The effects of climate change on storm surges around the United Kingdom

J.A Lowe, J.M Gregory

Abstract

Coastal flooding is often caused by extreme events, such as storm surges. In this study, improved physical models have been used to simulate the climate system and storm surges, and to predict the effect of increased atmospheric concentrations of greenhouse gases on the surges. In agreement with previous studies, this work indicates that the changes in atmospheric storminess and the higher time-average sea-level predicted for the end of the twenty-first century will lead to changes in the height of water levels measured relative to the present day tide. However, the details of these projections differ somewhat from earlier assessments.

Uncertainty in projections of future extreme water levels arise from uncertainty in the amount and timing of future greenhouse gas emissions, uncertainty in the physical models used to simulate the climate system and from the natural variability of the system. The total uncertainty has not yet been reliably quantified and achieving this should be a priority for future research.

Keywords:

1. Introduction

Tide gauge measurements suggest that global average sea-levels rose by between 1 and 2 mm yr−1 during the twentieth century (Church et al. 2001). Between 1993 and 2000, satellite altimetry indicated that the rate of rise was approximately 2.5 mm yr−1 (Cabanes et al. 2001), which may indicate a recent acceleration of the long-term trend or it could be the result of interdecadal variability. In the future, global average sea-level is expected to increase more rapidly as a result of anthropogenic climate change. The Third Assessment Report of the Intergovernmental Panel on Climate Change concluded that between 1990 and 2100, the global average sea-level rise is likely to be between 9 and 88 cm (Church et al. 2001). While this century-scale rise in sea-level will threaten many low-lying unprotected coastal areas, it is the extremes of sea-level associated with storm surges1 that will cause much of the damage (for instance McCarthy et al. 2001).

Extreme water levels will also increase as time-average sea-levels rise. In addition, changes in the number, path and strength of atmospheric cyclonic storms may alter the formation and evolution of storm surges. However, because individual surge events depend on the driving meteorology, we cannot predict individual surges more than, at most, a few days ahead. Therefore, for 50 or 100 years into the future, the best we can do is to predict how many events will, on average, exceed a given height in a fixed length of time. This can also be expressed as predicting the average time between a given surge height being exceeded, and is referred to as the return period for the height in question.

There are two main methods for estimating future changes in storm-surge heights: statistical and dynamical. In the statistical method, relationships between large-scale driving meteorology and local storm-surge heights are developed from observations or simulations of the recent past and present day. These relationships must be capable of explaining a significant fraction of the surge height variability, and for predictive purposes, they are assumed to remain applicable even if the climate changes in the future. Next, a projection is made of future large-scale meteorology using a global climate model and the future storm-surge characteristics are estimated from these using the statistical relationships (Von Storch & Reichardt 1997). In the alternative dynamical approach, physical models are used to derive fine scale winds and atmospheric surface pressure from the large-scale climate simulated by the coarse global climate model. These are then used to drive a dynamic model of storm surges. The remainder of the paper will describe the results of the dynamic method applied over the European continental shelf region.

Flather & Smith (1998) used the ECHAM 3 climate model to drive a storm-surge model for two 5 year time-slices, which represented present day and future climate conditions. While the surge extremes were different for future and present day simulations, the differences were mostly within the range of natural variability, estimated from a longer storm-surge simulation driven by surface forcing from meteorological analyses of the period 1955–1994 (Flather et al. 1998). Flather & Williams (2000) improved on earlier work by using longer (30 year) time-slices, a more policy-relevant emissions scenario and an improved climate model, ECHAM 4. A three-parameter generalized extreme value (GEV) distribution was fitted to 10 surge maxima from each year in order to extrapolate the results to 50 year return periods. At some locations, the authors identified statistically significant differences between the present day and future time-series. Lowe et al. (2001) used a similar methodology to Flather & Williams (2000), with the same surge model and a policy relevant scenario. The atmospheric changes were taken from Hadley Centre global and regional climate models, HadCM2 and HadRM2, which provided a finer spatial resolution over the European shelf region than the ECHAM model (50 km instead of approximately 100 km) and a higher temporal resolution. The Gumbel distribution, which is a special case of the GEV distribution and has only two parameters, was used to extrapolate the time-slice results to 50 year return periods. The Lowe et al. study suggested that changes in storminess would increase the height of storm surges around much of the UK coastline but the pattern of change was found to be very different from that of Flather and Williams. The effect of including a time-average sea-level rise of 50 cm was to change extreme water levels, expressed relative to the present day tide, by between approximately 45 and 55 cm. The deviations from 50 cm were due to nonlinear effects. Debernard et al. (2002) recently ran a storm-surge model for two 20 year time-slices, representing present day and a future period centred on 2040, using surface forcing from the HIRHAM regional atmospheric simulations model. As in the HadRM2 regional climate model, this has a resolution of around 50 km. Significant changes in extreme water levels were found in the southern and western North Sea in the autumn season but these did not greatly affect the annual results.

