Challenges in the attribution of river flood events

Advances in the field of extreme event attribution allow to estimate how anthropogenic global warming affects the odds of individual climate disasters, such as river floods. Extreme event attribution typically uses precipitation as proxy for flooding. However, hydrological processes and antecedent conditions make the relation between precipitation and floods highly nonlinear. In addition, hydrology acknowledges that changes in floods can be strongly driven by changes in land‐cover and by other human interventions in the hydrological system, such as irrigation and construction of dams. These drivers can either amplify, dampen or outweigh the effect of climate change on local flood occurrence. Neglecting these processes and drivers can lead to incorrect flood attribution. Including flooding explicitly, that is, using data and models of hydrology and hydrodynamics that can represent the relevant hydrological processes, will lead to more robust event attribution, and will account for the role of other drivers beyond climate change. Existing attempts are incomplete. We argue that the existing probabilistic framework for extreme event attribution can be extended to explicitly include floods for near‐natural cases, where flood occurrence was unlikely to be influenced by land‐cover change and human hydrological interventions. However, for the many cases where this assumption is not valid, a multi‐driver framework for conditional event attribution needs to be established. Explicit flood attribution will have to grapple with uncertainties from lack of observations and compounding from the many processes involved. Further, it requires collaboration between climatologists and hydrologists, and promises to better address the needs of flood risk management.


| Problem statement and objectives
The interest for attributing an extreme event commonly originates in the severity of the societal impacts (Stone et al., 2021), rather than in its remarkable meteorological features.For instance, there is frequent discussion of a potential role of extreme event attribution in the Loss and Damage Mechanism of the United Nations Framework Convention on Climate Change (Olsson et al., 2022;Parker et al., 2015).Extreme event attribution is typically based on analysis of meteorological variables.During an initial step, called "event definition" (Philip et al., 2020), researchers define the intensity and domain of the meteorological event that contributed to the disaster (van der Wiel et al., 2017).For example, for the pluvial floods of July 28, 2014 in the Netherlands, the event was defined as equal or surpassing 132 mm of daily precipitation over the whole country (Eden et al., 2018).However, in the case of river floods, which we consider here, it is not the extreme precipitation that directly causes the socioeconomic impacts.Impacts scale rather with the magnitude of the flood hazard, commonly defined as the "temporary covering by water of land not normally covered by water" (Barredo, 2007).While extreme precipitation does not necessarily lead to flooding, flooding can be generated by moderate precipitation, when other factors are at play (Berghuijs et al., 2019;van der Wiel et al., 2020).Thus, the intensity of precipitation is only a proxy for the magnitude of flood impacts.The validity of this proxy depends on the local context and specific event, but is generally not examined.To bring event attribution closer to the impacts of flooding, two solutions have been attempted: (1) using hydrological models to convert precipitation into discharge (e.g., Schaller et al., 2016; see Section 2); (2) directly relating the magnitude of the economic impacts to that of the precipitation event, thus bypassing the complications of solution 1 (Frame et al., 2020).Both approaches are infrequent.
Non-climatological changes can also affect river flood hazard, that is, regional and local hydrological changes (Boulange et al., 2021;Munoz et al., 2018;Sebastian et al., 2019;Syvitski & Brakenridge, 2013).The effects of these changes can in some cases oppose those of climate change.For example, dam construction can lower flood occurrence on a given location, even when climate change would have increased it; landscape change from forest to urban can increase flood occurrence at the site and downstream, even when climate change would have lowered it.Furthermore, hydrological change can alter sediment fluxes, in turn leading to geomorphic responses, such as in-channel sedimentation and channel enlargement, that affect the flow capacity of river channels and thus its tendency to flood (Hoffmann et al., 2010).
The science of extreme event attribution has quickly advanced, and methods are becoming standardized (Philip et al., 2020;van Oldenborgh et al., 2021).But an explicit attribution of floods to their multiple drivers is still unattempted.In this Perspective: we explain how the scientific community has attempted to include hydrological processes in the attribution of flood events (Section 2); we examine the challenges to flood event attribution, differentiating between near-natural cases and cases where substantial hydrological change has occurred (Section 3); and we argue that the expansion of the existing probabilistic event attribution framework can address the earlier cases, and that a new multi-driver framework for the conditional event attribution can address the latter (Section 4).
Extreme event attribution that includes representations of hydrological processes and changes could better isolate the influence of climate change on flood occurrence, yielding a more accurate climate attribution.Accordingly, it would also inform on the influence of other key drivers of floods.This addresses the mechanisms behind changes in flood hazards, one of the most pressing questions in hydrology and flood risk management (Blöschl et al., 2019).In the wake of a flood disaster, when the urgency of the problem is clearest to citizens and decision-makers, a multi-driver attribution of the event could offer a strong scientific basis for flood risk management.In fact, while stakeholders see general merit in attribution studies (James et al., 2019;Sippel et al., 2015), they seem doubtful about their present usefulness for the practice of climate adaptation and disaster management (Osaka & Bellamy, 2020).

