Climate change is likely to increasingly affect hydrological regimes and flood hazards over coming decades. Considerable changes in atmospheric temperature, precipitation, humidity and circulation are expected, which may result in increasing extreme events, including floods (IPCC, 2013). According to Rojas et al. (2013), flood frequency in Europe will increase due to climate change, with significant socio-economic implications for the region. Blöschl et al. (2017, 2019) conclude that the timing and magnitude of European floods have shifted due to climate change, and its consequences are not uniform across the region, with northwestern Europe experiencing earlier and higher flood peaks. Modelling and understanding of catchment-scale flood hazard projections is therefore an important endeavour for informing adaptation strategies.

Climate change impact assessment is subject to considerable uncertainties (Wilby and Dessai, 2010; Smith et al., 2018); Blöschl et al. (2019) recently highlighted uncertainty in hydrology as one of the 25 challenges in hydrological science. Traditional or top-down climate change impact assessments typically follow a modelling chain where output from global climate models (GCMs), forced with scenarios of future greenhouse gas concentrations, are extracted and scaled to represent a study catchment. Hydrological models, calibrated for current conditions, are then forced with these GCM outputs to create discharge series spanning multiple decades into the future. When assessing flood hazard, extreme value distributions are typically fitted to samples of extreme events (e.g. annual maximum flood series) representing current and future climates to evaluate changes in the characteristics of flooding. This modelling chain is replete with uncertainties that propagate and interact, resulting in potential large ranges in projected change at the catchment scale (e.g. Meresa and Gatachew, 2019) that can impede decision-making (Smith et al., 2018).

Numerous studies have attempted to quantify the uncertainties in future climate change impacts due to the use of different climate models (e.g Knutti and Sedláček, 2013), natural variability (e.g. Giuntoli et al., 2018; Hughes et al., 2011), bias correction techniques (e.g. Kay et al., 2009; Saini et al., 2015; Soriano et al., 2019), downscaling approaches (e.g. Fowler et al., 2007; Gutmann et al., 2014), hydrological modelling uncertainties (e.g. Wilby and Harris, 2006; Bastola et al., 2011a; Addor et al., 2014; Meresa and Romanowicz, 2017; Broderick et al., 2019) and the application of different extreme value distributions for flood hazard estimation (e.g. Meresa and Romanowicz, 2017; Lawrence, 2020). Findings from such studies confirm that the dominant source of uncertainty varies across studies. For example, Jobst et al. (2018) concluded that the GCM structure is the dominant contributor followed by emission scenario, bias correction and hydrological model structure in New Zealand catchments. Peleg et al. (2015) found that the hydrological regimes are highly sensitive to changes in convective precipitation in eastern Mediterranean catchments. Gosling et al. (2011) evaluated seasonal flow changes in UK catchments and found greater uncertainty in future estimates due to climate models than hydrological models. In Norwegian catchments, Lawrence (2020) highlighted that climate models and frequency models are dominant sources of uncertainty in future flood projections. Therefore, the key components of the uncertainty cascade and their contributions tend to vary depending on the region, precipitation type and catchment.

Recognition of the uncertainties inherent in future climate change impacts has also given rise to novel approaches to decision-making that embrace uncertainties (e.g. Wilby and Murphy, 2019; Clark et al., 2016). Rather than aiming to derive precise assessments of future risk, such approaches aim to stress test and evaluate adaptation options, either in static or dynamic ways (Mazzorana et al., 2012), to identify actions that are functional across a range of plausible future conditions rather than optimised to a certain outcome (e.g. structured decision-making (SDM), robust decision-making, decision scaling and adaptation pathways). Shepherd et al. (2018) recommend a storyline approach for navigating uncertainty, whereby examination of recent, notable extremes and the development of plausible narratives of change in their drivers enables a novel assessment of future changes, placing emphasis on physical processes. Each of these approaches requires the cascade of uncertainty to be evaluated or navigated in ways that can better inform adaptation. For example, Broderick et al. (2019) employ a scenario neutral framework (Prudhomme et al., 2011) to evaluate design allowances for flood defences, taking into account uncertainties derived from a large climate model ensemble, natural variability and hydrological models.