The results described here are an update of those in Lowe et al. (2001) using more up-to-date emissions scenarios, an improved set of climate models and similar statistical methods to Flather & Williams (2000).

2. Methodology

Coupled global climate models cannot yet simulate storm surges directly because the spatial resolution of their atmospheric components (≈100–300 km) is too coarse to adequately represent the winds and the gradients of atmospheric surface pressures that drive the extreme surges. In addition, the ocean components of coupled global climate models are usually too coarse to represent storm surges directly and many use a rigid lid formulation, which means that sea surface height changes are not represented explicitly. Regional climate models have a higher spatial resolution (≈50 km or finer) in order to better represent the smaller scale features of the atmospheric circulation, but many current generation models do not have an ocean component. Consequently, in the current work, a global atmospheric model (HadAM3H) was used to provide boundary conditions for a regional atmospheric model (HadRM3H). The atmospheric winds and pressure from the regional model were then used to force a 35 km model of the UK shelf seas, which was provided by the Proudman Oceanographic Laboratory (POL).

The global coupled climate model (HadCM3) is a grid point model with a horizontal atmospheric resolution of 2.5×3.75° and 19 vertical levels. The ocean component has a horizontal resolution of 1.25×1.25° and 20 vertical levels. The global atmospheric model (HadAM3H) has a resolution of 1.25×1.88° and is driven at its ocean boundary by a combination of observed surface temperatures from the HadISST dataset (Rayner et al. 2003) plus anomalies from the coarser HadCM3 model. HadCM3 and HadAM3H are described in more detail by Gordon et al. (2000) and Jones et al. (2001), respectively. The regional climate model (HadRM3H) has a horizontal resolution of 0.44×0.44° (approximately 50 km), providing a much more realistic representation of orographic and other small-scale features than the global model. As in the global model, the equations governing climate are expressed in spherical polar coordinates, but in the regional model the North Pole is rotated to 38°N and 190°E in order to obtain quasi-uniform resolution over Europe. A one-way nesting procedure is used, with lateral boundary conditions for the region being taken from the global atmosphere model every 6 h. Instantaneous winds and atmospheric surface pressure, needed by the surge model, are output and stored every 3 h. A more complete description of HadRM3H is provided by Jones et al. (2001).

The climate models used in this study have many advantages over the earlier work of Lowe et al. (2001), with a better representation of climate physics, including the parametrization of layer cloud and precipitation. Combined with a higher global ocean model resolution, these improvements enabled the physically unrealistic flux corrections, which were used in the earlier study in order to maintain a stable climate in the global coupled model, to be omitted. The inclusion of an intermediate model stage between the coupled global model and atmospheric regional model led to the regional model being driven by higher resolution driving data. The use of observed sea surface temperature in the intermediate model also allowed the north Atlantic storm track to be more realistically placed in the present day simulation.

