Interactions between soil moisture and terrestrial evaporation affect water cycle behaviour and responses between the land surface and the atmosphere across scales. With strong heterogeneities at the land surface, the inherent spatial variability in soil moisture makes its representation via point-scale measurements challenging, resulting in scale mismatch when compared to coarser-resolution satellite-based soil moisture or evaporation estimates. The Cosmic Ray Neutron Probe (CRNP) was developed to address such issues in the measurement and representation of soil moisture at intermediate scales. Here, we present a study to assess the utility of CRNP soil moisture observations in validating model evaporation estimates. The CRNP soil moisture product from a pasture in the semi-arid central west region of New South Wales, Australia, was compared to evaporation derived from three distinct approaches, including the Priestley–Taylor (PT-JPL), Penman–Monteith (PM-Mu), and Surface Energy Balance System (SEBS) models, driven by forcing data from local meteorological station data and remote sensing retrievals from the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor. Pearson's correlations, quantile–quantile (Q–Q) plots, and analysis of variance (ANOVA) were used to qualitatively and quantitatively evaluate the temporal distributions of soil moisture and evaporation over the study site. The relationships were examined against nearly 2 years of observation data, as well as for different seasons and for defined periods of analysis. Results highlight that while direct correlations of raw data were not particularly instructive, the Q–Q plots and ANOVA illustrate that the root-zone soil moisture represented by the CRNP measurements and the modelled evaporation estimates reflect similar distributions under most meteorological conditions. The PT-JPL and PM-Mu model estimates performed contrary to expectation when high soil moisture and cold temperatures were present, while SEBS model estimates displayed a disconnect from the soil moisture distribution in summers with long dry spells. Importantly, no single evaporation model matched the statistical distribution of the measured soil moisture for the entire period, highlighting the challenges in effectively capturing evaporative flux response within changing landscapes. One of the outcomes of this work is that the analysis points to the feasibility of using intermediate-scale soil moisture measurements to evaluate gridded estimates of evaporation, exploiting the independent, yet physically linked nature of these hydrological variables.
Land surface evaporation and soil moisture play major roles in defining the water cycle behaviour of landscapes as well as controlling the feedback from the land surface to the atmosphere at a range of spatial and temporal scales (Manfreda et al., 2007; Seneviratne et al., 2010). The coupling between soil moisture and the overlaying atmosphere has been a topic of intense investigation in recent years. Koster et al. (2006) compared multiple atmospheric general circulation models with regard to the strength of land–atmosphere couplings and reported that while the coupling strengths varied widely for the models, most models agreed upon certain locations of high land–atmosphere coupling. Dirmeyer (1994) used a simple biosphere model to evaluate the effect of soil moisture and vegetation stress on the climatology of drought, while Martens et al. (2016) showed that assimilating satellite-derived soil moisture improved model estimates of terrestrial evaporation at the continental scale. Land–atmosphere coupling studies have also investigated, among others aspects, the impact of soil moisture on precipitation (Eltahir 1998; Koster et al., 2004; Schär et al., 1999) and how this knowledge can be an indicator of climate change (Seneviratne et al., 2006); the links between soil moisture and cloud cover (Betts, 2004); and also how the ENSO cycle influences the coupling and the surface–atmosphere feedbacks (Miralles et al., 2014).
Land surface evaporation (sometimes also referred to as ET) comprises the processes of plant transpiration, evaporation from the soil and evaporation from canopy-intercepted rainfall (Kalma et al., 2008), and has been estimated to return up to 70 % of precipitated water back to the atmosphere (Hanson, 1991; Trenberth et al., 2011). In arid and semi-arid regions, this value can be much larger. Although coupling between evaporation and soil moisture is expected to be high in arid and semi-arid regions, the dynamics of surface–atmosphere feedbacks are not well understood in such environments (Wang et al., 2012). Several studies have attempted to describe these links, with the aim of predicting one variable through knowledge of the other (Mintz and Walker, 1993; Wetzel and Chang, 1987) or to use developed relationships to inform upon linked hydrological responses such as evaporation (McCabe et al., 2005; Stisen et al., 2011), soil moisture (Liu et al., 2012), drought (Entekhabi et al., 1992; Fischer et al., 2007; Oglesby and Erickson, 1989), precipitation (Findell et al., 2011; Held et al., 2005), and even vegetation response to soil moisture stress (Liu et al., 2011). A common feature of such studies is the use of model estimates of terrestrial evaporation across a wide range of study areas and land cover and biome types. The reliance on these model estimates necessitates a more critical examination of both the models used and an evaluation of their performance.
