Developing and evaluating decision support indicators (DSIs) of climate change impacts on flood and drought: a case study in Western Norway

In the recent decade, Norway has experienced several destructive hydrological extreme events, such as floods in 2014, 2017, 2018, 2022 and 2023 and drought in 2018. These events led to significant negative impacts on agricultural production, hydropower production, infrastructure, etc., and raised great public attention on hydrological hazards. Yang & Huang (2023) assessed the attribution of the historical flood and drought events in 50 Norwegian catchments and found that more than 62% of the events in the recent decade were magnified by climate change. Such a finding is not surprising because the northern high latitudes have experienced the strongest warming since 1980 among all regions in the world (IPCC 2021), and changes in hydrological extreme characteristics have already been observed in Norway, e.g., an increasing trend in flood frequency (Mangini et al. 2018) and shift in spring flood timing (Vormoor et al. 2016; Blöschl et al. 2017).

Motived by the historical changes and the provision of a future warming climate, numerous climate impact studies on hydrological hazards have been conducted in Norway (Wong et al. 2011; Lawrence 2020). Ideally, the knowledge gained from research should serve as the basis for climate actions to alleviate emerging risks. However, big gaps exist that hinder effective communication between the scientific community and stakeholders, such as the gaps between the scientists' use of theoretical concepts and stakeholders' reality, uncertainties in predictions of climate impacts, the difference in the geographic scale of climate data and stakeholder needs, and the need to manage natural climate variability, etc. (Klein & Juhola 2014).

To bridge the gaps between the scientific community and stakeholders, indicators have been proposed as a foundational decision support product that shows the key impacts of climate change and is relevant for target stakeholders. They can be status, rates of change or trends of a phenomenon based on measured data and modelled data, to assess scientific understanding, to communicate and inform decision-making and to denote progress in achieving management objectives (Kenney et al. 2016). Various climate indicators have been developed based on global datasets for a relatively quick and effective analysis of hydrological impacts (Merks et al. 2022). Due to the coarse resolution of data, the global indicators focus more on mean state, such as mean temperature, mean precipitation, mean runoff, mean discharge, mean soil moisture and mean aridity. At the national or continental level where high-resolution data is available, indicators were developed to account for high- and low-flow characteristics as well as for extremes (Samaniego et al. 2019; Peters-Lidard et al. 2021).

In order to assess future changes in extreme events, numerous flood and drought indicators have been developed based on observed or modelled data. A review of flood indicators can be found in Maranzoni et al. (2023), and 50 drought indicators that are being applied across drought-prone regions were summarized by WMO & GWP (2016). Among them, flood magnitude, flood frequency and Palmer Hydrological Drought Index are frequently used indicators to analyse river flood and river drought in literature.

Based on the various indicators used in literature, Sörensen et al. (2024) further developed the concept of decision support indicators (DSIs) for extremes, i.e., indicators that are specifically designed to inform local and regional stakeholders on the characteristics of a predicted or ongoing event to facilitate decision-making. They classified the DSIs into three classes: conventional DSIs, impact-based DSIs and event-based DSIs. According to the classification, most of the flood and drought indicators in literature can be considered as conventional DSIs. The advantage of the conventional DSIs is that they are easy to calculate based on the observed or modelling data, but the disadvantage is that they are hard for non-experts to understand. The indicators, which assess the impacts of flood and drought, are defined as impact-based DSIs. This class of DSIs, in contrast, is easy to understand by non-experts but often difficult to calculate because it requires extensive data on the impact on society, ecology and the economy.

To overcome the disadvantages of the conventional and impact-based DSIs, Sörensen et al. (2024) proposed a new class of DSIs, event-based DSIs, which scale a projected state (e.g., the duration of a drought event or the peak flow of flood event) relative to a locally well-known historical event. This class of DSIs is intended to be more tailored to decision-makers' needs than the conventional DSIs and requires considerably less data than impact-based DSIs. The concept of the event-based DSIs is simple. First, certain historical events that would still be prominent in the minds of the local population and stakeholders should be defined as baselines. Second, any future changes or events are expressed relative to these baselines. However, the work of Sörensen et al. (2024) remains to date conceptual as the authors neither present a methodology to calculate the new class of DSIs nor evaluate whether the proposed event-based indicator outperforms existing and readily available indicators. Here, we see a need to develop methodologies to calculate the event-based DSIs for climate impact assessment because it is not straightforward to link an observed historical event with model-based simulation results driven by climate model output. In addition, it has so far not been proven that local stakeholders and decision-makers perceive event-based DSIs as useful and potentially superior to existing indicators. Hence, it is important to hear the opinions of the end-users.