The POL storm-surge model is a depth average shallow water model (Flather & Smith 1998). The height of the water surface above mean sea-level and the depth average currents are determined using continuity and momentum conservation equations. The surge model equations are solved using a finite-difference method on a regular latitude–longitude grid of approximately 35 km resolution. The meteorological forcing from the regional climate model is linearly interpolated in space onto the surge model grid and linearly interpolated in time between the 3 hourly results stored from the regional climate model. Tidal forcing for eight harmonic components was applied along the lateral boundary. For each surge model experiment, two separate integrations of the surge model were carried out. The first was driven by both tidal and meteorological forcings, while the second was driven by tidal forcing alone. Surge heights were determined by subtracting the tide-only results from the tide-plus-meteorological forcing results. This approach makes it possible to include the nonlinear interactions between the tide and surge components in projections of surge height. Depth average currents and surface elevations, relative to mean sea-level, were output from the surge model as instantaneous values at hourly intervals.

The coupled global climate model was integrated from pre-industrial times up to the end of the twenty-first century. Prior to 1990, historical greenhouse gas concentrations were used. Later, the greenhouse gas concentrations resulting from the Intergovernmental Panel on Climate Change's Special Report on Emissions Scenarios (SRES) A2 and B2 emissions scenarios were applied (Nakicenovic et al. 2000). The higher resolution global atmospheric model, the regional model and the storm-surge model were used to simulate two 30 year time-slices, present day (1961–1990) and future (2071–2100). Two separate surge model experiments were carried out for each future emissions scenario; the first to examine the effect of future changes in storminess alone and the second to look at the combined effect of future changes in storminess and the greater water depths associated with twenty-first century time-average sea-level rise. Statistical distributions were fitted to the results and used to extrapolate them to long return periods. Here, we follow Flather & Williams (2000) in choosing the GEV distribution, which can provide a good fit to events drawn from the tail of a population (Ferreira 1997). The five most extreme events from each year were used and, to ensure that the events were independent, consecutive extreme surges were required to be separated by a minimum of 5 days. A sensitivity test at five locations revealed that the GEV parameters were unchanged as the number of extreme surge events was increased from 5 to 15 events. Increasing the number of annual maxima includes more information in the calculation of the GEV distribution and this was found to be important when only a small number (less than 5) of annual maxima were used. However, as the number of extreme surge events from each year is increased, there is a possibility of including results that are not really from the tail of the population of surge heights, as occurred with more than 15 annual maxima. The full experimental set up is shown diagrammatically in figure 1.

Figure 1

Schematic of experimental set up.

3. Results

(a) Present day storm-surge simulation

High frequency observations of sea-level from tide gauges are available for a number of sites around the UK coastline. Where these time-series are long enough, it is possible to remove the tidal component and estimate an observed 50 year storm-surge height. The heights of surges simulated by the climate and storm-surge modelling system are generally less than those measured by the tide gauges. Nevertheless, the geographical pattern of simulated surges for present-day climate, with elevated values at the southern end of the North Sea, compares well with that of the observations (figure 2).

Figure 2

Pattern of simulated present day 50 year return period storm-surge height (m).

Additional validation was performed by comparing the present day storm-surge events with results obtained by driving the same surge model with 40 years of meteorological analyses data covering the period 1955–1994 and assembled by the Norwegian Meteorological Institute (Flather et al. 1998). The simulated spatial pattern of 50 year return period storm-surge height in the current work is qualitatively similar to the analyses forced simulation, but slightly lower in magnitude. This could be because the extreme winds simulated by the driving climate models are either too weak, or occur in a different place or direction than the observations. It may also partially be due to very long period variability.

(b) Twenty-first century climate change

The average global temperature rise between present day and the 2080s was approximately 3.5 and 2.5 °C in the A2 and B2 simulations, respectively. In the future climates (2080s), the synoptic situation over the UK in winter continues to be dominated by cyclonic storms arriving from the west. The number of low-pressure storm systems (with a minimum pressure below 1000 mb) which cross the UK during winter is predicted to increase from approximately five per winter in the present day simulation to eight during the 2080s in the A2 simulation (McDonald 2002). This increase is accompanied by a strengthening of the winter winds by as much as 6% over some south of England locations. In the far north of the country there is a slight reduction in the wind speed. These changes are consistent with a southward movement across the UK of the average winter storm tracks. The predicted changes in storminess in the B2 scenario experiment are less than those in the A2 case.