| STATE-OF-THE-ART: HYDROLOGICAL MODELING IN EVENT ATTRIBUTION
Models of hydrology and of hydrodynamics can represent the processes relevant to the generation of floods.A handful of probabilistic extreme event attribution studies include hydrology in their analysis.Pall et al. (2011) propose runoff as a better proxy for flooding than precipitation, and use a simple precipitation-runoff model, fed with daily precipitation input from global climate model simulations, to produce daily runoff series.They do not compare their results to results that would have been obtained from precipitation alone.They calculate, however, the changes in precipitation that would follow thermodynamically from the warming of air masses, concluding that the fraction of attributable risk is similar across the two approaches.Kay et al. (2011) use the same meteorological input as Pall and colleagues, and expand on that study in two respects.They realize 1-year-long continuous simulations of precipitation-runoff, using the probability-distributed model PDM for six smaller catchments and the semi-distributed model CLASSIC for two larger catchments in England.They thus capture the effects of antecedent conditions in the basin and of spatio-temporal variations in precipitation.Further, they include snow-related processes resulting from temperature change, showing that these have a notable effect on runoff peaks.Schaller et al. (2016) also perform continuous simulations, for 4 years, and include snow processes, with the same CLASSIC model, this time coupled with the hydrodynamic model JFlow+, based on very high resolution (5 m), LiDAR-enabled digital terrain.They highlight the importance of multi-year antecedent conditions to attribute discharge peaks.Moreover, they expand the scope toward flood impacts, and employ an empirical relationship to infer flooded property from peaks in discharge, assuming absence of flood defenses.Kay et al. (2018) apply a similar approach to the larger domain of Great Britain, and also highlight the importance of including flood modeling and snow processes.Philip et al. (2019) aim to systematically assess the difference between attribution either based on precipitation only and with inclusion of modeling of river discharge.For the first time, they use multiple hydrological models: PCR-GLOBWB, SWAT, LISFLOOD, and JULES (in combination with RFM, for the river routing component).They find that results of discharge attribution differ from those based on precipitation.Sebastian et al. (2019) use hydrological model Vflo ® to simulate discharge in the city of Houston resulting from extreme precipitation associated with Hurricane Harvey.Notably, they separate the effects of urban development and climate change in changing peak discharge, finding that urban development had a larger effect.No event attribution study, to our knowledge, has explicitly modeled flood hazard, that is, the flood metrics like flooded area or flood depth.

| CHALLENGES TO FLOOD EVENT ATTRIBUTION
We argue that two main challenges limit our capacity to accurately attribute river flood events (Figure 2): (1) Flooding relates to the triggering precipitation event in highly nonlinearly ways, because of the mediating hydrological and hydrodynamic processes, even in the absence of changes to the hydrological system.(2) Changes to the hydrological system have occurred during the period of climate change, adding their influence on the occurrence of floods.These changes include natural and human changes in land cover, other human hydrological interventions, and the geomorphic responses of the river channel.In this section, we examine these two challenges.