In order to fill these gaps, this study aims to develop methodologies for calculating event-based DSIs and to evaluate the usefulness of all three classes of DSIs for climate impact assessment and climate actions by learning about users' perceptions. We focus on western Norway as a case study and include four conventional flood indicators, one event-based flood indicator, three conventional drought indicators, two event-based drought indicators and one impact-based drought indicator. All the indicators are calculated based on an ensemble of hydrological projections driven by three combinations of GCM (global climate model) and RCM (regional climate model) outputs under two RCP (representative concentration pathway) scenarios. The definitions, methodologies and results of all indicators were summarized in questionnaires and sent to the stakeholders in the region to evaluate the understandability, importance, plausibility and applicability of the indicators. The methodologies applied for the conventional and event-based indicators as well as the evaluation method in this study can be easily transferred to other regions of the world. The evaluation results can also serve as a useful basis for selecting flood and drought indicators for communication with stakeholders.

Due to the high amount of precipitation, Vestland county frequently experiences rainfall and snowmelt events that produce flooding and landslides, causing considerable threats to human health, local communities and infrastructure. Recent flood events occurred in 2014 and 2018. Vestland county also experienced severe and prolonged water deficit periods in recent years, such as 2018, 2006, 2002/2003 and 1995/1996, with substantial impacts on water supply and hydropower production.

In this study, we selected 29 unregulated catchments to calculate the DSIs (Figure 1 and Table 1). Note that three gauges located outside of Vestland county were also included in this study because they are included in the model for estimating energy inflow in this region. The gauges at the outlet of these catchments have long-term observed daily discharge data from 1983 to 2020 with less than 5% missing data. The size of the catchments varies from 3.3 to 1,091 km². The catchments have a seasonal variation in snow cover and several catchments receive a considerable runoff contribution from glaciers.

Since this study focuses on river flood and river drought, all DSIs are calculated based on the observed and modelled river discharge using the spatially distributed version of the HBV (DistHBV) model (Beldring et al. 2003). The DistHBV was developed for calculating the water balance for 1 × 1 km² grid cells, which can be subdivided into lake area, glacier area and up to three vegetation types. The model calculates subgrid scale accumulation and ablation of snow, glacier ice melt, interception storage, distribution of soil moisture storage, evaporation, groundwater storage, runoff response and lake evaporation.

Snowmelt and ice melt are calculated by degree-day methods. Instead of the temperature-based method for computing potential evaporation in the version of Beldring et al. (2003), the present DistHBV calculates potential evaporation using the Penman–Monteith method (Huang et al. 2019). The actual evapotranspiration is calculated based on potential evaporation, field capacity and permanent wilting point. The runoff is simulated by two non-linear parallel reservoirs representing direct discharge and the groundwater response.

Since the DistHBV model assumes that glacier-covered areas do not change with time, although glacier mass balance is negative, ice melt water can be overestimated under warming climate scenarios. To overcome this problem, catchments with at least 5% of the area covered by glaciers had their glacier-covered areas reduced manually following the approach used in the ‘Climate in Norway 2100’ report (Hanssen-Bauer et al. 2017).

• Model simulations for the control period (1976–2005): 100% of the present glacier area in all grid cells was used.

• Model simulations for the near future (2031–2060): For RCP4.5, 80% of the present glacier area in all grid cells was used. For RCP8.5, 50% of the present glacier area in all grid cells was used.

• Model simulations for the far future (2071–2100): For RCP4.5, 40% of the present glacier area in all grid cells was used. For RCP8.5, 20% of the present glacier area in all grid cells was used.

To set up the DistHBV model for the studied catchments, we used a digital elevation model (DEM) from the Norwegian Mapping Authority, and land cover raster maps derived from the National Land Resource Map (Ahlstrøm et al. 2014) and a remote sensing-based forest map (SAT-SKOG, Gjertsen & Nilsen 2012) with 1 km horizontal resolution.

