A hydrological perspective on evaporation : historical trends and future projections in Britain

Evaporation is an important component of the hydrological cycle. Potential evaporation (PE) from a vegetated surface is the amount of water that would be lost to the atmosphere were the supply unlimited; actual evaporation (AE) is a fraction of PE dependent on soil wetness. Many formulae exist for estimating PE from meteorological data. PE is usually a required input, with rainfall, for hydrological modelling, but PE accuracy is generally considered less important than rainfall accuracy for model performance. Few studies investigate historical evaporation trends in Britain, but generally indicate increases. Most studies presenting future PE projections for Britain indicate increased annual PE, but some suggest small decreases in some months. Limited consensus on the best formulae to derive PE projections from climate model data is further complicated by possible changes in plant behaviour (transpiration and growth) under higher carbon dioxide concentrations. Appropriate PE estimation could be particularly important in regions where precipitation and PE are in close balance, but PE uncertainty could be less important than climate model uncertainty for hydrological impacts. Further research is needed into which PE formulae are likely to be most reliable when applied with climate model data, and into climate change and plant feedbacks. doi: 10.2166/wcc.2013.014 s://iwaponline.com/jwcc/article-pdf/4/3/193/374900/193.pdf A. L. Kay (corresponding author) V. A. Bell E. M. Blyth S. M. Crooks H. N. Davies N. S. Reynard Centre for Ecology and Hydrology, Maclean Building, Crowmarsh Gifford, Wallingford, Oxfordshire, OX10 8BB, UK E-mail: alkay@ceh.ac.uk


INTRODUCTION
Evaporation transfers water from the land-surface to the atmosphere, so is an important part of the hydrological cycle.While there is ongoing debate on terminology (Lhomme ; Cain et al. ), here the terms evaporation and evapotranspiration each encompass transfer via both evaporation (loss of water lying on a surface, e.g., lake, soil, leaf) and transpiration (loss via plant stomata).Potential evaporation (PE or PET) is generally considered as the amount of water that would be lost to the atmosphere if there were no limits to soil-moisture supply (Federer et al. ).So actual evaporation (AE or AET) can be estimated as a fraction of PE dependent on soil wetness; it can be less than PE if soils are dry, but generally cannot be greater than PE.PE is difficult to measure directly, although several techniques attempt to measure AE.Some of the complexities of measuring and modelling evaporation are described by Shuttleworth ().
Four meteorological variables influence PE: radiation (or sunshine), temperature, humidity (or vapour pressure) and wind speed.Further variables influence the transpiration component: plant height, rooting depth, leaf area and vegetation roughness.Thus PE varies for different plant types.To simplify matters, the 'reference crop' concept was introduced, with average crop parameters provided for estimation of reference PE.The reference crop is often short grass (Pereira et al. ).Many formulae have been developed for estimating PE, some for particular reference crops and some where crop parameters can be specified.
The simplest, empirical formulae involve a single meteorological variable (e.g., the temperature-based formulations of Thornthwaite (), Hamon () and Oudin et al.
(a)), while the most complex and physically based formulae involve all four meteorological variables (e.g., Penman-Monteith; Monteith ), with a range in between (e.g., Blaney-Criddle, involving

BACKGROUND Penman-Monteith PE, MORECS and MOSES
Penman-Monteith PE for short grass (m/s) is given by: with λ latent heat flux (J/kg), ρ w water density (kg/m 3 ), ρ a air density (kg/m 3 ), c p specific heat of air (J/kg/

HISTORICAL TRENDS PE and AE trends in Britain
There

Global context
The shortage of information on evaporation trends is not Their correlation analyses suggest the recent decline is due to increasing soil moisture limitation, but the time periods studied are too short to imply trends; differences could be due to decadal variability.were found for all latitudes, GCMs and formulae, and broadly followed corresponding temperature changes.However, significant differences were found between formulae; There is little consensus on the best approach for deriving future PE projections; some authors believe that formulae including all meteorological variables influencing PE must be applied, while others believe that sensitivity of such formulae to data quality makes this inadvisable, or that the choice will make little difference in subsequent hydrological modelling.The dilemma is summarised by Kingston et al.Most studies presenting PE projections for Britain indicate increases in annual PE, although some studies suggest (usually small) PE decreases in some months (Table 1).

