Transient scenarios for robust climate change adaptation illustrated for water management in The Netherlands

The applied transient climate change scenarios consist of daily time-series of the period 2001–2100 for three relevant boundary conditions of the Rhine delta: (1) the Rhine discharge at Lobith, (2) precipitation and (potential) evaporation for six regions in the Netherlands and (3) sea levels at two key locations along the Dutch coast.

To construct the time-series daily weather information on temperature, precipitation and evaporation, and sea water levels for 1961 to 1995 were used as reference period (Rhine are derived from the so called CHR-OBS data, see Görgen et al 2010; sea level at Dutch coast: https://www.watergegevens.rws.nl/; Dutch stations: www.knmi.nl). Using the 1961–1995 reference period two synthetic time-series of 1000 years of daily temperature and precipitation were generated for the Rhine basin (figure 2, top-left) with a rainfall generator specifically developed for this basin⁷ (Beersma 2002). The same (resampled) 1000 year sequences of historical dates were used to get daily time-series of precipitation and evaporation in the Netherlands that are consistent with those for the Rhine basin. The two 1000 year time-series were split into time-series of 100 years, resulting in an ensemble of 20 members. These 20 members are equally plausible and only differ as a result of natural climate variability. They serve as the baseline for an ensemble of transient time series in which all members have the same climate forcing but in which the members differ again as a result of natural variability. In this way we derived 60 transient precipitation and temperature time-series (20 for the no climate change scenario, 20 for the G scenario and 20 for the W+ scenario).

Transient climate change scenarios were constructed by gradually transforming each ensemble member according to two so-called KNMI'06 climate change scenarios of the Royal Netherlands Meteorological Institute; a moderately warm scenario with a temperature rise of 1 °C in 2100 and a warm scenario with a rise of 2 °C (respectively denoted as G and W+; Van den Hurk et al 2007). The KNMI'06 scenarios represent an equilibrium and thus stationary climate for two projection 'years' (being 2050 and 2100). The monthly changes for these scenarios were used to adapt the synthetic 100 year sequences according to a classical delta-method (see e.g Lenderink et al 2007, Te Linde 2007), in which each daily value of the time-series is perturbed with the scenario-dependent change for that specific calendar month. To make these perturbed time-series transient for the period 2001 to 2100, the transformation coefficients for 2050 were linearly scaled between 2001 and 2100.

Each time-series was were used as input for a river basin model HBV (Bergström and Forsman, 1973, Lindström et al 1997) for the Rhine basin⁸ upstream of Lobith (Berglöv et al 2009) yielding transient climate impacted daily river flows for the Rhine at Lobith (figure 2). Figure 3 shows the yearly discharge maxima of four of these transient scenarios. Note that, for the experiments in example 1 the ensemble of transient time-series of the river discharges was constructed slightly differently: first equilibrium discharge time-series for each equilibrium climate change scenario were simulated with the HBV model (Te Linde et al 2010) and in a second step these equilibrium discharge time-series were made transient by applying a classical delta method to the discharge time-series and again linearly interpolating the monthly deltas (Haasnoot et al 2012).

Zoom In Zoom Out Reset image size

Sea level time-series along the Dutch coast were obtained through a (balanced) bootstrap technique (Efron and Tibshirani 1994). By sampling with replacement complete years from the 35 yr reference period, ten 100 yr series were constructed in which the day-to-day persistence is essentially preserved. These time-series were transformed into transient sea level time-series both for the upper and lower estimate of the sea level rise for each of the two selected KNMI'06 climate scenarios. To account for nonlinearity as a result of accelerated sea level rise as described in the KNMI'06 scenarios, sea level was increased linearly between 2001 and 2050 and between 2051 and 2100, with a higher rate in the latter period.

Appendix A describes which (combinations of) transient scenarios are used in the experiments.