| Nonlinear relation between extreme precipitation and floods
Shifts in precipitation extremes are not clearly reflected in global flood observations (Berghuijs & Slater, 2023;Blöschl et al., 2017;Do et al., 2017;Kundzewicz et al., 2014;Slater et al., 2021;Zhang et al., 2022).To examine where trends of extreme precipitation and of extreme discharge converge and diverge in sign, we calculated significant trends in annual multi-day extreme precipitation based on the ERA5 climate reanalysis dataset (Hersbach et al., 2020), and in extreme discharge based on the dataset GloFAS (Harrigan et al., 2020), which is modeled based on ERA5 meteorological forcing, for the period 1979-2020 (Figure 3).Although over most locations the sign of trends in discharge extremes matches that of precipitation extremes (orange and green colors), for many locations an increase in precipitation extremes is matched by a decrease in discharge extremes (blue).Locations where decrease in precipitation extremes is matched by an increase in discharge extremes (purple) are very rare.ERA5 and GloFAS are the only datasets suitable for this global trend detection.However, it must be noted that these are reanalysis products, and have discrepancies with respect to observations.ERA5 precipitation, and therefore also discharge in GloFAS has errors, especially in the tropics (Lavers et al., 2022).

| Mechanisms of river flood generation
The discrepancies in trends are due to nonlinear relations between precipitation and flooding, generated by multi-step mediating processes.In a first step, part of precipitation is converted into runoff.In a second step, runoff accumulates into river discharge.In a final step, a portion of discharge can be converted into flood water during an extreme event when river bankfull conditions are exceeded.The first two steps are mediated by processes related to hydrology, like evapotranspiration and infiltration.These are determined by the characteristics of the land surface: slope, soil, vegetation, land-cover, and river network.The second and third steps are mediated by hydrodynamic processes, determined by the hydraulic characteristics of the river channel and of the floodplain.
The spatial and temporal pattern in which these processes play out is essential in determining their outcome.A key determinant is the state of the relevant components of the water cycle at the time of the precipitation event: the antecedent conditions.Key antecedents are: the amount of snow priorly accumulated in the mountainous part of the basin and the timing of its thawing (Berghuijs et al., 2016;Huntingford et al., 2014;Musselman et al., 2018); the level of moisture of the upper parts of the soil (Neri et al., 2019;Tramblay et al., 2019;Wasko & Nathan, 2019); for large-scale basins and events, the level of groundwater.In geographies where these phenomena are seasonal, flood occurrence will typically have strong seasonality (Rottler et al., 2021).For example, the same precipitation event can more likely result in flooding during springtime than summer (Schaller et al., 2014), due to the higher infiltration capacity of summer soils, which can present a moisture deficit due to higher temperatures and evapotranspiration.More recently, attention is raised to drought as an aggravating antecedent factor for floods (Rashid & Wahl, 2022), whereby soil permeability is reduced by protracted dry conditions (Alaoui et al., 2018).An example of this phenomenon unfolded in spring 2023, over vast parts of Northern Italy (NASA_Earth_Observatory, 2023).

| Hydrological change has occurred
Hydrological changes reduce flood hazard, or increase it; some changes still reduce it at one location while increasing it at another.The key problem for flood attribution is that often these changes have taken place during the period of climate change.As such, hydrological changes can amplify or counterbalance the effect of climate change on flood occurrence, and failing to take them into account vitiates the attribution.
Effects of land-cover change on river discharge and flood are difficult to predict (Kirchner et al., 2020).Changes in land-cover can be natural or anthropogenic; in the latter case they are called land-use change.Observations show that deforestation in 56 developing countries increased flood occurrence during the last decades (Bradshaw et al., 2007).Similarly, Anderson et al. (2022) show that urbanization and re-forestation have respectively increased and decreased extreme streamflow, in the context of 729 US catchments.Similar indications emerge from many modeling studies (e.g., Du et al., 2012).
Other types of human intervention on hydrology have taken place over a large part of the world's rivers (Grill et al., 2019), altering hydrological and hydraulic properties relevant to flooding.Key interventions are: dam construction and management, river bed encroaching, levees and dikes, channeling and water expansion areas, civil structures like roads, bridges and drainage networks, irrigation and groundwater abstraction, and other flood management measures.While most of these interventions are explicitly meant to have a local hydrological effect, for example, building a levee to reduce local flood hazard, some have unintended hydrological effects, for example: irrigation lowers the water table in the soil; river training may increase flood hazard further downstream (Munoz et al., 2018;Vorogushyn & Merz, 2013).

