Abstract
In a warming climate, the water budget of the land is subject to varying forces such as increasing evaporative demand, mainly through the increased temperature, and changes to the precipitation, which might go up or down. Using a verified, physically based model with 55 years of observation-based meteorological forcing, an analysis of the water budget demonstrates that Great Britain is getting warmer and wetter. Increases in precipitation (2.96.0 ± 2.03 mm yr–1 yr–1) and air temperature (0.20 ± 0.13 K decade–1) are driving increases in runoff (2.18 ± 1.84 mm yr–1 yr–1) and evapotranspiration (0.87 ± 0.55 mm yr–1 yr–1), with no significant trend in the soil moisture. The change in evapotranspiration is roughly constant across the regions, whereas runoff varies greatly between regions: the biggest change is seen in Scotland (4.56 ± 2.82 mm yr–1 yr–1), where precipitation increases were also the greatest (5.4 ± 3.0 mm yr–1 yr–1), and the smallest trend (0.33 ± 1.50 mm yr–1 yr–1, not statistically significant) is seen in the English Lowlands (East Anglia and Midlands), where the increase in rainfall is not statistically significant (1.07 ± 1.76 mm yr–1 yr–1). Relative to its contribution to the evapotranspiration budget, the increase in interception is higher than the other components. This is due to the fact that it correlates strongly with precipitation, which is seeing a greater increase than the potential evapotranspiration. This leads to a higher increase in actual evapotranspiration than the potential evapotranspiration, and a negligible increase in soil moisture or groundwater store.
I Introduction
Evapotranspiration affects many important physical aspects of the land: the dryness of the soil, vegetation growth, the temperature of the air and the amount of water in the rivers and groundwater reserves. Therefore, it is important to understand how evapotranspiration responds to changing climate.
For Great Britain (GB), studies based on the National River Water Archive (NRFA) have shown an overall increase in river flow of 1.6 mm yr–1 yr–1 over the last five decades (Hannaford, 2015). Meanwhile, an analysis of the meteorological data for the years 1961–2012 has shown an overall increase in potential evaporation for GB over that time of 0.76 ± 0.75 mm yr–1 yr–1 (Robinson et al., 2017b). With an increase of rainfall at 2.86 ± 2.22 mm yr–1 yr–1 (Robinson et al., 2017b) and assuming the evapotranspiration increases roughly in line with potential evaporation (Kay et al., 2013), this would imply an increase in soil moisture or groundwater recharge of about 0.5 mm yr–1 yr–1.
There is uncertainty in two of these estimates: firstly, the river flow record is based on a subset of rivers and the remainder – particularly in the Scottish Highlands, which contribute a large portion of the water – is gap-filled using a simple model. Secondly, the actual evapotranspiration may not follow the potential evaporation. The estimate made by Kay et al. (2013) was using a simple model with little representation of vegetation processes.
To answer the question about how the actual evapotranspiration has changed over the last 55 years, we would ideally have a direct observation of it both at the long term and over a representative large scale. However, evapotranspiration is difficult to measure. Direct observations of evaporation in GB from flux-tower data exist over short time periods or over small areas, but there are no large-scale, long-term observations of this elusive flux.
Since there are few long-term observations of evapotranspiration, the only way to study its evolution and response to changes in climate is by using a model.
However, evapotranspiration is also difficult to model as it depends on the trio of soil, vegetation and atmospheric conditions and the interactions between them. Key processes include the availability of soil moisture to plants during dry spells, the response of plants to temperature, sunshine and soil moisture, and the interception of rainfall by plants. Assumptions about the processes involved and their interactions (for a review of methods, see Wang and Dickinson, 2012) can have a significant impact on the resulting modelled evaporation (for an overview of model differences, see Schellekens et al., 2017). For instance, using too shallow roots for the vegetation will result in the premature reduction of evaporation in a dry spell (Teuling et al., 2009), ignoring interception and infiltration processes will mean that the surface runoff component will be insensitive to rainfall intensity (Dolman and Gregory, 1992), and including water use efficiency processes can alter the long-term response of the land to climate change (Prudhomme et al., 2014).
A previous study using a simple soil-moisture-stress-based model (Kay et al., 2013) indicated that there has been an increase in evaporation over the last few decades. However, this modelled trend is only due to changes in soil moisture stress and evaporative demand. The overall trend might be altered by other aspects of vegetation–atmospheric interactions such as rainfall interception and transpiration responses to changes in meteorology and soil moisture stress.
In this paper, a comprehensive land surface model will be used to diagnose the changes in evapotranspiration in GB over 55 years (1961–2015), including the analysis of the different components of evapotranspiration (soil surface, transpiration, interception).
Although the model is complex, there are always simplifications and assumptions made. The model used here is the Joint UK Land Environment Simulator (JULES), which is used by the UK Met Office in the Unified Model. It is also used by a community of land surface and hydrology researchers. The version used in this paper is the new standard configuration used for the UK. The paper is intended to act as a benchmark against which model developments and new research can be compared.
In order to interrogate the performance of the model and highlight new research agendas, the following questions will be addressed: Is the evaporation of GB and the regions increasing or decreasing? Which components of the evaporation are contributing to the trend? What meteorological changes are driving these changes?
We have not included land cover change in this analysis. There has been a 5% increase in forest cover over GB (www.fao.org), which, in practise, would make a small difference to the water budgets. But the model analysis for this benchmark paper would be made more complex by this additional variable. Future work would benefit from a comparison of the effect of climate change versus land cover change on the water budgets.
In Section II, the methodology, the model, the ancillary and the driving data will be described. In Section III, the model outputs will be presented and evaluated with available data, and the resulting trends in the model outputs will be analysed to address the three questions formulated above. Section IV then discusses the implications of these results and Section V presents the conclusions about the study.
