1. Introduction
Representation of drought in land surface models (LSMs) is important for accurate weather forecasts, assessing water resources, and projecting the possible impacts of droughts under climate change. LSMs have been used to investigate both the changing frequency of drought under climate change and the potential for future droughts to have increased societal impacts (Prudhomme et al. 2011, 2014; Dai 2013). The trends of changing drought frequency and intensity are dependent on which model is used so that it has been recommended that ensembles are used to check the robustness of any signals (Prudhomme et al. 2014; Tallaksen and Stahl 2014). The LSM used by the Met Office for forecasting and climate projections is known as JULES (Joint U.K. Land Environment Simulator; Best et al. 2011). JULES and seven other LSMs run offline were assessed for their ability to represent evapotranspiration (ET) during meteorological drought with observations from six flux tower sites (Ukkola et al. 2016). This showed that in general LSMs are poor at representing the duration, magnitude and intensity of droughts. As well as problems representing ET, Ukkola et al. (2016) showed that soil moisture and soil hydrology are important components of the modeling.
During the summer of 2018 there was an unusually prolonged spell of dry weather that led to a substantial dry-down of the C3 grass field at the Met Office research unit at Cardington, Bedfordshire, United Kingdom. This has provided an opportunity to exploit the wide range of field measurements for comparing the course of the dry-down in terms of ET and soil moisture with simulations by JULES. Since the meteorological drought in 2018 led to senescence of the grass at Cardington, the conditions approached those associated with semiarid climates. Critically, the observations include profiles of both the near-surface atmosphere and of soil moisture down to the water table. These data allow us to focus on aspects of the modeling that could lead to improvements in JULES during drought conditions. Potentially the results from this study could be informative for a major field campaign designed to improve LSMs that is planned in the semiarid Ebro basin in Spain (Boone et al. 2019).
Within semiarid climates dewfall can provide a critical source of moisture for some plant types (Uclés et al. 2014). Dewfall is a frequently neglected component of the water cycle within LSMs. There have been a few measurement campaigns that include estimation of dewfall with a variety of techniques (e.g., Veste and Littmann 2006; Moro et al. 2007; de Roode et al. 2010; Hanisch et al. 2015). Microlysimetry can be used to measure rates of dewfall, fog droplet deposition, frost accumulation, precipitation rate, and ET (Price and Clark 2014; Uclés et al. 2014). Dew meters with two types of canopy were used in this study 1) artificial grass and 2) grass living on top of 5 cm of soil. The real grass dew meter normally provides measurements of transpiration, evaporation from the canopy surface and bare-soil evaporation. The paper concentrates on the years 2018 and 2019. During the course of the summer of 2018, the living grass on the real grass dew meter died. This meant that a measured decrease in mass represented evaporation from bare soil and from the dead canopy but not transpiration. Instrumentation including the use of dew meters is discussed in section 2.
For modeling purposes we have used JULES as employed in the first midlatitude Regional Atmosphere and Land (RAL1-M) science configuration of the U.K. 1.5-km variable resolution operational forecast model (UKV; Bush et al. 2020). The UKV itself is a limited area, or regional, version of the global Met Office Unified Model (Tang et al. 2013). In section 3 we describe small modifications to the UKV configuration that are designed to more appropriately represent the local site characteristics. In section 4 we describe the derivation of soil hydrological properties based on observations and compare them to the UKV values.
Section 5 is devoted to discussing the results. Section 5a(1) discusses the JULES output soil moisture and ET compared to observations in 2018 and 2019. Next, water budgets are considered in terms of cumulative plots of observed rainfall and observed- and modeled-ET in 2018 compared to 2019 [section 5a(2)]. The progression of the drought in 2018 in terms of greenness, and changes in the size of latent heat errors from JULES is considered in section 5b(1). Section 5b(2) considers the modeling of evapotranspiration from dry soils. Section 5c(1) discusses the physical conditions that lead to dewfall at night based on very short-term (5-min) observations of friction velocity, temperature and humidity profiles, plus dew meter data. Section 5c(2) discusses the observed and modeled latent heat flux during dewfall overnight, whereas section 5c(3) considers latent heat during the days following dewfall. Section 5d(1) considers how observed and modeled soil moisture responds to rainfall after drought. The final results in section 5d(2) considers bare-soil evaporation after rainfall on to dry ground.
2. Field site and instrumentation
The Met Office research site at Cardington (52.105°N, 0.424°E, 29 m above sea level) has been making continuous subsoil and near-surface meteorological observations since 1995. The 18-ha site is laid mainly to manicured grass kept at 5–10-cm height throughout the year. The prevailing southwesterly direction is the most open and flat of all the wind direction sectors. The ground is generally flat within 1 km of the site. The surrounding area 1 km to the northwest is partially urbanized by housing, and 4–6 km northwest is the town of Bedford (population 164 000 in 2014). Two large airship hangars, 250 m long × 52 m high, lie 400 m to the north and cover wind directions between 350° and 30°. The other sectors are open farmland of varying crops, interspersed with hedges, shrubs, and small trees.
Details of equipment used at the Cardington site can be found in Horlacher et al. (2012) and Osborne et al. (2014). Rainfall is observed with a Met Office tipping-bucket Mk5 gauge using a 0.2-mm accuracy. Permanent flux towers were installed at 10, 25, and 50 m in 2005 and at 2 m in 2011 and are equipped with Gill HS50 3D ultrasonic anemometers, plus Vaisala HMP155 relative humidity and Vector Instruments T302 temperature sensors, both housed in aspirated screens. At the 10-m level there is a Licor Li-7500 open-path hygrometer for measuring high-frequency water vapor fluxes. Latent heat flux is measured at 10 m whereas the sensible heat and momentum fluxes are measured at all of the flux tower heights based on 10-Hz data. An aspirated and screened grass canopy measurement of temperature and relative humidity of air between 0- and 5-cm height uses a Rotronics Hydroclip2.
Shortwave radiation of wavelength between 0.3 and 3 μm is measured as global downwelling, diffuse downwelling, and upwelling components using conditioned clear-domed Kipp and Zonen CM21 pyranometers. Shortwave downwelling radiation is used as input to JULES and represents the sum of the visible and near-infrared radiation parts (VIS + NIR). In addition, red-domed pyranometers using RG715 glass are deployed in the same manner so that the near-infrared portion of the solar spectrum (NIR) can be measured. Therefore, VIS is calculated as the difference between the clear- and red-domed pyranometers. Normalized difference vegetation index (NDVI) is estimated here as (NIR − VIS)/(NIR + VIS). Kipp and Zonen CG4 pyrgeometers measure the downwelling and upwelling longwave radiation (4.5–40 μm). In-house intercomparisons of the pyranometer and pyrgeometers with secondary standard devices indicate accuracy within 3%.
A Heitronics KT15D thermometer provides the radiometric (8–14 μm) surface temperature looking at 1 m2 of grass from a height of 2.5 m [called the infrared temperature (IRT)]. Measurements at Cardington provided an estimated positive bias in the IRT, which increases from 0.5°C at a surface temperature of 0°–2°C at a surface temperature of 20°C (Edwards et al. 2011). We have linearly extrapolated the apparent bias of Fig. 1b of Edwards et al. (2011) up to 45°C surface temperature and used this for correction to the IRT values.
Delta-T ML3 theta probes measure volumetric soil moisture at depths of 10, 22, 57, and 160 cm (Table 1), expressed below as a percentage, to a manufacturer’s claimed accuracy of ±1%. An additional set of soil moisture measurements are made starting in July 2018 with a Delta-T PR2/6 profile probe at depths of 10, 20, 30, 40, 60, and 100 cm. There is an unexplained gradual instrumental drift in these measurements, but they are useful for assessing responses to rainfall events at different depths. A Delta-T ST2-396 thermistor probe measures soil temperature at 1 cm to a claimed accuracy of ±0.1°C. We use Hukseflux self-calibrating heat flux plates buried at 2-cm depth to measure ground heat flux. These data are corrected for the heat storage in the top 2 cm of soil based on the change in the 1-cm soil temperature with time. Water table depth is determined by two methods using bore holes: automatically with a Druck 1830 pressure transducer and manually once a week using a dip-stick technique.
JULES soil layer characteristics and soil moisture observation sensor depths.
In-house designed portable microlysimeters were used to measure dewfall (Price and Clark 2014). The Mk 2 dew meters consist of a load cell to measure the weight of a flat, water-tight pan of 34-cm diameter that contains a canopy of choice. They are positioned in holes so that the tops of the pans are flush with the surrounding soil. They have a precision of 0.5 g m−2 or 0.0005 mm of water. We had two running, one with a real grass canopy and one with an artificial polymer grass canopy. The real grass canopy was cut from turf within the site with grass 3–4 cm high, soil about 5 cm thick, and a total dry mass of 1.7 kg.
