Climate Dynamics

, Volume 42, Issue 7, pp 1715–1732

Sensitivity of a coupled climate model to canopy interception capacity


    • School of Geographical SciencesUniversity of Bristol
  • P. J. Valdes
    • School of Geographical SciencesUniversity of Bristol
  • C. D. Jones
    • Met Office Hadley Centre
  • J. S. Singarayer
    • Department of MeteorologyUniversity of Reading

DOI: 10.1007/s00382-014-2100-1

Cite this article as:
Davies-Barnard, T., Valdes, P.J., Jones, C.D. et al. Clim Dyn (2014) 42: 1715. doi:10.1007/s00382-014-2100-1


The canopy interception capacity is a small but key part of the surface hydrology, which affects the amount of water intercepted by vegetation and therefore the partitioning of evaporation and transpiration. However, little research with climate models has been done to understand the effects of a range of possible canopy interception capacity parameter values. This is in part due to the assumption that it does not significantly affect climate. Near global evapotranspiration products now make evaluation of canopy interception capacity parameterisations possible. We use a range of canopy water interception capacity values from the literature to investigate the effect on climate within the climate model HadCM3. We find that the global mean temperature is affected by up to −0.64 K globally and −1.9 K regionally. These temperature impacts are predominantly due to changes in the evaporative fraction and top of atmosphere albedo. In the tropics, the variations in evapotranspiration affect precipitation, significantly enhancing rainfall. Comparing the model output to measurements, we find that the default canopy interception capacity parameterisation overestimates canopy interception loss (i.e. canopy evaporation) and underestimates transpiration. Overall, decreasing canopy interception capacity improves the evapotranspiration partitioning in HadCM3, though the measurement literature more strongly supports an increase. The high sensitivity of climate to the parameterisation of canopy interception capacity is partially due to the high number of light rain-days in the climate model that means that interception is overestimated. This work highlights the hitherto underestimated importance of canopy interception capacity in climate model hydroclimatology and the need to acknowledge the role of precipitation representation limitations in determining parameterisations.


InterceptionGCMLand–atmosphere interactionsEvapotranspirationCanopy storage capacity

1 Introduction

Evapotranspiration from the Earth’s surface is an important part of the Earth’s climate system. It affects the balance of sensible and latent heat fluxes that help to determine surface air temperature and thus the thermodynamics and dynamics of the atmosphere. Evapotranspiration itself is an amalgamation of three processes: evaporation of water from the surface of wet vegetation canopies; evaporation from the soil; and transpiration from plants. The size of the stores of water in the canopy and soil partly determines the rate of total evapotranspiration, which in turn contributes to precipitation rates. The canopy evaporation represents the quickest return of precipitated water to the atmosphere, rather than entering the soil or becoming runoff. Therefore it is especially important in sustaining continental precipitation (Scott et al. 1995; Savenije 2004).

The canopy interception capacity (Cm) is one of the main parameters controlling how much water is evaporated, since it controls how much water is available for evaporation from the canopy. The canopy interception is key to evaporation and precipitation in land surface models and without it, their representation suffers (Desborough et al. 2001). The canopy interception also affects the ratio of evaporated to transpired water, which has been shown to have an important impact because of changes to cloud and vegetation feedbacks (Wang and Eltahir 2000).

The amount of canopy evaporation, the size of the canopy interception store and the ratio of transpiration to evaporation are issues that are not yet resolved globally. However, there is now a substantial body of Cm data available (Eckhardt et al. 2003; Breuer et al. 2003). Moreover, new techniques and more extensive studies have rectified the paucity of global data on the partitioning of evapotranspiration and its relationship with precipitation (Miralles et al. 2010; Jasechko et al. 2013; Kool et al. 2014). These new data make better assessment of Cm parameterisation now possible.

This paper aims to: (a) establish that compared to Cm measurements, the Cm values used in the land surface models in climate models are often lower than measured values and have a small range (Sects. 1.1, 1.2); (b) show that realistic ranges of Cm values result in significant impacts on the mean annual global temperature compared to the control (Sect. 3.1); (c) show that in one particular climate model, HadCM3, reduced Cm values improve the representation of evapotranspiration in the model (Sect. 3.2); and (d) suggest a new value for the parameterisation for Cm in HadCM3 (Sect. 4).

1.1 Canopy interception capacity measurement values

The size of Cm according to field measurements is very variable (see Table 1). Different vegetation types (Breuer et al. 2003), regions, seasonal effects (Link et al. 2004; Macinnis-Ng et al. 2014), inter-annual effects (Gerrits et al. 2010), precipitation intensity (Keim et al. 2006) and the length of dry periods (Klaassen et al. 1998) all affect the size of Cm. The measured range of Cm is large, with a range between less than 0.1 mm to more than 10 mm. Even accounting for some differences in values from measurement methods, Cm is highly spatially and temporally variable (Klaassen et al. 1998; Dunkerley 2000). This spatial and temporal variability is a challenge for models, especially climate land surface models that have a relatively low spatial resolution.
Table 1

Canopy interception capacity (Cm) ranges (in mm) and interception as a percent of precipitation (%) from the literature for three general vegetation types


Cm (mm)































To simplify, crops have been included as grasses, under story is included as shrubs, and broadleaf and coniferous trees are considered together. Min minimum, Max maximum. Mean values only available for temperate regions and are taken from combined Breuer et al. (2003) values. All values given to one decimal place. Data collated from: Breuer et al. (2003): temperate; Dykes (1997): tropical trees; Dunkerley and Booth (1999): arid; Dunkerley (2000): arid; Garcia-Estringana et al. 2010: arid shrubs; Motahari et al. (2013): arid trees; Herwitz (1985): tropical trees