II Method
This study uses a physically based land surface model, JULES (Best et al., 2011; Clark et al., 2011), driven by observation-based driving data for 55 years (see Section II.1). Mean monthly model outputs are extracted for four sites (details in Table 1; locations shown in Figure 1). The model outputs are extracted as annual and seasonal averages, averaged over GB and four regions (Scotland, Wales, England and the English Lowlands, which are a subset of England; Figure 1). Observational data are extracted at the same time and spatial scales for comparison. Where trends are quoted, they are calculated as the slope of the linear least squares fit of the annual time series for the different water budget variables. The reported trend errors are calculated as the difference between the trends and the 95% confidence interval of the linear fit, assuming a non-zero lag-1 autocorrelation (Robinson et al., 2017b).
Sites used to compare the CHESS data with JULES runs, with details of land cover.
*Land cover used in the single dominant vegetation cover runs (Figure 4).
Map showing the GB domain and four defined regions to be used in Section III: green is Wales, blue is Scotland, orange and pink together are England, orange alone is the English Lowlands. Site locations correspond to the positions of the flux sites used for validation.
1 JULES model description and setup
The JULES model includes many of the processes that are likely to affect changes in water loss (see Appendix 1). It is used here in a new assessment of the GB water balance for the years 1961–2015. The model output, ancillary files and meteorological forcing data are collectively referred to as CHESS (Climate, Hydrological and Ecological research Support System).
Using land cover maps (Centre for Ecology and Hydrology (CEH) Land Cover 2000: Fuller et al., 2002), soil maps (Harmonised World Soil Database (HWSD): FAO/IIASA/ISRIC/ISS-CAS/JRC, 2012) and a new spatially disaggregated meteorological driving dataset (Robinson et al., 2017b), the model has been run at a landscape scale of 1 km. This spatial scale is a compromise between being small enough to make reasonable comparisons with observational datasets (such as flux towers), and large enough to be able to obtain reasonable meteorological driving data and with a feasible computer cost.
The model used a fixed vegetation map based on observations in the year 2000 (CEH Land Cover 2000: Fuller et al., 2002) to prescribe the land cover fractions of the eight different categories: broadleaf (BL) trees, needle leaf trees, grass, crops, shrub, water, bare soil and urban. Parameter values for these land covers are part of the model configuration (see Appendix 2).
The phenology for each month was prescribed for the deciduous vegetation and the crops. The soil hydrology component of JULES is based on the Darcy–Richards Equations (see Appendix 1 for a summary) with vertical discretization into four layers. The parameters used in the equations depend on the soil type, and maps of these were derived from the HWSD (see Appendix 2 for a description of their provenance).
The meteorological dataset used in the simulations is described in Robinson et al. (2017b). This dataset is a combination of daily precipitation based on observations (Gridded Estimates of daily and monthly Areal Rainfall, GEAR: Keller et al., 2015) and other meteorological data mostly derived from the observation-based product MORECS (Hough and Jones, 1997; Thompson et al., 1981). The MORECS data are produced at 40 km resolution and, for CHESS, the data are downscaled to 1 km using information about topography. Robinson et al. (2017b) analysed the CHESS data and showed a positive trend in the air temperature as well as a positive trend in short-wave radiation, which is due to increasing short-wave radiation in the spring months. The short-wave radiation is based on sunshine hours and the CHESS data include a spatially varied impact of aerosol loading that affects the relationship between sunshine hours and short-wave radiation. However, it does not explicitly include any time variation of the aerosols (dimming and brightening). To some extent, the dimming and brightening is included implicitly as the sunshine hours are above a certain threshold of intensity, and, therefore, overall lower light levels will result in fewer hours. This is discussed in more detail in Section IV. The data have since been extended to 2015 (Robinson et al., 2017a).
The model represents a significant upgrade to the current product available, which employed a much simpler water balance model – the Met Office Regional Evaporation Calculation Scheme (MORECS) (Hough and Jones, 1997; Thompson et al., 1981) – that does not represent photosynthetic processes, has a simple soil-physics routine and runs on a 40-km grid square on a daily time step. The MORECS evapotranspiration is used widely and provides an interesting point of comparison for this new evapotranspiration data product (see Section III.2).
III Results
The analysis starts by evaluating the model results with observations at site and regional scales (Section III.1) and other models (Sections III.2 and III.3). Then we address the questions posed in the Introduction. The trends in annual and seasonal evapotranspiration in GB are calculated (Section III.4), and next we move on to study the regional differences in evapotranspiration in the regions of GB (Section III.5). These two sections answer the first question about evaporation and its trends in GB and the regions. The trends in the different components of evapotranspiration for the whole of GB are quantified to answer the second question about the components’ contribution to the trend in total evapotranspiration (Section III.6). The third question about meteorological drivers is addressed by studying the correlation between the evapotranspiration in the regions and the different drivers of change: precipitation and solar radiation (Section III.7).
Figure 2 presents maps of the time-average quantities of modelled evapotranspiration, sensible heat, soil moisture and soil temperature, and Figure 3 shows the spatial average annual values of runoff, evapotranspiration, soil moisture and soil temperature. The rest of the paper describes the provenance of these figures, evaluates the outputs and analyses the implications of the results and their trends (see Table 2).
Modelled estimates of mean evapotranspiration, sensible heat, soil moisture and soil temperature averaged over 1961–2015 for GB.
Modelled estimates of annual mean runoff, evapotranspiration, soil moisture and soil temperature, averaged over GB.
Annual average and seasonal average variables and trends for: precipitation, evapotranspiration, runoff, soil moisture, soil temperature (for the 3 m column), air temperature and potential evapotranspiration (PET).
1 Evaluating the model results with observations
This paper makes no attempt to calibrate the model further than has already been done for global simulations. But it is instructive for this analysis to quantify the performance of the model with a standard configuration. Ideally, we would be able to evaluate the model with observations at the 1 km scale with daily or monthly data. However, it is only possible to make direct observations of evapotranspiration using eddy-correlation systems which observe the hourly fluxes over an area of about 100 m. At the other extreme, river flow data can be used in combination with the precipitation to imply the evapotranspiration of a catchment (∼100 km) over a year. In this section, the results of the modelled actual evapotranspiration are compared to data from four eddy-correlation systems and the river flow records of the UK.