| POSSIBLE SOLUTIONS FOR FLOOD EVENT ATTRIBUTION
The challenge of nonlinear relations between extreme precipitation and flooding applies to all flood cases, and thus also to cases of near-natural hydrology, where hydrological changes during the time interval of interest can be neglected (Figure 2).We argue that, for these simpler cases, an approach that can be explored is the expansion of the existing framework of probabilistic extreme event attribution, to include hydrological processes and antecedent conditions that drive floods.On the other hand, cases where relevant hydrological changes have occurred, that is, land-cover changes and human hydrological interventions, present the additional challenge that the effect of these flood drivers overlaps with the effect of climate change.For these more complex cases, we argue that it is necessary to establish a multi-driver framework for conditional event attribution.

| Possible solution for near-natural cases
The methods of probabilistic event attribution can be expanded to include representation of the relevant hydrological and hydrodynamic processes, by explicitly using flood data and flood modeling.This allows overcoming the problem posed by the nonlinearity between precipitation and floods, in two main respects.First, explicit flood attribution overcomes the errors associated with the initial step of event definition.When using precipitation as a proxy for floods, the precipitation event needs to be defined in terms of intensity and extent in space and time, such that it most closely captures the generation of the flood.This involves arbitrary choices (van Oldenborgh et al., 2021).For example, after consultation with local experts and consideration of impacts, it can be defensible to define the triggering event as the cumulative precipitation over either 2 or 5 days, and over an area of either 500 or 2000 km 2 .However, the results of the attribution may strongly diverge in either case (Angélil et al., 2018), requiring sensitivity tests (Luu et al., 2021).If, instead, the definition is an observed metric of flood hazard, this problem is largely negated.Second, explicit flood attribution addresses the issue of antecedent conditions.Multi-year hydrological simulations, followed by flood simulations, can adequately reproduce the state of, for example, soil moisture and snow pack.
In the following, we explore possible solutions for: the definition of an appropriate flood metric; the availability of suitable flood data; and the flood modeling.

| Flood metric
Adding explicit consideration of floods to attribution means that all analysis is based on a metric of flood hazard, instead of precipitation.As established in event attribution science, the event definition should be closely informed by the socio-economic impacts of the event.For any type of exposed element, the impact is primarily determined by the depth of the flood waters.Other relevant quantities are the flow velocity, flood duration, and any pollution or sediment carried by the water (Vogel et al., 2018).The total impact of the event is determined by the sum of the impacts at each point over the flooded area.Therefore, a flood metric that reasonably relates to the total impact is the total flood area.However, since exposure widely varies over the territory, to better estimate impacts, a finer analysis should take into account the location of population and of valuable and critical assets, and the local flood depth.