To calibrate and validate the model in the historical period, we used the precipitation data from the seNorge2 precipitation dataset (Lussana et al. 2018), the downscaled Tmax, Tmin and wind data from the Norwegian ReAnalysis 10 km (NORA10) product (Reistad et al. 2011), at the same 1 km horizontal resolution from 1982 to 2020. Based on the Tmax, Tmin, P and wind data, we applied the Mountain Microclimate Simulation Model (MTCLIM) (Bohn et al. 2013) to generate the remaining climate variables required by HBV (Tmean, SWR and VP) for the same period. More details on the observed historical forcing data are available in Huang et al. (2019).

In order to analyse the climate change impacts on streamflow, the calibrated hydrological models were then driven by climate projections for one control period 1976–2005 representing present-day climate and two future periods 2031–2060 and 2071–2100. The climate projections were generated by three combinations of GCMs and RCMs from the Coordinated Regional Climate Downscaling Experiment (CORDEX) project, i.e., EC-EARTH/KNMI-RACMO, IPSL-CM5A/SMHI-RCA4, MPI/CCLM, for two emission scenarios representing a moderate-emissions pathway (RCP4.5) and a high-emissions pathway (RCP8.5). The RCM results were bias-adjusted and downscaled to the 1 km² resolution based on observations using the empirical quantile method (Gudmundsson et al. 2012).

The selection of flood and drought DSIs was based on a workshop in 2019 with stakeholders, who represent different sectors. The results from the workshop showed that the stakeholders were most concerned about flash floods in small catchments, drainage problems on forest roads and changes in precipitation seasonality that cause drought stress in forests and challenges for the dimensioning and operation of reservoirs. Some stakeholders also mentioned that 50-year and 100-year extreme events are relevant to their operations. It also revealed that there is a lack of critical information or services to support and inform decision- and policymaking in the region. The detailed results from the workshop were documented by Beldring et al. (2020).

Based on these findings, a list of DSIs was proposed for climate impact studies in Vestland county as presented in Sörensen et al. (2024). This list includes four conventional DSIs and one event-based DSI for floods, and three conventional DSIs, two event-based DSIs and one impact-based DSI for droughts. Note that there is no impact-based flood DSI because we did not have sufficient data and tools to estimate the impacts for the studied catchments. The details of the proposed DSIs are explained below.

In this study, flood events were extracted from the daily discharge time series using a peak-over threshold (POT) approach for the control period and two future periods, respectively. The POT approach leads to a more comprehensive selection of events in comparison with the block maxima approach (e.g., annual maximum floods) as there is more than one flood event per year in most catchments in the study area (Vormoor et al. 2016). The thresholds were selected so that on average two independent peaks per year are represented. As a result, 60 flood events were selected for each catchment and in each period and served as the basis for all flood DSI calculations. In this section, we will briefly introduce the DSIs that are widely used in literature but give more attention to the new DSIs in this study. More detailed information on the definition and methodology of each DSI is presented in the Supplementary material.

Four conventional DSIs were selected for climate impact assessment on river flood: the future return period of today's 100-year and 50-year floods, the change in the magnitude of 100-year and 50-year flood events, the shift in the mean day of flood occurrence, and the shift in dominant flood generation processes (FGP). The first two DSIs were calculated based on flood frequency analysis, which was widely used for flood analysis in Norway (Lawrence 2020). The selected flood events were fitted with a generalized Pareto distribution (GPD) to generate the flood frequency curve (FFC) for each period. Present-day 50- and 100-year flood levels were estimated based on the FFC for the control period, and the future return period as well as the magnitude of today's 50- and 100-year flood were estimated based on the FFC for the future periods. The calculation of the mean date of flood occurrence was based on the method used by Blöschl et al. (2017) for all selected flood events in one catchment in different periods. The indicator ‘shift in the mean day of flood occurrence’ is calculated as the difference between the mean date of flood occurrence in present-day climate and the mean date of flood occurrence under future climate.