1.1 Evaporation from eddy-correlation systems
There are several flux sites across GB. Out of these, four have been selected, based on the following criteria: a good energy closure, running for several years when the CHESS data is available and representation of contrasting land cover types (trees and grass) and regional variation (Scotland and England). Their locations are shown in Figure 1, and Table 1 lists the data that are available, the dates and location coordinates of each site.
The observations at each site are compared with the output from the CHESS grid square in which the site lies. As each CHESS grid square contains a range of vegetation types, two model results are shown: with the grid square covered entirely by the vegetation at the site of each flux tower and the grid square with the fractional vegetation cover as used in CHESS (see Table 1). The original flux data are measured at 30-min intervals; the data are checked for quality and gap-filled before daily averages are calculated. From the daily averages, mean monthly values are calculated and presented here.
Figure 4 shows the comparison of the model output with the mean monthly observed data. The first thing to note is that CHESS tends to simulate higher evaporation than the observations for all of the sites. Following Blyth et al. (2010), we compare the evaporative fraction (the ratio of evapotranspiration to the sum of evapotranspiration and sensible heat flux) in Table 3. In this case, the model overestimates the total evapotranspiration at Griffin Forest and Cardington, while at Alice Holt the model underestimates evapotranspiration and no difference is found at Easter Bush. On average, however, there is a systematic overestimation of the evaporation in JULES, which has been noted previously (Van den Hoof et al., 2013).
Climatological mean monthly latent and sensible heat fluxes for each of the sites (black solid), compared to the CHESS square (green solid), as well as the same square rerun with the single dominant vegetation cover (green dashed). (For interpretation of the references to colours in this figure legend, refer to the online version of this article)
Summary of model results at the four flux sites: annual average evaporative fraction (modelled and observed) and precipitation (P). Annual average fluxes of modelled transpiration (Tr), soil surface evaporation (BS) and interception (I) as % of total evapotranspiration (Etot) and precipitation (P).
1.2 The seasonality of evapotranspiration
For the forest sites (Alice Holt and Griffin Forest) the main discrepancy is in the winter when the model substantially overestimates the evapotranspiration compared to the data: while the observations indicate values of latent heat around zero to 10 W m–2, the model has values of about 20 W m–2 (Figure 4). In the case of Griffin Forest, the energy for the modelled winter evapotranspiration is coming from the negative sensible heat flux (around –40 W m–2 compared to an observed value of around –10 W m–2). In the case of Alice Holt, the negative winter time sensible heat flux is reasonably matched by the observations for the run with single dominant vegetation cover (BL trees). The energy required by the model for the winter evapotranspiration must, therefore, be due to an overestimate of the net radiation balance.
In the case of the grass sites (Cardington and Easter Bush), the winter time fluxes are reasonably well modelled, but the summer modelled evapotranspiration is too high. In Cardington, this overestimation in latent heat is matched by an underestimation of sensible heat. In Easter Bush, there is no simultaneous reduction in sensible heat and, therefore, energy required for the high latent heat must be due to an overestimate of the net radiation.
The conclusions are the same for both types of model runs (single dominant vegetation cover and mixed vegetation cover), apart from Alice Holt, where the high winter latent heat fluxes are for the BL run. This can be explained by the fact that the CHESS square contains a substantial fraction of crops and grasses (62%; see Table 1).
1.3 Analysis of modelled components of evapotranspiration at the flux sites
Figure 5 presents the different components of the modelled evapotranspiration (soil surface evaporation, transpiration and interception). A summary of the results is shown in Table 3.
Seasonal variation in evapotranspiration observed (black solid) and modelled (green solid), and components – interception (cyan dashed), soil surface evaporation (orange dashed) and transpiration (green solid). (For interpretation of the references to colours in this figure legend, refer to the online version of this article)
There are no observations of these components at the four flux sites, so it is not possible to evaluate them directly. However, there are estimates of the three components for different vegetation types in the literature, which can be used to assess the model results. In order to translate between the quoted figures and the model, it is important to note that some authors present the results as the fraction of the total evaporation and others as the fraction of the total precipitation. For instance, Van den Hoof et al. (2013) summarizes work from a range of studies in Europe (Choudhury and DiGirolamo, 1998; Verstraeten et al., 2005; Wilson et al., 2001), which focus on the former and quote values of forest interception to range from 13% to 25% of the total evaporation, while for grasses it is closer to 10%. Meanwhile, Nisbet (2005) summarizes a large body of work by scientists studying interception and transpiration in the UK (Calder, 1990; Calder et al., 2003; Roberts, 1983), and focusses on the percentage of rainfall that is intercepted: about 20% of rainfall for BL trees and 35% for needleleaf. The evidence suggests that the annual fraction of rainfall intercepted by trees is fairly constant across a wide range of annual rainfall regimes, although increases slightly at low annual rainfalls (<500 mm yr–1). The data are sparser for grass, and no values are given in this report, although interception rates of heather and bracken are quoted at about 20% of rainfall.
The model results in Table 3 confirm the analysis of Nisbet (2005). The fraction of rainfall that is lost through interception is fairly constant over the four sites, ranging from 11% to 17%. This translates into a large fraction of total evaporation that is due to interception: 18% to 52%. At Griffin Forest, the model evaporative fraction is much higher than the observation (95% compared to 61%). Some 52% of the model evaporation is due to interception. Compared to the Van den Hoof et al. (2013) figures of interception, this might be considered to be too high. However, this only represents 17% of the precipitation in this high-rainfall area. Many of the studied sites from Van den Hoof et al. (2013) are in regions with much lower annual precipitation. The value of 17% is low compared to other estimates of interception loss in needleleaf trees, as reported in Nisbet (2005). In the much lower rainfall regime where Alice Holt is sited, the modelled interception is only 22% of the evaporation budget, which is dominated by the transpiration (54%). This analysis suggests that the model has a reasonable representation of interception, capturing its conservative relationship to the precipitation.