Price and Clark (2014) established that noise is introduced into the dew meter data when wind speeds exceed 2.5 m s−1. However, we have found that the raw 1-min observations from the dew meters are noisy at all times. Therefore, to improve the signal-to-noise ratio, the data have been smoothed using a Hanning window and the results are shown here at 30-min time steps. The dew meters were deployed near the center of the grass site away from hedges and buildings and so represent open aspect measurements. We have taken the dew formation rate (g m−2 s−1) and converted it to a latent heat flux (W m−2) to provide comparison to the eddy-covariance method at 10 m. Allowing for the noise and Hanning smoothing, the uncertainty of the dew meters is estimated to be up to ±5 W m−2. As will be shown below, this is a novel method for estimating negative latent heat fluxes at night during very low turbulence when the eddy-covariance method suggests values around zero.
3. The JULES surface scheme
The JULES surface scheme is used operationally as a component of the Met Office Unified Model but is run here as a standalone model (Best et al. 2011). Specifically, the configuration of JULES used is derived from the RAL1-M version of the Unified Model (Bush et al. 2020), referred to as the control run. JULES is a tiled scheme using separate skin temperatures, solar and thermal radiative fluxes, latent and sensible heat fluxes, ground heat flux, and canopy water over five vegetation types (broadleaf tree, needleleaf tree, C3 grass, C4 grass, shrub) and four nonvegetation types (urban, lake, bare soil, ice). Combining tile fractions allows subgrid land heterogeneity to be represented. Four unknowns, i.e., three heat fluxes and skin temperature (tstar), are obtained in JULES via four equations that are solved simultaneously and describe the (i) energy balance, (ii) sensible heat flux, (iii) latent heat flux, and (iv) ground heat flux (Best et al. 2011). JULES has been driven using 30-min Cardington meteorological observations of surface barometric pressure, downwelling shortwave- and longwave-radiative fluxes, near-surface temperature (at 1.2 m), specific humidity (at 1.2 m), and wind speed (at 10 m), and rainfall. The time step of the model is 30 min, and output was obtained as mean values over each time step.
The UKV grid box (1.5 km × 1.5 km) containing the site is 90% C3 grass with small contributions from the urban and shrub tiles. For our standalone runs at a single point, however, the surface type at Cardington is represented as 100% C3 grass. The JULES C3 grass plant functional type includes crops as well as normal grass, so that the stipulated canopy height of 1.46 m in the UKV model is much higher than the manicured grass at Cardington (0.05–0.1 m).
Within UKV, leaf area index (LAI) varies between about 2.8 and 1.1 through the year to represent varying crop cover. Since the grass at the site is maintained at a near-constant height, it is reasonable to assume in normal years a constant LAI. Observed NDVI in 2018 prior to the dry-down is about 0.70. Using Eq. (6) based on Beer–Lambert’s law from Sun et al. (2013) this implies a LAI of 2.36. During the dry-down NDVI decreased to 0.55, implying a LAI of 1.62. To represent the effect of the drought on LAI for JULES, we have used a fixed value of 2.36 from 1 January to 19 June 2018, followed by a linear decrease to 1.62 from 19 June to 5 July, a fixed value of 1.62 from 5 July to 15 October, then a linear increase back to 2.36 from 15 October to 1 December 2018. Thereafter LAI is fixed at 2.36 until the end of 2019. This variation in LAI is combined with the use of the UKV canopy height (canht) of 1.46 m in the control configuration.
Three additional sensitivity tests, or model runs, have been investigated where we tune JULES to the specific conditions at Cardington obtained by modifying the model configuration. These additional tests are designed to allow exploration of the factors controlling modeled soil moistures and latent heat fluxes. JULES within UKV obtains forcing data from the bottom level of the modeled atmosphere, whereas for the standalone runs they are taken as the observed meteorological data from the site. Canopy height within JULES is defined using canht and this controls the roughness length and hence the parameterized turbulence. The turbulence in the real world is related to friction velocity, which is observed at the site. To determine the most appropriate canopy height for the standalone runs, we varied canht between 1.46 and 0.1 m for the time interval including the drought period April–September 2018. Selection of the optimum canht was based on the minimum root-mean-square error of the JULES versus observed friction velocity; canht under this criterion is set at 0.2 m.
Within JULES, layer 1 is able to lose water from the soil to the atmosphere via transpiration and evaporation from bare soil and, after rainfall or dewfall, from the canopy. As a result layer 1 can dry out below the wilting point. In contrast, layers 2, 3, and 4 can only lose water to the atmosphere via transpiration (and by drainage when saturated).
Plant roots in JULES are distributed exponentially with depth, with the e-folding rooting depth called rootd. The depths to the bases of the four JULES soil layers are 0.1, 0.35, 1.0, and 3.0 m. In control, rootd = 0.5 m so that layer 1 has 33%, layer 2 has 42%, and layer 3 has 23% of the roots; Eq. (50) of Best et al. (2011) provides the fraction of roots in each soil layer. The UKV rooting depth value of 0.5 m for C3 grass with canht of 1.46 m is not appropriate for the short grass at the Cardington site where we reduced canht to 0.2 m. With rootd = 0.5 m, 23% of roots reach soil layer 3, so that even if layers 1 and 2 have soil moisture at the wilting point, the simulated grass readily carries on photosynthesizing and transpiring. A revised rootd value of 0.2 m has been used as potentially more appropriate in the model run called canht-rootd. This revised rootd means that layer 1 has 63%, layer 2 has 34%, and layer 3 has 3% of the roots (Table 1).
With modeled C3 grass there is a certain implicit fraction of bare soil that interacts with the atmosphere and radiation. Within JULES this fraction is controlled by the optical extinction coefficient (kext) that also determines the albedo and the thermal coupling between the surface and the atmosphere. Hence the change in kext from the UKV value of 1.0–2.0 should allow the grass to heat more during the day and cool more efficiently overnight affecting the diurnal range in skin temperature (tstar); this is relevant when discussing the measurement of dewfall. At present, JULES has a hardwired value of the bare-soil fraction. As a result, changes in kext have little effect on bare-soil evaporation. The final sensitivity test is called canht-rootd-kext.
4. Soil hydrology parameters at Cardington
The observations of the soil temperature and moisture are used to initialize the four JULES soil layers, as is the skin temperature. We have not used spinup in JULES because the initial values have been taken from the observations, but after the first time step these variables evolve freely in the simulations. The availability of modeled soil moisture is dependent on various soil hydrological parameters as implemented using the Brooks and Corey (1964) soil hydraulics. Excess water runoff during saturation was determined with the probability distribution model (PDM; Moore and Clarke 1981).
We have used observations at Cardington to obtain values of the soil parameters and compared these values to those used in UKV. The soil at Cardington down to 1-m depth is medium clay loam with low organic carbon content, with about a third each of sand, silt, and clay: sand = 34.6%, silt = 31.0%, clay = 29.8%, and organic matter = 4.6% (Burton 1999). Sand and gravel content increases with depth below this, with the gravel in well-defined thin layers that can assist in lateral water advection. This change in composition is ignored for depths below 1.6 m where the soil is permanently saturated. Using the univariate method of Cosby et al. (1984) this soil composition leads to the soil parameters b, sathh, satcon, and sm_sat as named in Table 2. The parameters hcap and hcon are estimated using the model from Johansen (1975). The derived values of these six variables agree within error (±1 standard deviation) with the UKV values in Table 2. The thermal conductivity is shown for dry soil in Table 2 yet the modeled conductivity increases with soil moisture and will therefore potentially vary as such between the different JULES runs (control, canht, canht-rootd, and canht-rootd-kext). The sm_crit and sm_wilt values (Table 2) are calculated using the new values of b, sm_sat, and sathh together with the limits of hydraulic suction of 1.5 and 0.033 MPa for the wilting and critical points, respectively, following Cox et al. (1999). The same soil moisture limits in UKV (sm_sat, sm_wilt and sm_crit, Table 2) are used in all four soil layers, and we use this approach here.
Two sets of soil parameters—from the operational UKV model and as derived from observations at Cardington; the latter are used in the standalone JULES runs. Soil moistures are expressed here as volumetric fractions, but in the text percentages are used, i.e., m3 m−3 × 100. SM = soil moisture.