Global mean sea-level is predicted to increase by approximately 33 and 25 cm for the A2 and B2 scenarios, respectively. These projections comprise contributions from the thermal expansion of the ocean, melting of small glaciers and changes in the mass balance of Greenland and Antarctica ice sheets (Gregory & Lowe 2000).

(c) Future changes in the height of storm surges

Changes in atmospheric storminess have the potential to cause the height of storm surges to change. The height of a storm surge with a 50 year return period increases along much of the UK coastline for both A2 and B2 scenarios (figure 3), with the largest increases predicted to occur off the southeast coast. A decrease in the storm-surge height is predicted for the Bristol Channel, although the 35 km resolution of the storm-surge model means that we have less confidence in the results for narrow channels. The changes are greater in the A2 simulation than in the SRES B2 case, which is expected given the larger temperature rise and the greater change in storminess in the former.

Figure 3

Change in the height (m) of a 50 year return period due to changes in atmospheric storminess only, between the present day and the 2080s for the (a) A2 scenario and (b) B2 scenario.

The predicted future changes in storm-surge height that result from the changes in atmospheric storminess differ considerably from the earlier simulation of Lowe et al. (2001). In the current work, and viewed from a UK perspective, the largest increase in 50 year return period storm-surge height is approximately 0.7 m (in the A2 scenario) and occurs off the southeast coast. In the previous study, the maximum was approximately 0.5 m and occurred off the south coast. The predicted changes around the Welsh coast are also very different. In the current work these are slightly negative, but in the Lowe et al. (2001) study they were around +20 cm.

Increases in extreme water levels, measured relative to a point on the land, are also changed by increases in time-average sea-level and vertical land movements. The changes in time-average sea-level obviously contribute directly to an increase in extreme water level. However, they may also have an indirect effect by changing the surge height and propagation. While both of these effects are included in the current study, the earlier work of Lowe et al. (2001) suggests that around the UK, this indirect term is small compared with the direct sea-level rise. The time-average sea-level rise can also alter the tides by changing both dissipation and resonance effects, although Flather et al. (2001) estimated that the change in tidal range is also relatively small. Increases in time-average sea-level during the twenty-first century are not expected to be spatially uniform (Gregory & Lowe 2000). However, this effect has not been included here because predictions of the spatial pattern of change are currently very uncertain (Gregory et al. 2001).

Vertical land movements occur naturally for a number of reasons, such as the ongoing adjustment of the Earth's crust to the deglaciation at the end of the last ice age, tectonic activity and localized sediment consolidation. For the UK as a whole, the first cause is dominant. The northern part of the country, which was covered with an ice sheet during the last glacial maximum, is rising relative to the level of the sea, whereas much of southern Britain, which was situated on the forebulge at the edge of the ice sheet, is sinking (Masselink & Hughes 2003). For the current work, the observational land movement dataset of Shennan (1989) was applied. This has rates of vertical movement that range between approximately −2 mm yr−1 (southeast England) and +2 mm yr−1 (northwest Scotland), and which are assumed to remain constant over the twenty-first century.

The predicted change in the height of a 50 year return period water levels around the UK, measured relative to the present day tide, when changes in storminess, a rise in global sea-level, and vertical land movements are all included, is shown in figure 4. The inclusion of vertical land movements tends to further increase the already large rise off the southeast coast but reduces the changes around northwest Scotland. Consequently, the largest rise, 1.2 m in the A2 scenario, occurs along the southeast coast of England.

Figure 4

Change in the height (m) of a 50 year return period extreme water level event, measured relative to the present day tide, due to changes in atmospheric storminess, an increase in mean sea-level and vertical land movements. Results are shown for the (a) A2 scenario (b) B2 scenario.

It is desirable to understand whether changes in the height of storm surges are significant. A statistically significant difference means there is only a small probability that the present day and future surge heights at a given location are sampled from populations with identical characteristics. However, even if the distributions of extreme surge heights are different in future it still remains to be established whether the difference is due to interdecadal natural climate variability or whether it is a long-term trend. Doing this comprehensively requires knowledge of long-period surge variability, which needs much longer storm-surge simulations than were available here.