With this in mind, the Global Energy and Water Cycle Exchanges (GEWEX) LandFlux project (McCabe et al., 2016) and the related Water Cycle Multi-mission Observation Strategy – Evapotranspiration (WACMOS-ET) project (Michel et al., 2016; Miralles et al., 2016) reflect ongoing efforts to develop strategies for the prediction of land surface fluxes at regional and global scales. As part of these activities, studies were undertaken to compare the remote-sensing-derived evaporation products with tower-based measurements: a standard approach to flux evaluation (Ershadi et al., 2014). However, such comparisons suffer from both spatial- and temporal-scale mismatches, making robust evaluations inherently challenging. Generally, the spatial footprint of satellite-based sensors is much larger than the fetch of an eddy-covariance tower. Furthermore, while tower-based sensors routinely record information at intervals of between 15 and 30 min throughout the diurnal cycle, many satellite observations used in hydrological studies are generally instantaneous retrievals that may only be available at a daily interval. While questions on the suitability of comparing large-scale gridded evaporation estimates to fine-scale tower observations have been raised previously (McCabe et al., 2016), they remain largely unresolved. As such, identifying complimentary observation sources that can be used to improve the evaluation of a variety of hydrological processes is a much-needed objective (McCabe et al., 2008). This critical need forms a key motivation of this work where we look for an answer to the following question: are independent hydrological datasets available that can be used to inform upon linked elements of the hydrological cycle?
From an observational perspective, a range of approaches have been employed to obtain soil moisture values at multiple resolutions (Jana and Mohanty, 2012; Jana et al., 2008; Vereecken et al., 2007). Generally, soil moisture measurements are made using either in situ devices that are ground-based or via air- and satellite-borne sensors. While in situ measurements tend to represent a spatial scale on the order of centimetres, remotely sensed soil moisture products have resolutions on the order of several hundred metres (airborne) to tens of kilometres (satellite based) (Vereecken et al., 2007). Unfortunately, field-scale spatial variability of soil moisture is generally smoothed out in large-scale soil moisture estimates (Manfreda et al., 2007). While technically a number of point-scale measurements can be collected and then spatially averaged over a domain, such an approach is infeasible for continuous monitoring over multiple fields. Establishment of critical zone observatories in recent years has provided valuable insight into hydrological process due to their intensive instrumentation (Lin et al., 2011). However, they involve significant outlays involved with regards to finance, physical effort, and time. Although such observatories are invaluable in providing data to understand the underlying processes, it remains impractical to implement a large number of sensors across any and every field of interest. As such, there is a clear need for resolving field-scale heterogeneity at scales that can be monitored from remotely observing platforms or conversely, providing ground-based observations that better reflect the scale of satellite systems.
In recent years, the challenge of obtaining intermediate-resolution (between point-scale and satellite resolution) soil moisture has been addressed by the use of Cosmic Ray Neutron Probes (CRNPs) (Zreda et al., 2012). Based on determining the neutron density of cosmic rays, CRNPs are able to make measurements of soil moisture at spatial resolutions of a few hundred metres. The CRNP footprint, with a radius of approximately 130 m (tropical climate) to 240 m (arid/semi-arid climates) (Kohli et al., 2015), is comparable to that of airborne remote sensors, while being ground-based and capable of continuously recording soil moisture over long periods of time. Moreover, the effective measurement depth of the CRNP sensor ranges from 12 to 76 cm, depending on the degree of saturation. This depth allows the sensor to capture the root-zone soil moisture dynamics to a great extent (Desilets et al., 2010).
Development of such approaches to capture the field-scale dynamics of soil moisture brings with it the ability to more closely explore the interactions between the surface and atmosphere, and their linking mechanisms. Soil surface evaporation and plant transpiration processes are influenced significantly by the root-zone soil moisture: transpiration rates depend upon the amount of plant-available water in the root zone, while evaporation plays a regulatory role in governing the dynamics of the surface soil moisture. It is well recognized that soil moisture is a limiting factor in the evaporative process (Seneviratne et al., 2010), playing an important role in modulating plant stress and vegetation response. With improved sensing of the root-zone soil moisture, it is expected that any modelled relationship between evaporation and soil moisture will be more robust. From an observational standpoint, however, it has been challenging to explore these links directly due to the mismatch in data scales. Using CRNP soil moisture data collocated with gridded model estimates of evaporation may provide some insight into these processes and relationships.