The method of identifying FGP classes combined the information selection method used by Berghuijs et al. (2016) and the classification criteria for Norwegian catchments suggested by Vormoor et al. (2016). Four FGP classes were defined for this region: rainfall-driven flood, snowmelt-driven flood, ice melt-driven flood and mixed-processes floods. For each flood event, we selected the following information from the climate model outputs and the HBV model simulation results of the corresponding catchment:

• The total amount of rainfall within 4 days before the event.

• The total amount of snow reduction in the week before the event.

• The total amount of glacier melts in the week before the event.

Then we summed the water amount from the three sources and calculated the share of each source. If the share of rainfall is larger than 0.667, then it is a rainfall-driven flood. If the share of snowmelt is larger than 0.667, then it is a snowmelt-driven flood. If the share of ice melt is larger than 0.667, the flood is ice melt driven. If none of the above-mentioned processes dominates, it is a mixed-processes flood.

The last flood DSI is an event-based indicator and aims to assess the direction of change (decrease/increase) in the return period for a given event size (reference event). In this study, the flood of 2014 was used as the reference because this flood was the highest on record since 1908 in some catchments in Vestland and caused damage of more than 100 million Norwegian crowns NOK. Since the control run of climate models only statistically reproduces the climate conditions of the 20th century, the 2014 flood event is not explicitly represented in the control simulations. Hence, we applied extreme event statistics to link the observed floods with the simulated floods in the control runs, which is a similar method linking the observed flood damages with the simulated floods used by Hattermann et al. (2014). Firstly, we estimated the return period of the 2014 flood based on the FFC fitted on observed flood events. Secondly, we estimated the flood level based on the return period of the 2014 flood and the FFC fitted on the flood events from the control simulation. Lastly, we estimated the return period based on the flood level from the last step and the FFC for future scenarios. The illustration of the methodology is presented in the Supplementary material.

In this study, three conventional DSIs were chosen to inform planning and climate change adaptation to drought in Vestland county: change in average drought duration, change in average drought severity and change in seasonal low flow. The first two DSIs have been widely used for climate impact assessment on droughts in Norway (Wong et al. 2011; Yang & Huang 2023). The same as the previous drought studies in Norway, we only focused on river droughts in the snow-free season from May 15 to October 15 which are caused by low precipitation and high evaporation. Following Wong et al. (2011), drought events were identified when flow is below a specific threshold, which was defined as the discharge level that exceeded 80% of the time in the control period, termed Q80 in the following. Q80 is an established and widely used drought threshold. A value with an 80% exceedance probability has a low probability of occurrence (20%) and will be a sufficiently low discharge to define the start/end of a drought period. For each catchment, the Q80 threshold was calculated independently for each day of the year (DOY). Daily Q80 thresholds were subsequently smoothed with a moving average filter (two-sided, window size: 31). A 7-day moving average procedure was applied on the daily discharge to pool dependent droughts and remove minor droughts. The annual maximum droughts were accounted for to calculate the average drought duration and severity (deficit volumes) in the control and future periods. The seasonal low flow was defined as Q80 in autumn (September to October), winter (December to February), spring (March to May) and summer (June to August). Detailed descriptions of these DSIs are presented in the Supplementary material.

Two event-based DSIs that relate to a severe event in the historical period were developed for drought in this study: (1) change in maximum drought magnitude between the control and future periods and (2) change in the return period of the 1996 drought. For each catchment, we determined the most severe drought event in terms of deficit volume both under present-day climate (control period) and under future climate. In order to make deficit volumes comparable between catchments, deficit volume was divided by the 30-year mean daily discharge of the respective time period in each catchment. The resulting normalized deficit volume (NDV) can be interpreted as the number of days with mean discharge that would be required to offset the drought's accumulated deficit volume. The indicator ‘change in maximum drought magnitude’ is finally calculated as the percentage difference between NDV under present-day climate and NDV under future climate.

The 1996 drought was selected as the reference historic drought event because it was the most severe drought event in 19 catchments (out of 29) in the study period. The drought frequency curves were determined by fitting a generalized extreme value (GEV) distribution to the annual maximum deficit volumes in different periods. As with the event-based flood indicator, the return period of the 1996 drought event was estimated based on the drought frequency curves fitted on observed drought events. Then the deficit volume of this return period was estimated in the drought frequency curve for the control simulation and linked to the drought deficit volume in the drought frequency curve for the future periods. The illustration of the methodology is presented in the Supplementary material.