Whether employing traditional impact-led or novel decision-centric approaches, it is critical that key components of the cascade of uncertainty are adequately included. Research over the past 2 decades clearly shows that methodological choices made in assessing future climate change fundamentally impact on the portrayal of climate risk (Clark et al., 2016; Melsen et al., 2019). Top-down approaches must a priori decide on the components of the modelling chain to include. Including all relevant components is typically a very resource expensive task (Smith et al., 2018), and it may not be apparent at the outset which components of the uncertainty cascade to include for specific catchments. For bottom-up approaches, such as the scenario-neutral method (Broderick et al., 2019), the modeller is forced to choose which aspects of uncertainty to include in assessing sensitivity to future changes. Again, this may not be apparent from the outset. It is widely acknowledged that climate models exhibit biases in their outputs (Krinner and Flanner, 2018; Giorgi and Gao, 2018), with numerous studies highlighting the need to post-process climate model output before use in simulating hydrological response (Ehret et al., 2012; Teng et al., 2012; Osuch et al., 2017; Meresa and Romanowicz, 2017). Dobler et al. (2012) suggest that the dominant components of the modelling chain may further vary, depending on which part of the flow regime (low, mean and high flows) is of interest. They highlight, for example, that the importance of bias correction approaches to the overall uncertainty in hydrological response increases for high flows. However, there is still no consensus on which bias correction techniques are most effective, nor in how bias correction techniques can modify future climate change signals. For instance, Teutschbein and Seibert (2013) and Yang et al. (2010) showed distribution mapping based on theoretical distributions outperforms other bias correction methods. Similarly, Berg et al. (2012) and Chen et al. (2013) show that theoretical distribution mapping performs similar to, or marginally better than, empirical quantile mapping. On the contrary, Gudmundsson et al. (2012), Gutjahr and Heinemann (2013) and Lafon et al. (2013) show that empirical quantile mapping demonstrates higher skill than theoretical distribution mapping in systematically correcting precipitation.

Moreover, deep uncertainty can arise by virtue of the ad hoc ways in which components of possible modelling chains are assembled and characterised within the overall modelling framework adopted (Wilby and Murphy, 2019). Yet there are currently no established ways of sampling from the hierarchy of models used to evaluate impacts of climate change (Clark et al., 2016). Therefore, an important step in better integrating future climate risk into decision-making is quantifying and partitioning the contribution of different components of uncertainty and their interactions to help scientists and decision-makers better navigate the cascade of uncertainty by identifying contributing sources that should be more fully explored (Smith et al., 2018). In this regard, ANOVA (ANalysis Of VAriance)-based techniques, which can be used to decompose sources of uncertainty and their interaction, offer considerable utility (e.g. Kay et al., 2009; Vetter et al., 2016; Hattermann et al., 2018; Meresa, 2020). In this study, we explore uncertainties in future flood hazard for four catchments, representative of different flood response types, in Ireland and employ additive chains and variance decomposition to quantify and examine the contribution of various sources of uncertainty, together with their interaction, to the overall uncertainty in flood hazard projections. In doing so, we place emphasis on evaluating the uncertainties derived from (i) climate models in the newly available Coupled Model Intercomparison Project Phase 6 (CMIP6) ensemble (Wyser et al., 2020), (ii) widely used bias correction techniques, (iii) hydrological model parameter uncertainty and iv) the use of different extreme value distributions. The remainder of the paper is organised as follows: Sect. 2 outlines the study design and data/methods employed, and Sect. 3 presents key results, exploring uncertainties for key steps in the modelling chain examined, together with their contributions and interactions to the full range of projected change in flood risk. Section 4 provides a discussion of key insights, limitations and future directions before drawing main conclusions in Sect. 5.

Our study design is illustrated in Fig. 1. We quantify uncertainties and their interaction in projected flood hazards using 12 climate models (CMs) contained in the Coupled Model Intercomparison Project Phase 6 (CMIP6) ensemble (https://esgf-node.llnl.gov/search/cmip6/, last access: 6 June 2020), forced using three Shared Socioeconomic Pathways (SSPs) scenarios. We use five bias correction techniques (change factor – CF; double gamma distribution quantile – DGQM; Birnbaum distribution quantile mapping – BQM; single gamma distribution quantile – SGQM; empirical quantile – EQM) to post-process climate model outputs (daily precipitation and air temperature time series). The daily bias-corrected precipitation and temperature data are used as input to the Génie Rural à 4 paramètres Journalier (GR4J) hydrological model (Perrin et al., 2003). The generalised likelihood uncertainty estimation (GLUE) technique is used to quantify GR4J parameter uncertainty. In assessing future flood hazard, the sensitivity of results to different extreme value distributions is examined. In addition to evaluating each source of uncertainty independently, we use variance decomposition to quantify the contribution of each methodological choice and their interactions to the overall uncertainty in assessing flood risk. The climate change impact is evaluated based on relative future changes in the magnitude of floods for the 2020s (2010–2039), 2050s (2040–2069) and 2080s (2070–2099) with respect to the reference period (1976–2005). The following sections provide further details on the study catchments and each stage of the modelling chain employed.