Given this, the objective of this study is to investigate the potential of using CRNP soil moisture retrievals to evaluate modelled evaporation estimates derived from a combination of tower-based and remote sensing inputs. The capacity to indirectly monitor surface flux responses using such data offers a mechanism through which land–atmosphere couplings can be explored and provides an additional constraint on coupled water and energy cycle modelling at the land surface. In order to achieve this objective, we conducted an exploratory study that examined the link between the soil moisture retrievals from CRNP and evaporation estimates collected over a semi-arid grassland. The underlying hypothesis to be tested here is that rainfall input in such landscapes will be well reflected by the soil moisture values, which in turn should be strongly coupled to evaporation.
The study was conducted over a pasture site near Baldry, a rural township in
the central west of New South Wales (NSW), Australia. The site is classified as a
semi-arid region with latitude
Location of the Baldry study area in the central west of NSW Australia, along with a photograph of the study area.
Of the three evaporation models evaluated in this study, the Priestley–Taylor
Jet Propulsion Laboratory (PT-JPL) model uses the least number of
meteorological and remote sensing input data, including air temperature,
humidity, net radiation, and a vegetation index (Fisher et al., 2008). The
model has been used to estimate actual evapotranspiration at local (Ershadi
et al., 2014) and global scales in various studies (e.g. Badgley et al.,
2015; Ershadi et al., 2014; Vinukollu et al., 2011), including the recent
LandFlux (McCabe et al., 2016) and WACMOS-ET (Miralles et al., 2016) efforts.
Detailed descriptions of the model are provided in those references. The
PT-JPL is a three-source model that uses net radiation (
The Penman–Monteith-based model developed by Mu et al. (2011) (PM-Mu) is
another three-source scheme that has been used in a range of applications for
estimating terrestrial fluxes (Mu et al., 2013), including forming the basis
behind the global evaporation product (MOD16) (Mu et al., 2013). The PM-Mu
model computes total evaporation as the sum of the three components: soil
evaporation, canopy transpiration, and evaporation of the intercepted water in
the canopy, i.e. (
The Surface Energy Balance System (SEBS) model (Su, 2002a) is a physically based scheme that has been widely used in estimating evaporation across a range of scales (Elhag et al., 2011; McCabe and Wood, 2006; Su et al., 2005a). The model utilizes commonly available hydrometeorological variables, including net radiation, land surface temperature, air temperature, humidity, wind speed, vegetation phenology, and height to calculate both latent and sensible heat surface fluxes. The model first calculates land surface roughness parameters, including roughness lengths for momentum and heat transfer using a method developed by Su et al. (2001). These roughness parameters are then applied to a set of flux-gradient equations along with temperature gradient and wind speed data to compute the sensible heat flux. The flux-gradient equations quantify the heat transfer between the land surface and the atmosphere. SEBS uses either the Monin–Obukhov similarity theory (MOST) or the bulk atmospheric similarity theory (BAST) equations (Brutsaert, 2005), based on the height of the atmospheric boundary layer. SEBS then determines the sensible heat flux under hypothetical extreme wet and dry conditions to calculate the evaporative fraction. Latent heat flux is estimated as a component of the available energy based on the calculated evaporative fraction. Further details regarding the SEBS model and its formulation can be found in the works of Su (2002a) and Ershadi et al. (2013b). A summarized version is presented in Appendix C at the end of this article.
The following is a brief review of the data that has been used in the analysis, as well as a description of the standardization process employed to allow comparison of these distinct datasets.