The transpiration, on the other hand, is more controlled by the available energy and has a more uniform relationship with total evaporation (see Roberts, 1983) ranging from 35% to 60%, and a wider range of fractions of precipitation, from 11% to 45%. The values of transpiration are lower than the values given by Van den Hoof et al. (2013) and Nisbet (2005), who quote values for trees from 53% to 70% of total evaporation being due to transpiration. It is possible that the total value of transpiration is reasonable, but the fraction is low due to the overestimation of the total evaporation.
The soil surface evaporation compares reasonably well with data quoted in the literature, although figures are rather low. The values range from 13% to 24% of the total evaporation and correspond to values quoted in Van den Hoof et al. (2013) of 20% and 30%. The low bias might be due to the fact that the UK sites are in relatively high rainfall areas with high interception, so that even if the total evaporation is correct, the percentage is low.
1.4 River flow
At the annual timescale, changes in water storage in a UK catchment are negligible and the area-average evapotranspiration can be estimated as precipitation minus river flow (Holmes, 1984). Daily river flow records for the UK are available in the NRFA, which is hosted at CEH, Wallingford. A subset of these records have been chosen for their length of record, accuracy, little or quantifiably affected by abstractions. The annual runoff from these catchments is then combined with a model to scale up to the regional and GB scale to present net annual and monthly runoff. The method for this is described in Hannaford (2015) and Marsh et al. (2015). Hannaford (2015) has summarized the results, which are reproduced here in Figures 6 and 7 and presented in Table 4. The challenge addressed here is to identify whether using a model to fill in the gaps introduces a bias to the results.
Comparison of CHESS annual runoff and precipitation–evaporation (P–E) to the observed GB annual river flow (NRFA; data available up to 2011).
Annual mean evapotranspiration in the four regions for CHESS and observed precipitation-runoff (P-Q). Here, the observation-based products used are GEAR for precipitation and NRFA for runoff.
Annual average water balance for the four regions.
Figure 6 plots the total GB annual river flows from these processed observations. The GB runoff and precipitation minus the evapotranspiration (the difference being the soil moisture) from the model is also shown. It is apparent from this figure and Table 4 that the modelled runoff is slightly lower than that observed, which is commensurate with the analysis in the previous subsection that the modelled evapotranspiration is too high. However, the percentage difference is very small (0.15%), which is not consistent with the previous estimates of the bias (roughly 10%). An analysis of the regions shows where this discrepancy occurs.
The definitions of different regions used by the NRFA are shown in Figure 1. They represent different GB climate types: Scotland is wet and cold; Wales is wet and warm; England is dry and warm; and English Lowlands is very dry and warm (these adjectives are relative to each other, not to a global standard).
Figure 7 shows the model evapotranspiration and observation-based estimate of evapotranspiration from precipitation and river flow in these regions, while Table 4 gives a summary of the annual average values for the regions. Table 4 shows that in Wales, England and the English Lowlands, the model overestimates evapotranspiration by 5% to 20%. This corroborates the results of the analysis with the flux data, which show an overestimation of about 10%. However, in Scotland the comparison indicates the opposite: the model has a lower evapotranspiration than the precipitation-runoff product. This anomalous outcome is probably a result of the way the river flow observations were sampled and then extrapolated to the regional scale. As already explained, the estimate is made with ‘Index Catchments’ chosen for the length of record (see Marsh et al., 2015), which in the Scottish region are mainly in the low-lying East of the region. Due to the short record length, the Highland and West of Scotland runoff data are not used and, instead, the area contribution was estimated by a model. The evapotranspiration in this wet region will be near to the potential evapotranspiration (PET), so the value of PET used in the model will dominate the result.
The PET product used for the gap-filling was from MORECS, which is based on observations and calculated at the 40-km grid scale. In flat terrain, it is a valid assumption that PET will not vary over a 40-km grid and that it can be used at the 1-km grid scale. However, in hilly terrain such as western Scotland, that is no longer the case. As indicated in Blyth (1999), since is it always windier on the tops of the hills where PET is lower (due to the cold temperatures), the area-averaged PET is lower in a hilly region than would be implied using topographic-mean data. This may explain the overestimation of PET and underestimation of the river flows in Scotland in the observation-based product of Hannaford (2015).
If the results for Scotland are ignored, the comparison suggests an overestimate of the model of the order of 10% to 15%.
2 Evaluating the model by comparison with other modelled estimates
There are several available large-scale evapotranspiration estimates to compare to the model. All of these estimates are model-derived. It is not expected that they will be more accurate than CHESS and this is not an evaluation exercise. However, it is interesting to make the comparison. Table 5 describes the datasets: their derivation and the annual mean evapotranspiration for GB and the four regions in Figure 1. They are derived from models of differing complexity, either driven by satellite-derived data or ground-based weather data. Figure 8 shows the maps of the annual mean evapotranspiration from the four products (CHESS scaled up to 40 km, eartH2Observe, GLEAM and MORECS), while Figure 9 shows how the GB average varies over time.
Description of the three large-scale evapotranspiration estimates for GB, and annual averages over GB and the four regions: Scotland (S), Wales (W), England (E) and English Lowlands (EL).
Maps of mean evapotranspiration for eartH2Observe, GLEAM, MORECS and CHESS (averaged up to 40 km).
Annual mean evapotranspiration for eartH2Observe, GLEAM, MORECS and CHESS.
It can be seen from the Figures 8 and 9 and from the Table 5 that the CHESS evapotranspiration is higher than GLEAM and lower than MORECS, and close to the ensemble product from eartH2Observe.
To understand the evapotranspiration in all of these estimates, it is important to know the assumed role and values of PET (or equivalent) in the model. Figure 10 shows the value of PET for GLEAM, MORECS and CHESS (not available for E2O). It is clear from this figure that the MORECS PET is higher than the CHESS PET, commensurate with the difference in actual evapotranspiration. This suggests that the difference between CHESS evapotranspiration and MORECS evapotranspiration is due to the PET. A comparison of CHESS PET with GLEAM PET shows that they have a similar magnitude. This indicates that the reason the GLEAM evapotranspiration is lower is due to model differences, not to the PET.
Annual mean potential evapotranspiration for GLEAM, MORECS and CHESS.