5. Results
a. Comparing 2018 and 2019
1) Comparison of soil moistures and evaporation in 2018 and 2019
Figure 1 shows an overview of the water cycle at Cardington during the whole of 2018 and 2019. Rainfall events occur throughout the period, but June and July 2018 are notable for the lack of rain. Thus, total rainfall in June 2018 was 2.2 mm compared to the 2005–19 June average of 43.2 mm. Similarly, July 2018 was very dry with just 9.3 mm compared to a July average of 42.5 mm. The screen-level temperatures in 2018 were also relatively high with June 0.6°C and July 2.1°C warmer than the 2005–19 Cardington average.
(a)–(e) Cardington observations (obs) and JULES simulations of water balance components at 30-min steps during 2018 and 2019. The time intervals covered in Figs. 3–8 are noted in (a). In (b)–(e) observations in black are compared to JULES control and canht-rootd. The brown horizontal dashed lines indicate the soil moisture limits (sat = saturation point, crit = critical point, wilt = wilting point) calculated from observed soil characteristics following Cosby et al. (1984) as used in the JULES simulations. (f) Two sets of water table observations are shown. (g) The observed and modeled daytime maximum upward latent heat flux.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
(a)–(e) Cardington observations (obs) and JULES simulations of water balance components at 30-min steps during 2018 and 2019. The time intervals covered in Figs. 3–8 are noted in (a). In (b)–(e) observations in black are compared to JULES control and canht-rootd. The brown horizontal dashed lines indicate the soil moisture limits (sat = saturation point, crit = critical point, wilt = wilting point) calculated from observed soil characteristics following Cosby et al. (1984) as used in the JULES simulations. (f) Two sets of water table observations are shown. (g) The observed and modeled daytime maximum upward latent heat flux.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
(a)–(e) Cardington observations (obs) and JULES simulations of water balance components at 30-min steps during 2018 and 2019. The time intervals covered in Figs. 3–8 are noted in (a). In (b)–(e) observations in black are compared to JULES control and canht-rootd. The brown horizontal dashed lines indicate the soil moisture limits (sat = saturation point, crit = critical point, wilt = wilting point) calculated from observed soil characteristics following Cosby et al. (1984) as used in the JULES simulations. (f) Two sets of water table observations are shown. (g) The observed and modeled daytime maximum upward latent heat flux.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
The soil moisture probes show high and constant values from late January to early May 2018 at 10, 22, 57, and 160 cm. The values at 10 cm are just below the sm_sat value used in JULES and are interpreted as saturated or near saturated. Since the probe at 160 cm lies below the water table depth throughout most of 2018 and 2019, we interpret the near-constant values at this depth as indicating saturation as well. Although at 22 and 57 cm the values from late January to early May 2018 are near-constant and appear to represent near saturation, they are considerably lower than the Cosby et al. (1984) sm_sat value (44.5%). The probe values indicating saturation appear to decrease with soil depth occurring at 43%, 38%, 33%, and 33%. This progression is consistent with the soil porosity decreasing with depth due to compaction.
In May, June, and July 2018 the dry weather led to a reduction in soil moisture down to at least 60 cm except for a brief period in late May when rain led to an increase in soil moisture at 10 cm. During the dry-down—i.e., the first part of the drought—the deeper layers took longer to respond than the top layer of soil (cf. the soil moisture observations in Figs. 1c and 1d with those in Fig. 1b). The 10-cm observations decrease to values well below the JULES sm_wilt value. The 22- and 57-cm soil moistures decrease to values very close to the JULES sm_wilt values. Despite the return of rain in late July, the observed soil moistures at 10 and 22 cm remained low until mid-October. During the autumn and winter, soil moistures increase with brief saturation by mid-March 2019 at 10 cm. The rest of 2019 represents a typical pattern of generally drier soil in the summer, but with rain events causing brief periods of wetting. Rain in autumn and winter allowed the soil to return to saturation.
The water table is very shallow but of varying depth between January and April 2018 at a time when the overlying soil is recorded as saturated. Despite the delays in drying at 22 and 57 cm compared to the surface, the water table drops continuously from May to mid-October 2018. Relatively small amounts of rain between October 2018 and March 2019 meant that soil moistures increased relatively slowly and the water table did not return to shallow depths in the winter. On the other hand, the drop in water table in 2019 until mid-November was followed by a rapid increase in soil moisture and a similar rapid return of the water table to typically shallow depths by mid-December. Both observations and JULES outputs for 2017 show essentially the same patterns of soil moisture content and water table depth as for 2019 (not shown).
There is a strong contrast in the observed latent heat flux variations between 2018 and 2019. In 2018, maximum daytime latent heat flux exceeding 150 W m−2 occurred between mid-April and early June, whereas in 2019 the maximum fluxes span from May to the end of August (Fig. 1g). These differences are clearly associated with the meteorological drought from mid-June to early August 2018.
There is overall quite good agreement between JULES and soil moisture observations in layers 2, 3, and 4. Using our JULES standalone configuration, we find that the model is unable achieve sustained soil saturation in early 2018 in layer 1; this is found in many other years between 2005 and 2019. Modeled soil moistures in layers 2 and 3 vary too sluggishly following rainfall events compared to observations.
On the other hand, the simulated soil moistures in layer 1 are a good match with the observations at 10 cm during years without drought, as exemplified by 2019 in Fig. 1b. During the dry-down in 2018, the JULES soil moistures in layer 1 become drier than the observations by late July. Additionally, there is a marked disparity between soil moistures in layer 1 and the 10-cm observations from mid-August until early December in 2018. Specifically, the layer 1 soil moisture in JULES responds dramatically to rainfall events in later summer unlike the observations [section 5d(1)]. In Fig. 1g, the maximum daytime latent heat flux in JULES control is consistently higher than the observations between late June and mid-August 2018, but this is improved using canht-rootd (next section).
2) Rainfall versus ET in 2018 and 2019
Figures 2a and 2b shows cumulative plots of observed rainfall and observed and modeled ET in 2018 and 2019. Although the total accumulated rainfall in 2018 was similar to 2019, the distribution through the year was very different. By the end of May 2018, about 240 mm (i.e., 240 kg m−2) of rain had fallen contrasting with 130 mm in 2019. From 1 June to the end of December 2018, including the drought of June and July, a further 242 mm of rain fell. In 2019, from the beginning of June to the end of December, 390 mm fell.
(a),(b) Cumulative observed rain in 2018 and 2019 together with cumulative observed and modeled ET. (c),(d) The observed and modeled (cumulative rain − cumulative ET) differences for 2018 and 2019.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
(a),(b) Cumulative observed rain in 2018 and 2019 together with cumulative observed and modeled ET. (c),(d) The observed and modeled (cumulative rain − cumulative ET) differences for 2018 and 2019.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
(a),(b) Cumulative observed rain in 2018 and 2019 together with cumulative observed and modeled ET. (c),(d) The observed and modeled (cumulative rain − cumulative ET) differences for 2018 and 2019.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
The cumulative ET in Fig. 2 is expressed in equivalent units of water (kg m−2). There is a difference in the observed cumulative ET between the years, being much higher in 2019. The cumulative ET rates in 2018 were far lower than the rainfall rates (Fig. 2a). In 2019, the observed cumulative ET approximately matches the cumulative rainfall with close to parallel evolution until the end of September.
Cumulative ET in JULES is dependent on the specific sensitivity run used. Relative to the control configuration (red in Fig. 2), canht-rootd (blue in Fig. 2) uses a much lower canopy height and much shallower rooting depth that results in reduced rates of evapotranspiration, especially in 2018. The year-end cumulative ET for canht-rootd was almost the same in 2018 and 2019 (Figs. 2a,b). The substantial differences in cumulative ET for control between the two years are related to the differences in soil moisture.
Between April and July 2018, both control and canht-rootd show similar drying in layers 1 and 2 (Fig. 1c). In the same period, in control with a deep rootd, transpiration removed water from layer 3 starting from the sm_crit point until soil moisture was close to the sm_wilt point. The canht-rootd run, with a shallow rootd, primarily accessed water from layers 1 and 2 (Table 1) so that layer 3 never reached sm_wilt (Fig. 1d). The lack of water available from layer 3 for canht-rootd limited the total amount of ET in 2018 and 2019 (Figs. 2a,b). For control in 2018, a considerable amount of moisture was available from layer 3 so cumulative ET was higher than for canht-rootd. For control in 2019, layers 2 and 3 started much drier so the cumulative ET was less than 2018.
We next consider the difference between cumulative rainfall and cumulative ET as shown in Figs. 2c and 2d. As a typical year, in 2019 rain minus ET starts positive, hovers near zero in the summer and returns to positive at the end of the year. Both JULES runs are reasonably successful in reproducing the observations in that year. In 2018 the observations show a very wet start to the year, a decrease in cumulative rain minus cumulative ET during the dry-down and drought that was followed by an increase toward the end of the year. At no point for the observations does the cumulative ET exceed the cumulative rain (Fig. 2c). Evapotranspiration for control is far higher than the observations (Fig. 2a) and cumulative rain minus cumulative ET becomes strongly negative in June 2018. These differences in canht-rootd are closer to the observations but cumulative ET is still too high.