Table 1 presents the characteristics of the three examples. The results are presented in the next section. In example 1 the scenario ensemble for the Rhine river discharge was used. Two types of experiments were done in this example. First, participants were informed on historical extreme discharge events and possible future climates that may result in more extreme low and high flows. Then, they were shown a 100 year ensemble member in subsequent 25 year periods (2001–2025; 2001–2050 etc). For this experiment, we selected four time-series that differ in climate variability (one with peaks in the beginning and one with no peaks in the first 50 years) and climate change (a lower and an upper estimate). Appendix B presents the selected time-series in combination with the other ensemble members. After each period, participants were asked whether or not they would take flood risk actions. A second experiment was done with the simulation game Sustainable Delta (Deltares, online Valkering et al 2012). In this game a group of participants had the assignment to develop a water management plan for the fictitious river stretch. As the future unfolded stepwise, the participants experienced changing boundary conditions and impacts in the delta (e.g., floods, droughts, socio-economic development) as a result of one of the transient scenarios, while they did not know whether or not it was driven by an underlying changing climate. Based on their experience and societal responses, participants decided whether or not to implement adaptation actions. The IAM model for stylized river (Haasnoot et al 2012) returned direct feedback to the participants on the impacts of the transient scenario and their policy actions.

For the second example, impacts of transient scenarios were assessed to identify whether or not and when adaptation is needed in the Rhine delta using the IAM model for the Rhine delta (Haasnoot et al 2014) that was driven by all climate related boundary conditions. The performance of the reference case (which assumes no adaptation) was evaluated for different scenarios against a-priori specified objectives. When a mismatch arises between the objectives and the time-dependent performance, an adaptation tipping point (ATP Kwadijk et al 2010) occurs and new actions are needed to achieve objectives again. This analysis yielded for each ensemble member a moment that an ATP occurs, i.e. the 'use-by' year of the status quo depending on how the future unfolds. Taking into account a lead time for implementation of action(s), decision makers can assess when they need to adapt. Likewise, after implementation of actions a new ATP might occur after a period of time. Therefore, similar assessments were done for a range of adaptation actions.

The third example addresses the identification of early warning signals for adaptation. Here, the river discharge was used as a signpost, i.e.—the information that one needs to monitor to assess the need for adaptation (Dewar et al 1993, Walker et al 2001). We then searched for threshold values for this signpost (referred to as triggers Walker et al 2001). Comparing observed values of signposts with their pre-specified trigger-values, enables one to decide whether adaptation decisions need to be taken (Hermans et al 2014). Although climate change can best be monitored early in the impact chain, for example by measuring temperature change, in practice adaptation actions are generally based on the (potential) impacts later in the impact chain that are closer to objectives, such as impacts potential casualties, flood damage, and loss of habitats. In the Rhine delta, the river discharge is a signpost that is closely related to water management objectives. Different trigger values, frequencies of occurrence and time slices were investigated and evaluated. A signal is given if the value for the scenario realization is outside the range obtained from the realizations without climate change. Ideally, a trigger value gives a justified and reliable signal for adaptation as soon and clearly as possible without false positive alarms.

The third transient scenario in figure 3 was shown step by step to the participants. After showing the period until 2025 and also after extending it to 2050, almost all participants responded that they would not take actions. After extending it to the year 2075 all participants responded that they would take action(s) immediately, as two peak discharges occurred. At some occasions, participants identified a trend in the river discharges that was attributed to climate change (between 2040 and 2060). After showing the entire time-series participants were surprised—even disappointed—not to see any peak discharges in the last period. Some participants then concluded that they had invested in adaptation too late (i.e. after the peak flows) and/or mis-invested as the damage already had occurred.

Next, the first 50 years of all four transient scenarios were shown to discuss the influence of climate variability and climate change on implications for decision-making. The first two time-series present (in 2016 and 2020) a similar situation to the major peak flows that subsequently occurred in the Netherlands in 1993 and 1995, triggering large-scale evacuation. Such a situation might offer a window of opportunity to implement rigorous anticipatory measures for a future where climate change may result in a larger occurrence of such events. However, events might occur at any time, as shown by the entire ensemble. All realizations are equally likely; 'early warning' peak discharge events might not occur before a major event takes place, while conversely, extreme peak events might not do so in the second half of the century. Such sequential issues are independent from a change in underlying climate. Accordingly, different adaptation pathways arose for the graphs shown in figure 3. Without extreme events the sense of urgency for adaptation may disappear. However, when peak discharges occur the impacts may be high. The time-series in figure 3 also show that for the nearby period the differences between the realizations with and without climate change are remarkably small: only on the long term the differences become visible (for the W+ scenario). Moreover, the first 50 years of the time-series at the bottom demonstrate that it is possible that climate change is happening, but that we do not see this in the occurring discharges.