We examined future changes in flooding for four catchments in Ireland, including within our modelling chain the uncertainties derived from 12 CMIP6 CMs, five commonly used bias correction techniques, hydrological model parameters and three flood frequency distributions. We found that maximum precipitation in all catchments is projected to increase over the coming decades; however, changes are not uniform across catchments. The largest change in annual maximum precipitation was found for the Newport catchment, with ensemble median increases becoming progressively larger for the 2020s, 2050s and 2080s (27 %, 40 % and 60 %, respectively). In the Slaney catchment, increases in annual maximum precipitation between the 2050s and 2080s are not statistically significant. In the Blackwater catchment, most of the climate models indicate a positive change, but these have a wide range of uncertainty relative to other catchments (from 0 % to 55 % in 2050s and 0 % to 60 % in 2080s). We find projected changes in flood magnitudes to be broadly proportional to changes in maximum precipitation in all catchments. Projected changes in flood magnitude are higher using the SSP585 and lower using SSP126 climate scenarios. Overall changes in precipitation and air temperature are broadly consistent with previous findings using CMIP5 (Broderick et al., 2019) in Ireland.

Our findings demonstrate the large uncertainties associated with projected flood magnitudes. Over recent years, much research has sought to explore the dominant sources of uncertainty in climate change impact assessments. Similar studies have identified climate models as being the dominant source (e.g. Sulis et al., 2012; Hattermann et al., 2018). Indeed, Addor et al. (2014) highlight that there seems to be a general agreement in the literature on the dominant contribution of climate models to the uncertainty in discharge projections. In contrast, our findings demonstrate the challenges of generalising about the dominant components of the cascade of uncertainty at the catchment scale, important for understanding future flood risk. Using both additive chains and variance decomposition to examine components of the modelling chain employed, our results show that not only does the contribution and importance of different components of the modelling chain vary across catchments, but that the dominant source of uncertainty can be completely different. Moreover, the dominant sources of uncertainty vary depending on which approach is used to evaluate uncertainty. Using additive chains, in two of our study catchments the climate model ensemble was the dominant source of uncertainty, as was expected. However, for the Boyne and the Slaney catchments, the extreme value distribution employed for flood frequency analysis and the bias correction method used to adjust raw GCM output contribute greater uncertainty to future flood risk than the ensemble of CMs employed. Using variance decomposition allows for the interaction of components of the modelling chain to be evaluated and quantified and offers a fuller assessment of uncertainty. Our results show that in the Slaney and Blackwater catchments the climate model ensemble used dominated the cascade of uncertainty. In the Slaney catchment, the second most important component is the bias correction technique used, while in the Blackwater catchment the frequency distribution model is the second most important component. In the Newport catchment, the flood frequency distribution model ranks as the most important component of the cascade of uncertainty, followed by the climate model ensemble. Finally, in the Boyne catchment, the climate model ensemble ranks as only the third most important component of the uncertainty cascade in simulating flood quantiles, with the bias correction methods coming first and the flood frequency distribution second.

We hypothesise that catchment characteristics may influence the dominant components of the cascade of uncertainty in each catchment. The Slaney and Blackwater catchments, where climate models dominate, are among the largest in our sample, with groundwater contributing significantly to runoff in both. In the Blackwater catchment, topography has a large bearing on flood response time within the catchment, with flood peaks moving more rapidly from elevated sub-catchments into the main channel, and may be a reason as to why the flood frequency model ranks as the second dominant component in the cascade of uncertainty. The Slaney catchment drains the southern component of the Wicklow Mountains before reaching shallower elevations as it flows south through Carlow and Wexford. The variation in topography and, hence, precipitation from upper to lower reaches may be a reason as to why bias correction methods emerge as the second most important component of the uncertainty cascade in this catchment. The Newport catchment is the smallest in our sample, with the flood response influenced more strongly by varied land use cover (forestry and rotational cropping) and the presence of a lake in the upper reaches of the catchment. The flood frequency distribution dominates the cascade, followed by the climate model ensemble used. Finally, the Boyne catchment has been heavily influenced by arterial drainage works, resulting in the insertion of field drains and the deepening and widening of the river channel (Harrigan et al., 2014). These interventions have resulted in a faster rainfall–runoff response. In the Boyne catchment, therefore, projected changes in future rainfall are likely to have a more direct bearing on the flood response, with uncertainty in bias correction and the flood frequency distribution dominating the cascade of uncertainty.