An eddy-covariance system provided measurement of heat fluxes and radiation
components for periods throughout the duration of the CRNP installation. The
system comprised of a Campbell Scientific 3-D sonic anemometer (CSAT-3,
Campbell Scientific, Logan, UT, USA) along with a Li-COR 7500 (Li-7500,
Li-COR Biosciences, Lincoln, NE, USA) for high-frequency water vapour and CO
Information required for the different evaporation models (see Sect. 2.2) were obtained using both the tower-based observations as well as the Moderate Resolution Imaging Spectroradiometer (MODIS) sensor on board NASA's Terra and Aqua satellites. Land surface temperature data required for the SEBS model were derived from the daily MOD11A1 and MYD11A1 products of the Terra and Aqua satellites (Wan, 2009). Normalized difference vegetation index (NDVI) data (used by all models) were obtained from the MOD13Q1 product (Solano et al., 2010).
The intermediate-scale soil moisture data used in this study were obtained
from the COSMOS repository
(
Volumetric soil moisture obtained via the CRNP is reported in units of
cm
Standardized soil moisture (CRNP and TDR), evaporation (modelled and observed), and precipitation signatures for the entire duration of record.
Given that modelled evaporation estimates are generally validated against
observations from eddy-covariance towers (Ershadi et al., 2014), an initial
step in our study was to query the relationship between the CRNP soil
moisture retrieval and the tower-based evaporation observations. Soil
moisture and land surface evaporation are both processes with inherent
stochasticity in their determination, due in part to the imprecise nature of
physical measurement, i.e. field-based observations of these processes are
inferred rather than measured in an absolute sense. In such situations, it is
often more appropriate to analyse the quantities for their similarity of
statistical distributions rather than deducing a lack (or presence) of a
relationship based on point-to-point statistics such as correlation.
Figure 2a shows a scatter of the raw (non-standardized) CRNP soil moisture
daily averages plotted against the corresponding 24 h daily evaporation
values observed at the eddy-covariance tower on site. The two quantities have
a Pearson's correlation coefficient close to 0.40. Figure 2b shows the
scatter of change in soil moisture on a daily scale vs. the corresponding
change in observed evaporation across a daily interval for the entire period
of record at the Baldry site. The plot is clustered around the zero-change
values, with a correlation coefficient of
Figure 3 shows the precipitation data at the study site, together with the standardized CRNP soil moisture and evaporation estimates from the PT-JPL, PM-Mu, and SEBS models. The average of the three TDR soil moisture measurements, and the evaporation observed from the in situ eddy-covariance tower, are also shown. As can be seen, the CRNP and TDR soil moisture series reflect similar responses in their variability and seasonality, with the CRNP data indicating greater fine-scale variability. While it might be expected that the TDR data should display greater variability in response, the CRNP measurements have higher variability in soil moisture values. This could be due to factors such as the variability in the measurement depth of the CRNP with change in the saturation, and higher sensitivity of the CRNP to near-surface moisture as compared to deeper layers (Bogena et al., 2013). High-frequency variations at the soil surface may also be attenuated in the TDR signal since it is integrated over the 30 cm probe depth.
Quantile–quantile (Q–Q) plots of standardized CRNP soil moisture vs. standardized evaporation estimates and observations (at EC tower) for the entire period of record.
With the aim of investigating the link between the CRNP soil moisture and the modelled evaporative response, non-parametric quantile–quantile (Q–Q) plots and box plots were used. Figure 4 shows the Q–Q plots for the standardized soil moisture data vs. the standardized model-derived evaporation estimates. The inter-quartile range is highlighted in grey and the red line denotes the extrapolation of the slope of the inter-quartile range. Good agreement between the Q–Q plot and the red expectation line indicates that the two quantities have been sampled from the same distribution. It can be seen that for all three models, the evaporation and soil moisture relationships follow the expectation (red line) closely, particularly in the inter-quartile range. This indicates that the two independent datasets (modelled evaporation and CRNP soil moisture) are sampled from distributions that are very similar to each other. However, beyond the inter-quartile range, a deviation of the plot away from the expectation line is seen in all cases. This is more apparent at the extremes. The SEBS estimates deviate more from the expectation on the higher side of the inter-quartile range, as compared to the other two models. However, this behaviour is similar to that of the measured evaporation from the eddy-covariance tower, as can be seen from the last plot in Fig. 4. Towards the higher end of the data range, the PT-JPL and PM-Mu values, while close to the expectation, are underestimated when compared to the tower measurements.
The above observation is additionally supported by the boxplots resulting
from the one-way ANOVA test, which are shown in
Fig. 5. In general, an ANOVA test generates a box plot along with a table of
statistics, the most important of which is typically the
ANOVA boxplots of standardized CRNP soil moisture, and standardized evaporation estimates and observations for the entire time series.