Subsequent sections will be used to investigate the details of the disparity between JULES ET and the observations in 2018.
b. Dry-down in 2018
1) Development of the dry-down
Figure 3a illustrates the 2018 dry-down and the drought in terms of the soil moisture observed at 10 cm and as modeled for layer 1 in JULES. Following rainfall in late May, simulated soil moisture is a good match to the observations. During this period, the grass turned from green to yellow and then brown and it ceased photosynthesising and therefore transpiring. The estimated greenness index (NDVI) as measured at the site decreased substantially between the end of May and early July, and then remained constant at a low level (Fig. 3b). Satellite-derived NDVI based on the MODIS monthly Terra MOD13C2 product was obtained from the EOS Data Gateway (https://earthdata.nasa.gov/). MODIS NDVI starts decreasing at a similar time to the site observed NDVI, but carries on decreasing until early August. The differences between MODIS and site-measured NDVI is partly because the satellite pixel size is larger than the Cardington site so it includes surrounding crops. The site grass did not revive to a green state until the end of September 2018 (not shown), which coincided with the pyranometer observations of NDVI rising to above 0.6.
(a) The dry-down and drought during the summer of 2018 as described by decreasing soil moistures observed at 10 cm and as modeled in JULES layer 1; (b) vegetation greenness estimated by satellite and local measurements of NDVI; (c),(d) observed and modeled upward latent and sensible heat fluxes; (e) observed and modeled skin temperature; and (f) observed and modeled downward ground heat flux (ht = heat, temp = temperature, IRT = infrared temperature).
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
(a) The dry-down and drought during the summer of 2018 as described by decreasing soil moistures observed at 10 cm and as modeled in JULES layer 1; (b) vegetation greenness estimated by satellite and local measurements of NDVI; (c),(d) observed and modeled upward latent and sensible heat fluxes; (e) observed and modeled skin temperature; and (f) observed and modeled downward ground heat flux (ht = heat, temp = temperature, IRT = infrared temperature).
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
(a) The dry-down and drought during the summer of 2018 as described by decreasing soil moistures observed at 10 cm and as modeled in JULES layer 1; (b) vegetation greenness estimated by satellite and local measurements of NDVI; (c),(d) observed and modeled upward latent and sensible heat fluxes; (e) observed and modeled skin temperature; and (f) observed and modeled downward ground heat flux (ht = heat, temp = temperature, IRT = infrared temperature).
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
During the drought all JULES runs have an imposed decrease in LAI from 2.36 to 1.62 as discussed in section 3. The decrease in LAI has a relatively small impact on latent heat flux [section 5b(2)]. At the start of the dry-down, latent heat flux is overestimated by control, but by late June it is severely overestimated by about 100 W m−2 (Fig. 3c). Table 3 summarizes average JULES performance in terms of the mean differences between JULES and observations in energy fluxes, Bowen ratios and skin temperatures (i) before the dry-down, (ii) during the dry-down, and (iii) during the drought. Also shown in Table 3 are the mean absolute Bowen ratios for the same three time periods. As expected, sensible heat compared to latent heat increases substantially as conditions move to meteorological drought.
Mean differences in heat fluxes and skin temperature between the control and canht-rootd JULES simulations and observations (obs) for three time periods shown in Fig. 3. These periods coincide with high NDVI (before dry-down), decreasing NDVI (dry-down), and low NDVI (drought). Values for latent heat, sensible heat, and Bowen ratio only use daytime data. The 95% confidence intervals are shown in parentheses. Also shown are the mean absolute values of Bowen ratio (= sensible heat flux/latent heat flux).
Overall during this period, canht-rootd latent heat flux is a good match with peak daytime observations except for a period from mid- to late June where it underestimates compared to the observations. Both the control and canht-rootd runs overestimate the sensible heat flux with canht-rootd being slightly worse (Table 3). This overestimation could be explained by the UKV configuration having a low albedo.
Sensible heat flux in JULES is related to the difference between air temperature and skin temperature (tstar) multiplied by air density and specific heat capacity and divided by aerodynamic resistance; Eq. (2) of Best et al. (2011). The diurnal range in tstar in control is underestimated compared to the corrected grass IRT (i.e., too cool during the day, and too warm at night, Fig. 3e). During mid-May and also late June, canht-rootd is too warm in the middle of the day by up to 3°–4°C, but later in Fig. 3e the daytime tstar values are quite good. Both control and canht-rootd have nighttime values that are too warm by 2°–4°C.
Figure 4 focuses on four days at the end of the drought (1–4 August 2018). Results from all four JULES runs are illustrated. Midday latent heat flux is far too high compared to observations for control and canht. However, reducing the rooting depth for canht-rootd produces a much closer match, and switching to canht-rootd-kext makes no further difference. Modeled sensible heat flux is slightly higher than observed around midday for all four runs (Fig. 4b). Skin temperature from control is up to 8°C too cold during the daytime. A lower canopy height in canht means a reduced roughness length, a decrease in turbulence and therefore extra heating of the surface in the day. Consequently for the interval shown in Fig. 4c the midday tstar is close to observed for canht, canht-rootd and canht-rootd-kext. In mid-May and late June, however, midday tstar is overestimated even with a reduced canopy height (Fig. 3e). Modeled overnight tskin is too warm in JULES until the optical extinction coefficient (kext) is increased for canht-rootd-kext. The doubling in kext causes less thermal coupling between the surface and the atmosphere allowing extra nighttime cooling.
Observed and modeled heat fluxes and skin temperature over four days of the drought in early August 2018; four JULES runs are shown: control, canht, canht-rootd, and canht-rootd-kext.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Observed and modeled heat fluxes and skin temperature over four days of the drought in early August 2018; four JULES runs are shown: control, canht, canht-rootd, and canht-rootd-kext.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Observed and modeled heat fluxes and skin temperature over four days of the drought in early August 2018; four JULES runs are shown: control, canht, canht-rootd, and canht-rootd-kext.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
The diurnal range of the ground heat flux is larger in all model runs compared to the observations (Figs. 3f and 4d). Unfortunately there is some doubt whether the range in observed ground heat flux is correct because the thermal contact of the instrument with the surrounding soil may not be optimized. Consequently, we cannot judge with confidence which of the JULES runs provides the better estimate.
2) JULES evapotranspiration from dry soils
Considering latent heat flux in JULES, we focus on two short time intervals in 2018 and 2019 where the observed soil moisture conditions in layers 1, 2, and 3 are comparable for both years. In early August 2018 and mid-September 2019 both JULES runs are below the wilting point in layer 1, but at the wilting point in layer 2. In layer 3, control is at the wilting point while canht-rootd is slightly above this point. The difference between the time intervals is that in the 2019 case the grass was alive and therefore photosynthesising and transpiring. Figures 5 and 6 illustrate heat fluxes for control with constant LAI of 2.36 (Figs. 5a and 6a) as well as control (LAI varying in 2018 only). Figures 5a and 5b are almost identical because the LAI is the same in 2019 but there are small inherited differences in soil moisture from the changes in LAI during dry-down in 2018. Figures 5a and 5b show that control substantially overestimates midday ET due to large amounts of transpiration. Figure 5c shows that in the run where we reduce the canopy height alone (canht) from 1.46 to 0.2 m, despite a decrease in turbulence the increase in daytime surface temperature causes a small increase in both transpiration and bare-soil evaporation.
Observed latent heat flux and modeled upward latent heat flux and its components in mid-September 2019 as represented by the JULES runs control with LAI fixed at 2.36, control, canht, and canht-rootd.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Observed latent heat flux and modeled upward latent heat flux and its components in mid-September 2019 as represented by the JULES runs control with LAI fixed at 2.36, control, canht, and canht-rootd.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Observed latent heat flux and modeled upward latent heat flux and its components in mid-September 2019 as represented by the JULES runs control with LAI fixed at 2.36, control, canht, and canht-rootd.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Observed latent heat flux and modeled upward latent heat flux and its components in early August 2018 as represented by the JULES runs control with LAI fixed at 2.36, control, canht, and canht-rootd.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Observed latent heat flux and modeled upward latent heat flux and its components in early August 2018 as represented by the JULES runs control with LAI fixed at 2.36, control, canht, and canht-rootd.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Observed latent heat flux and modeled upward latent heat flux and its components in early August 2018 as represented by the JULES runs control with LAI fixed at 2.36, control, canht, and canht-rootd.