In the experiments with the simulation game, the transient scenarios presented in figure 3 were played many times (>50 times). Although all sessions evolved quite differently due to the various backgrounds of the players and differences in negotiation results during the game, we did see some general steps in how the future unrolled in the game sessions. In most sessions with the transient scenario W+ ensemble member 8 (the second in figure 3) moderate actions were taken at time zero as the participants realized something needed to be done. Still, actions were limited to avoid spending too much money and due to the large variety in preferences for certain actions; the more far-reaching actions failed to get support for implementation during negotiations between the participants. At the next evaluation moment—after simulation of the first 20 years with the actions implemented—two peak discharges that caused flooding had occurred. Participants were then—surprised by these events—willing to implement major flood reduction measures. In the following simulation periods no flood events occurred; sometimes drought damage occurred as taken measures primarily targeted at flood reduction. Although participants then were satisfied about their actions, once we showed the discharges of other transient time-series they realized that they might have been lucky that in their realization no further major peak event had occurred.

In both experiments, despite the intention to act pro-actively and to anticipate on the future, adaptation actions were often determined in response to extreme events. Remarkably, even water policy professionals tended to respond reactively instead of pro-actively. For example, in a session with policy analysts involved in the Dutch Delta Programme, participants were confronted with W+ ensemble number 5 (fourth series in figure 3). After two periods of 25 years, without any severe impacts of flooding, one of the participants stated that nothing was happening, concluding that no measures needed to be taken. They were subsequently surprised by the two peak events in the period of 2055–2060, and then started to take rigorous measures. Some of the participants became quite frustrated by this scenario, even though it contained a plausible combination of a longer period without major events followed by events that well fall within the current design standards. Participants were often focused on the peak discharges, as these result in the most severe impacts in this case and they tended have less eye for the impacts of low flows. Only if they were satisfied about flood risk, focus was shifting towards drought risk management and impacts on nature.

These sessions raised the awareness that people—including water managers—tend to try to identify trends in the single transient scenario that they experience during a thought experiment—and what they will experience in the future. The occurrence of extreme peak flow events during a session was often seen an indicator that climate is changing, or in the evaluation after the game session the used discharge series was thought to be a realization with climate change. However, maximum yearly discharge is highly variable, as is natural climate variability and thus difficult for detecting trends, especially in a single realization. In contrast, the occurrence of low flow periods is less variable, and would be a better indicator of an underlying changing climate. However, due to peoples focus on floods, this indicator is generally overlooked.

With these experiments participants became aware that both climate change and climate variability are relevant for decision making. The willingness to take measures was remarkably driven by the occurrence of floods, and resulted in a responsive instead of anticipatory strategy. Also, participants were unable to see whether the transient scenario was with or without climate change, which they attempted to detect from the evolving time-series of river discharge, and in particular from the extremes.

Here, we illustrate the use of transient scenarios for identifying when to adapt for flood risk management in the Lake IJsselmeer in the Rhine delta. For this example the transient scenarios for sea level, precipitation and evaporation and river discharge were applied. The water levels in the Lake IJsselmeer are regulated to maintain summer and winter target levels by draining under gravity through the IJsselmeer closure dike (Afsluitdijk) into the Wadden Sea. As a result of climate change and sea level rise, it may be more difficult to maintain target water levels in winter and thus to ensure safety against flooding. Sea level rise will limit the period to drain water to the Wadden Sea under gravity during low tides and more precipitation in winter will increase the inflow of water both directly and through the Rhine river into the lake. Two main alternatives for flood risk management are available: (1) maintaining the current target water levels through additional gravitational discharge capacity or pump capacity, or (2) increase water levels to enable continuation drainage by gravity. Allowing the target water level to rise to +0.2 m MSL in winter would sustain lake drainage under gravity, but should be combined with an increase of the heights of the embankments along IJsselmeer. An ATP for the flood risk policy in the IJsselmeer is assumed to occur when a large event (which increases the lake level to more than 0.3 m MSL, causing flooding of surrounding areas) occurs, or when three small events (minor lake level increase to 0.1 m–0.3 m MSL), occur consecutively within a few years.