1.2 Interception, canopy evaporation and canopy capacity in models

When surface interception schemes were first implemented into land surface models there was some indications of changes to the surface energy balance (Pitman et al. 1990, 1993; Dolman and Gregory 1992). Therefore it is important to consider the whole system into which the Cm parameterisation fits, as the different aspects of the model are interconnected and can have subsequent effects when changed. In hydrological models, Cm controls the partitioning between intercepted water and throughfall (refer to Fig. 1), which can significantly affect evapotranspiration partitioning and rainfall (Wallace et al. 2013). The interception part of these models often combines the Cm with the potential evaporation rate and the amount of water in the canopy retained from the previous timestep.
Fig. 1

Representation of the surface hydrology in MOSES, showing the main flows, stores and partitioning of water. The precipitation is initially split into canopy interception and throughfall. The canopy interception is stored as canopy water content and cannot be any larger than the canopy interception capacity (Cm), which is the parameter examined in this study. In MOSES the canopy water includes the uppermost soil water, roughly equivalent to puddles. Note also that the soil evaporation and plant transpiration are combined (referred to here as soil-evapotranspiration)

Compared to interception models and Cm measurements, the representation of canopy evaporation and interception in land surface models used for climate modelling are necessarily simplified and altered. Interception models (e.g. Rutter et al. 1971; Gash 1979; etc.) require the canopy interception capacity to be a parameter which can be defined as the “minimum depth of water required to saturate the canopy expressed as millimetres per unit of projected area of canopy cover” (Lloyd et al. 1988; Miralles et al. 2010). By contrast, the MOSES land surface model (Essery et al. 2001) used here requires Cm to be “the amount of water which is freely available for evaporation from the surface” (Cox 2001). The difference in the parameter definition makes a fundamental difference between the model types. Where interception models consider canopy evaporation to be demand limited, the MOSES land surface model uses Cm to limit the supply of water for evaporation. For this reason, interception models tend to disregard the very high Cm values because they represent a super-saturated canopy (i.e. more than the minimum depth of water required to saturate the canopy). However, as represented in MOSES, a super-saturated canopy is a legitimate value because canopy evaporation is limited by the available water in Cm.

Land surface models also tend to use a smaller range of values than those measured. Most land surface schemes uniformly parameterize Cm based on the leaf area index (LAI) or as universal global value (e.g. 0.5 mm in HTESSEL, Teuling et al. 2009), ignoring the different water holding characteristics of different plant functional types and species. A multiplier of around 0.1 or 0.2 with the LAI is typical, though some are much smaller (e.g. 0.0002 LAI in JSBACH, Roeckner et al. 2003). As well as the LAI, some models acknowledge the role of trunk capacity and additionally include a stem in the Cm calculation (Lawrence et al. 2007). This results in a large number of parameter structure combinations that actually have a relatively small range of real values (up to a maximum of ~2 mm), compared to the measured values (which can be as much as 10 mm). Land surface models may also include a parameterisation of surface water or ‘puddling’, combined within the Cm or separately, as it is water immediately available for evaporation. The values of ground surface capacity in models vary considerably. The land surface model SiB2 uses a maximum ground surface capacity of 0.2 mm (Sellers et al. 1996), MOSES and JULES use a universal static ground surface capacity of 0.5 mm (Cox 2001; Essery et al. 2001; Best et al. 2011) and many other models exclude this element altogether.

One of the major reasons for the disjoint between the measured values of Cm and those used in climate models is that the variability of Cm and precipitation is subgrid scale (Wang and Eltahir 2000). In the land surface model MOSES the statistical approach used to combat the subgrid problem means that convective rainfall only falls on 30 % of the tile, whereas large scale rainfall falls on the entire tile (Gregory et al. 1994). This is combined with the Warrilow et al. (1986) representation of throughfall, which is slightly different to the widely used Shuttleworth (1988) scheme. When the canopy is not yet full, the throughfall increases proportionally with the canopy saturation. I.e. when the canopy is 40 % saturated, 40 % of precipitation is throughfall, when the canopy is 80 % saturated, 80 % of precipitation is throughfall, etc. Other land surface models may require the canopy to become saturated before throughfall occurs [such as Shuttleworth (1988)] or have a fixed proportion (10–30 % is typical) or a dataset which varies temporally and spatially (Wang and Eltahir 2000). However, subgrid heterogeneity is still a common problem in land surface and atmosphere models. Since precipitation is the largest control on the hydrological cycle, the subgrid heterogeneity of precipitation tends to be more important than that of Cm and therefore in many cases the Cm or throughfall is effectively tuned to the model’s precipitation regime.

The interception and Cm are particularly important to the representation of evapotranspiration partitioning. A study of the effect of a number of different interception schemes in the land surface model CABLE coupled to the CSIRO climate model found that Cm affects the partitioning of evapotranspiration but not the temperature, precipitation or the overall climate (Mao et al. 2011). However, this work doesn’t extend the analysis to different values as well as model structures, resulting in a relatively small range of real values. Other analyses of land surface models have taken a simpler approach and reduced the magnitude of Cm in order to improve evapotranspiration representations, including in the models CLM3 (Lawrence et al. 2007; Oleson et al. 2008) and JULES (Demory 2009; Van den Hoof et al. 2013). These studies show intriguing potential for the importance of Cm, but are limited by availability of data; these land surface models can be run globally but they are verified using individual site observations. Further, all these previous studies have only considered a single different Cm value, from either another model or a speculative change. No work so far has done a sensitivity analysis of Cm using a full range of the available measured values.