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Figure 5d indicates that additionally shortening the rooting depth from 0.5 to 0.2 m causes a large drop in JULES midday latent heat flux. Hence the canht-rootd run approximately reproduces the correct latent heat flux with substantially lower transpiration compared to the other JULES runs, but essentially unchanged bare-soil evaporation. During the time interval shown in Fig. 5, soil moisture in layer 1 in the simulations is well below the sm_wilt and for canht-rootd layer 2 is at sm_wilt (Figs. 1b,c). Therefore water for photosynthesis can only be accessed in canht-rootd from layer 3 but the reduction in rooting depth has restricted the supply of the water and thus the amount of transpiration (Table 1). As a result, for canht-rootd soil moisture in layer 3 decreases during the dry-down but it remains well above the wilting point (Fig. 1d).
Figure 6 focuses on the ET during the last three days that are illustrated in Fig. 3, which is the driest time in terms of 10-cm soil moisture observations. Figures 6a and 6b show that the effect of reducing LAI for control during the dry-down is to decrease the maximum daytime latent heat flux by up to 30 W m−2. For control, latent heat flux generally exceeds the observed daytime values. Latent heat consists of a small contribution from bare-soil evaporation with the rest from transpiration. Transpiration in control shows a distinct daytime pattern on 5 August with a reduction to almost zero at midday. For canht-rootd this midday drop is seen on all three days and is associated with parallel drops in photosynthesis.
There are three situations where JULES is designed to stop photosynthesis: when soil moisture drops to zero, when there is zero photosynthetically active radiation flux, or where the near-surface specific humidity gradient exceeds a threshold (dqcrit). Nevertheless, during the daytime none of these limits are crossed on the days shown in Fig. 6. An additional parameter, Tupp, sets the upper limit of the optimum leaf temperature range for photosynthesis (Clark et al. 2011). JULES skin temperature, tstar, exceeds Tupp of 36°C only on 5 August for control, whereas tstar exceeds it on 3, 4, and 5 August for canht (Fig. 4c) causing a midday dip in transpiration (Fig. 6c). On 1 and 2 August, midday tstar exceeds Tupp for canht, causing the same suppression of photosynthesis and transpiration and thus leading to a drop in latent heat flux (Fig. 4a). In September 2019 on the days illustrated in Fig. 5, tstar for control, canht, and canht-rootd never exceeds Tupp, so there is no midday drop in latent heat flux. Examining data for 2005–19 reveals multiple cases where JULES transpiration is suppressed around midday when tstar exceeds Tupp. The midday modeled suppression of photosynthesis and thus transpiration seems to be inappropriate. Hence the value for Tupp of 36°C for C3 grass recommended by Clark et al. (2011) and 32°C by Harper et al. (2016) should be reviewed.
In Fig. 6 there is a small decrease in bare-soil evaporation going from canht to canht-rootd, but a large drop in transpiration. Unlike mid-September 2019, in early August 2018 despite these changes, during the drought canht-rootd still has ET that is too high early morning and late afternoon compared to the observations (Fig. 6d). Since in reality the grass is senescent and not transpiring during the drought, the real world ET derives only from bare soil and evaporation of dew from the canopy. Despite the imposed reduction in LAI during the drought, JULES should not be representing any transpiration at this time with or without midday suppression. Yet on average the bare-soil evaporation in canht-rootd is approximately 20 W m−2 too low compared to the observations. The difference between observed midday latent heat and canht-rootd bare-soil evaporation might be partly explained by insufficient canopy water evaporation (ecan) due to evaporation of dew.
c. Dewfall during drought
1) Conditions leading to dewfall
The conditions allowing dewfall involve radiative cooling of the surface at night and consequently the overlying air. Provided that the surface has cooled to or below the dewpoint, implying a temperature inversion at the surface, the air in contact with the canopy will be at 100% relative humidity and dew will form by condensation. During dewfall, the specific humidity of the air above the canopy is larger than the saturated specific humidity at the skin temperature
Figure 7 shows four nights toward the end of the drought in 2018 when dew was forming. Friction velocity as a measure of turbulence is very low on the first three nights but on the fourth night the turbulence picks up slightly just before midnight. Figure 7b shows that overnight there are temperature inversions between the surface and 25 m. As expected, this is associated with a larger gradient in specific humidity between the surface and 25 m than between the surface and the canopy measurements at 0.05 m (Fig. 7e). The dew meter measurements confirm accumulation of dewfall when SHG is negative.
(a)–(d) Four nights of dew formation in early August 2018 showing observations and simulations of associated thermodynamic variables; qz is the specific humidity at height z;
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
(a)–(d) Four nights of dew formation in early August 2018 showing observations and simulations of associated thermodynamic variables; qz is the specific humidity at height z;
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
(a)–(d) Four nights of dew formation in early August 2018 showing observations and simulations of associated thermodynamic variables; qz is the specific humidity at height z;
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Figure 9 of Price and Clark (2014) shows a nighttime thermal infrared photograph where the dew meter with an artificial canopy is on average 1.1°C cooler than the dew meter with the real grass canopy. Our artificial dew meter measurements show dewfall starting earlier in the night than the real dead canopy, consistent with a cooler artificial canopy. Due to differences in the temperatures of the two dew meters, the dead grass canopy is more likely to give an accurate indication of the duration of the grass site dewfall than the artificial canopy. Nevertheless, both dew meters indicate the same maximum negative evaporative flux by the end of the night.
In general, times of vertical temperature inversion coincide with negative dew meter evaporative flux and negative SHG. Just before sunrise on 1 and 3 August, however, the recorded canopy temperature coincides with grass IRT. Consequently, the SHG calculated using the canopy temperature increases to near zero (black line in Fig. 7e) despite continued dewfall according to the dew meters. At the 1.2-, 10-, and 25-m levels, the SHG remains as negative as it was earlier in the night. Therefore the canopy temperature measurements are apparently erroneous when they coincide with grass IRT.
In detail, the times of 100% relative humidity of the air within the canopy (0–5 cm) do not coincide precisely with the times of negative evaporative flux from the dew meter and negative SHG (Fig. 7). The real grass dew meter measurements and SHG become negative before humidity rises to 100% in the air above the canopy. On the morning of 1 August the dew meter values and SHG become positive, indicating the evaporation of dew, before the relative humidity between 0 and 5 cm drops below 100%. Therefore the air measured within the canopy reaches 100% relative humidity up to an hour after dewfall commences, and at sunrise the evaporation of dew sometimes precedes the measured drop in humidity.
2) Latent heat fluxes during dewfall
Measurements of dewfall from both dew meters indicate maximum negative evaporative fluxes of about 10–20 W m−2 during the drought (equivalent to a dew deposition of up to 0.03 mm h−1). Measurements of latent heat flux at 10 m indicate negative values overnight of only a few watts per square meter. On the nights of 31 July and 1 and 2 August, turbulence indicated by friction velocity was generally less than 0.1 m s−1 (Fig. 7a). On these three calm nights, the corresponding mean wind speed at 10 m was 1.5 ± 0.1 m s−1 (±95% confidence interval) with a range of 0.2–3.2 m s−1. At those times, latent heat flux as determined with eddy covariance at 10 m is, within error, measured as 0.0 W m−2.
On the night of 3 August, initially when the friction velocity was below 0.1 m s−1 a strong negative SHG developed, but just before midnight turbulence increased so the friction velocity increased above 0.1 m s−1. On the night of 3 August the mean wind speed was 1.9 ± 0.1 m s−1 ranging from 0.4 to 3.1 m s−1. Wind speed on this night was somewhat higher than on the previous three nights but turbulence increased considerably (Fig. 7a). During the first part of this night, according to measurements at 10 m, latent heat fluxes were essentially zero but when the turbulence increased there was a measurable negative flux. Just after midnight it appears that the increased turbulence led to a reduction in the negative SHG and 10-m latent heat observations returned to about zero. Toward the end of the night, despite the higher turbulence, a negative SHG redeveloped and this was associated with measurements of slightly negative latent heat flux at 10 m. These observations show that the eddy-covariance system only measures negative latent heat flux when there is sufficient overnight turbulence. When the formation of dew due to the radiative cooling of the surface occurs in calm conditions, the eddy covariance measurements fail to represent the negative evaporative flux indicated by the dew meters.