Across each of our catchments and in the context of the modelling chain employed, the uncertainty from hydrological model parameters is the smallest component of total uncertainty in the estimation of future flood hazard magnitude. Bastola et al. (2011a, b) identify hydrological model uncertainty as being important in a previous assessment of change in flood risk, while Mizukami et al. (2016) rank hydrological model uncertainty to be of greater importance than bias correction approaches in USA catchments. We note that we do not consider hydrological model structure uncertainty in our modelling chain, and that our decision to select only the top 300 behavioural parameter sets, rather than all those identified, may be a factor here. This highlights how subjective decisions in the application of hydrological models may have important implications for uncertainty in future projections (Melsen et al., 2019). We also find that the contribution to total uncertainty from the interaction of components within the modelling chain is substantial – up to one-quarter of the uncertainty range considered in two catchments.

While uncertainty decomposition may be useful in helping scientists navigate the cascade of uncertainty by prioritising the key components of the uncertainty cascade to sample (Clark et al., 2016) for a particular catchment, our small sample makes it difficult to determine how catchment characteristics may be informative in identifying the dominant sources of uncertainty a priori. There are some indications that catchment properties and the characteristics of observed precipitation may influence which uncertainties may dominate. Vetter et al. (2017) used variance decomposition to examine the contribution of modelling chains (GCM, emissions scenario – in the form of Representative Concentration Pathways or RCPs – and hydrological model) to total uncertainty in different parts of the flow regime. In their study, they examine 12 large catchments globally. While Vetter et al. (2017) highlight the dominance of CMs in total uncertainty across many catchments, they also highlight that the share of emissions scenario and hydrological model contributions to overall uncertainty can differ between catchments. Similar to our findings, Vetter et al. (2017) highlight the importance of interacting components of the modelling chain employed.

As our study has shown, the evaluation of uncertainties and their interaction in future flood risk assessment with climate change is a critical step in understanding future impacts and informing adaptation decision-making. However, results depend on the modelling chain employed and the method used to quantify uncertainties. Given the possibility of examining interactive components of a modelling chain, we recommend the use of ANOVA-based techniques to decompose the cascade of uncertainty. We highlight, however, that our modelling chain is not complete, and that the work presented here is illustrative rather than exhaustive. We do not, for example, include uncertainties in future flood estimates derived from different hydrological model structures with previous work, as indicated above, showing that hydrological model structure and complexity may be an important component of future uncertainties. Other potential sources of hydrological model uncertainty are also omitted. For example, we neither evaluated the transferability of parameter sets to represent the future rainfall–runoff response (e.g. Broderick et al., 2016) nor did we evaluate the influence that selection of different objective functions may have on the identification of behavioural parameter sets. Westerberg et al. (2020) also highlight the potential importance of uncertainty in the discharge data used to calibrate hydrological models. In assessing different flood quantiles, we use only 30 years of annual maxima to extrapolate to return periods outside our record length. Such extrapolations should be treated cautiously but are prevalent in many studies (e.g. Kay et al., 2006; Lawrence, 2020; Meresa and Romanowicz, 2017), while rarer floods are also associated with design standards (e.g. 100-year flood) and are therefore of interest to decision-makers. We also assume that flood processes remain stationary within the 30-year windows used to evaluate changes, while other factors that can influence rainfall–runoff response, such as land use change, remain unchanged from the calibration period used to train our hydrological models.

This study evaluates changes in future flood magnitude with climate change for four Irish catchments using a modelling chain incorporating 12 CMs comprising the CMIP6 ensemble, five bias correction techniques, hydrological model parameter uncertainty and the use of extreme value distributions. Our findings suggest increasing flood hazards over the coming decades in all catchments, with changes in flooding largely consistent with changes in maximum precipitation. However, uncertainties in future flood changes are large and increase with time and flood quantile. Using ANOVA, we decompose uncertainties in future flood quantiles to examine how individual components of our modelling chain and their interactions contribute to overall uncertainty. Our results show how the dominant sources of uncertainty may vary on a catchment basis, highlighting the inability to generalise on the dominant components of uncertainty across catchments. Across all four catchments, the climate models, bias correction methods and the extreme value distributions used to evaluate flood return periods were differentially dominant, while the uncertainty derived from the interaction of various components was substantial in all catchments. Our work shows the value of ANOVA methods in visualising and quantifying the uncertainty cascade at a catchment level, which will be helpful in navigating the uncertainties associated with future flood risk for better adaptation decision-making. While our sample is small, the dominant components of uncertainty in future flood risk may be related to catchment characteristics. We, therefore, recommend that future work seeks to better understand the link between the key components of the cascade of uncertainty and catchment characteristics. The ability to do so would offer considerable advantages in knowing a priori which sources of uncertainty should be targeted as priority.