Here we use Cm values from the literature to re-parameterize the MOSES land surface model to fully understand what the sensitivity of the climate and surface hydrology is to changes in Cm. We will also establish how well the model performs compared to the near global datasets of canopy evaporation and transpiration now available. This will enable a recommendation of what could be a good parameter value for Cm in MOSES and increase understanding of the role of Cm in climate models.

2 Methods

2.1 Model description

The model used in the simulations in this study is the UK Met Office Hadley Centre’s HadCM3. HadCM3 is a three dimensional, fully coupled, fully dynamic ocean, non-flux adjusted global climate model (Collins et al. 2001). The atmosphere component, HadAM3, has a cartesian grid with a horizontal resolution of 2.5° × 3.75°, 19 vertical levels and a time step of 30 min (Pope et al. 2000). The ocean and sea-ice component has a 1.25° × 1.25° horizontal resolution, with 20 vertical ocean levels. Though not from the latest generation of climate models, HadCM3 remains an extensively used model for many research applications around the world due to its speed, which mean that long integrations and many ensemble members can be run. Its computational efficiency continues to make it ideal for performing multiple runs to explore the sensitivity to a range of parameter values (e.g. Murphy et al. 2007; Liu et al. 2012; Booth et al. 2012; Jackson and Vellinga 2013).

The land surface scheme used for the atmosphere component of HadCM3 is the Met Office Surface Exchange Scheme, MOSES 2.1 (Gregory et al. 1994; Cox et al. 1999). MOSES is run in these simulations with an additional vegetation and terrestrial carbon model, top-down representation of interactive foliage and flora including dynamics (TRIFFID) (Cox et al. 1998; Cox 2001). TRIFFID predicts the vegetation based on plant functional types using a competitive, hierarchical model. Here we use the TRIFFID equilibrium mode, which quickly brings the vegetation cover into equilibrium by running 50 years of TRIFFID for each 5 years of the climate model run. TRIFFID and MOSES have nine land surface types, five of which are vegetation: broadleaf trees, needle leaf trees, shrubs, C3 grasses and C4 grasses, which have different LAI, maximum Cm and other phenological differences in the model. Soil moisture in the model is represented on four layers of thicknesses (measured from the top) of 0.1, 0.25, 0.65 and 2.5 m (Essery et al. 2001).

JULES is the latest version of the land surface scheme MOSES and is used as both as an offline land surface model (Best et al. 2011; Clark et al. 2011) and within the latest Hadley Centre climate models, such as HadGEM2 (Martin et al. 2011; Collins et al. 2011). The representation of Cm in JULES is relatively little changed compared to MOSES 2.1 (Best et al. 2011; Clark et al. 2011; Van den Hoof et al. 2013) and hence we believe our results are applicable to this model as well.

2.1.1 Canopy interception capacity representations in MOSES

Cm is the arbiter of throughfall in MOSES and therefore the main control on whether precipitation reaches the soil (see Fig. 1). The amount of water intercepted is dictated by the balance of the Cm, the current canopy water content, the canopy saturation and the rainfall amount and duration. Though Cm usually only refers to the water caught by the plant canopy, MOSES includes 0.5 kg m−2 as the ground surface capacity (‘puddling’) on all vegetated and bare soil surfaces within Cm (Essery et al. 2001). For vegetated surfaces, canopy capacity is calculated linearly from the LAI:
$$C_{m} = C_{g} + C_{C} LAI$$
where Cm = canopy capacity (kg m−2), Cg = 0.5 kg m−2 is the ground surface capacity and CC = 0.05 is the multiplier with LAI to give the vegetation canopy capacity. The ground surface capacity (Cg) is in effect the minimum canopy capacity in kg m−2 (Cox 2001; Best et al. 2011). The 0.5 mm ground capacity is separate and in addition to the soil moisture evaporation. Soil moisture evaporation is represented in two places (additional to the ground surface capacity). On individual tiles, areas of bare soil evaporate and this is available as a separate diagnostic. The bare soil tile Cm is 1.0 mm (Gregory et al. 1994). On vegetated tiles there is a proportion of soil evaporation that is grouped with transpiration and therefore is not available as a separate diagnostic. This soil evaporation under the vegetation combined with transpiration is referred to here as ‘soil-evapotranspiration’.

Cm of vegetation is a spatially and temporally variable parameter within MOSES. The Cm for each surface type is calculated from the LAI and then averaged across the tile according to the proportions of each surface type (Gregory et al. 1994). There are no differences in Cm between different plant functional types, other than the LAI, i.e. a broadleaf tree with an LAI of 4 has the same Cm as a C3 grass with an LAI of 4. As noted in Sect. 1.1, this is in contradiction to the observations which show that different plant types can have very different Cm values, even when the LAI is very similar.

