1. Introduction
Soil moisture is a key variable influencing meteorological conditions such as near-surface air temperature and humidity, boundary layer mixing, clouds, and precipitation through its impact on the partitioning of net radiation into sensible and latent heat fluxes (Bonan 2008; Seneviratne et al. 2010; Berg et al. 2014). Soil moisture thus has an essential role in the energy and water budgets of continental surfaces. It also has a strong influence on crop growth, vegetation distribution, and on drought and flood risk (Stanhill 1957; Teuling et al. 2010).
However, there is still a limited understanding of soil moisture spatiotemporal distribution and variability at either regional or global scales because of the heterogeneity of several environmental factors, such as climate, soil texture, orography, and vegetation. For example, ground measurements offer us direct measurements of soil moisture per se, but they are at point scale, expensive, and sparse in time and space. Furthermore, land surface models (LSMs) with observation-based forcing are efficient tools to study soil moisture at regional or global scales, but their resolution is typically 10–50 km, the results are greatly dependent on forcing data quality, and there are large discrepancies among models (Seneviratne et al. 2010). In this context, the launch of the National Aeronautics and Space Administration’s (NASA) Soil Moisture Active Passive (SMAP) satellite in January 2015, operating at L band, will expectedly help improve global soil moisture estimation of weather and climate models and thus improve their predictive capability. However, SMAP satellite products will have a 9-km resolution, which may not be enough to capture fine spatial variabilities in soil moisture, and, although the coverage is global, no data will be available under dense vegetation cover.
High-resolution hydrology modeling has previously been used to capture soil moisture heterogeneity in order to contribute to ground network design (Chaney et al. 2015) and satellite calibration–validation (cal–val) efforts (Crow et al. 2005). In preparation for the validation of SMAP satellite products, the Surface Prediction System (SPS) from Environment Canada (EC) is run at hyperresolution (100 m) over Saskatchewan (Canada) and compared to data from one of the SMAP ground-measurement sites during the summer of 2014. Indeed, as part of the model-based validation of SMAP surface soil moisture products, SPS would be used in direct validations of SMAP products (at 3-, 9- and 36-km resolution) and of in situ SMAP sites, as well as in scaling sparse in situ point samples to SMAP grid cells. SPS would further be used in triple collocation (Scipal et al. 2008; Dorigo et al. 2010; Miralles et al. 2010), which allows the assessment of the degree to which root-mean-square differences between SMAP retrievals and sparse ground-based observations are inflated by spatial representativeness error in the ground-based observations. Thus, the purpose of this paper is to evaluate the performance of SPS at hyperresolution with respect to the evolution and the variability in time and space of soil moisture and soil temperature.
2. Model system, experimental design, and datasets
a. Surface Prediction System
The land surface modeling component of the SPS [formerly known as Global Environmental Multiscale Surface (GEM-SURF); Leroyer et al. 2011; Bernier et al. 2014; Rochoux et al. 2015] in this study is the Soil, Vegetation, and Snow (SVS) scheme, which is described in Alavi et al. (2015, manuscript submitted to J. Hydrometeor.). This scheme is developed and tested as a replacement to the Interactions between Soil, Biosphere, and Atmosphere (ISBA) scheme (Noilhan and Planton 1989) in the Meteorological Service of Canada’s (MSC) operational numerical environmental prediction applications. As compared to the Canadian implementation of ISBA (Bélair et al. 2003a,b; Carrera et al. 2010), SVS calculates the energy and water budgets for vegetation, two snowpacks, and ground underneath vegetation and snow with a new tiling approach. The scheme also has improved parameterization of the vegetation thermal coefficient and includes photosynthesis processes in order to evaluate the surface stomatal resistance. It includes new formulations for land surface albedo and emissivity and a new snowpack under vegetation. It also includes root density function depending on the vegetation type and multilayer water vertical transport in the soil and represents freeze–thaw states for each of its seven soil layers.
b. Experimental design
SPS is run in offline mode from 1 March to 30 September 2014 over Kenaston (Saskatchewan, Canada) at a 100-m resolution with a 36 km × 72 km domain (Fig. 1) and a 30-min time step. Forcing data, as described below, are given to SPS with an hourly time step, and thus they are interpolated to the model time step. To limit the impact of initial conditions on results, the analysis presented in this study focuses on the period from 1 June to 8 September 2014 (100 days).
(a) Topography (m) of the study domain along with the location of the 36 measuring stations from the Brightwater Creek network (black-filled circles), (b) the geographical location of the study domain, and (c) its primary land-cover type at each grid point.
Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0070.1
c. Atmospheric forcing
Atmospheric forcing data from EC’s operational numerical weather prediction (NWP) model, the 15-km version of the regional Global Environmental Multiscale (GEM) model (Mailhot et al. 2006), is used to drive the evolution of the land surface state simulated by SPS in external mode. Downwelling shortwave and longwave radiation and surface pressure are taken at the surface, while air temperature, specific humidity, and wind are taken from GEM’s lowest vertical level, which is roughly at 20 m for temperature and humidity and 40 m for the wind. As detailed in Bernier et al. (2014), the downscaling procedures of forcing data are embedded within SPS, including interpolation to the high-resolution surface of air temperature, surface pressure, and specific humidity.