FUTURE PROJECTIONS
However, there is considerable variation in the magnitude of projections, caused not just by the PE formula applied temperature and sunshine (Blaney & Criddle ), and Priestley-Taylor, involving temperature and radiation (Priestley & Taylor )).Oudin et al. (a) provide a useful summary of 17 variations found in the literature.Penman-Monteith is recommended by the United Nations Food and Agricultural Organisation (FAO) for deriving grass reference PE (Pereira et al. ), and is used by the UK Climate Projections 09 (UKCP09) weather generator (Jones et al. ).The UK Met Office Rainfall and Evaporation Calculation System (MORECS; Thompson et al. ; Hough et al. ) and the Met Office Surface Exchange Scheme (MOSES; Cox et al. ) use modified versions of Penman-Monteith.PE estimates are a required input for hydrological modelling, alongside rainfall.PE changes, on their own or in combination with rainfall changes, can contribute to changes in hydrological indices like mean monthly river flows.This paper provides some background on AE and PE in Britain, influence of PE in hydrological modelling, and the ways in which environmental change can affect PE.A review of historical evaporation trends in Britain is presented, with a global context.Future PE projections in Britain are reviewed, with estimation difficulties and effects of PE uncertainty on hydrological climate change impacts discussed.Finally, a discussion and conclusions are presented.Although the focus here is the hydrological perspective, evaporation is of interest in other areas (e.g., agriculture and ecology; Fisher et al. ), and many of the same issues will apply.