| Flood data
Probabilistic extreme event attribution requires the following categories of data: (1) observed magnitude of the event, for the event definition; (2) time-series of observations, for trend detection; (3) model-based time-series, representing both factual and counterfactual climates, for the actual attribution (Philip et al., 2020).
For the magnitude of the flood event, it is ideal if collaboration with the local relevant institutions is established; the chance is then higher that data from remote sensing, gauges or field surveys can be accessed.Sometimes, local measurements are compromised by the flood (Kreienkamp et al., 2021).In the absence of local flood data, recent flood products at continental or global scale are potentially suitable replacements: • Global Drought and Flood Catalogue et al., 2020) For the attribution of events that will happen in coming years, ongoing developments in remote sensing are promising.New flood observation data are becoming available, for example, in the Fractional Water data from NASA's Soil Moisture Active Passive satellite (Du et al., 2021).New algorithms may soon reconstruct near-real time observations of flood area, integrating imaging from satellite and aerial sensors with elevation maps (Muñoz et al., 2021).
In contrast to probabilistic attribution of extreme weather events, for explicit flood attribution the step of trend detection in observational time series is not essential.The reason for this is that the different climatic and hydrological drivers explained in Section 3.1.1,and their changes and interactions in time (e.g., in the case of snow accumulation and melt) will likely overlap, confounding any observed trend.However, time series of past floods are useful for the contextualization of the event under study, for model validation, and for the interpretation of the results.Complete, uniform time-series of floods are very rare.For a few locations, the data listed in the previous section may contain a short time-series of historical floods.Failing that, the next best option is to resort to a flood proxy better than precipitation: river discharge.If that is also not available, the best approach is to use modeling reconstructions that are based on observations or on climate reanalysis.The key datasets are: • Global Runoff Data Centre (GRDC).Comprises observations for 9900 stations globally of daily or monthly discharge.

| Flood modeling
Flood modeling should be used to produce time-series of flood events, based both on observed and on modeled boundary conditions.In probabilistic attribution, the former are needed in the step of trend detection, and the latter are needed in the model-based attribution step that compares floods in factual and counterfactual climates.
To produce an observation-based flood time-series, the modeling chain needs to include both a hydrological and a hydrodynamic modeling step.The hydrological model uses meteorological observations on precipitation and temperature, typically at daily time-step; solve key processes like evapotranspiration, infiltration, exchanges between storage in snow, soil and groundwater; and produce time-series of discharge or runoff at a suitable resolution.In turn, the discharge or runoff series are used in the hydrodynamic model, which solves processes related to the surface flow of water more accurately than the hydrological model, and yields peak discharge and flood metrics.
To produce the entirely model-based flood time-series representing factual and counterfactual climate, the modeling chain is the same, but will take precipitation and temperature from climate models.These could come from the simulations coordinated by the Coupled Model Intercomparison Project Phase 6 (CMIP6).CMIP6 includes simulations called "piControl", that is, "preindustrial control", reflecting the counterfactual climate of AD 1850 (Eyring et al., 2016), unaffected by anthropogenic greenhouse gas emissions, using constant forcing along the whole duration, of at least 100 years.The factual climate is addressed by the "historical" simulations of CMIP6.These simulations are transient, meaning that increasing levels of greenhouse gas are applied to reflect the history of anthropogenic emissions during 1850-2014.To extend the series until the present, results of the "scenarioMIP" experiments of CMIP6 can be used.CMIP6 results are available at https://esgf-node.llnl.gov/projects/cmip6/.Such results have been used to globally simulate high-resolution floods for the preindustrial (counterfactual) climate (Scussolini et al., 2020), and both preindustrial and modern (factual) climates (project ISIMIP2b;Lange et al., 2020).
Additionally, to force flood models at higher resolution, results from regional climate models can be used.A group of dynamical downscaling experiments using different regional climate models is coordinated by CORDEX (Diez-Sierra et al., 2022; https://cordex.org/data-access/).These experiments cover 14 continental domains, at resolution between 12 and 50 km.Simulations start at 1950, and thus do not include the preindustrial, fully counterfactual climate.Also, they are presently based on CMIP5 global climate models, but results based on CMIP6 will become available in the near future.
However, it is necessary to consider whether the meteorological input from global or regional climate models has adequate resolution and skill, especially with respect to convective precipitation events.These fine-scale events are not adequately represented by parameterization schemes in those models (Coppola et al., 2020).This can be overcome by the recent emergence of convection-permitting climate models, with resolution finer than 4 km (Luu et al., 2022;Manola et al., 2018).Such input into the flood modeling can improve skills in reproducing flood properties.However, this comes at massive computing costs; hence cannot be performed over a large area or for a large ensemble and their application is rare (Pichelli et al., 2021).
Hydrological simulations for attribution should be carried out continuously for multiple years, as first evidenced by Schaller et al. (2016).This is the only way to capture antecedent conditions, as it allows the hydrological model to adjust to long term effects on, for example, storage of snow, lakes and groundwater (Ajami et al., 2014).Thus, results will also include floods generated by moderate precipitation, for example, when it coincides with snow-melt, or when it occurs after prolonged cold/wet conditions that have saturated the soil in the basin.Ideally, simulations should be run continuously over the whole interval covered by the multi-decadal climate time-series available.If that is computationally too expensive, and if long-term changes in water storage are negligible, a reasonable choice is to apply a threshold to extract extreme events from the precipitation time-series, and to simulate hydrology and floods during multi-year intervals culminating with these events.
The step of model evaluation should include, besides precipitation and temperature, comparison of modeled floods and discharge with the respective observations, both for the event and for historical time-series.For this, the datasets mentioned above can be used.