2.2 Experimental set up

We ran a set of sensitivity simulations to explore the effect of different Cm values. The single model parameter ensemble (hereafter ensemble) uses different canopy and ground surface capacity values (see Table 2). Combined with their potential LAI the ensemble gives a range of potential Cm values (see Fig. 2). The ensemble explores three values for ground surface capacity: 0 mm (L), 0.5 mm (N) and 2.0 mm (H). These values replace Cg in Eq. 1. For CC (the canopy capacity which varies with LAI), five values are used: 0.005 (VL), 0.025 (L), 0.05 (N), 0.25 (H), 0.5 (VH).
Table 2

The ensemble combinations run


CC = 0.005 (VL)

CC = 0.025 (L)

CC = 0.05 (N)

CC = 0.25 (H)

CC = 0.5 (VH)

Cg = 0.0 (L)






Cg = 0.5 (N)






Cg = 2.0 (H)






The ground surface capacity (in mm) is indicated by Cg and ranges from 0 to 2 mm (L low, N normal, H high). The canopy capacity (in mm/LAI) in indicated by CC or can, has a range from 0.005 mm/LAI to 0.5 mm/LAI (VL very low, L low, N normal, H high, VH very high). Together, Cg and Cc make Cm (see Sect. 2.1.1)
Fig. 2

The potential Cm (canopy interception capacity) of each ensemble member for LAI values between 0 and 9. This plot shows only the potential range of Cm values given a maximum LAI attainable of 9. The actual Cm is calculated within the model. The LAI is calculated dynamically in the EqVeg ensemble and uses a set annual pattern in the FixVeg ensemble

The ensemble results in a range of mean global Cm values from 0.014 to 3.2 mm (see Fig. 3a). The mean global Cm for the control run (canNgN) lies approximately in the lower third of the ensemble, at 0.58 mm (see Fig. 3a). The ensemble members Cm ranges are shown in Fig. 2. The ensemble range and mean Cm values are well within the plausible range of measured Cm values (see Sect. 1.1; Table 1), though don’t extend up to the high maximum Cm values. The range includes around the same number of ensemble members above and below the 1.2 mm ± 0.4 used by Miralles et al. (2010) in the calculation of a canopy evaporation satellite product, which is used for comparison in the results section.
Fig. 3

a Absolute values of mean global Cm (canopy interception capacity) in mm for each of ensemble members, including the control (canNgN). b Anomalies in mean annual daily global surface hydrology in mm day−1 (over land only from 77° North to 55° South) for each of the ensemble, compared to the default parameterisation which is used as the control (canNgN). The soil moisture is in mm, (not a rate of change as in the other variables) and is included to give an indication of the amount of overall change in the soil moisture amount. The positive and negative indicators above the anomaly bars indicate the scale of change in Cm of that ensemble member, compared to the default. They are categorized as a small increase (+), a small decrease (−), a big increase (++) or a big decrease (−−)

We ran two versions of the ensemble using HadCM3. Both ensembles start nominally at spun-up conditions for 1,850 for both climate and vegetation. In the EqVeg ensemble the vegetation is able to adapt to the climate, whereas the FixVeg remains with the same vegetation cover throughout. The EqVeg ensemble was run with TRIFFID in equilibrium mode. The FixVeg was run with TRIFFID turned off and the vegetation does not change from the initial prescribed vegetation. They are run for 200 years to reach climatic equilibrium. The climate means of the last 30 years are used in the analysis. Global mean values given refer to the land only from 77° North to 55° South, unless otherwise specified.

Results presented here are from the EqVeg ensemble, unless noted otherwise. For clarity, we concentrate on six ensemble members compared to the normal model setup (canNgN). These are: the highest Cm member canVHgH; the lowest canVLgL; a high canopy capacity with a normal ground capacity, canHgN; a normal canopy capacity with a low ground capacity, canNgL; a very high canopy capacity with a low ground capacity, canVHgL; and a low ground capacity with a high canopy capacity, canLgH.

3 Results

3.1 Climate and hydrological changes

The most immediate effects from a change in Cm would be expected in the hydrological cycle (see Fig. 3b), and more specifically the canopy evaporation, which is directly affected by the change in the size of the potential storage, Cm. However, although the size of changes in Cm and the range in the ensemble are relatively large, the canopy evaporation changes are proportionally smaller. The control ensemble member has a mean global canopy evaporation of 0.39 mm day−1, and the largest Cm ensemble member (canVHgH) has a canopy evaporation slightly less than doubled at 0.69 mm day−1, for a quintupling of mean Cm. The smallest canopy evaporation (canVLgL) is also not as substantial a change as might be expected, at 0.071 mm day−1. This is because a large proportion of the total precipitation is convective precipitation, which only ‘falls’ on 30 % of the tile, in effect increasing the throughfall and making Cm less important. Moreover, because the Warrilow et al. (1986) throughfall scheme allows increasing proportions of throughfall with increasing canopy saturation, the value of Cm is mainly important in regard to how saturated the canopy is. The higher Cm values particularly are less effective than they would be otherwise because the Warrilow scheme always allows throughfall, albeit at a lower rate when the canopy is less saturated. The net effect of the interception scheme is that a large change in Cm translates into a much smaller change in canopy evaporation.

The change in canopy evaporation is almost matched by a change in the soil-evapotranspiration of the opposite signal (see Fig. 3b). For the most part, a very similar amount of water is evaporated, either from the soil and transpiration or from the canopy. The majority of the change in canopy evaporation is cancelled out by changes in soil-evapotranspiration, but the residual changes the total evapotranspiration. Therefore the sign of total evapotranspiration follows the canopy evaporation, increasing where canopy evaporation increases.