In the canht JULES simulation, the skin temperature increases to more realistic values compared to control during the daytime, but nighttime tskin remains too warm compared to the observations (Fig. 4c). Changing from canht to canht-rootd has no significant effect on diurnal tskin range so the nighttime temperatures remain too warm. The doubling of the canopy optical extinction coefficient to 2.0 in canht-rootd-kext, however, leads to an improved and cooler nighttime tstar (Fig. 4c). Specific humidity gradients calculated by JULES and shown in Fig. 7e refer to the drive level of 1.2 m. Therefore the SHG from canht-rootd-kext (purple) should follow the gray line in Fig. 7e.
On the first night (31 July–1 August), canht-rootd-kext-kext SHG follows the observed SHG fairly well (Fig. 7e), but on subsequent nights tstar is warmer than the grass IRT, so the JULES SHG does not remain negative for as long as the observations. Formation of dew on the canopy in JULES is represented by negative ecan values and therefore negative latent heat flux (since bare-soil evaporation and transpiration are zero). Nevertheless, all the JULES simulations have latent heat flux very close to zero on the first three nights consistent with the 10-m observations (Fig. 7d). Toward the end of the night of 3–4 August, when there was appreciable turbulence, JULES has a slightly negative latent heat flux, consistent with the observations of latent heat flux at 10 m. Note that friction velocity according to JULES on this night is far lower than the observations (Fig. 7a). This demonstrates that at times JULES fails to correctly parameterize turbulence.
Comparing the patterns of changes in overnight SHG for z = 1.2, 10, and 25 m, there is a clear similarity with the changes in negative evaporative flux in the dew meter data. On the morning of 1 August the values remain flat and on 4 August there are more negative values before and after midnight. On the mornings of 2 and 3 August there are clear changes toward more dewfall and more negative SHG. These observations suggest that rates of dewfall are related to the magnitude of negative SHG, depending on the reference height.
Since dewfall is a process related to radiative cooling of the surface, typically in calm surface conditions, the eddy-covariance measurements of latent heat flux at 10 m almost invariably fail to represent the times, magnitudes and changes in negative heat flux overnight (Fig. 7d). For the same reasons, JULES is currently unable to simulate appropriate negative latent heat flux during dewfall conditions. Despite the improvement in JULES SHG overnight when using canht-rootd-kext compared to control, and also despite the fact that the patterns are correct, the SHG values are often less negative than those observed at 1.2 m (Fig. 7e). On the night of 31 July, the minimum in tstar for canht-rootd-kext matches the grass IRT (Fig. 4c). Additionally, the JULES SHG matches the drive-level observations of SHG at 1.2 m but latent heat flux is still simulated as essentially zero because the turbulence is minimal and critically the transfer of moisture to the surface in these conditions is not parameterized.
De Roode et al. (2010) also demonstrated that during clear, stable nights the eddy correlation technique provides latent heat fluxes that are erroneously small. Furthermore, Xiao et al. (2017) analyzed use of four equations to compute dewfall, but they recommended that even their most favored equation (Penman–Monteith) should not be used without comparison to high-precision lysimeter (dew meter) data. Parameterizations of turbulent heat fluxes in JULES are based on the Penman–Monteith equation, such that latent heat is specified as the positive or negative SHG multiplied by air density and divided by the sum of aerodynamic and stomatal resistances; Eq. (3) of Best et al. (2011). However, aerodynamic resistance tends to infinity as the wind speed approaches zero so that the JULES-calculated latent heat tends to zero (Allen et al. 1998). We have demonstrated that dew formation on clear, calm nights is independent of measurable turbulence. We believe that this inability to represent dewfall associated with radiative cooling in calm conditions is common to all land surface models that parameterize energy fluxes via turbulence.
3) Daytime latent heat flux following dewfall
Following dewfall, the dew meters after sunrise indicate a rapid change from negative to positive evaporative flux (Fig. 7d). The artificial grass canopy has a clear brief interval of a few hours of evaporation of the previous night’s dew. Values during the rest of the daylight hours hover close to zero but probably have an error of less than ±5 W m−2. Unlike the artificial grass canopy, the real dead grass dew meter shows positive evaporative flux throughout the day. This is interpreted as recording both initial evaporation of dew as for the artificial canopy, and bare-soil evaporation.
The daytime ET values from the real dead grass dew meter agree fairly well with the latent heat flux measured at 10 m. Therefore, this dew meter provides a reasonable representation of the daytime evaporative flux of the surrounding site. Just before the time interval indicated in Fig. 7 there were a few millimeters of rainfall that caused a significant increase in mass on both dew meters (Fig. 1a). The rainwater evaporated very fast from the artificial grass dew meter. However, the water from the rain was still present as soil moisture, albeit rapidly decreasing, on the real dead grass dew meter during the first few days of August. Nevertheless, it is possible that some of the positive evaporative flux from the real dead grass dew meter represents a small component of water that had been adsorbed overnight on to the soil particles.
The change from canht-rootd-kext to canht-rootd-kext causes virtually no change in daytime bare-soil evaporation or latent heat flux (section 3); for example compare 3 August in Figs. 6c and 7d. As discussed in section 5b(2) latent heat flux from canht-rootd in early August includes about 50% transpiration, whereas in reality the grass is senescent. In early August the modeled daytime latent heat flux should exclude transpiration and only include canopy and bare-soil evaporation. Latent heat observations at 10 m during the daytime apparently include the contribution from the evaporation of dew. If the nighttime dewfall could be improved within JULES then potentially this would lead to an increase in latent heat flux during the day associated with canopy water evaporation (ecan). Hence, an increased ecan, due to improved modeling of overnight dewfall, added to the daytime bare-soil evaporation will lead to better modeled latent heat flux at the end of the drought (Fig. 6c).
d. Dry ground, soil moisture, and evaporation
1) Response of soil moisture to rainfall on dry ground
In normal soil conditions at Cardington, tens of mm of rain within a few days would cause the soil moisture at 10-cm depth to increase substantially. For example, in Fig. 1b between 24 and 29 September 2019 the soil moisture at 10 cm increased by 20% from 9.8% to 29.8% after a total of 32.0 mm of rain. The long dry spell of 2018 was broken by a similar amount of rain (29.7 mm) between 9 and 12 August (Fig. 8a). However, the soil moisture at 10 cm, of similar dryness to September 2019, increased gradually by just 1.4% from 10.3% to 11.7% after the rain according to the theta probe (Fig. 8b). The alternative PR2 probe, deployed 5 m away from the theta probe, showed the soil moisture rising in shallow steps by 5.9% from 7.2% to 13.1% over the same period.
Response to rainfall after the drought of the 10-cm observations and JULES layer 1 soil moistures and of the modeled and observed heat flux components (bare-soil = bare-soil evaporation, ecan = canopy-only evaporation).
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Response to rainfall after the drought of the 10-cm observations and JULES layer 1 soil moistures and of the modeled and observed heat flux components (bare-soil = bare-soil evaporation, ecan = canopy-only evaporation).
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
Response to rainfall after the drought of the 10-cm observations and JULES layer 1 soil moistures and of the modeled and observed heat flux components (bare-soil = bare-soil evaporation, ecan = canopy-only evaporation).
Citation: Journal of Hydrometeorology 22, 2; 10.1175/JHM-D-20-0148.1
The different responses to rainfall at 10 cm of the soil moisture theta probe (gradual increase by 1.4%) and the alternative PR2 probe (stepped increased by 5.9%) is interpreted here as an indication of spatial heterogeneity in the soil moisture characteristics. During the drought a few mud cracks became very large, with holes up to 10-cm diameter and 10 cm deep. In nondrought conditions, rainfall events produce larger responses (percent increase in soil moisture) at 10 cm compared to 22 cm (Figs. 1b,c). At the end of August, end of September and middle of October 2018, however, there were rain events that caused much larger increases in soil moisture at 22 cm than at 10 cm. These observations can be explained by rainwater bypassing the top layer through macropores (i.e., heterogeneous or preferential flow).
The observations of soil moisture at 10 cm in 2018 demonstrate that after the drought the soil moisture response to rainfall was far smaller than during normal conditions. This is a result of the medium clay loam having far lower permeability and therefore lower hydraulic conductivity—apart from the macropores.
The amount of water infiltration in JULES can be diagnosed from throughfall minus surface runoff. The fixed infiltration proportion in the UKV configuration of JULES during the rain events accounts for about 80% of the precipitation (Fig. 8a). Unlike the observations, the rainfall events of 9–12 August 2018 lead to the simulated layer 1 soil moisture rising rapidly above the wilting point (Fig. 8b, Table 2). For both model runs, layer 1 soil moisture increased by 19%, from 9.5% to 28.5% for control and for canht-rootd from 7.5% to 26.5%. On the other hand, at this time there is no impact on the soil moisture content in layers 2, 3, and 4 (Fig. 1).