This result corroborates the results of the analysis in Schellekens et al. (2017), which showed that GLEAM had much lower evapotranspiration than a wide spread of models in the temperate regions.
3 Summary and discussions of model evaluation
The evaluation of the model show that the overall evaporation is overestimated by of the order of 10%. In winter, this overestimate is greater for trees than for grass, whereas for grass the overestimate is in the summer. According to analysis of the observations, the model underestimates the interception from trees, especially conifer trees.
The winter evaporation for deciduous trees tends to be overestimated. This is probably due to the known issue in the model of using a single aerodynamic resistance for the trees and underlying soil surface evaporation.
Analysis of river flows suggests that the use of a fine grid (1 km compared to 40 km) for evaluating evaporation results in more accurate estimates of the water budget over areas of high terrain.
Although comparison with observations is not perfect, the model displays reasonable allocation of evaporation across the different components compared to observational evidence, and we therefore deem it good enough to proceed with the subsequent analyses.
4 Trends in GB water balance: annual average and seasons
In the previous subsections, the annual average of total evapotranspiration has been presented and compared to observations (Table 4). Here, the seasonal variation in evapotranspiration is explored. The seasons are defined as Winter (December to February), Spring (March to May), Summer (June to August) and Autumn (September to November).
Table 2 and Figure 11 show that there is a strong seasonality to the evapotranspiration, with summertime values over four times the wintertime values. This seasonality is strongly driven by the seasonal variation in PET, which has a six-fold increase from the winter values to the summer values. Soil moisture control reduces the summer evapotranspiration to about 75% of the PET, while the winter and autumn evapotranspiration exceed the PET by, on average, 25%. The reverse is true of the river flow, which responds to a low seasonal variation in precipitation (only 1.35 variation between winter and summer) but exhibit some soil moisture control, with summer runoff 2.5 times smaller than winter runoff.
Annual and seasonal water budgets and trends (evapotranspiration and runoff). Trends are represented on the right y-axis.
The largest trend in evapotranspiration is seen in the spring months (1.12 mm yr–1 yr–1). This result corroborates the results of an analysis of PET by Robinson et al. (2017b), who demonstrated that the largest increase in the PET for GB was in spring due to an increase in spring sunshine hours and a decrease in relative humidity. In Section IV we discuss the impact of this on river flow.
Unlike the PET results, however, the smallest trend of evapotranspiration is seen in the autumn – this may be due to soil moisture control of evapotranspiration.
The trends in runoff are, overall, two times larger than the evapotranspiration trends. This is due to the very large increase in winter runoff due to the large positive trend in precipitation in Scotland. The other seasons have a similar upward trend of about 1 mm yr–1 yr–1.
5 Regional trends
The water balance of the different regions can be summarized as follows: the water balance is runoff-dominated in Scotland (runoff is 70% of rainfall) and evaporation-dominated in the English Lowlands (runoff is 30% of rainfall).
Table 6 and Figure 12 show that the trend in evapotranspiration is roughly constant between regions, in the same way that the total evapotranspiration is fairly constant. A map of the trend in evapotranspiration in Figure 13 (left) shows the spatial distribution of the positive trend. However, the runoff trends are more varied between regions. Scotland dominates the trend with an increase in runoff of 4.56 mm yr–1 yr–1, and the English Lowlands have the smallest trend of 0.33 mm yr–1 yr–1. The spatial distribution of the trend in runoff in Figure 13 (right) is a direct response to the patterns in the trend of the driving precipitation (see Robinson et al., 2017b: fig. B1c).
Annual average GB and regional fluxes and trends in the fluxes of precipitation, evapotranspiration, runoff, potential evapotranspiration (PET) and soil moisture. Bottom row shows annual observed averages from NRFA runoff product.
Annual GB and regional water budgets and trends (evapotranspiration and runoff). Trends are represented on the right y-axis.
Maps of trends in evapotranspiration and runoff.
6 Totals and trends in the components of evapotranspiration for GB
There are three components of the evapotranspiration: soil surface evaporation, transpiration and interception. The JULES model calculates each explicitly and their totals and trends for GB are given in Table 7. The percentage of evapotranspiration that is due to interception is about 30%, due to soil surface evaporation is about 20% and due to transpiration is due to about 50%. Despite the possible errors in the modelled allocation of evapotranspiration between the components reported in Section III.1, it is likely that the trends of the components are well represented. A key factor is the contribution of the trend of each component to the trend in total evapotranspiration. These vary between components and they are slightly different to the contributions they make to the annual average. This means they are not all increasing at the same rate: the interception is increasing more rapidly than the other two components. This is discussed in more detail in Section IV.
Annual average and trends in GB evapotranspiration and components. For each component, the percentage of total evaporation in shown in brackets.
Figure 14 displays how the components vary across the regions. Although transpiration generally dominates, the interception is nearly equal to the transpiration in Scotland which experiences very high rainfall rates.
Annual mean evapotranspiration components in GB and regions.
7 What meteorological variable is driving the trend?
To understand the reason for the trend in evapotranspiration, we calculate the correlation between the annual evapotranspiration and its two main drivers: precipitation and radiation. The analysis is carried out for the regions and for the separate components.
Figure 15 shows the results of the correlations. There is always a strong correlation between precipitation and interception and always a strong correlation of transpiration with radiation. Hence, each one of the two main drivers correlates strongly with interception or transpiration, while showing no correlation with the other. From the analysis, soil surface evaporation seems to have similar correlations with both precipitation and radiation. Further details on the distribution of these correlations are given with the correlation maps in the Figure S1 (online supplemental material).The implications for these results are discussed in the following section.
Correlation of annual precipitation (P) and short-wave radiation (SW) with Total evapotranspiration (Etot), Interception (I), Transpiration (Tr) and Soil surface evaporation (BS) for GB and the regions. Error bars show the 95% confidence interval for the correlation.