The other surface hydrological terms change less than the evapotranspiration. Top level soil moisture follows soil-evapotranspiration changes quite strongly, increasing and decreasing by around 0.22 mm in the extremes of the ensemble. This is to be expected, as it is the main reservoir for the soil-evapotranspiration and makes up the majority of total evaporation. Precipitation is affected by the changes in total evapotranspiration and therefore also approximately follows the trend of the canopy evaporation (see Fig. 3b). However, the global signal is very small and the largest mean annual global change is 0.015 mm day−1 in canVHgL, rather than one of the two extreme ensemble members. Runoff and precipitation are not strongly affected by the change in Cm because the majority of the change in Cm is displaced to changes in soil-evapotranspiration and soil moisture. The high level of synchronicity between the soil-evapotranspiration and canopy evaporation suggests that water not evaporated from the canopy is very likely to be evaporated from the soil or transpired. This infers that a change in the infiltration parameter between the soil levels would have a major impact on the results of this study. However, there are also significant changes to the total evapotranspiration from the change in Cm, which are predominantly in the tropics where the largest changes to canopy evaporation occur and have significant impact on local precipitation.

The precipitation changes are small on the global scale, but are focused in the tropics, along the inter-topical convergence zone (see Fig. 4). Reduced canopy capacity gives a northward shift in the precipitation band from the Amazon basin to the Congo region of West Africa. The changes in precipitation correlate with the changes in total evapotranspiration (Fig. 5) with the largest anomalies compared to the control over the Amazon region and west Africa (refer to Figs. 4, 5). This suggests a causal link: a decrease in Cm decreases the total evapotranspiration over these regions and provides decreased water availability for convective cloud formation and thus precipitation. The FixVeg ensemble supports this hypothesis, as it shows a very similar change in precipitation (see supplementary information 1). However, the direction of interaction between evapotranspiration and precipitation is difficult to establish clearly .
Fig. 4

Results from the last 30 years of the EqVeg simulations for mean annual daily precipitation anomalies. The anomalies are shown in mm day−1, compared to the control (canNgN). Results with percentage change in precipitation have a very similar distribution. Only areas with p < 0.01 using a Wilcoxon rank sum test are shown. See Supplementary Information 2 for details of the statistical testing used here
Fig. 5

Results from the last 30 years of the EqVeg simulations for the mean annual daily total evapotranspiration anomalies. The anomalies are shown compared to the control (canNgN), in mm day−1. Results with percentage change in total evapotranspiration have a very similar distribution

As well as affecting precipitation, the changes to Cm also affect the global and regional temperature. The reduced Cm ensemble members give a significant mean annual warming signal over land of up to 1.9 K locally, and 0.43 K globally (see Fig. 6) which is outside of the interannual variability (see supplementary information 3). The larger increase in Cm gives a proportionally larger cooling, −0.64 K globally and locally as much as −1.9 K in the canVHgH ensemble member. The temperature changes in the FixVeg ensemble are very similar in size and distribution in the EqVeg ensemble. The differences between the two ensembles do not show any statistically significant differences that are spatially consistent, and the overall spatial distribution of temperature is very similar between the two ensembles (see supplementary information 4). Further, the mean annual temperature differences in the EqVeg ensemble are very similar to the FixVeg ensemble. For example, the lowest Cm ensemble member, canVLgL, has a mean annual global temperature increase of 0.44 K in the EqVeg ensemble and 0.48 K in the FixVeg. This suggests that although vegetation change has some impact, it is not the main driver of the temperature change in this instance.
Fig. 6

a Mean annual global temperature anomalies for each of the ensemble members, compared to the control (canNgN). The positive and negative indicators above the anomaly bars indicate the scale of change in Cm of that ensemble member, compared to the default. They are categorized as a small increase (+), a small decrease (−), a big increase (++) or a big decrease (−−). b Mean annual temperature anomaly results, over the last 50 years of the simulations. Results shown are the mean annual temperature anomalies compared to the control (canNgN). Only areas statistically significant at p < 0.01 using a Wilcoxon rank sum test are shown. See supplementary information 2 for details of the statistical testing used here

The temperature anomaly is closely related to changes the partitioning of the turbulent fluxes. The change in the evaporative fraction (latent/latent + sensible heat) in particular has a good correlation to the temperature change in both the tropics and the mid to high latitudes (compare Figs. 6b, 7). In the evaporative fraction changes, it can be more clearly seen that although the absolute changes to total evapotranspiration are small, they do have significant repercussions for sensible heat and temperature. Changes to the Bowen ratio (sensible/latent heat) show a broadly similar picture, though are confused by very large changes in arid regions, making the evaporative fraction a clearer metric when considering the spatial patterning.
Fig. 7

Results from the last 30 years of the EqVeg simulations for the mean annual daily evaporative fraction (latent/latent + sensible heat) anomalies

We can diagnose which the strongest influences on temperature are using a linear model for each grid box, using values across the whole ensemble (see Fig. 8). In the mid latitudes, the top of atmosphere albedo explains the change of temperature best (see supplementary information 5), whereas in the tropics the evaporative fraction is the best predictor of temperature change. In the EqVeg ensemble, some of the temperature change can be accounted for by surface albedo changes caused by changes in the vegetation distribution, especially some die back in broad leaf trees in the boreal forest area of north Eurasia due to decreased temperature. However, both ensembles show a similar distribution of the largest factor explaining temperature changes (Fig. 8). The top of atmosphere albedo changes are related to changes in the cloud albedo, rather than cloud quantity, as shown by the lack of total cloud amount being a good explanatory factor for temperature in Fig. 8. Similarly, the evaporative fraction is a better explanatory factor than either total evapotranspiration or Bowen ratio changes in most areas. The r squared values for all of the factors considered in Fig. 8 are very high (>90 %) when averaged across latitudinal bands. Further, all of the factors are related, making it difficult to assert that one particular factor is more influential than others. Previous studies have shown that changes to the evaporation to transpiration ratio can affect the low level cloud and the total heat flux (Wang and Eltahir 2000). Therefore we suggest that it is the change in evaporative fraction and top of atmosphere albedo which creates this effect, driving changes in temperature.
Fig. 8