The reduction in potential hydropower production (PHP) caused by drought events between the control and future periods is used as an impact-based indicator. This indicator informs on the effect of drought years, expressed as the percentage reduction in annual PHP compared with present-day mean annual PHP as a sum over the entire region. Hydrological model results for daily streamflow were combined with weekly energy production and catchment-specific energy equivalents to determine energy inflows. Changes in reservoir storage volumes, flood losses and losses caused by minimum flow requirements were considered, and the resulting daily time series thus depicts net energy inflow. The details of the method are described by Holmqvist (2017) and in the Supplementary material.

All the above-mentioned DSIs were summarized in one questionnaire, including the definition of DSI, methodologies to calculate the DSI, some examples of the results, a short explanation of the information in Norwegian and a questionnaire table (see Supplementary material). The questionnaire was designed such that each DSI was presented on one page. Considering that the stakeholders have different background knowledge, we tried to avoid specific scientific terms and explain the DSIs as simply as possible. Hence, the texts in the questionnaire may not read the same as the texts here. The methodology, which was the most theoretical part of the questionnaire, was explained at the back side of each page to avoid overloading non-experts, but still providing all relevant information to those who were interested.

In order to avoid overwhelming the stakeholder with a large number of questions, we opted for covering all relevant aspects of the single DSIs (i.e., understandability, importance, plausibility and applicability) in one general question each. The questionnaire table included four statements for each DSI and the stakeholders were required to give their opinions on a total of 44 statements (11 DSIs * 4 statements). The four statements were also formulated as simply as possible to ensure all stakeholders with different hydrological backgrounds can understand them. The first statement ‘The explanation of the indicators and the results are easy to understand’ aims to investigate whether the explanations of the DSIs and the illustration of the results are clearly presented in the questionnaire. The second statement ‘The indicators are important’ aims at quantifying the importance of the DSIs from the stakeholders' point of view. The third statement ‘The projected climate impact results are plausible’ aims to know whether the stakeholders would trust the indicators' results. Finally, the fourth statement ‘You will consider the indicators and the projection results for decision makings’ investigates whether the stakeholders would apply the DSIs in future planning processes. There are five options for the stakeholder to express their opinions on these statements: ‘totally agree’, ‘partly agree’, ‘don't know’, ‘partly disagree’ and ‘totally disagree’.

A short description of the study area, hydrological modelling and data was sent to the stakeholders together with the questionnaire (see Supplementary material). The stakeholders were asked to fill out a Google online questionnaire, which only contained the statements and answer options. Based on their requests, we also arranged one meeting to explain the details and discuss their work on flood and drought adaptations directly.

We selected the stakeholders from the public and private institutions, which are responsible for hydropower production, management of water resources and mitigation of natural hazards on infrastructures and society for the whole region. We sent the questionnaire to seven key stakeholders, but only six stakeholders gave us feedback. These stakeholders include Statkraft Hydropower Company, Norwegian Forum for Natural Hazards, Norwegian Public Roads Administration, BaneNor Railway Infrastructure Manager, Voss Municipality Administration and Lærdal Municipality Administration.

The DistHBV model was calibrated and validated against daily discharge for each catchment separately. Table 1 summarizes the results of the three statistics criteria: NSE, LNSE and BIAS. In general, the model shows good model performance for most catchments in the calibration period. About 19 and 29 of the catchments exhibit an NSE and LNSE higher than 0.7. The model also shows good estimates on water amount with BIAS within ±0.05 for 25 catchments. In the validation period, LNSE is still higher than 0.7 for 26 catchments, indicating good performance in terms of low flows. However, the model performance is degraded for high flow and water amounts, with NSE larger than 0.7 and BIAS within ±0.05 in 14 and 10 catchments, respectively. Nevertheless, the model still shows satisfactory performance for most catchments, with NSE larger than 0.5 and BIAS within ±0.10 in 27 and 22 catchments, respectively. The poor model performance in some catchments may be attributed to the complex terrain in this region, where the gridded forcing data cannot capture the horizontal variation, especially for small catchments.