Q–Q plots of standardized soil moisture vs. standardized
evaporation for each defined period of analysis.
In order to get a better understanding of the observed dynamics between the soil moisture and evaporation signatures, we divided the data time series into four distinct sub-periods based on short-term trends and the level of correspondence between the soil moisture and evaporation records. Period 1 ran from day of record (DoR) 55 to 144. In the course of this period both the soil moisture and the modelled evaporation had a steady descending tendency (see Fig. 3). The soil moisture and evaporation signatures behaved consistent with each other in both increasing and decreasing tendencies during Period 2, running from DoR 321 to 410. Period 3, between DoR 410 and 500, covers a series of wetting and drying cycles of soil moisture, while no such corresponding changes were observed in the evaporation signatures. Period 4, covering the period between DoR 501 and 590, exhibits signatures that oppose each other in behaviour such that an increase in soil moisture corresponds to a decrease in the estimated evaporation. The four sub-periods were selected to examine the relationship between the CRNP soil moisture and the model-derived evaporation estimates under diverse hydrological conditions. Data from these sub-periods were individually standardized using the mean and standard deviation for each particular period.
Q–Q plots of CRNP soil moisture against the model-derived evaporation
estimates for each of the four sub-periods are shown in Fig. 6. The
corresponding Pearson's correlation value (
It is possible that under certain conditions of vegetation type and density, the CRNP measurement might include canopy-intercepted water along with the soil moisture. Canopy interception in densely vegetated surfaces could also be a significant contributor to the total evaporative flux. In such cases, the TDR measurements of soil moisture may better represent the root-zone wetness conditions. However, in the semi-arid grassland environment of the current study, canopy interception is unlikely to comprise a significant component of the terrestrial evaporation. In order to test if the TDR measurements were more representative of the root-zone soil moisture, we performed a Q–Q analysis using the average of the three TDR measurements at the study site instead of the CRNP measurements. We found that there was no significant difference in the plots, and thus these results are not reported here.
The Q–Q plots for sub-period 4 show that there is a marked incongruity between the distributions of soil moisture and the PT-JPL evaporation estimates. The plot deviates from the line representing the slope of the IQR, particularly within IQR where the bulk of the data would be expected to lie. This behaviour indicates that during this sub-period, the modelled evaporation was de-coupled from the soil moisture signature. The other three-source model (PM-Mu), also exhibits this de-coupling, although to a lesser extent. However, the de-coupling is not seen in the flux estimates of the SEBS model. The PT-JPL and PM-Mu estimates also show negative correlation with the soil moisture during this period, while the SEBS estimate is positively correlated. Period 4, corresponding with the Australian winter (May–July 2012), received frequent rainfall events (Fig. 3) that resulted in elevated soil moisture levels while the cloud cover probably limited the energy available for the evaporation process.
ANOVA boxplots of standardized CRNP soil moisture, and standardized evaporation estimates and observations for each defined period of analysis.
Q–Q plots of standardized soil moisture vs. standardized
evaporation for each season.
Comparisons between the ANOVA boxplots of the model-derived evaporation to the CRNP soil moisture and the observed flux from the eddy-covariance tower, for each period of the trend-based analysis, are shown in Fig. 7. During Period 4, the modelled evaporation distributions are significantly different from the eddy-covariance measured at the tower, especially for the PT-JPL and PM-Mu models. The observed evaporation has a distribution that is closer to that of the CRNP soil moisture, while the distributions of the PT-JPL and PM-Mu estimates are skewed to the left. The PT-JPL and PM-Mu methods are based primarily on the available energy (Rn-G) of the system, with soil moisture being implicitly accounted for by adjusting the air humidity. The Penman (1948) and Penman–Monteith (Monteith, 1965) combination equations that form the theoretical basis for the PT-JPL and PM-Mu models were developed for and tested (Rana and Katerji, 1998; Shahrokhnia and Sepaskhah, 2011; Sumner and Jacobs, 2005) in situations where energy limitations were not present. In contrast, the SEBS model follows a more physically based approach dependent on the turbulent mixing theory (Brutsaert, 2013), which is valid in energy-limited situations similar to those observed during Period 4. It is also likely that low temperatures and additional hydro-meteorological factors could have caused a de-coupling of the soil moisture from the air humidity. Due to the low temperatures, the air humidity would be lower, while the frequent precipitation ensures high soil moisture content. This creates a steep gradient for the moisture at the soil–air interface. In such a scenario, regardless of the presence of abundant soil moisture for evaporation, the models which use air humidity (or vapour pressure) as a surrogate for soil moisture may report lower estimates compared to those observed from the eddy-covariance tower. These factors may represent physical constraints on the application of the PT-JPL and PM-Mu methods and require further investigation.