Results of a linear model for each grid box used to explain the mean annual temperature anomaly from the anomaly of mean annual top of atmosphere albedo; total cloud amount (with random overlap); evaporative fraction; total evaporation; and bowen ratio. Each ensemble member is plotted as an anomaly compared to the default parameterisation, giving a sample size of 15. Each variable is linearly modeled separately against temperature. The variable with the highest r2 value is plotted. Where the highest r2 value is has p > 0.01, no colour is plotted. The two plots are for the (left) EqVeg ensemble and (right) FixVeg ensemble

The relationship between the change in temperature and Cm is approximately linear (see Fig. 9). A global mean Cm increase of 1 mm gives −0.30 K change in global mean annual temperature. However, temperature is more sensitive to a decrease in Cm than an increase. For both temperature and precipitation, the relationship changes near the default Cm parameterisation. This result is robust to how the relationship is plotted (absolutes, anomalies, regions, etc.). The Cm relationship with precipitation is globally approximately logarithmic (see Fig. 9), showing a much stronger pivot point around the default parameterisation than temperature. The change in precipitation is almost exclusively due to changes in precipitation in the tropics. Precipitation is sensitive at low levels of Cm, but once the mean global Cm is greater than ~1.2 mm the precipitation remains approximately the same as Cm rises. The changes in the global soil-evapotranspiration to total evapotranspiration also follow the same pattern of decreasing change at high Cm parameter values (see Figs. 3, 10). It is unclear what the physical reason for the default parameterisation being a pivot point might be. Because the Warrilow et al. (1986) scheme isn’t strongly affected by canopy saturation (since throughfall is allowed at a gradually increasing rate) this is unlikely to be the basis of the change in sensitivity. Similarly, the change in sensitivity of temperature suggests that this is not solely due to precipitation variability.
Fig. 9

Mean annual global Cm anomalies, compared to the control (canNgN) for each of the ensemble for: a mean annual global temperature anomalies and b mean annual global precipitation anomalies
Fig. 10

The global mean annual soil-evapotranspiration as a percentage of total evapotranspiration for each of the ensemble members, including the control (canNgN). The positive and negative indicators above the bars indicate the scale of change in Cm of that ensemble member, compared to the default. They are categorized as a small increase (+), a small decrease (−), a big increase (++) or a big decrease (−−)

3.2 Data comparisons

The changes in total evapotranspiration appear to be mainly caused by alteration of the ratio of canopy evaporation and soil-evapotranspiration. The real world partitioning between transpiration and evaporation is still not agreed upon, but recent work suggests that transpiration makes up around 90 % of evapotranspiration globally (Jasechko et al. 2013). It is notable that this study refers to transpiration but the model used here only considers soil evaporation combined with transpiration. Therefore the two terms are not directly comparable and the soil-evapotranspiration in the model would be expected to have be a larger proportion of total evapotranspiration than suggested by Jasechko et al. (2013). However, in the model the proportion of soil-evapotranspiration to total evapotranspiration is relatively small; this is the opposite of what we would expect. In the control scenario the mean global soil-evapotranspiration is 71 % of total evapotranspiration (see Fig. 8). Increasing the Cm further decreases the soil-evapotranspiration to total evapotranspiration ratio to as little as 53 %. The global range remains largely the same (around 20–80 %). By comparison, the reduced Cm scenarios give are more in line with Jasechko et al. (2013). The most extreme reduction in Cm, canVLgL, results in a mean global soil-evapotranspiration to total evapotranspiration ratio of 94 %. Even removing just the 0.5 mm ground surface capacity Cm in canNgL gives a significant improvement of this ratio to 81 %. However, the ratio of transpiration to evapotranspiration is very variable for natural vegetation (Kool et al. 2014). Since the uncertainties in the actual evapotranspiration ratio are large, it is difficult to draw strong conclusions. Neither is the spatial distribution illuminating. The climate model evapotranspiration ratio relies heavily on the available energy at the surface, giving it a strong latitudinal bias to the pattern. The evapotranspiration ratios from isotope data (e.g. Jasechko et al. 2013) is often only available for irregularly shaped or very small hydrological basins. Therefore an empirical comparison of these two datasets spatially is difficult. However, it is evident that decreased Cm brings the model more in line with current understanding that transpiration makes up the majority of total evapotranspiration.