The timing of the ATPs (i.e. the use-by years) for all ensemble members of the reference, G and W+ scenarios were identified using modelling results.

Figure 4 shows boxplots of use-by years of the status quo and various adaptation actions for all 50 transient realizations and for the W+ climate change ensemble. For some actions the time span of the sell-by date is large, while for others there is little difference between the scenarios. The current situation without any adaptation actions reaches an ATP after ∼55 years (median) in the transient scenarios without climate change, after ∼30 years in Scenario G and after ∼25 years in Scenario W+. This gives an indication on when adaptation is needed. Doubling the gravitational discharge capacity delays an ATP in most of the realizations, but not for many of the W+ realizations (median ∼80 years) or for several outliers in the realizations without climate change (earliest after 60 years). Additional pumping capacity of 500 m³ s⁻¹ is not sufficient to prevent outliers which may result in an ATP after ∼55 years at earliest. A risk aversive policy maker could implement a pumping station with a higher capacity (e.g., 1000 m³ s⁻¹); in that case water levels will rarely exceed the threshold value for the tipping point, even in the W+ scenario with the largest sea level rise. Allowing the target water level to rise to +0.2 m MSL in winter and increase the levee heights for example by 0.5 m, an ATP is reached after ∼80 years in W+ (median value for all realizations). Based on the ATP's an adaptation pathways map was generated for flood risk management in the IJsselmeer (figure 45 in Haasnoot 2013).

Zoom In Zoom Out Reset image size

In this example the transient scenarios were used to evaluate trigger values for the river discharge on their performance as 'early warning' signal (as a signpost) that climate change is affecting river discharge and that adaptation is needed. Using twenty transient realizations for river discharge for each climate scenario we explored various types of triggers, such as threshold values, mean values and return flows.

The timing of the signal for the evaluated trigger values is given in table 2. Figures 5 and 7 present the results over time. For some triggers the bandwidth is very large illustrating the large influence of climate variability. For example, the 1:10 year discharge has a large variability and varies thus largely between the different ensemble members for the same climate change and also within one ensemble member. For the number of days that the discharge is below 1200 m³ s⁻¹ the natural variability is much less apparent. The values for the G scenario do not deviate enough from the ensemble for the current climate to detect a climate-induced change in river flow. Only for the number of days the Drought Committee will be active some deviation is shown for the G scenario, but not enough to get a clear signal for climate adaptation. The values for the average discharge in the summer half year and the discharge deficit deviate well from the values for the ensemble for the current climate. These trigger values may thus be good indicators for the need of climate adaptation.

Zoom In Zoom Out Reset image size

In reality, we will experience only a single future (comparable to one ensemble member in our study). Therefore, it is useful to know the spread of the timing of the signal and also whether there may be false warnings. To quantify the signal's timing and to get an idea of the chance a trigger value will timely give a signal, the number of ensemble members that are outside the range of the ensemble of the current climate was calculated (table 2, figure 6). In correspondence with the above, the trigger values do not give a signal in many of the ensemble members of the G scenario. For the W+ scenario, figure 6 shows that in general farther away into the future and hence with increasing climate change, more ensemble members give a signal. The average discharge for the summer half year seems the best trigger with respect to sharpness and reliability. Already around 2024 half of the ensemble members are outside the reference range and would thus give a signal. This will be in time to take decisions on the next actions in the adaptation pathways as it is well before the ATP as described by the Dutch Delta Programme (Delta Programme 2015). The 1:10 year low flow varies greatly and is neither sharp nor reliable and may confuse decision makers in the sense that they get a signal and later this signal is gone. The number of days the Drought committee will be active has a stable signal but the signal manifests itself much later than the average discharge during the summer half year.

Zoom In Zoom Out Reset image size

The triggers values for high flows (figures 7 and 8) do not show a clear signal despite a clear trend of the ensemble mean. Even the average value for the wet season does not appear to be a good trigger. Only the number of days the flow is above 6000 m³ s⁻¹ gives a signal in a lot of ensemble members, but triggers very late in the century and probably too late to adapt.

Zoom In Zoom Out Reset image size