To understand the seasonal patterns in the relationship between the CRNP soil
moisture and the evaporation data, the 2-year record of data was
partitioned according to seasons. The period between December and February
corresponds to the Austral summer, while autumn is from March to May, winter
from June to August, and spring from September to November. The summer and
spring seasons experienced the greatest number of precipitation events
(defined here as rainfall greater than 1 mm day
Corresponding Q–Q plots for the four seasons are shown in Fig. 8. In the autumn, the PM-Mu evaporation estimates correspond most closely with the CRNP soil moisture retrievals in the inter-quartile range (IQR), followed by the SEBS estimates. PT-JPL performed the poorest. Beyond the IQR, and overall, the SEBS estimates were the closest match to the soil moisture distribution in this season. In winter, the PT-JPL estimates were the closest to the soil moisture distribution within the IQR, while overall the SEBS model again performed best, relative to these specific metrics. In the spring, the SEBS model estimates were distributed most similarly to the soil moisture, both within the IQR and overall. PM-Mu estimates were the least similarly distributed.
The summer season shows that all three evaporation estimates depart from the
expectation within the IQR, with the SEBS estimates being least similar and
the PT-JPL estimates most similar to the soil moisture distribution. As
mentioned above, the site experienced 34 rainfall events out of a total of
164 days of record. However, there were also long periods with no rainfall
events. Combined with the higher temperatures of summer, this leads to
greater non-monotonic variations in the soil moisture signature, thus
creating a disconnect with the evaporation patterns. There are more switches
between moisture-constrained and energy-constrained conditions during this
season. It has been demonstrated previously that the occurrence of hot and
dry periods leads to de-coupling of soil moisture and evaporation (Pollacco
and Mohanty, 2012). The soil moisture profile in such situations becomes
heterogeneous in that the process driving the surface soil moisture
variability (mainly soil evaporation) no longer influences the deeper layer
soil moisture variability (mainly due to transpiration). Further explanation
of this de-coupling process can be found in Pollacco and Mohanty (2012).
Evaporation variability in summer is driven more by the precipitation
patterns than the soil moisture. With an abundance of energy, and severe
limitation of soil moisture, any influx of moisture due to precipitation is
quickly evaporated back to the atmosphere. Despite this, in an example of the
“correlation does not imply causation” maxim, it is observed that the
evaporation estimates for this season exhibit the highest correlation
(
From Fig. 8, it is also seen that in most cases the PT-JPL and PM-Mu models underestimate the evaporation at the higher end of the scale, at least when compared with the eddy-covariance tower measurements. The SEBS model generally performs better in this regard, as also seen in the previous analysis of the shorter time series (Fig. 5). Previous studies have shown that SEBS-based evaporation estimates were found to correspond well with tower-based measurements when there is a short, homogeneous canopy (McCabe et al., 2016) and the Baldry grassland site meets this criterion.
In this study, we exploit the physical mechanism that makes soil moisture a key driver of the evaporation process. As such, it makes perfect sense to evaluate models using observations that govern or influence that process to a significant extent. The evaporation models examined here do not make use of soil moisture as an input, and neither do the majority of the satellite-based evaporation models, in general. Hence, the evaporation and soil moisture datasets are statistically independent, although physically linked. Given the general lack of observation data concerning any specific process, it is important that independently observed, yet physically linked variables, be used to aid in the evaluation process, as demonstrated by the results of this study.
Correlation analyses are based on a one-to-one comparison between datasets.
Q–Q plots, on the other hand, are a measure of the similarity of
distribution. While there may be low point-to-point correlation between two
datasets, it is very much possible that the two quantities are sampled from
the same (or similar) distributions (Jana et al., 2008). Such correspondence
at the distribution level, rather than at the point level, is much more
meaningful for stochastic variables such as soil moisture and evaporation.