| Possible solution for complex cases
When relevant hydrological change has occurred, changes in flood occurrence have more than one driver.This is not contemplated in the probabilistic attribution framework, where the question is a variant of "has anthropogenic climate change increased the frequency of events like this one?"(NASEM, 2016).

| A multi-driver framework for conditional event attribution
A multi-driver framework should be formulated, to address driver-specific questions that condition the attribution on the states of the other drivers (Figure 4).By systematically including and excluding, in the hydrological and hydrodynamic modeling steps, the effect of land-cover change and of human hydrological interventions, explicit flood attribution can separate the relevance of each driver, including that of climate change.Frameworks that embrace multiple hydrological drivers have been proposed.For example, Harrigan et al. (2014) proposed the method of "multiple working hypotheses" to attribute the observed increase in average streamflow in the Boyne catchment, Ireland.A multi-driver framework for conditional event attribution also takes inspiration from emerging literature on storyline event attribution for flood management (de Bruijn et al., 2016;Goulart et al., 2023;Sillmann et al., 2021).
The core steps should be: event definition; definition of hydrological drivers; evaluation of the modeling chain; conditional model-based attribution.The event definition and the evaluation of the modeling chain present the same challenges as in the near-natural cases discussed in Section 4.1.We here discuss: the available data about hydrological F I G U R E 4 Conceptual representation of changes through time in drivers of river floods.Global warming, and associated changes in temperature and precipitation, often overlaps with changes in the hydrology, such as land-cover change, construction of levees and largescale irrigation.Global temperature data are from the National Centers for Environmental Information of NOAA.Flood occurrence data is hypothetical.The bottom box illustrates the questions pertinent to a multi-driver framework for conditional attribution.
changes, to inform the definition of the relevant hydrological drivers; and how to represent hydrological changes in models, to enable the conditional model-based attribution.
To judge which hydrological changes should be included, it is necessary to know the hydrological history of the basin (Figure 4).Different relevant hydrological changes have likely occurred at different times, and can be very old, as in the case of the Netherlands, where water management has a famously long history (Hoeksema, 2007).Restricting the analysis to the period coinciding with most of climate change, that is, the last century, makes the problem more tractable, and seems more relevant from the perspective of flood risk management.In the following, we review information on land-cover change and on other human interventions.