Figure 2
Figure 2(a) shows 1961-1990 monthly means of these variables for three contrasting catchments (modelled with CLASSIC): wet/energy-limited, energy-water balanced and dry/water-limited.In summer, the wet/energy-limited catchment shows PE < rainfall with near-constant AE close to PE, while the energy-water balanced catchment has PE close to rainfall and a slightly larger difference between PE and AE.In the dry/water-limited catchment PE > rainfall for the middle part of the year, leading to higher SMDs and AE < <PE, with reduced AE as summer progresses (peak AE precedes peak PE).A given PE change for a wet/energy-limited catchment is likely to result in a similar AE change, whereas the same PE change for a dry/water-limited catchment is likely to result in little AE change (assuming little/no rainfall change).Thus
Figure 3 | MORECS PE (solid line, red) and AE (dashed line, green) for 1961-2012 (for short grass and median AWC soils), averaged over England and Wales (left) and Scotland (right).In each case a fitted linear trend line is shown (dotted).For references to colour see the online version of the paper: http://www.iwaponline.com/jwc/toc.htm.
ging and are expected to change in future, single-variable PE formulae should be avoided when estimating PE (historical or future) trends 'due to the inherent nature of the trend passed to PE from the variable'.A view echoed by Donohue et al. () as 'the greater the number of the four key variables … in a formulation, the more realistic the trends from that formulation become', although this clearly depends on the sensitivity of a formula to each variable and the relative strengths and directions of any trends in those variables (likely to have some level of inter-dependency; Fisher et al. ).The sensitivity of PE formulae cannot necessarily be predicted directly from the meteorological variables included; Bormann () compared PE changes given by 18 formulae, for six locations in Germany, and found as much variability for formulae of the same type as for different types.However, Shaw & Riha () show that the temperature-only Hamon and Thornthwaite formulae are much more temperature-sensitive than the more complex Penman-Monteith and Priestley-Taylor formulae.This should be interpreted carefully, since changes in other variables were neglected but are likely to co-vary with temperature (Fisher et al. ; Chun et al. ), and their inclusion could increase the response of more complex formulae.Interestingly, Shaw & Riha () show that the temperature-sensitivity of the Wilby & Harris () use data from four GCMs (2080s A2 and B2) with statistical downscaling for the Thames.They obtain PE increases of 5-43% in winter and 11-22% in summer, but for two GCMs, the summer changes are larger than winter, with the opposite for the other two GCMs.Wilby et al. () use a similar method for the Kennet, using three of the same GCMs, but find much smaller PE increases (winter 3-9%; summer 5-16%) presumably due to a different regression model.Diaz-Nieto & Wilby () showed that projected PE increases for the Thames derived via statistical downscaling (using temperature and specific and relative humidity) were roughly half those derived using Penman-Monteith.Chun et al. () use data from the HadCM3 GCM (1950-2099 A2) to estimate PE at 25 sites across Britain using Penman-Monteith and a GLM with four variables (radiation, temperature, wind and humidity).They show annual percentage increases almost everywhere by the 2080s, with greater increases in the south/east than north/west, but smaller increases using the GLM than Penman-Monteith.However, the site-specific parameter calibration required by the GLM may make its use under climate change questionable, as for Blaney-Criddle (Sperna Weiland et al. ).Christierson et al. () use UKCP09 probabilistic temperature data (2020s A1B) to calculate Oudin PE percentage changes for each river-basin region in the UK.They present maps of the central estimate of PE changes, which show an increase across the country, with the largest changes in winter.The range of monthly PE changes, for two example catchments, almost always show increases, usually 0-30%, with a greater variation in winter than summer (but only a 20-member subset of the full 10,000 UKCP09 set is shown, selected via latin-hypercube sampling).These PE changes are shown to be generally consistent with an older set of scenarios, and with those calculated using temperature data from the 11-member UKCP09 RCM ensemble in the Oudin formula.Similarly, Kay & Jones () use UKCP09 probabilistic temperature data (2080s A1B) to calculate Oudin PE changes for nine catchments in Britain.Again, the PE changes are almost always positive, and generally larger (with greater variation) in winter than summer, but the range of 5-60% is larger than that of Christierson et al. (), because of the later time-horizon.Prudhomme & Williamson () use data from one member of UKCP09 RCM ensemble (2050s A1B) and compare PE changes across Britain for 12 formulae for a month representative of each of the four seasons.PE almost always shows an increase, although the magnitude varies by season, location and formula.Kingston et al. () use data from five GCMs, for a scenario representing a 2 W C rise in global mean temperature, and compare latitudinally averaged annual PE changes (for 60 W S to 60 W N) for six formulae.PE increases