Hence, we emphasize the agreement in the Q–Q plots rather than the
Other potential causes of errors could be uncertainties in the observed data from the CRNP instrument and the eddy covariance, as well as cloud cover resulting in inaccurate MODIS observations, which could further lead to inconsistencies in the model outputs. Additionally, uncertainties in the meteorological forcings and other model inputs have not been explored in this study. The model structure and variations in model parameterization could also affect the analyses. We have used the structure and parameterization described by Ershadi et al. (2014) as they have been shown to correspond well with the tower observations. Other model parameterizations may improve (or degrade) the correspondence with the soil moisture, but that investigation is beyond the scope of this preliminary study.
This study shows that intermediate-resolution soil moisture can be used to validate and constrain models for land surface evaporation. Importantly, the soil moisture distribution can act as a guide for validating the model evaporation estimates in cases where eddy-covariance data are either unavailable or of poor quality. Further, considering that the footprint of the tower observations is at a much finer scale in comparison with the gridded model estimates of evaporation, it may be more prudent to evaluate evaporation models using the CRNP soil moisture, which is at a comparable resolution. Obviously, further analyses across different biomes and hydroclimatic regimes is necessary before a robust relationship between the model evaporation estimates and the CRNP soil moisture can be established. However, the outcome of this study encourages such an effort to be made. As shown in earlier studies to validate gridded evaporation products (Ershadi et al., 2014; McCabe et al., 2016) no single evaporation model consistently performed better than others across all conditions, whether that was seasonal or based on climate or biome type. This suggests that an ensemble modelling approach with model weights assigned according to, among other factors, their established relationship with soil moisture may be more suitable.
Relationships between soil moisture observations from a CRNP sensor and evaporation estimates derived from three distinct model structures using a combination of tower and satellite-based data were examined across a semi-arid grassland site. Standardized daily evaporation and CRNP soil moisture data were compared, with an analysis performed over different hydrological regimes, as well as an examination of seasonal-scale variations. As theorized, the two hydrological variables displayed significant correspondence with each other over the entire time series, indicating that there is a strong and hydrological consistent connection relating them. It was also established that a relationship exists between the intermediate-scale soil moisture measurements and the modelled evaporation estimates across most of the defined analysis periods. It was observed that the PT-JPL and PM-Mu model estimates behaved contrary to expectation in conditions where high soil moisture existed with colder temperatures. SEBS model estimates presented a similar disconnect from the soil moisture distribution, but in the summer season during long dry spells. These deviations are attributed to the model structures and reflect previous works identifying the geographic and temporal variability of model performance. Overall, no single model estimate of evaporation fully reproduced the CRNP soil moisture across all conditions. However, the outcome of this study indicates that the intermediate-scale soil moisture could be employed as a useful constraint to validate gridded evaporation estimates derived from models.
In the PT-JPL model (Fisher et al., 2008), total evapotranspiration is
partitioned into canopy transpiration (
In the PM-Mu model, total evaporation can be accounted as the sum of the
evaporation from the intercepted water in the wet canopy (
Evaporation of intercepted water from a wet canopy (
Transpiration from the canopy (
Evaporation from the soil (
The aerodynamic resistance at the soil surface (
The SEBS model includes routines for calculating the sensible heat flux
(
SEBS uses the stability-correction functions proposed by Beljaars and
Holtslag (1991) for stable conditions and the functions proposed by Brutsaert (2005)
are used for unstable conditions. The roughness length for momentum
and heat transfer (
SEBS uses a correcting method to scale the MOST-derived sensible heat flux
between hypothetical dry and wet limits based on the relative
evapotranspiration concept. Finally, this scaled sensible heat flux can be
used to calculate the evaporative fraction (
Research reported in this publication was supported by the King Abdullah University of Science and Technology (KAUST), Saudi Arabia. The CosmOz instrument was supported by the Commonwealth Scientific and Industry Research Organization (CSIRO). Instrumentation at the Baldry site was funded and commissioned as part of the Australian government's National Collaborative Research Infrastructure Strategy (NCRIS) and the University of New South Wales. Ali Ershadi was supported by the Australian Research Council Discovery Project (DP120104718).Edited by: H.-J. Hendricks Franssen Reviewed by: H. Bogena and one anonymous referee