| Land-cover change
For flood attribution, land-cover (land-use) maps need to be available for the present and also for past periods.Ideally, to ascertain whether changes overlap with climate change, maps should cover ca. the last century.Other relevant aspects are accuracy, resolution and the number of land-cover classes that are differentiated.Fine resolution is especially useful for detailed hydrodynamic modeling in urban contexts, whereas for hydrological modeling of large basins, resolution can be coarser.If adopting land-cover maps from different sources, for example, for the past and for the present, it is also important to consider coherence across sources: namely, in the method of production and in the classes that are distinguished.The priority should be to access any local data curated by regional institutions, which will typically be finer, more accurate, and may extend further back in time.Should this not be available, potentially useful global and continental datasets are: • Global Land Survey.Global maps curated by the US Geological Survey, for years 1975, 1990, 2000, 2005, 2010, at 30 m of resolution.Available at https://www.usgs.gov/landsat-missions/global-land-survey-gls.• Global Land Cover and Land Use Change (Potapov et al., 2022).Global maps from satellite imagery, for period 2000-2020, at 30 m of resolution.https://glad.umd.edu/dataset/GLCLUC2020. • Climate Change Initiative Land Cover V2.Global maps curated by the European Space Agency, from each year between 1992 and 2015, at 300 m of resolution.https://www.esa-landcover-cci.org/.• Land-Use Harmonization (Hurtt et al., 2020).Global modeled maps for period 850-2100, at 0.25 of resolution.
If different datasets have comparable merits, including multiple datasets in the modeling could enable quantifying the uncertainty relative to the land-cover.When land-cover changes affect large areas of the river basin, they should be included in the hydrological modeling step.When they affect the urban areas adjacent to the flood, it may be appropriate to include them in the hydrodynamic modeling step.Most distributed hydrological models have internal representation of land-cover for processes as evapotranspiration, canopy interception, infiltration, irrigation (e.g., Horton et al., 2022).However, the associated parametrizations are not evident and are model-dependent; the hydrological effects of land-cover types are captured across different models with different levels of accuracy.Therefore, it must be noted that this step will be associated with uncertainties and errors, which are likely hard to quantify.Another aspect that should be considered is whether the land-cover changes had implications for soil properties.Deforestation, for example, is known to cause loss of soil, especially on steep terrain.If this is the case, soil changes should be included in the modeling, either using direct available information, or resorting to assumptions.

| Other human interventions on hydrology
Flood attribution requires information on human hydrological interventions: their key features relevant to the modeling, and the timing of their realization.Similarly, progressive changes to the channel geometry and channel bed elevation as a result of fluvial aggradation or incision need to be assessed, when relevant.As with land-cover, often the best information should be accessed in collaboration with local authorities.However, there are a few global datasets that can function as alternatives: • Global Dam Watch.This can be used to assess the presence of dams and reservoirs in the basin upstream of the study area, and the date of construction thereof.http://globaldamwatch.org.• Global Water Watch (Donchyts et al., 2022).This contains time series of surface area for 71,000 reservoirs.https:// www.globalwaterwatch.earth/.• Historical Irrigation Database (Siebert et al., 2015).This contains information on which global areas are equipped for irrigation, in time-series covering 1900-2005, at 5 0 of resolution.https://aquaknow.jrc.ec.europa.eu/en/content/historical-irrigation-dataset-hid.• OpenDELvE (Nienhuis et al., 2022).This contains maps, locations and metadata of known levees over global river deltas.https://www.opendelve.eu/.
These datasets might still miss data for data-poor regions.Additionally, we are not aware of global datasets that document changes in geomorphological aspects like channel geometry.It could be therefore valuable to try to assess the presence and extent of interventions indirectly, using more general datasets that have recently emerged.For example, HydroATLAS (Linke et al., 2019) contains information on hydrological, climatological, environmental and anthropogenic characteristics of river basins and segments, at 15 00 of resolution; the database of free-flowing rivers (Grill et al., 2019) contains information about human pressure on river segments, notably on: degree of fragmentation and of regulation by dams; on urban areas, enabling assumptions on the presence of levees and river confinement structures.
The hydrological and hydrodynamic modeling will then need to adopt methods to represent these interventions in the simulations, often by "burning" them into the elevation map (Wing et al., 2019), and drawing from the large experience documented in the literature (e.g., Remo et al., 2018;Zhao et al., 2016).

