial and temporal patterns of rainfall may vary with event severity, and this could be used to inform estimates of dam failure risks. Furthermore, it can be used for climate impact assessments by sampling storms based on characteristics associated with a warmer climate e.g. higher depths and shifting spatio-temporal pattern distributions. With this storm database we believe it will enhance hydrological risk assessments performed for both present and future climate scenarios and deepen understanding of the role of spatio-temporal distributions on extremes.

Flood estimates used in engineering design are commonly based on intensity–duration–frequency (IDF) curves derived from historical extreme rainfall. Under global warming, extreme rainfall is increasing, threatening the capacity of existing infrastructure. Hence, there is a need to update our methods of engineering design, namely our design rainfall intensities, for climate change.One way of adjusting our design inputs for climate change is to incorporate covariates into the fitted probability distributions that describe extreme rainfall. To this end, here we evaluate which large-scale climate driver is best for modelling non-stationarity in IDF curves up to the 100-year design return level. The climate drivers we evaluate include global and continental mean temperature, continental diurnal temperature range, continental dewpoint temperature, continental precipitable water, the Indian Ocean Dipole, the El Niño Southern Oscillation, and the Southern Annular Mode.Based on the Akaike Information Criteria, precipitable water is the superior covariate, irrespective of storm duration. However, when quantile changes across the historical period are inspected, we find that global temperature is best able to adequately capture the variability in changes across both storm duration and annual exceedance probability. We finish with presenting a case study where extreme rainfalls are projected using a global mean temperature covariate. The implications for flood risk are that, under 4ºC of global warming, flood risk increases by a multiple of eight.

It is now well established that climate change is increasing the intensity of extreme rainfall. What is less well established is how best to model these changes. Most literature considers non‐stationarity in extreme rainfall using a Generalized Extreme Value (GEV) model with either the location, or scale, or both parameters varying with either time or some climatic covariate, and it is assumed that the shape parameter will not vary. Here we present evidence that the rainfall quantile increases for rare events are greater than those for frequent events, and these relative changes increase with shorter rainfall durations. Further, we demonstrate that this behavior can only be correctly captured if the shape parameter is non‐stationary. We do this by considering annual rainfall maxima at 48 stations in Australia with durations varying from 6 min to 7 days, for events up to the 1 in 100 Annual Exceedance Probability. The results show that a GEV model with non‐stationarity in all parameters is able to capture the variation in changes across both duration and frequency, whereas a model with a constant shape parameter is not. Finally, we apply this approach to calculate non‐stationary intensity‐duration‐frequency curves and their associated uncertainty. We conclude that while a non‐stationary GEV shape parameter is required to capture the greater relative increase in the rainfall depths for rare events compared to frequent events, this increase in model flexibility comes at the expense of considerably larger uncertainty.

Atmospheric Rivers (ARs) play a pivotal role in precipitation dynamics, often leading to hazardous flood events. While much research has investigated the relationship between ARs and precipitation in the northern hemisphere, the role of ARs in modulating Australia's precipitation has received little attention. The ambiguity surrounding AR definition and identification criteria has resulted in the emergence of various algorithms, differing in prerequisite thresholds. This study focuses on refining AR detection algorithm choice with a specific emphasis on the precipitation-AR relationship for Australia. Using data from 1980 to 2019, we examine seven AR selection algorithms, investigating the impact of Integrated Vapor Transport (IVT), a key indicator of AR activity, as well as ARs, on precipitation activity across Australia. We find a robust positive relationship, with a correlation of 0.5 to 1, between annual IVT and precipitation anomalies across Australia. Our findings also reveal a positive correlation between ARs and precipitation on an annual scale. While ARs consistently align with the top 25 % of precipitation, indicative of their association with heavy precipitation events, our study highlights a substantial difference between detection algorithms in the likelihood of AR occurrence on days with and without precipitation - varying from approximately 10 % to almost 40 %. Our results demonstrate that, although ARs are typically associated with heavy precipitation, this association varies greatly between detection algorithms. We anticipate these findings will have implications for our understanding of the role of ARs in hydrologic assessments, particularly related to heavy precipitation events leading to flooding.

Atmospheric rivers (ARs) are narrow corridors of intense water vapor transport in the atmosphere. While the link between atmospheric rivers and extreme precipitation has been established across many regions of the world, the relationship between atmospheric rivers and flooding, the ultimate hazard resulting from extreme precipitation, remains poorly understood. Utilizing 467 Hydrologic Reference Stations (HRS) across Australia, the contribution of ARs to extreme precipitation and flooding is investigated by calculating the probability of occurrence of an AR on peak over threshold (POT) event days using different lag periods. By examining the tail behaviour of heavy precipitation and flooding caused by ARs, using the Generalized Pareto distribution (GPD), the magnitude to which ARs impact extreme events, and how this varies with event severity, is also quantified. Here, we find that southeast Australia has the highest AR concurrence (around 75-100 %) with extreme precipitation and streamflow events. The median magnitude of extremes is 20-70 % higher in the presence of an AR. In addition, the return periods of extreme flood and precipitation events of a given magnitude are on average 2 to 12 times shorter when they coincide with an AR compared to when they do not coincide with an AR. Our study highlights that ARs are a major factor in significantly increasing the frequency of extreme weather events in different regions of Australia. This suggests a need to incorporate AR impacts in hydrological modelling to enable better water resource management and flood risk assessment.

Action needs to be taken in response to the changes in future flood risk due to the impact of global warming on the magnitude and frequency of extreme rainfalls. Projected changes in extreme rainfalls can be used to estimate the associated changes in design flood estimates using Intensity-Frequency-Duration (IFD) curves in combination with event-based flood models. IFD curves are estimated from records of historical annual maxima across different storm durations and exceedance probabilities. Past studies investigating changes in extreme rainfall across Australia have been limited in scope as they have focused on single durations, single exceedance probabilities, or limited regional extents. This means that we do not yet have a comprehensive understanding of how projected changes in extreme rainfalls impact on IFD curves.Here, to fill this gap, we investigate the changes in extreme rainfall changes across different storm durations and exceedance probabilities across 42 stations which span the entire continent of Australia. We begin with examining the trend in annual maximum rainfall across 16 different storm durations (6 min to 7 day) using the Theil-Sen slope estimator, testing for statistical significance using the Mann-Kendall test. To extrapolate 1% annual exceedance probability, we fit non-stationary Generalized Extreme Value Distributions (GEVs) at each site. Non-stationarity was assessed by varying the location parameter, varying the scale parameter, and varying both the location and scale parameters as a linear trend in time.We find that the short duration (1 hr and 1 day) show fewer positive trends with some sites exhibiting negative trends. Based on Akaike Information Criteria, the GEV models which varied either the location parameter, or both the scale and location parameters, were found to be superior. However, when changes in quantile estimates were examined for rare exceedance probabilities (up to the 1 in 100 AEP), it was found the GEV model which only varied the location parameter was unable to capture the increased rate of change in extreme rainfalls. Accordingly, we conclude that changes in extreme rainfalls is best represented by non-stationary models that incorporate changes in both location and scale parameters.