Similarly to transpiration, it is only relatively recently that global estimates for canopy evaporation have become available. However, comparison of canopy evaporation in climate models with the available canopy evaporation satellite product by Miralles et al. (2010) is complicated, as the satellite product only accounts for interception by the forest canopy (where >50 % of the land cover is trees, according to the MODIS MOC12C1 product). MOSES, on the other hand, includes interception by the ground surface (see Sect. 2.1) as well as all plant functional types. Neither does the model used here have the same vegetation distribution as the one used by Miralles, which makes individual areas more difficult to directly compare. Because of these differences in scope, the satellite product has relatively low global mean canopy evaporation to precipitation ratio (interception loss) of around 7 % and to achieve this ratio in the model, a reduced Cm value is required (see supplementary information 6). For the same reason, it is to be expected that the canopy evaporation values of the control ensemble member would be higher than the satellite product, especially for areas of bare soil or grasses. However, areas of a large proportion of tree cover (such as 40°–60° North and the tropics around 10° North to 20° South) should be comparable as they have the same conditions that are assumed by Miralles (a high proportion of vegetation cover, especially trees).

The near global satellite product provided by Miralles et al. (2010) gives a range of ~0–650 mm year−1 canopy evaporation (see Fig. 11). The control range is quite close to this, at 0–615 mm year−1, but has much larger areas of the higher values than the satellite product. Therefore the mean global value of the control is much further off, some 100 mm year−1 higher than the global mean of the Miralles dataset. The spatial distribution of canopy evaporation in the control model compared to the satellite product suggests that the model overestimates canopy evaporation in the non-tropical regions. In the tropical regions, the model overestimates the total area of high canopy evaporation, but does better overall.
Fig. 11

The mean annual canopy evaporation, in mm year−1, for a selection of the ensemble members including the control and for ‘satproduct’, the Miralles et al. (2010) satellite canopy evaporation product. The satellite product has been re-gridded to 96 longitude, 73 latitude grid boxes to match the climate model output in the ensemble. The satellite product is only available for −60° to +60° latitude because of satellite coverage limitations, but is plotted on the same map proportions as the climate model output for easy visual comparison

The mean canopy evaporation value is improved (compared to the Miralles dataset) by the decreased Cm in ensemble member canNgL (see Fig. 11). The global mean canopy evaporation anomaly for canNgL is just 40 mm year−1 more than the satellite product. Moreover, the canopy evaporation over the northern hemisphere mid latitudes is substantially reduced and much closer to the very low values found in the satellite product. The extremely reduced Cm scenario (canVLgL) is even closer, with the global mean very close to the satellite product, only 17 mm year−1 less. However, canVLgL gives a considerable underestimation of the range (at a maximum of 110 mm year−1). The mean of canVLgL agrees with the satellite product because the satellite product gives no canopy evaporation to grass covered areas. Since the model does include grass canopy evaporation, it should be higher. Legitimately higher values in the model might also be caused by the model using simulated pre-industrial equilibrium vegetation as predicted by the model, which has some differences from the modern distribution of vegetation used in the satellite product. Notably, the model’s preindustrial vegetation distribution has more forested areas in both the tropics and the mid to high latitudes. It is also possible that the lower LAI for needle leaf trees in JULES (see Table 2) would rectify the overestimation of canopy evaporation at the mid to high latitudes, but since the satellite product excludes canopy evaporation from grasses, they will never completely agree. Bearing these differences in mind, the canNgL ensemble member represents canopy evaporation well.

4 Discussion

The results from this ensemble of a range of plausible Cm values show a surprising result in terms of the sensitivity of different aspects of climate. Cm values are generally associated with changes to precipitation. However, this research shows that a range of realistic Cm values have a major impact on the land–atmosphere coupling with temperature.

When comparing climate model data to satellite products, such as that used here by Miralles et al. (2010), a further complication is that different processes are assumed to drive canopy evaporation. In climate models canopy evaporation is limited by the radiative energy available (i.e. the surface energy balance must balance). The satellite product on the other hand uses an algorithm to combine a precipitation and vegetation product and does not account for evaporation being limited by energy availability. Therefore in regions where the canopy evaporation may be energy limited (for instance the mid to high latitudes) the canopy evaporation from the satellite product could be an overestimate. That the control model predicts higher canopy evaporation over the mid to high latitudes than the satellite product therefore suggests that there is a significant issue with the control model’s representation of canopy evaporation.

One of the key issues in this work is the conflict between the measured values of Cm taken from the literature and the substantially lower values that are used in the model. These low values in the model are primarily due to limitations in the model’s representation of precipitation. Global and regional models still struggle to accurately represent precipitation, tending to have more frequent, lighter rainfall events than measured values suggest and underestimating large precipitation events (Dai 2006; Sun et al. 2007; Crétat et al. 2013). This bias is because the moist convection tends to simulate rainfall more frequently than would occur in reality (Dai 2006). For the same Cm value, frequent light precipitation will give much larger canopy evaporation than a small number of large precipitation events, because the former has more opportunities to evaporate and less chance of saturating the canopy. Thus a reduced Cm is required in a model with ‘drizzly’ rainfall in order to represent canopy evaporation adequately. A corollary of this is that when the precipitation representation is improved (by for instance, increased resolution (Boyle and Klein 2010; Kendon et al. 2012), the Cm will need to be increased, else the evapotranspiration ratio may be compromised. Therefore solving the light rain problem could therefore also help lessen the conflict between measured Cm values and model Cm values. It is possible that changes to other parts of the interception scheme (such as the statistical approach to convective precipitation) could also be used to counteract the effects of an increase in Cm. However, in an interconnected system there may be unforeseen feedbacks.