IV Discussion
1 Trends in the annual GB water budget
Hannaford (2015) suggests that the overall runoff of GB is increasing by 1.6 mm yr–1 yr–1, while Robinson et al. (2017b) show an increase in precipitation of 2.86 ± 2.22 mm yr–1 yr–1 (updated analysis of data including later years shows 2.96 ± 2.03 mm yr–1 yr–1; Table 2) If this was taken as the only evidence for a change in actual evapotranspiration, we could calculate an increase of 1.36 mm yr–1 yr–1. However, Robinson et al (2017a) indicate an increase in PET, which represents an upper limit of transpiration, of only 0.76 ± 0.75 mm yr–1 yr–1 (updated results in Table 2 show an increase of 0.74 ± 0.66 mm yr–1 yr–1) and Kay et al. (2013) have suggested that the overall trend in evapotranspiration of GB is close to the trend in PET, which they estimate as 0.7 mm yr–1 yr–1.
If all this were true, then there would be an increase in soil moisture or groundwater store of the order of 0.5 mm yr–1 yr–1.
The results of this study give slightly different results. The overall trend in evapotranspiration is 0.87 ± 0.55 mm yr–1 yr–1 and the trend in runoff is larger than that given by Hannaford (2015) at 2.18 mm yr–1 yr–1. This then almost balances the water budget, leaving negligible increase in soil moisture or groundwater (Table 8).
Overview of previous and new estimates of trends in the water budget for GB.
PET: potential evapotranspiration.
We have already demonstrated (Section III.1.4) that the Hannaford (2015) estimates of river flow are too low in Scotland. This was probably due to the use of large-scale area-average PET over hilly terrain to gap-fill the data, which tends to overestimate the evaporative loss, leading to runoff estimates that are too low. It is possible that the trend in runoff, rather than the total, is affected by the lack of interception trend (discussed in Section IV.4) in the estimated runoff. A deeper analysis of the flows in Scotland should be carried out to understand these trends.
The reason why the evaporative loss might be greater than the potential could be due to the large fractional component of interception (30%): in the wet and windy areas (West Scotland), there is a higher fraction of evergreen needleleaf trees, which have a high interception capacity (see Calder, 1990). The evaporation from a wet forest often exceeds the PET, drawing down energy in the form of negative sensible heat (i.e. cooling the air) to drive it (Stewart, 1977).
If the drivers of interception are increasing faster than the PET, then with such a high interception fraction, there might be a larger trend in the evapotranspiration than the potential evaporation. This is discussed in Section IV.4.
It should be noted that the trends calculated form these model results are subject to the model biases that are explored in Section III. The overall evaporation is low compared to the observed data, and the interception is particularly low compared to the estimates from the Forestry Commission (Nisbet, 2005). If it is true that the trend in evapotranspiration is higher than the trend in PET due to the presence of interception, then it is possible that the true trend might be higher still if that is being underestimated.
2 Impact of seasonal variation of trend in evapotranspiration on river flows
The results of this study indicate a clear increasing trend in spring evapotranspiration, driven by an increase in PET.
It is interesting to note the impact of this on the river flows. In their review, Watts et al. (2015) suggest that there is no clear evidence of how actual evaporation has changed over the last few decades. Hannaford (2015) notes that, while there is an observed increase in winter, summer and autumn river flows from 1961 to 2010, there is a decrease in the spring river flows (see Table 1). This is reiterated in Harrigan et al. (2017), who note that we need an ‘improved understanding of the drivers of these seasonal changes in river flow’.
This study clearly indicates that some of the springtime drop in runoff will be due to increased evapotranspiration in addition to a reduction of rainfall.
3 Temporal changes in short-wave radiation over the study time period in GB
There is an overall increase in downward short-wave radiation apparent in the observations used in this study of 1.1 ± 0.7 Wm–2 decade–1 over GB.
There are two possible reasons for an increase in short-wave radiation: one is that there is less cloud and the other is that the air is more transparent due to a reduction in aerosols (pollution).
The aerosol effect has been widely studied and is referred to as ‘dimming and brightening’, since the aerosol amount increased throughout the 20th century until about the 1970s when, due to legislation, it began to fall. Observations reported by Wild (2009) illustrate this at the global scale. Across Europe, the dimming resulted in a decrease in the total short-wave radiation of 1.4 Wm–2 decade–1 from the beginning of the century to the mid-1970s, and the brightening to an increase of 2.2 Wm–2 decade–1 up to the mid-1990s. This change in radiation has been shown to have a significant impact on river flows in the region (Gedney et al., 2014).
GB has a somewhat different temporal pattern of dimming and brightening to mainland Europe. Using an ensemble of computer model runs, Folini and Wild (2011) simulated the aerosols in Europe. Their results agree with the overall result of Wild (2009) for Europe, but also note some interesting regional variations: in the case of GB, from 1960 onward, there is no dimming, only brightening. For the all-sky conditions (i.e. including clouds; see Folini and Wild, 2011; Figure 9) the increase in radiation is about 0.8 Wm–2 decade–1. This tallies with an observational study by Stjern et al. (2009), which included a long-term record at Aberdeen (the only GB station in the study) where there was no dimming from the 1960s, only a brightening of 1 Wm–2 decade–1. This is also consistent with the increase of short wave (0.9 ± 1.1 Wm–2 decade–1 for the years 1979–2012) given in the observation-based meteorological forcing dataset WFDEI (Watch Forcing Data ERA Interim) (Weedon et al., 2014), which included explicit aerosol effects.
The short-wave radiation data used in CHESS are based on sunshine hours. There is a spatially varying aerosol effect included, but it is based on data from 1978 and does not change in time. As reported in Robinson et al. (2017b), it may be that the aerosol effect is implicit in the sunshine hours record as there is a threshold for bright sunshine of 120 Wm–2, and the number of hours above that threshold will be affected by the pollution levels, although the changes in cloud cover are likely to be dominant in this signal.
The biggest increase in short wave is in the spring. This is consistent with data from across Europe (Sanchez-Lorenzo et al., 2008) and may be due to changes in the Atlantic Multidecadal Oscillation, which has resulted in decreased spring precipitation in Northern Europe (Sutton and Dong, 2012).
It is concluded, therefore, that the CHESS data give a reasonable representation of the short-wave radiation over GB for the study period.
