Use of very high resolution climate model data for hydrological modelling: estimation of potential evaporation

There is increasing concern about the potential impacts of climate change on the hydrological cycle (Stocker et al. 2013). Modelling the possible hydrological impacts is particularly important as changes in the water cycle can affect people and ecosystems both directly (e.g. via changes in water availability and flood frequency) and indirectly (e.g. via changes in food and energy production) (Jiménez Cisneros et al. 2014).

Using Regional Climate Model (RCM) data as an input to hydrological models allows investigation of how climate change may affect river flows (e.g. Ott et al. 2013; Kay & Jones 2012). Many hydrological models require inputs of potential evaporation (PE) from vegetated surfaces, alongside precipitation (Bartholomeus et al. 2015). Unlike precipitation, PE is usually not available directly from RCMs, so has to be estimated from other variables. The meteorological variables that influence PE are temperature, radiation, humidity and wind speed. In addition, vegetation factors such as leaf area and roughness affect the transpiration component (Kay et al. 2013). Many formulae exist for estimating PE, ranging from the physically based Penman–Monteith formula (Monteith 1965) to much simpler empirical formulae like that of Oudin et al. (2005), and the choice can affect the results of subsequent hydrological modelling (e.g. Kay & Davies 2008; Seiller & Anctil 2014). There is much disagreement on the best approach when deriving PE from climate model data for future periods, with concerns about empirical formulae not explicitly including changes in all the influencing variables, but also concerns about data quality when using more complex formulae (see discussion by Kay et al. (2013)).

As part of a recent Natural Environment Research Council (NERC) Changing Water Cycle project, CONVEX, the Met Office ran a very high resolution (1.5 km) RCM for southern Britain, nested in a 12 km RCM driven by global atmospheric reanalysis (ERA-Interim) boundary conditions (1989–2008). They also ran Current (1996–2009) and Future (∼2100 s) climate simulations, with both aerosol climatology and full aerosol modelling setups, nesting the RCMs in a Global Climate Model (GCM). Kendon et al. (2012) found that rainfall in the 1.5 km RCM is more realistic than in the 12 km RCM. In the 12 km RCM, there is a tendency for heavy rain events to be too persistent and widespread, and not heavy enough. Conversely, the 1.5 km RCM has a tendency for heavy rain to be too intense, but it still gives a much better representation of duration and spatial extent.

This paper uses the RCM data from the CONVEX project to estimate PE for short grass, using the Penman–Monteith formula. The following questions are considered: how do RCM estimates of PE compare with observation-based PE; how do the 12 and 1.5 km estimates of PE compare; and how might PE change in the future due to climate change, and what factors influence this PE change?

PE estimates from the ERA-driven RCM runs are compared against observation-based PE from the Met Office Rainfall and Evaporation Calculation System (MORECS) (Hough et al. 1996), as MORECS is the closest to an observational estimate of PE and is widely used by the hydrological community in Britain. Also, the RCM PE estimates are required for an investigation of the use of very high resolution data for hydrological modelling, in which the hydrological models to be used are tuned using MORECS PE along with observed precipitation data (Bell et al. 2012; Crooks et al. 2014). However, MORECS PE is only produced on a 40 × 40 km grid of squares across the UK. The RCM data thus provides the opportunity to estimate PE using a much finer resolution. The comparison includes an assessment of several choices available within the PE estimation method.

Future changes in PE are investigated using the GCM-driven Current and Future RCM simulations. Relatively few studies have looked at potential future changes in PE in Britain, and even fewer have looked at historical changes, either in Britain or globally, but the studies that do exist generally suggest increases (Kay et al. 2013). However, most of these studies have calculated PE changes only from changes in (some of) the meteorological variables; PE can also be affected by increases in the atmospheric concentration of CO₂ via changes in stomatal resistance (Bell et al. 2011; Pan et al. 2015). The effect of changes in stomatal resistance is considered here, as is the influence of the method of including aerosols in the RCMs.

Although the focus of this paper is use of RCM data to produce PE estimates that will subsequently be used to drive hydrological models, the issues highlighted will be of wider interest. For example, PE can be an important component in crop modelling (Lovelli et al. 2010) and ecological modelling (Fisher et al. 2011).

For the Current and Future GCM-driven runs, there are both aerosol climatology and fully coupled aerosol modelling runs. The aerosol modelling runs differ from the aerosol climatology runs in that aerosol concentrations are calculated interactively, with advection and deposition of aerosols described. In the case of the 12 km RCM, the aerosols are fully coupled to the microphysics, so that aerosol concentrations determine the number of cloud condensation nuclei. A comparison is made of PE from both pairs of model runs for the 12 km RCM, as data are available for the full period of ∼13 years. The equivalent 1.5 km RCM aerosol modelling runs are only about 5 years in length (Table 1) and are therefore not included in this analysis.