(d) Case study

As a part of the analysis procedure, extreme value distributions were fitted to the model results at every point in the model domain. For an example location, Immingham, on the east coast of England, the return period curves for the present day and the 2080s are presented (figure 5). The results are again expressed relative to the present day tide and include the effects of changes in storminess and mean sea-level rise for the A2 scenario and vertical land movements. An extreme water level of 1.5 m is predicted to be exceeded every 120 years on average (green line and black curve) in the present day climate. However, by the 2080s, it is projected that this level will be exceeded once every 7 years on average; a 17-fold increase in the exceedence frequency (green line and red curve). A 120 year return period event is projected to exceed around 2.2 m in the future climate.

Figure 5

The height of simulated extreme water levels at Immingham, measured relative to the present day tide, for the present day and the 2080s (A2 scenario). Increases in extreme water level are due to changes in atmospheric storminess, an increase in mean sea-level and vertical land movements.

4. Uncertainties in estimates of climate-driven changes in storm surges

Projections of the height of extreme storm surges in a future climate are subject to uncertainty, and being able to quantify this uncertainty will lead to improved probabilistic risk assessments for flood defences. The total uncertainty can be split into several components: the uncertainty in the emissions of greenhouse gases, the uncertainty in the science and methodology used to estimate the response to greenhouse gases, and natural climate variability.

Future emissions from human activities depend upon socio-economic factors such as population, economic growth and technological development, and we do not, and never will, know how these will change in the future. The best we can do is to anticipate several plausible ways in which the world might develop in the future and then estimate the emissions that are consistent with these futures. The Intergovernmental Panel on Climate Change's SRES carried out this exercise in 1999, resulting in a range of 40 different futures. The greenhouse gas emissions associated with these futures diverge immediately and rapidly, and by 2100 the carbon dioxide emissions range from less than today's emissions up to a fourfold increase on today's values. The SRES also indicates that all of the emissions scenarios are plausible and it is not possible to put relative probabilities on the scenarios. In the current work, two emissions scenarios have been used: a medium–high scenario (A2), representing a world in which the population increases to 15 billion and cumulative emissions of carbon dioxide (relative to 1990) reach 1850 GtC by 2100, and a medium–low scenario (B2) with a smaller population increase and cumulative emissions of 1150 GtC. This range of emissions covers approximately 60% of the span between the upper and lower SRES scenarios.

In order to estimate future changes in storm-surge height, the effect of future emissions on the climate system and of climate changes on storm-surge height have been simulated. Uncertainties in this process are referred to as science or methodological uncertainty but we are not able to quantify this component at present. However, we can estimate a minimum range for the atmospheric storminess driven component of surge height variations, the range of time-average sea-level rise, and the range of spatial patterns of sea-level change predicted by global coupled climate models.

Figure 6 shows the predicted changes in 50 year return period height due to changes in storminess from the earlier study of Lowe et al. (2001), which was driven by older versions of the Hadley Centre models, and the results from the STOWASUS project (Flather & Williams 2000), which were driven by the ECHAM climate model. These can be compared with the current results (figure 3), which use an improved set of Hadley Centre climate models. While all three patterns suggest that the changes will vary with location, they disagree on the details of the pattern of change. In fact, there are very few features that are common to all three simulations. The differences are probably due to the different changes in storm track characteristics. For instance, in the current work, the storm track appears to move south, whereas in the STOWASUS study there is a northwest movement. It should be noted that the emissions scenarios were also slightly different in these scenarios, although a comparison of the A2 and B2 results in figure 3 indicates that this is more likely to change the magnitude of the response than the relative pattern of change.

Figure 6

Comparison of future changes in the storminess driven component of 50 year return period storm-surge height (m) from the same surge model but using driving data from different climate model simulations. (a) HadCM2/HadRM2 and (b) ECHAM4.