4 Analysis of correlations
It is generally accepted (Teuling et al., 2009; Wang and Dickinson, 2012) that in moisture-limited regions, there will be a strong correlation of evapotranspiration with precipitation, while in wet, cloudy regions, there will be a stronger correlation with radiation. However, in the analysis presented by Teuling et al. (2009), most of GB is deemed to be radiation limited (confirming its wet and cloudy climate) apart from west Scotland and central England. The result for west Scotland is a counter-intuitive result since it is the wettest region of GB.
In the correlation analysis of annual evapotranspiration with precipitation and short-wave radiation, the same patterns emerged: the correlation of evapotranspiration and precipitation was high in Scotland. To understand why the evapotranspiration correlates more strongly with precipitation than radiation in Scotland, the correlations for the four regions for each of the components of evapotranspiration against precipitation (P) and short-wave (SW) radiation were made.
This analysis shows that there is always a strong correlation between rainfall and interception (Figure 14). This tallies with the results of the analysis in Section III and the evidence from Nisbet (2005) that interception is near to a fixed fraction of rainfall. Interception is not ‘limited’ by rainfall, but it strongly correlates with it because it is unlimited.
Transpiration, however, is generally limited by energy in GB since there is usually enough water and often not enough sunshine. This results in a strong correlation of transpiration with radiation (Figure 14).
Soil surface evaporation has similar correlations with both precipitation and radiation. These results show why the evapotranspiration–precipitation correlation dominates in Scotland: it is because the rainfall is high and, so, interception is nearly as large as the transpiration component of the evapotranspiration (see Figure 14). English Lowlands provide the opposite case. The evapotranspiration is dominated by transpiration (see Figure 14) and it is equally water limited and radiation limited.
5 Importance of interception to GB hydrology
As discussed in Section IV.4, this study highlights the importance of interception to the overall hydrological balance and trends in GB. It may also have implications for floods.
Floods represent a major hazard for the UK and recently there has been effort to identify natural ways to alleviate the floods. Dadson et al. (2017) and Stratford et al. (2017) both review the evidence of whether the presence of trees have an impact to reduce flood peaks. They both conclude that there can be an impact of mature trees to reduce smaller floods, probably due to the interception.
Using a physically based model (JULES) with detailed representations of all the evaporation components, including interception, enables us to quantify the effect of land cover on hydrology at the GB scale. The interception in the model has some uncertainties, and possibly underestimates the effect (see Section III.1.3). Some of the uncertainty is due to the size of the interception store, some to do with the efficiency of evaporation (the aerodynamic resistance) and some to do with how the rainfall intensity distribution is represented (see Appendix 1.2) both in time and space. All of these aspects would benefit from further investigation. This paper is only able to highlight the importance of the role of interception in the modelled trend, thus motivating new research. However, the results are consistent with the observations and can be used as evidence for this increasingly important issue.
6 Possible impact of rising CO2 levels on water balance of GB
Over the 55 years of the study period, there has not only been a change in the physical climate, but a 50% increase in the atmospheric concentration of CO2, from 300 ppm to 410 ppm (Keeling and Keeling, 2017). It is possible that this has affected the water balance of GB due to the response of vegetation to CO2. There is evidence (Leakey et al., 2009) at the canopy scale that plants will transpire less in increased atmospheric CO2 levels. However, models differ substantially in their prediction of the large-scale, long-term response to elevated CO2, which includes changes to ecosystems and vegetation structure (De Kauwe et al., 2013).
The JULES model includes the effect of an increase in CO2 on evapotranspiration, although it might reduce the transpiration too much (Prudhomme et al., 2014). As a result of this uncertainty, in the version of CHESS reported here, we used the CO2 level half way through the period (in 1986) and kept it constant.
However, it is instructive to analyse to what extent the water balance would be predicted to change in the presence of a 50% increase in CO2 according to the model. The model showed that overall evapotranspiration and runoff were unchanged. There was, however, a change in the trend: the evapotranspiration trend was reduced to 0.60 ± 0.47 mm yr–1 yr–1 (from 0.87 ± 0.55 mm yr–1 yr–1), while the positive trend in runoff increased to 2.33 ± 1.88 mm yr–1 yr–1 (from 2.18 ± 1.84 mm yr–1 yr–1, with no CO2 increase). Most of the difference in trend comes from a lower positive trend in transpiration of 0.11 mm yr–1 yr–1 (compared to 0.47 mm yr–1 yr–1 in the constant CO2 run). In the earlier part of the run, when transient CO2 levels were lower than the constant CO2 run, the stomata were more open, so stomatal conductance was higher, while the reverse was true in the later part of the run, with higher CO2 levels. This leads to the same mean transpiration over the run, but a weaker trend.
V Conclusions
This study set out to explore the long-term evolution of the water budget of GB, including how and why it is changing. The model used is the UK community model JULES, with the new standard configuration for the UK.
A study of the river flows (summarized in Hannaford, 2015) suggest that the runoff from GB is increasing with time (1.6 mm yr–1 yr–1 over the last 50 years), with a decrease in springtime (Harrigan et al., 2017). It is possible that the increase is an underestimate due to the methods of gap-filing used.
The PET, which expresses the likely increase of a well-watered grass due to changes in wind, radiation and temperature, has been estimated in this study to be 0.74 ± 0.66 mm yr–1 yr–1, and by Kay et al. (2013) to be 0.7 mm yr–1 yr–1. In order to examine the actual evapotranspiration, we used a validated model of the land surface (JULES), which was driven by observation-based meteorological data (Robinson et al., 2017a) over five decades. The modelled evapotranspiration increased at a rate of 0.87 ± 0.55 mm yr–1 yr–1, which is greater than the increase in PET.
The analysis showed that the interception was a large component of the overall evapotranspiration (30%) and increasing at a slightly higher rate (proportionally) than the other components.
In this paper, it was demonstrated that annual interception correlates strongly with annual precipitation rather than solar radiation. Over the last five decades, precipitation has increased faster (2.96 ± 2.03 mm yr–1 yr–1) than the PET (0.74 ± 0.66 mm yr–1 yr–1). This increase in precipitation, combined with the high interception rates in GB (due to the wet conditions) explains why the trend in evapotranspiration is higher than the trend in PET.