(
), who considers 'whether more reliable estimation of changes in PET can be obtained from physically based methods (e.g.Penman-Monteith) with uncertain data quality, or more empirical methods (e.g.Hargreaves) with more reliable input data'.Vorosmarty et al. () argue that 'Although these [physically based PE] methods are attractive on theoretical grounds, the degree to which the necessary input data sets can be successfully assembled … remains an open question.Use of more physically realistic evaporationfunctions must be weighed against potential inaccuracies in, and inconsistencies among, the several climatic forcing fields used by these methods'.The choice is further complicated when considering possible changes in plant behaviour under higher CO 2 concentrations (stomatal closure, increased plant growth, etc.).
but by the climate model, emissions scenario and time-horizon, and by location.These factors make it difficult to compare PE projections between studies, as do the varied ways of describing changes (i.e., percentage or absolute; monthly, seasonal or annual).It has been suggested that the choice of PE formula could be particularly important in regions where precipitation and PE are in close balance (e.g.,Kingston et al.   ).The maps in Figure1, based on MORECS data, show that this is likely to include parts of Britain, although changes in rainfall as well as PE make it difficult to predict precisely which areas are likely to be most affected.Bormann () suggests calculating the change in annual climatic water budget (annual precipitation minus annual PE) between baseline and future climates, to test the sensitivity of a given location to choice of PE formula; the choice is probably not crucial if changes are relatively consistent between formulae, but needs greater consideration if there is more variation.However, likely changes in the seasonality of rainfall under climate change (Murphy et al. ) mean that such annual water balance tests may mask important seasonal changes.Plots of monthly mean rainfall and PE under climate change may be more informative.In some circumstances, it may be necessary to extend the sensitivity study to the hydrological modelling, as the importance of PE is likely to vary according to the hydrological aspect under investigation as well as catchment location (e.g., Kay & Davies ).Nevertheless, several studies have shown that uncertainty due to climate model structure is greater than that due to PE formulation, implying that, where there is limited capacity for hydrological model runs, the priority should be to cover climate model uncertainty more comprehensively.The question remains: which PE formulae are likely to be more (or less) reliable when applied with climate model data?For each region of the globe, the answer requires investigation of: (1) which meteorological variables are most important for PE changes (i.e., the necessary level of complexity), and (2) the reliability of each of these meteorological variables when taken from different climate models.Together, these could enable the derivation of improved PE projections, and so improved hydrological projections.Further investigation is also needed into feedbacks between climate change and plant transpiration and growth.Clarification of the best ways to measure/model evaporation would also be useful for detection and attribution, as evaporation trends are easier to detect and attribute than precipitation or runoff trends (Ziegler et al. ; Douville et al. ).
) did not initially include wind speed, as wind data were not available from all the GCMs used in the statistical methodology(Sexton & Murphy ).Fisher et al.
Issues with PE estimation under climate changeProjections of PE from vegetated surfaces are generally not produced directly by Global or Regional Climate Models (GCMs or RCMs), so have to be made offline.But which PE formulae are likely to perform best when using climate model data, as opposed to observed weather data?One consideration is whether the climate model PE estimates will be used directly, so the absolute values need to be reasonable, The use of more complex formulae with climate model data is not necessarily straightforward, since some variables may be less reliable(Vorosmarty etal.; Kingston et al. ), or simply not available from all climate models.For example, the probabilistic UKCP09 projections (Murphy et al. mates (Shaw & Riha ), although some level of extrapolation may be acceptable where formulae perform well for a wide range of locations and climatic conditions (Sperna Weiland et al. ).Irmak et al. () suggest that, as many climate variables affecting PE have been chan-

Table 1 |
Sometimes, bias-correction is applied to allow for perceived biases in climate model data(Piani et al. ).Summary of PE projections for Britain Oudin formula is mid-way between that of the other two pairs of formulae.OudinPE is temperature-based but also includes extraterrestrial radiation (dependent only on Julian greaves PE (Hargreaves & Samani ) is an empirical formula using temperature, extraterrestrial radiation and diurnal temperature range (a humidity proxy).Sperna Weiland et al. () compared Penman-Monteith, Priestly-Taylor, Hargreaves and Blaney-Criddle that this was due to lower reliability of some of the extra variables required for Penman-Monteith.Ekstrom et al. () similarly found that Penman-Monteith PE derived from climate model variables for a baseline period was lower, in every month, than PE derived from weather observations.),andthatthesamebias applies for future periods.PE projections for BritainTable1summarises the PE changes for Britain presented in several references, described in more detail below.that percentage changes in Oudin PE are positive throughout the year, and larger in winter than summer, whereas changes in Penman-Monteith PE can be negative for some months and some climate models, and show greater monthly variability.Increases in annual Oudin PE are than Oudin in the north, but generally higher in the south.Ekstrom et al. () use data from the HadRM3H RCM over Europe (2080s A2) and show seasonal absolute differences between future and baseline Penman-Monteith PE. cal modelling of a catchment in north-west England they used Blaney-Criddle instead of Penman-Monteith PE, as the latter gave a spread of daily values too large compared with observations, and percentage increases that were too high (up to 80%) in summer (Blaney-Criddle gave a