The uncertainty range for twenty-first century global time-average sea-level rise has been estimated to be 9–88 cm (Church et al. 2001). However, this includes both the science uncertainty and the uncertainty in the future emissions of greenhouse gases. The ranges for the A2 and B2 scenarios are 15–75 and 12–65 cm, respectively. In practice, it is not the global average sea-level rise that is important, but the local time-average sea-level rise. In the Hadley Centre model, the range of twenty-first century sea-level changes is approximately twice the global average value. However, the model intercomparison of Gregory et al. (2001) showed that, even for the same ‘business-as-usual’ emissions scenario, there is little agreement between different model simulations of the sea-level rise patterns and so there is currently only a low confidence in any particular model result. For the UK, the twenty-first century sea-level rise predictions range from around 10 to 60 cm.

We have not attempted to combine the scientific uncertainties, as the terms we have listed are not independent. For instance, part of the uncertainty in global average sea-level rise is included in the uncertainty in time-average spatial sea-level rise patterns. However, it is still useful to compare the relative magnitude of the uncertainty terms for sample locations (figure 7). In this figure, the storminess term is estimated from figures 3 and 6a,b; the mean sea-level uncertainty is taken from Church et al. (2001), as described above; the regional sea-level uncertainty is estimated from the figures in Gregory et al. (2001); and the uncertainty due to emissions is estimated from figure 4a,b. Clearly, the relative importance of the different terms varies with location.

Figure 7

Comparison of the minimum range of uncertainty from the effect of changes in storminess on surges, the effect of global average sea-level rise, the spatial pattern of time-average sea-level rise, and the uncertainty in future emissions. (a) Southend, (b) Immingham.

In addition to the emissions and science uncertainties, in any future period natural ‘chaotic’ variability of climate can either add to or subtract from any long-term trend. This variability acts on both short and long periods. Short period changes can be seen in the day-to-day and week-to-week variations of the weather, while long-period climate variability is evident in instrument records, proxy climate measurements and model simulations (Collins et al. 2002). A proportion of the apparent science uncertainty in figure 7 may actually be due to natural variability.

In future, it may be possible to quantify the science uncertainty and the effect of natural variability for any greenhouse gas emissions scenario. Estimating the science uncertainty will involve running a large number of climate model simulations, each with a slightly different version of the climate model. The differences need to be chosen to span the plausible range of model versions. The resulting simulations, which will simulate a period from the recent past into the future, can be weighted according to their success at simulating past and present climate. Estimating the effect of natural variability will involve performing a number of simulations from different, independent, starting conditions with a single version of the model.

5. Conclusions

In this study, new estimates of future changes in extreme water levels have been made for the European shelf sea region. These show that changes in storminess will increase the height of storm surges at most coastal locations around the UK coastline, however the size of the change will vary with location. Increases in time-average sea-level will also tend to increase the height of extreme water levels, measured relative to present day tidal level, at all locations around the UK coastline. In addition, vertical land movements will alter the relative height of extreme water events measured on the land. The largest increases in relative surge height are predicted to occur off the southeast coast of England, where the changes in storminess will have their biggest effect and where the land is sinking most rapidly. An alternative way to interpret these results is in terms of the return period of a given extreme water level being exceeded. The present work suggests that, in the future, the return period of large events could decrease by more than an order of magnitude at some locations.

The sources of uncertainty in projections of future storm-surge height have been highlighted. While minimum ranges for some of these uncertainties were presented, these should not be assumed to span the full uncertainty range. In order to reliably quantify the uncertainty in storm surges, a large number of storm-surge simulations will be needed, and this will require a large amount of computing resources. However, given the importance of being able to plan for future coastal flooding, this should be seen as a research priority.

Acknowledgments

Roger Flather at the Proudman Oceanographic Laboratory developed the storm-surge model and provided figure 6b. The regional climate model simulations were carried out by David Hassell at the Met Office. This research was supported by funding from the UK's department for the environment, food and rural affairs under contract PECD 7/12/37.

Footnotes

  • One contribution of 14 to a Theme ‘The Big Flood: North Sea storm surge’.

  • Storm surges are temporary increases in sea-level, above the level of the tide, caused by low atmospheric pressure and the force exerted on the sea surface by strong winds. The water level may be increased further by the geometry of the coast, which can cause a funnelling effect. Surges are potentially most damaging when they occur at high tide.

References

View Abstract