The climate model variables used for the calculation of PE are thus 1.5 m temperature, 1.5 m relative humidity, 10 m wind speed and net surface downward longwave and shortwave radiation (which sum to R_n).

PE can be affected not just by changes in meteorological inputs, but also by changes in the behaviour of vegetation. In particular, higher CO₂ concentrations can lead to plant stomata opening less widely, resulting in higher stomatal resistance, but can also enhance plant growth, leading to a greater leaf area and more stomata (e.g. Bunce 2004; Keenan et al. 2013). Here, following Kruijt et al. (2008), changes in crop resistance are considered but changes in leaf area are not, as it was found by Bunce (2004) that there is a lack of response in LAI due to elevated CO₂ concentrations in all functional types, except trees; only grass PE is considered here.

To estimate monthly PE, two different averaging methods were compared (for 1990 data): (a) calculating daily PE from daily meteorological variables and then averaging, and (b) calculating monthly averages of the meteorological variables and then calculating monthly PE from them. The two methods yield almost identical results (not shown) so daily PE are made and then the monthly average is calculated (the monthly plots in Figure 2 use this method). This test was done as, although daily data were available here, it can be the case that only monthly data are available, or that it is impractical to obtain all the required data at a daily rather than monthly time-step. In such situations, use of monthly data to calculate monthly PE is unlikely to cause problems (as also suggested by Allen et al. (1998)).

Estimates of PE from high-resolution (12 and 1.5 km) RCM data have been produced, using the physically based Penman–Monteith formula, with the aim of making PE suitable to use as input for hydrological modelling. Using ERA-driven RCM data, and a comparison with the observation-based MORECS PE, choices were made about the method to use. The comparison shows that both the 1.5 and 12 km RCM PE estimates are very similar to each other and comparable to MORECS, spatially and at selected locations, at daily and monthly time-steps.

PE estimates from Current and Future RCM runs driven by GCM boundary conditions have been used to investigate potential future changes in PE, and how certain factors affect those changes. It is found that the seasonal mean PE and its change are in close agreement between the 1.5 and 12 km models; this is perhaps not surprising because the large scale changes in humidity, temperature and circulation are common to both RCMs (inherited from the driving GCM). It is also found that future PE is likely to be larger than at present, with the largest increases in summer, autumn and (to a lesser extent) winter. This PE change is influenced by the changes in temperature, relative humidity, wind and radiation as well as stomatal influences through increased CO₂. Not accounting for the stomatal influences can inflate the future PE estimates, consistent with the findings of Bell et al. (2011) for UK annual and seasonal mean PE, and such differences can affect subsequent flow projections (e.g. Bell et al. 2012; Prudhomme et al. 2014).

PE projections are also shown to vary to some extent according to how aerosols are modelled in the RCMs. The PE differences from aerosol differences are in the summer/early autumn and this could potentially be important for hydrological modelling as autumn PE that is too high could delay the rewetting of soils in the lead-up to the main flood season in Britain (Bayliss & Jones 1993). The smaller percentage changes in PE for the aerosol modelling run appear to be related to smaller changes in radiation. The potential hydrological importance of aerosol concentrations, via their influence on radiation and so evaporation, is also shown by Gedney et al. (2014), who identify a link between solar dimming, due to rising atmospheric concentrations of aerosols from 1980, and increases in runoff.

Use of the Penman–Monteith PE formulation enabled specific application of changes to crop resistance alongside future changes in meteorological variables. This is not possible (at least in such a straightforward way) with most of the simpler PE formulae, where empirically derived coefficients replace many of the factors present in the more physically based formulae (see for example the summary of 17 PE variations given by Oudin et al. (2005)). There are already concerns about the application of empirical formulae under changing climates (e.g. Donohue et al. 2010; Bartholomeus et al. 2015), and the need to also consider changes in canopy resistance is an added complication (Kay et al. 2013). The use of fixed crop coefficients to estimate crop PE from reference PE is an additional factor that requires consideration under climate change (Bartholomeus et al. 2015), both for hydrological modelling and crop modelling, as different crops may react differently to the same change in CO₂ concentrations (Kruijt et al. 2008). Even for hydrological models that do not specifically require PE inputs, similar considerations are likely to apply in the model's internal calculation of evaporation when applied under changing climatic conditions.

Future work will involve running CLASSIC-GB (Crooks et al. 2014) with the high-resolution RCM data to investigate the effect of model resolution, and future climate changes, on peak river flows in southern Britain. As the 1.5 km full aerosol modelling run is not full length, the aerosol climatology run will be used to get sufficient length to look at flood frequencies.

The authors would like to thank Elizabeth Kendon and Richard Jones (Met Office Hadley Centre) for providing the data and help with analysis of results. This work was funded by the Natural Hazards science area of the NERC-CEH Water and Pollution Science programme.