There is also a tension between trying to make the land surface parameterisation values as realistic as possible, and obtaining realistic model outputs. In this case, a less realistic parameterisation gives a more realistic model output. The mean measured interception storage capacity of the vegetation canopy alone is likely to be around 1–2 mm (Breuer et al. 2003) but in MOSES the vegetation canopy interception capacity is limited to 0.45 mm. Even accounting for the ground surface capacity, it is still an underestimate at 0.5–0.95 mm. Therefore it might be expected that increasing the Cm in the model would improve the representation of surface hydrology. However, increasing the canopy interception storage to a plausible level results in a serious deterioration of the representation of the surface hydrology and climate. Conversely, a decrease, which is not strongly supported by the measured values, results in improvements to the surface hydrology and temperature rises over the mid latitudes. The small temperature changes in ensemble member canNgL are not inconsistent with Pope et al. (2000) assessment that the model is a little cool over North America in the present day.

We suggest that the ensemble member canNgL is a feasible and simple modification to MOSES that improves the representation of the total evapotranspiration partitioning and canopy evaporation. It does not substantially affect the temperature or precipitation and thus this change is a good modification for those wishing to use HadCM3 or HadAM3 to examine the surface hydrology. The Cm recommended here (canNgL) is:
$$C_{m} = C_{C} LAI$$
where CC = 0.05. Compared to the original values, this new parameterisation effectively removes the ground surface capacity. Though not consistent with measured Cm values, this proposed value instead is an explicit acknowledgement that this parameter cannot at this time be made more ‘realistic’ because of the inevitable limitations of the model.

Our recommended reduction in Cm is larger than that suggested by Demory (2009) and Van den Hoof et al. (2013). This is partly a result of the different LAI values in JULES compared to MOSES. However, more important is the standard by which the models are being judged, particularly with regard to transpiration as a proportion of evapotranspiration. Only the ensemble member with almost no Cm (canVLgL), has over 90 % of evapotranspiration from transpiration as Jasechko et al. (2013) asserts. However, this high evapotranspiration ratio ensemble member represents a significant increase in global and regional temperature (see Fig. 6b), making it unsound for use. Ensemble member canNgL increases the transpiration ratio to 81 % without affecting the global climate substantially. Similarly, the ensemble member most similar to the parameterisations suggested by Demory (2009) and Van den Hoof et al. (2013) (canHgL) has significant cooling. While this cooling is globally smaller than that of canNgL, it is incompatible with previous assessments of the climatology of the model (Pope et al. 2000). Considering the canopy evaporation to precipitation ratio, canHgL again performs worse than canNgL, giving extensive overestimations in the tropics, whereas canNgL is on average within 40 mm year−1 worldwide. This is not to say that the Cm value proposed here would necessarily be a better improvement to JULES, but that consideration of the effect on the climatology of the model when it is coupled is required to establish what value is most appropriate.

5 Conclusions

We have examined the effect of different values taken from the literature of canopy interception capacity (Cm) on the surface hydrology and climate in a coupled climate model (HadCM3). The main result is that altered Cm significantly affects global mean annual temperature and tropical precipitation. The temperature effects are strongly related to changes in evaporative fraction. We find that decreased Cm improves the canopy evaporation to precipitation ratios as well as the transpiration to evapotranspiration ratios. Although the role of soil moisture and transpiration is well known in land–atmosphere coupling, the role of Cm parameterisation has been overlooked. These results suggest that the Cm parameterisation is important.

It has been previously shown that the structure of the interception scheme does not significantly affect the climate or hydrology (Mao et al. 2011). However, the same study also shows that there is a very small range of Cm values within the different interception model structures (Mao et al. 2011). We show further in Sect. 1.2 that most models use a similar way of calculating Cm with LAI and use similar parameter values overall. Since the Cm affects the partitioning of evapotranspiration, it seems possible that partitioning is similarly represented in many models. In HadCM3 the canopy evaporation affects the total amount of evapotranspiration and the evaporative fraction which affects the climate via the latent and sensible heat fluxes. How much the ratio of these fluxes change from the same stimulus is known to be very variable in climate models (de Noblet-Ducoudré et al. 2012). However, that the model structures are in practice very similar and use very similar parameter values is suggestive that other coupled climate models might also show some temperature sensitivity to the Cm value.

The overall effect that Cm has on the surface hydrology and turbulent heat fluxes in the coupled climate model used here is substantial. This effect is likely to be related to the precipitation characteristics (intensity, frequency, duration) in HadCM3. Therefore it is important to reassess Cm values in the land surface model when significant changes to the precipitation are made in the corresponding atmospheric model. This should help to avoid unforeseen effects via changes to the ratio of evapotranspiration. Given that many other coupled climate models suffer from high light rain-day frequency and therefore may have low Cm values to counteract this, it may be more widely applicable too.

The Cm has not traditionally been considered important in climate models or indeed in hydrological models. However, temperature changes shown here demonstrate that the effect of partitioning of evapotranspiration and value of Cm are not trivial. Whether this sensitivity is a common feature of coupled climate models and what the sub-daily mechanisms behind it are should be the subject of further investigation.


We would like to thank John Gash for his comments on a draft version of the manuscript, as well as the two anonymous reviewers for their comments on the submitted manuscript. We gratefully acknowledge Diego Miralles for providing the canopy evaporation satellite product for comparison. CDJ is supported by the Joint DECC/Defra Met Office Hadley Centre Climate Programme (GA01101). TDB is supported by Natural Environment Research Council Dtg NE/J500033/1.

Supplementary material

382_2014_2100_MOESM1_ESM.pdf (2 mb)
Supplementary material 1 (PDF 2033 kb)

Copyright information

© Springer-Verlag Berlin Heidelberg 2014