Since the main objective of this study is to develop and evaluate DSIs rather than to assess climate impacts on floods and droughts systematically, we present all results for both future periods and RCP scenarios in the Supplementary material. In general, the effects of climate change are most pronounced under the RCP8.5 scenario and in the far future. Hence, in this section, we will mainly show the ensemble median results under the RCP8.5 scenario in the far future period to illustrate the strongest changes in floods. In addition, more attention will be paid to the DSIs that have been rarely applied in previous studies.

The conventional DSI, changes in FGP, has not been extensively applied in the climate impact assessment in Norway yet, but it provides insights on the causes of floods in the future. Figure 3(d) summarizes changes in the number of flood events caused by rainfall, snowmelt, ice melt and mixed processes under both RCP scenarios and in the near and far future periods, with respect to the number of the events in the control period, respectively. The results show that there will be more rainfall-driven floods and fewer snow-driven floods under both scenarios in both periods, mainly due to an increase in rainfall instead of snow in warmer conditions. The number of ice melt-driven floods and mixed-process floods is generally reduced except under RCP4.5 in the near future. This pattern of changes in FGP is consistent in almost all studied catchments (see Supplementary material).

As for flood DSIs, we will mainly focus on the ensemble median results under the RCP8.5 scenario in the far future period to show the largest changes in drought characteristics in this section. The results of all DSIs for both periods and RCP scenarios can be found in the Supplementary material.

Besides the DSIs for summer drought, we analysed the changes in seasonal low flows as conventional DSIs for droughts (Figure 5(c)) because low flow seasonality is an important indicator for hydropower production. We find a consistent and strong increase in winter low flow levels of an average of +163%. A pronounced regional pattern emerges in spring with decreases in low flow levels along the coast and increases in the interior catchments. We find a general decreasing trend in summer where low flow levels decrease by more than −25% in 18 out of 29 catchments while increases (of up to +22%) are found in only three catchments. For autumn, the results show a general trend towards increasing seasonal low flow levels with increases found in 22 out of 29 catchments, in 11 of them by more than 25%.

The second event-based DSI for drought uses the estimated return period of the 1996 summer drought as the reference (Figure 6(c)). Depending on catchment, the drought had a return period of 3–135 years with a tendency to higher return periods (more severe and rare events) in the southeast. In 15 out of 29 catchments, the event had a return period of 50 years or more. Under the high-emissions climate scenario RCP8.5, we observe a markable decrease in return period by the end of the 21st century in 27 out of 29 catchments (Figure 6(d)). According to our simulations, a drought event of similar magnitude to the 1996 drought will have a return period of less than 10 years in 25 out of 29 catchments and less than 5 years in 23 out of 29 catchments. It indicates that the 1996 drought may occur every few years in most of the studied catchments in the future.

The event-based flood DSI does not show higher understandability than the conventional DSIs of changes in flood magnitude and return period. Instead, the stakeholders from the hydropower sector doubted the importance and applicability of this DSI. Since the hydropower sector stakeholders have a thorough understanding of hydrology, they suggested that the event-based flood DSI may be more useful for communication with the general public or general stakeholders without a background in hydrology.

Regarding drought DSIs, the stakeholders from the railway authority are not interested in drought conditions at all and they gave the answer ‘don't know’ to all statements. Similarly, the stakeholder from the road authority answered ‘don't know’ in many cases because roads are not significantly affected by droughts. Excluding the ‘don't know’ answers from these stakeholders, the conventional drought DSIs we selected in this study seem more difficult to understand for the stakeholders than the flood DSIs, probably because drought itself is a more complicated process than a flood. Only one or two stakeholders can fully understand the conventional DSIs and the stakeholders from the Voss municipality could not understand almost all of them based on the information provided (one totally disagrees with two DSIs and one partly disagrees with one DSI). Interestingly, although this stakeholder could not understand the DSIs, they thought the DSIs related to changes in drought duration and severity were important and plausible and would consider them in decision-making. In contrast, the representative from the road authority could partly understand the conventional DSI of changes in seasonal low flow, but they would neither consider it important nor use it for future planning.