For the observations, it appears that it was the long interval without significant rain that has caused the topmost soil to reduce its permeability and hydraulic conductivity, i.e., in the soil above the 10-cm observation depth. The response of layer 1 JULES soil moisture to rainfall after the drought is far higher at 19% than the measured 10 cm increases of 1.4% and 5.9%. The responsiveness of the modeled soil moisture in layer 1 to rainfall events is the same regardless of the preceding rainfall and drying history.
In JULES and other LSMs, the hydraulic conductivity depends on the soil moisture and it is assumed there is a single valued relationship with capillary potential (Richards’ equation; e.g., Beven 2012). Studies of in situ and laboratory-based soil columns demonstrate that the soil–water characteristic curves can be different for soils that are wetting to those soils that are drying (Jaynes 1990; Pham et al. 2003, 2005). This is known as soil moisture hysteresis. The JULES soil moisture responsiveness could potentially be improved by including soil moisture hysteresis. For example, the Brooks–Corey soil–water characteristic curve could be used as the boundary drying curve from which the boundary wetting curve could be predicted (e.g., Pham et al. 2003).
2) Bare soil evaporation after rainfall on dry ground
The observed latent heat flux before the rainfall in early August 2018 was only a few tens of watts per square meter in the middle of the day due to bare-soil evaporation from very dry soil (Fig. 8d). During the rainy days, observed daytime latent heat flux increased to more than 150 W m−2. After the rain ceased, observed daytime latent heat decreased to 30–60 W m−2. Throughout this interval the grass remained senescent so that the observed latent heat flux can only be accounted for by evaporation from the canopy and from bare soil. Despite the large increase in bare-soil evaporation compared to 7–8 August, the almost static observed soil moisture indicates that the wetting front barely reached 10 cm (Fig. 8b). As explained in the previous section, after a long period of dry conditions, the soil prevented normal infiltration of rain.
Sensible heat flux as observed and modeled decreases substantially from the clear-sky conditions during drought (7–8 August) to the days with cloud cover and rainfall (9–12 August). Although changing from control to canht or canht-rootd has warmed tskin significantly for the better (Fig. 4c), it has had little or no effect on daytime sensible heat, which matches the observations relatively well (Fig. 8c).
Before the rainfall, as described in section 5b(2), latent heat flux from control is about double the latent heat from canht-rootd (Figs. 6a,c). The two components of the latent heat are bare-soil evaporation, which is almost identical between the two runs, and transpiration, which in control is about twice that of canht-rootd (Fig. 8e). After the rain starts on 9 August, latent heat is almost the same for both JULES runs and from 10 August higher than during the drought (Fig. 8d). On 9–10 August, the modeled latent heat flux is almost entirely accounted for by evaporation from the canopy (ecan, Fig. 8f). Canopy evaporation contributes a proportion of latent heat flux on 11–13 August, especially in the mornings, but then in the absence of rain, returns to zero by 14 August.
On 11–15 August, control and canht-rootd transpiration represents a significant fraction of the total latent heat flux. The difference in latent heat flux between control and canht-rootd has essentially disappeared after the rain starts because the transpiration in canht-rootd is far higher than before. Since the soil moisture in layers 2 and 3 for canht-rootd remains at the wilting point, it must be the increase in soil moisture above the wilting point in layer 1 that allows the large increase in transpiration. Since the grass is dead, however, the transpiration should not be modeled as contributing to latent heat flux. In contrast, the bare-soil evaporative flux from both JULES runs is much less than the observed latent heat flux on 11–13 August, and there is a small component of canopy evaporation (ecan, Figs. 8e,f). By 14–15 August, when the modeled canopy evaporation has dropped to zero, the modeled bare-soil evaporation essentially matches the observed latent heat flux at 10 m.
6. Discussion and conclusions
Observations of soil moisture and evapotranspiration (ET) during the meteorological drought in 2018 at Cardington (Bedfordshire, United Kingdom) have highlighted several issues related to modeling using the JULES surface scheme. The UKV configuration is used in the standalone mode of JULES for a single point (control run). This produces excessive ET that can be largely corrected by reducing canopy height and rooting depth within the model (canht-rootd). Soil moisture variability in JULES layer 1 (0–0.1 m) is a good match for observations, except after the drought when rain returns. Soil moisture in layers 2 (0.1–0.35 m) and 3 (0.35–1.0 m) provide a reasonable match to the observations in general but the responsiveness to rain events is too slow.
Skin and screen-level biases in temperature, i.e., insufficient diurnal ranges, are long-standing issues in both standalone JULES and operational forecasts of the Unified Model. These biases are greatest on clear radiatively cooled nights and sunny days (Bush et al. 2020). Our results show that the diurnal skin temperature biases in JULES within the UKV model can be improved by shortening the canopy height of grassland to 0.2 m, which allows increased daytime warming by 4°–7°C. Additionally, doubling the canopy optical extinction coefficient from 1.0 to 2.0, leads to increased cooling by 1°–2°C on clear nights and produces good agreement with the observations.
Reducing the canopy height compared to control leads to increased midday skin temperature so that sometimes tstar exceeds Tupp (fixed at 36°C). At such times, JULES suppresses photosynthesis and transpiration. When the grass is not senescent sometimes very high temperatures are observed in the middle of the day. For example on 23–25 July 2019 (not shown), temperatures are observed exceeding 36°C (Tupp) with grass IRT at 40.7°C, and with air temperatures at 0.05 m of 40.6°C and at 1.2 m of 36.5°C. On this nondrought but clear-sky day, both control and canht-rootd underestimate total latent heat by an average of ≈50 W m−2. This suggests that the value of Tupp in JULES should be reinvestigated for C3 grass.
During the drought of 2018, even using canht-rootd and with the imposed reduction in LAI, JULES represents transpiration but this should be zero as the grass was senescent. Within our model configuration canht-rootd, layer 1 correctly dries to below the wilting point and layer 2 reaches the wilting point at the end of the dry-down, but layer 3 still supplies water for transpiration. Experimenting by reducing rootd to less than 0.2 m leads to too little drying out of layer 3 (not shown), thereby disagreeing with the observations. We speculate that because the representation of soil water availability through roots is in slabs (JULES soil layers) rather than being continuous, any water present in layer 3 permits continued transpiration and photosynthesis. More accurately, the shallower parts of this model layer should dry out before the deeper parts so that all the soil layers with grass roots are at the wilting point and therefore the grass photosynthesis and transpiration should cease during the drought.
During the drought, excluding the erroneous contribution from transpiration from the canht-rootd run, modeled bare-soil evaporation underestimates observed midday latent heat flux by about 20 W m−2. A component of the missing latent heat is probably related to the underrepresentation of evaporation of dew from the canopy. Peters (2013) noted that for very dry soils, film and vapor flow need to be modeled otherwise bare-soil evaporation is underestimated. In particular, hydraulic properties for very dry soils associated with adsorptive water retention and film conductivity need to be represented. Experimentation with JULES demonstrates that implementing the parameterization of Peters (2013) leads to improved ground heat flux and land surface temperatures in arid and semiarid regions (J. M. Edwards 2020, personal communication).
Following the long drought, the soil properties of the topmost soil at Cardington changed so that despite a substantial amount of rain over four days almost no moisture reached a depth of 10 cm. This apparently indicates that there was a reduction in the permeability and hydraulic conductivity of the soil. In contrast, the top soil layer of JULES with the UKV configuration has the same rate of infiltration and hydraulic conductivity as in normal conditions. Modeling soil moistures could potentially be improved by modifying the soil water characteristic curves to allow for soil moisture hysteresis.
In terms of observations at night, as expected observed specific humidity gradients are negative at times when the evaporative flux measured by the real dead grass dew meter indicates dewfall. During the drought of 2018 maximum dewfall fluxes were measured as 10–20 W m−2 (deposition rates of up to 0.03 mm h−1). Since dew is formed during times of radiative cooling of the surface, when turbulence is often minimal, the eddy-covariance system fails to reliably record the times, intensity and variations in the negative latent heat flux. Additionally, the parameterization of turbulence in JULES and other land surface models based on using mean states to estimate evaporation are also unable to correctly estimate negative evaporative fluxes during calm conditions on clear nights. Correct modeling of dewfall during radiative cooling of the surface with very low or zero wind speeds, requires appropriate skin and air temperatures, specific humidity gradients and a new parameterization of the transfer of moisture to the surface.
Acknowledgments
We are grateful to colleagues at the Met Office research unit at Cardington who maintain the surface site instruments. Our thanks to Martin Best and John Edwards for helpful comments that improved an early draft, and to Lina Mercado for advice concerning photosynthesis and transpiration in JULES. We also thank three anonymous reviewers for their constructive comments.