| WHITHER FLOOD ATTRIBUTION?
We have proposed our view that explicit inclusion of flooding in extreme event attribution is necessary, but complicated by challenges at several levels.In all cases, hydrological/hydrodynamic modeling has to be included, as no existing dataset includes flood events under the factual and counterfactual climate conditions needed for the attribution.An area of advancement will therefore be to tackle these challenges in an expanded probabilistic attribution framework (Section 4.1).This can benefit from the huge progress recently made toward generating historical and near-real-time flood information (Paprotny et al., 2023).Such a framework should include validation and propagation of uncertainties over the additional steps of hydrological and flood modeling, due to the multiple processes that have to be represented in those models.We expect that uncertainties will grow considerably compared to attribution of precipitation.For most basins and river segments, hydrology is altered (Grill et al., 2019).Here, other drivers of flood occurrence need consideration, and future research should establish and test novel frameworks for attribution.Uncertainties here will likely be even larger, because of the additional source of uncertainty associated with information on hydrological changes and interventions, and their representation in models.We have illustrated the broad ideas of a multi-driver framework for conditional attribution (Section 4.2).As this framework enables isolating the hydrologic from the climatic drivers, it could also be adapted to include and isolate specific climatic drivers, to link the occurrence of the event to specific mechanisms.For example, thermodynamic versus dynamic drivers (Diffenbaugh et al., 2017), teleconnections with climatic oscillations, or sea surface temperatures.In the future, efforts could also be made toward a framework to also enable attribution of compound fluvial and coastal floods (Zscheischler & Lehner, 2022).
An interesting advantage offered by an attribution framework that includes representation of past hydrological change, is that with relatively small additional effort, it could also include simulations with possible future changes in land-cover and possible river/flood management measures, in addition to possible climatic scenarios.This would directly connect flood attribution with adaptation and with the policies of flood management (Lahsen & Ribot, 2022;Osaka & Bellamy, 2020).
Besides flood managers, flood attribution could benefit from collaborations with fluvial geomorphologists and paleoflood hydrologists.While dealing with the last decades or century makes the attribution more tractable, historical records or sedimentary evidence for extreme events can expand the limited window of modern observations, and inform on rare high-magnitude events that occurred in the past (Wilhelm et al., 2018).Usually, the offset between the past setting (land-use and river morphology) and the present requires assessment, which may be a challenge for heavily engineered Anthropocene rivers.Despite the uncertainty associated with discharge estimates from paleoflood studies, their inclusion could benefit flood risk attribution as they provide the precedent and the synoptic conditions for such extremes to have occurred in the past (St. George & Mudelsee, 2019).Besides, paleo flood information could enable attribution of older historical floods (Blöschl et al., 2020).

F
I G U R E 3 Convergence and divergence in the sign of significant trends in extremes of precipitation and of discharge, for period 1979-2020.This is calculated for annual maximum cumulative values of precipitation and of discharge during consecutive periods of 1-day (panel a), 2-day (b), and 20-day (c).Colors show where precipitation and discharge extremes both increase (green), both decrease (orange), where precipitation increases while discharge decreases (blue), and where precipitation decreases while discharge increases (purple).Precipitation data are from ERA5(Hersbach et al., 2020), and discharge data are from GloFAS(Harrigan et al., 2020).Trend significance is assessed with the Mann-Kendall test.
Harrigan et al., 2020)scharge and flood depth, from 2001 to the present with real-time update; the simulation resolution is 1/8 , and results are downscaled to 1 km.It takes precipitation input from several products to run the VIC hydrological model coupled with the DRTR flood model.http://flood.umd.edu/.•GlobalFloodAwarenessSystem (GloFAS;Harrigan et al., 2020).Contains daily discharge at 0.1 resolution, from 1979 to the present with real-time update.It takes runoff from hydrological model HTESSEL (part of ERA5 climate reanalysis) to run the hydrodynamic model LISFLOOD.https://www.globalfloods.eu/.•