The event-based DSI of change in the return period of the 1996 drought is the easiest drought DSI for the stakeholders to understand. In contrast, another event-based DSI, which is based on the maximum drought event in each catchment, has the lowest acceptance by stakeholders in terms of all statements. It indicates that the selection of events is very important and that well-known events are much more accepted as a reference than statistically important events. Although the DSI based on the 1996 drought can be understood by most stakeholders, it is not considered similarly important or as applicable as the conventional DSI of changes in drought magnitude and duration. Only two stakeholders would consider this DSI in their future planning while the others answered ‘don't know’ or ‘probably don't use it’.

The impact-based drought DSI (reduction in PHP) can be partly or fully understood by most stakeholders and it is considered as an important DSI by four stakeholders. However, it seems more difficult for the stakeholders to validate the projection results compared with the conventional DSIs related to changes in drought magnitude and duration. It is natural because additional calculation on the impacts brings more uncertainty to the results. The stakeholders from the road authority will not consider this DSI for future planning because the impact is not relevant to them. This indicates that applicability of the impact-based DSIs is restricted to the targeted sector while conventional DSIs are more universal and can be used by various stakeholders.

Based on the classification of DSIs proposed by Sörensen et al. (2024), the conventional DSIs are easy to calculate based on the observed or modelling data, but they are hard for non-experts to understand; the impact-based DSIs are easy for non-experts to understand but often difficult to calculate due to high demand of data; the new proposed event-based DSIs are supposed to overcome the problems of the former two classes of DSIs and should be more tailored to stakeholders. However, the feedback from the stakeholders in the Vestland region does not fully confirm the expected functions of these DSIs for climate impact studies. We will discuss what was confirmed and what was not confirmed by the stakeholders in the following.

Firstly, the event-based and impact-based DSIs are indeed easier for the stakeholders to understand, especially when the conventional DSIs are beyond their expertise. In this study, the stakeholders have a better understanding of the event- and impact-based drought DSIs than of the conventional ones. However, we found no differences between the understanding of the two conventional DSIs for flood magnitude and frequency and the understanding of the event-based flood DSI. It shows that conventional DSIs are not always difficult for the stakeholders and the selection of the appropriate conventional DSIs may overcome the understanding problem. In this study, the selected conventional drought DSIs were widely used in scientific research in Norway but they did not seem to be the most appropriate choice when communicating with stakeholders. Generally speaking, flooding is the more prominent type of extreme event than drought due to the humid to hyper-humid climate in this region. This might explain why the stakeholder group had difficulties interpreting the metrics chosen to quantify drought risks. Hence, indicators, which are easier to interpret and calculate, should be explored and evaluated with stakeholders in future studies, e.g., streamflow drought index, SPI (standardized precipitation index), SPEI (standardized precipitation evapotranspiration index), etc.

Secondly, although the event-based and impact-based DSIs are easier to understand, they are not readily implemented in decision-making processes by all stakeholders. Several stakeholders have already gotten used to applying some conventional DSIs in the long-term planning. They need to develop methodologies to incorporate the new DSIs and might not be motivated to do so. During the discussion with some stakeholders, they acknowledged that the event-based DSIs may be more appropriate to convey climate information to the general public or stakeholders with no or little hydrological background. In addition, these DSIs may be more useful for quick or short-term decision-making processes, e.g., flood and drought forecasting, and they should be evaluated further by the corresponding stakeholders.

Thirdly, although the most severe flood and drought events at the regional scale were selected as reference events to calculate the event-based DSI, we found that event sizes varied considerably between different catchments and some catchments did not even experience flood or drought in 2014 and 1996. As a result, the event-based DSIs in this study provide various reference information and cannot give any useful information for the local stakeholders in the catchments which were not hit by those events. This raises important questions about the appropriate scale for selecting reference events which is not discussed in Sörensen et al. (2024). In this study, the selected stakeholders need to consider the regional behaviour of events and DSIs, not only conditions in ‘their’ catchment or administrative area. Although the return period of events and results for specific DSIs vary within a region, the heterogenous and random behaviour of meteorological and hydrological processes warrants an understanding from stakeholders that the worst case may occur in all catchments with similar hydrological conditions within the region. For a local case study focusing on one specific catchment and targeting local stakeholders, one would naturally choose the most severe event in the specific catchment as a reference. For large-scale studies, where both local and regional stakeholders are involved, different events may be applied to satisfy the needs of all stakeholders.