REFERENCES
Allen, R. G., L. S. Pereira, D. Raes, and M. Smith, 1998: Crop evapotranspiration: Guidelines for computing crop water requirements. FAO Irrigation and Drainage Paper 56, 300 pp., http://www.fao.org/docrp/X0490E/x0490e00.htm.
Best, M. J., and Coauthors, 2011: The Joint UK Land Environment Simulator (JULES), model description—Part 1: Energy and water fluxes. Geosci. Model Dev., 4, 677–699, https://doi.org/10.5194/gmd-4-677-2011.
Beven, K., 2012: Rainfall-Runoff Modelling: The Primer. 2nd ed. Wiley-Blackwell, 457 pp.
Boone, A., and Coauthors, 2019: Land Surface Interactions with the Atmosphere over the Iberian Semi-Arid Environment (LIAISE). GEWEX News, Vol. 29, No. 1, International GEWEX Project Office, Silver Spring, MD, 8–10, https://hal.archives-ouvertes.fr/hal-02392949/document.
Brooks, R. H., and A. T. Corey, 1964: Hydraulic properties of porous media. Colorado State University Hydrology Papers 3, 37 pp., https://mountainscholar.org/bitstream/handle/10217/61288/HydrologyPapers_n3.pdf.
Burton, R. G. O., 1999: Assessment of soil characteristics at the Meteorological Office Research Unit experimental site. Cardington Tech. Note 49, 25 pp., https://digital.nmla.metoffice.gov.uk/IO_5dcd1143-a38f-4936-91ea-53d3c5bb1655/.
Bush, M., and Coauthors, 2020: The first Met Office unified model–JULES regional atmosphere and land configuration, RAL1. Geosci. Model Dev., 13, 1999–2029, https://doi.org/10.5194/gmd-13-1999-2020.
Clark, D. B., and Coauthors, 2011: The Joint UK Land Environment Simulator (JULES), model description – Part 2: Carbon fluxes and vegetation dynamics. Geosci. Model Dev., 4, 701–722, https://doi.org/10.5194/gmd-4-701-2011.
Cosby, B. J., G. M. Hornberger, R. B. Clapp, and T. R. Ginn, 1984: A statistical exploration of the relationships of soil moisture characteristics to the physical properties of soils. Water Resour. Res., 20, 682–690, https://doi.org/10.1029/WR020i006p00682.
Cox, P. M., R. A. Betts, C. B. Bunton, R. L. H. Essery, P. R. Rowntree, and J. Smith, 1999: The impact of new land surface physics on the GCM simulation of climate and climate sensitivity. Climate Dyn., 15, 183–203, https://doi.org/10.1007/s003820050276.
Dai, A., 2013: Increasing drought under global warming in observations and models. Nat. Climate Change, 3, 52–58, https://doi.org/10.1038/nclimate1633.
de Roode, S. R., F. C. Bosveld, and P. S. Croon, 2010: Dew formation, eddy correlation, latent heat fluxes and the surface energy imbalance at Cabauw during stable conditions. Bound.-Layer Meteor., 135, 369–383, https://doi.org/10.1007/s10546-010-9476-1.
Edwards, J. M., J. R. McGregor, M. R. Bush, and F. J. Bornemann, 2011: Assessment of numerical weather forecasts against observations from Cardington: Seasonal diurnal cycles of screen-layer and surface temperatures and surface fluxes. Quart. J. Roy. Meteor. Soc., 137, 656–672, https://doi.org/10.1002/qj.742.
Hanisch, S., C. Lohery, and A. Buerkert, 2015: Dewfall and its ecological significance in semi-arid coastal south-western Madagascar. J. Arid Environ., 121, 24–31, https://doi.org/10.1016/j.jaridenv.2015.05.007.
Harper, A. B., and Coauthors, 2016: Improved representation of plant functional types and physiology in the Joint UK Land Environment Simulator (JULES v4.2) using plant trait information. Geosci. Model Dev., 9, 2415–2440, https://doi.org/10.5194/gmd-9-2415-2016.
Horlacher, V., S. R. Osborne, and J. D. Price, 2012: Comparison of two closely located meteorological measurement sites and consequences for their areal representativity. Bound.-Layer Meteor., 142, 469–493, https://doi.org/10.1007/s10546-011-9684-3.
Jaynes, D. B., 1990: Soil water hysteresis–models and implications. Process Studies in Hill Slop Hydrology, Wiley, 93–126.
Johansen, O., 1975: Thermal conductivity of soils. Ph.D. dissertation, Norwegian University of Science and Technology, 291 pp.
Moore, R. J., and R. T. Clarke, 1981: A distribution function approach to rainfall-runoff modeling. Water Resour. Res., 17, 1367–1382, https://doi.org/10.1029/WR017i005p01367.
Moro, M. J., A. Were, L. Villagarca, Y. Canton, and F. Domingo, 2007: Dew measurements by eddy covariance and wetness sensor in a semiarid ecosystem of SE Spain. J. Hydrol., 335, 295–302, https://doi.org/10.1016/j.jhydrol.2006.11.019.
Osborne, S. R., S. J. Abel, I. A. Boutle, and F. Marenco, 2014: Evolution of stratocumulus over land: Comparison of ground and aircraft observations with numerical weather prediction simulations. Bound.-Layer Meteor., 153, 165–193, https://doi.org/10.1007/s10546-014-9944-0.
Peters, A., 2013: Simple consistent models for water retention and hydraulic conductivity in the complete moisture range. Water Resour. Res., 49, 6765–6780, https://doi.org/10.1002/wrcr.20548.
Pham, H. Q., D. G. Fredlund, and S. L. Barbour, 2003: A practical hysteresis model for the soil-water characteristic curves for soils with negligible volume change. Geotechnique, 53, 293–298, https://doi.org/10.1680/geot.2003.53.2.293.
Pham, H. Q., D. G. Fredlund, and S. L. Barbour, 2005: A study of hysteresis models for soil-water characteristic curves. Can. Geotech. J., 42, 1548–1568, https://doi.org/10.1139/t05-071.
Price, J. D., and R. Clark, 2014: On the measurement of dewfall and fog-droplet deposition. Bound.-Layer Meteor., 152, 367–393, https://doi.org/10.1007/s10546-014-9930-6.
Prudhomme, C., S. Parry, J. Hannaford, D. B. Clark, S. Hagemann, and F. Voss, 2011: How well do large-scale models reproduce regional hydrological extremes in europe? J. Hydrometeor., 12, 1181–1204, https://doi.org/10.1175/2011JHM1387.1.
Prudhomme, C., and Coauthors, 2014: Hydrological droughts in the 21st century, hotspots and uncertainties from a global multimodel ensemble experiment. Proc. Natl. Acad. Sci. USA, 111, 3262–3267, https://doi.org/10.1073/pnas.1222473110.
Sun, C., S. Sun, T. Liu, D. Guo, S. Mu, H. Yan, W. Ju, and J. Li, 2013: The validation of a model estimating the leaf area index of grasslands in southern China. Rangeland J., 35, 245–250, https://doi.org/10.1071/RJ12025.
Tallaksen, L. M., and K. Stahl, 2014: Spatial and temporal patterns of large-scale droughts in Europe: Model dispersion and performance. Geophys. Res. Lett., 41, 429–434, https://doi.org/10.1002/2013GL058573.
Tang, Y., H. Lean, and J. Bornemann, 2013: The benefits of the Met Office variable resolution NWP model for forecasting convection. Meteor. Appl., 20, 417–426, https://doi.org/10.1002/met.1300.
Uclés, O., L. Villagarcia, M. J. Moro, Y. Canton, and F. Domingo, 2014: Role of dewfall in the water balance of a semiarid coastal steppe ecosystem. Hydrol. Processes, 28, 2271–2280, https://doi.org/10.1002/hyp.9780.
Ukkola, A. M., M. G. De Kauwe, A. J. Pitman, M. J. Best, G. A. Bramowitz, V. Haverd, M. Decker, and N. Haughton, 2016: Land surface models systematically overestimate the intensity, duration and magnitude of seasonal-scale evaporative droughts. Environ. Res. Lett., 11, 104012, https://doi.org/10.1088/1748-9326/11/10/104012.
Veste, M., and T. Littmann, 2006: Dewfall and its geo-ecological implications for biological surface crusts in desert sand dunes (north-western Negev, Israel). J. Arid Land, 16, 139–147.
Xiao, H., R. Meissener, H. Borg, R. Wang, and Q. Cao, 2017: Agreement of four equations for computing dewfall in northern Germany. Water, 9, 607, https://doi.org/10.3390/w9080607.