Thus, the paper clearly demonstrates that a representation of all three components of evapotranspiration (interception, transpiration and soil surface evaporation) are needed to understand the trends in the GB water budgets.
In addition, the study demonstrates the importance of evapotranspiration to the annual and seasonal water budget of GB and its regions. For instance, the observed decreasing springtime runoff reported in Harrigan et al. (2017) is a combination of a decrease in precipitation and an increase in evaporation in this season.
Future directions for this research should include a focus on the evaporation processes that have been demonstrated to be so important: interception, the wintertime soil surface evaporation under deciduous trees and the summertime transpiration of grass. Narrowing our uncertainty in these key processes will help identify the role of land cover on the UK water budgets.
Supplemental material
Supplemental Material, FigureS1 - Trends in evapotranspiration and its drivers in Great Britain: 1961 to 2015
Supplemental Material, FigureS1 for Trends in evapotranspiration and its drivers in Great Britain: 1961 to 2015 by Eleanor M Blyth, Alberto Martínez-de la Torre and Emma L Robinson in Progress in Physical Geography: Earth and Environment
Supplemental material
Supplemental Material, tableb1 - Trends in evapotranspiration and its drivers in Great Britain: 1961 to 2015
Supplemental Material, tableb1 for Trends in evapotranspiration and its drivers in Great Britain: 1961 to 2015 by Eleanor M Blyth, Alberto Martínez-de la Torre and Emma L Robinson in Progress in Physical Geography: Earth and Environment
Footnotes
Data and model availability
This paper presents the results of the model in a particular code version and configuration. The version of a model refers to the codes that are used to solve the equations of the processes (this can be summarized by its Version Number). The configuration includes the selection of the options used (such as whether to use dynamic vegetation or not), the look-up tables of parameters used in the equations, the meteorological data used to drive the model and the ancillary data used to set up the maps of soils and land cover classification. This will also have a Version Number and a reference name. This information is given in 
.
Code availability
The JULES code is available freely and can be applied for through the JULES repository: https://code.metoffice.gov.uk/trac/jules. The version used for the CHESS runs based on JULES version 4.5, branch: r3488_albmar_spdm. For more detailed information about the JULES code, use and availability, visit .
Data availability
The data (drivers and initial inputs) of the configuration for running the model for CHESS is described in the rose suite number u-au394 (see Appendix 2). All the model outputs analysed here, and more (evaporation components, runoff, latent and sensible fluxes, soil moisture and temperature, surface temperature, carbon gross and net productivities), are publicly available as the CEH CHESS-land dataset (Martínez-de la Torre et al., 2018). The flux data used for evaluation are available upon request from the author.
Declaration of conflicting interests
The author(s) declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The author(s) received no financial support for the research, authorship, and/or publication of this article.
Supplemental material
Supplemental material for this article is available online.
Appendix 1
Appendix 2
The configuration (ancillary files, science options and parameters, and driving data) used for the CHESS simulation presented here are all documented in the rose suite u-au394 (https://code.metoffice.gov.uk/trac/roses-u/browser/a/u/3/9/4/). Instructions for how to access this are given in the website: http://jules.jchmr.org. A summary of the options and driver datasets is given here.
The model is run with eight surface tiles (five plant functional types: BL trees, needleleaf trees, grasses, shrubs and crops; and three non-vegetated types: open water, bare soil and urban) derived from the CEH Land Cover 2000 (Fuller et al., 2002) map at a resolution of 25 m. Table S1 shows how the Land Cover 2000 classes were mapped onto JULES land cover types. These were then aggregated to 1 km resolution, as fractions of the total grid box.
The soil hydrology component of JULES is based on the Darcy–Richards Equations (see Appendix 1 for a summary) and, in this configuration, the Van Genuchten (1980) approach, with the vertical discretization into four layers of varying depth: 0.0–0.1 m, 0.1–0.35 m, 0.35–1.0 m and 1.0–3.0 m. The soil hydraulic characteristics are assumed to be spatially uniform for each grid cell, and have been calculated for the model domain from the HWSD (FAO/IIASA/ISRIC/ISS-CAS/JRC, 2012), by classifying soils by their texture, then using the values from Wösten et al. (1999). We used a newly developed terrain slope dependency for the PDM scheme in the production of saturation excess runoff (Martínez-de la Torre et al., 2019), and the slope was derived from the GB 50-m resolution CEH-IHDTM (Institute of Hydrology Digital Terrain Model) database (Morris and Flavin, 1990, 1994).
The phenology for each month was prescribed for the deciduous vegetation and the crops. The dynamical vegetation scheme was switched off. A 10-layer approach is used for canopy radiation interception, including an exponential decline of leaf nitrogen with canopy height. For the run with increasing atmospheric CO2 concentration, we used annual values from the US National Oceanic and Atmospheric Administration Global Monitoring Division (https://www.esrl.noaa.gov/gmd/ccgg/trends/global.html).
A 10-year spin-up run was conducted to initialize the model. The meteorological data used to drive the model are publicly available: CHESS-met (Robinson et al., 2017a). This is a daily dataset based on observations from 1961 to 2015. The model is integrated at a half-hourly time step, using a daily disaggregation scheme (Williams and Clark, 2014) to disaggregate the driving data. In terms of precipitation, precipitation events start at a random time during the day and last for two hours in the case of convective precipitation, or five hours in the case of large-scale precipitation. The input rainfall is assumed to be convective for temperatures above 15°C, and covers the complete 1-km grid cell. The rest of the options and parameters (radiation, snow, vegetation, non-vegetated tiles) are available on the rose suite u-au394.
References
Supplementary Material
Please find the following supplemental material available below.
For Open Access articles published under a Creative Commons License, all supplemental material carries the same license as the article it is associated with.
For non-Open Access articles published, all supplemental material carries a non-exclusive license, and permission requests for re-use of supplemental material or any part of supplemental material shall be sent directly to the copyright owner as specified in the copyright notice associated with the article.