This study mainly focuses on the DSIs, which are used as tools to bridge the scientists' concepts and stakeholders' needs, but it does not tackle other challenges related to the communication of climate information. The stakeholders were only presented with ensemble median results for each RCP scenario because we were most interested in the stakeholders' perception of different types of DSIs. However, the stakeholders can be confused if they are confronted with uncertainty ranging from large ensembles of climate projections and it could be another challenge to communicate the projection results even though they understand the DSIs well. The stakeholders' perceptions of uncertainty in climate and climate impact projections lies beyond the scope of this study, partly because the ensembled size was rather small. Our next step will be to select the most appropriate DSIs based on the results of this study and apply them in the new generation of climate projections for the whole of Norway, which includes 10 GCM-RCM combinations for various RCP and SSP (shared socioeconomic pathways) scenarios. In addition, the effects of hydrological model structure and bias correction methods will also be considered to quantify the uncertainty of the whole modelling chain.

In this study, we applied different classes of DSIs of climate impacts on floods and droughts and evaluated them by analysing the perception of stakeholders in the Vestland region, Norway. The DSIs include four conventional flood DSIs, one event-based flood DSI, three conventional drought DSIs, two event-based drought DSIs and one impact-based drought DSI. Most conventional DSIs have been applied in previous studies in Norway and other countries in the world, while the event- and impact-based DSIs were specifically developed for the Vestland region in this study. We calculated the DSIs based on an ensemble of hydrological projections driven by three GCM-RCM outputs under two RCP scenarios and presented them in the form of fact sheets and questionnaires to stakeholders.

The feedback from the stakeholders was summarized in terms of the understandability, importance, plausibility and applicability of the DSIs. Regarding understandability, the three conventional DSIs of changes in flood magnitude, frequency and timing, the DSIs based on well-known historical events and the impact-based DSI are relatively easy to understand for most stakeholders, while the conventional drought DSIs we selected in this study are difficult for them. Excluding the ‘don't know’ answers from the railway and road authorities for drought DSIs, which are not relevant to these sectors, more than two-thirds of stakeholders agree or partly agree that all DSIs are important except for the DSIs of changes in flood occurrence and FGP. They also generally consider the projection results plausible for the conventional DSIs of changes in flood return period, magnitude, drought duration and severity as well as the impact- and event-based DSIs. However, they seem not to be well prepared to use the event-based DSIs in decision-making processes.

Based on the evaluation of DSIs, we conclude that the conventional DSIs are still preferred by stakeholders and an appropriate selection of conventional DSIs can overcome the understanding problems between the scientists and stakeholders. The DSIs based on well-known historical events are easy to understand but they are not readily implemented by all stakeholders in the decision-making process. The impact-based DSI is generally easy to understand and important but it is restricted to the relevant impact sectors so not all sectors can benefit from this DSI.

Although the event-based DSIs may not be implemented in decision-making processes by all stakeholders in this study, we acknowledge that they can be useful tools to convey climate impact information to the general public or stakeholders without sufficient background due to the advantage of easy understanding. In addition, they may be more useful for quick or short-term decision-making processes. The concept of event-based DSIs is new and further studies on their applicability and perception by stakeholders are required. The evaluation results of the DSIs demonstrate the gaps between stakeholders and the scientific community in this region more clearly and will serve as a basis to select the indicators for the next generation of climate impact projections in Norway. Future studies will also focus on the communication of projection uncertainties with stakeholders.

The authors thank the Research Council of Norway for funding the GlobalHydroPressure project no. 297298. The authors also thank the stakeholders from Statkraft Hydropower Company, Norwegian Forum for Natural Hazards, Norwegian Public Roads Administration, BaneNor Railway Infrastructure Manager, Voss Municipality Administration and Lærdal Municipality Administration for supporting this work.

This work was supported by The Research Council of Norway (Grant No. 297298).

All authors contributed to material preparation, data collection and analysis. Hydrological modelling was performed by S.H. The questionnaire was prepared by S.H. and S.E. The workshop was organized by S.E. and S.B. The contact with stakeholders was performed by S.B. The first draft of the manuscript was written by S.H. S.E. and S.B. commented on previous versions of the manuscript. All authors read and approved the final manuscript.

All relevant data are included in the paper or its Supplementary Information.

The authors declare there is no conflict.