With respect to precipitation, the forcing data come from two sources: the regional GEM model forecasts of hourly precipitation accumulation and the Canadian Precipitation Analysis (CaPA; Mahfouf et al. 2007). The 6-h CaPA product is disaggregated into hourly accumulation totals using the temporal structure provided by GEM hourly precipitation (Carrera et al. 2010).
d. Geophysical fields
Since SPS is run at hyperresolution (100 m), it is important to describe surface state with the greatest accuracy. Thus, several databases from different sources are used. For crops, which cover most of the domain, the fractional cover of each type during the summer of 2014 was obtained from Agriculture and Agri-Food Canada (AAFC) crop inventory (http://www.agr.gc.ca/atlas/data_donnees/agr/aafcCropTypeMapping/tif/, contains information licensed under the Open Government Licence–Canada). The seasonal cycle of the roughness length Z0 for each crop type was derived from crop height measurements taken throughout the growing season in 2014 over the Kenaston site (see section below). The seasonal cycle of the leaf area index (LAI) was obtained from the Canola Council of Canada for canola, from the United Kingdom’s Home Grown Cereals Authority (HGCA) for wheat, and from the Alberta Pulse Growers for lentils. Maximum rooting depth of each crop type, which remains constant throughout the simulation, comes from Cutforth et al. (2013). For other vegetation types, mainly needleleaf evergreen trees and grass, fractional cover comes from the European Space Agency’s (ESA) GlobCover (version 2.3; 2009) at 300-m resolution, and the LAI, rooting depth, and Z0 are obtained from lookup tables. Soil composition is extracted from the Soil Landscapes of Canada dataset (version 2.1) of Agriculture Canada, which has a 10-km resolution. Here, this dataset gives a homogeneous soil texture horizontally as well as vertically throughout the study domain. The orography of the region is obtained from the Canadian Digital Elevation Data (CDED50) geospatial dataset with a 20-m resolution over Canada.
e. Verification dataset
The Brightwater Creek soil moisture observation network (Champagne et al. 2010) was developed over southern Saskatchewan to support soil moisture validation activities for ESA’s Soil Moisture Ocean Salinity (SMOS) and NASA’s SMAP missions. The network consists of 36 observation stations situated within agricultural fields over a 1600 km2 region, each measuring soil moisture and soil temperature at depths of 5, 20, and 50 cm. Please note that the observation sites are mostly skewed to the western, lower-elevation portion of the model domain. The soil moisture monitoring network has been used in previous validation efforts for soil moisture products derived from passive microwave brightness retrievals (e.g., Champagne et al. 2010; Djamai et al. 2015).
3. Results and discussion
This section explores the performance of SPS when compared to observations from the in situ soil moisture network with respect to monthly averages, evolution in time, and temporal and spatial variability of both soil moisture and soil temperature. It should be noted that the simulated variables are taken as averages of 0–5-cm deep soil conditions, while the observations are taken at a 5-cm depth.
Monthly averages of soil moisture and soil temperature during the summer of 2014 are shown in Fig. 2. The spatial variability in both soil moisture and soil temperature is not the result of soil composition since it is uniform throughout the domain in this simulation, following the input data from the soil landscapes of Canada. However, both variables are sensitive to the type of crops and vegetation growing at the surface, underscoring the importance for accurately representing surface parameters to simulate the soil state, which has a large impact on surface–atmosphere interactions and fluxes. Color-filled circles (Fig. 2) represent the measurements at each observation station. Although the simulated soil moisture and soil temperature compare relatively well to observations, some differences can be seen, particularly with respect to soil moisture. These could be due to factors that are difficult to simulate, such as realistic soil texture and composition. In particular, observations of soil moisture in June 2014 show large spatial variability that is not captured adequately by the model. At the end of winter, the onset and rate of crop growth are strongly dependent on climate and planting date. However, in the model, all crops start to grow at a fixed date (1 May) and at a fixed rate, both defined based on previous studies and on field observations during 2014. This parameterization subdues spatial variability in soil moisture and soil temperature that arises from vegetation growth.
Monthly average of soil (a) moisture (m3 m−3) and (b) temperature (°C). The color-filled circles represent the observations on the same color-scale as the simulated data.
Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0070.1
Figure 3 shows soil moisture evolution in time as simulated by SPS and observed, as well as precipitation from CaPA. Figure 3a compares the simulated domainwide spatial average to the spatial average of observed data. In Fig. 3b, grid points corresponding to the observation stations in terms of location were selected and their spatial average is compared to the spatial average of observed data. Please note that the two observation stations that are outside the study domain (see Fig. 1) are not taken into account. This way, the comparison is performed with an equal number of simulated grid points and observed stations. Figure 3c is a station where both simulated soil moisture and soil temperature values are closest to observations, and Fig. 3d is a station where simulated values are farthest from observations.
Simulated soil moisture (m3 m−3; red) evolution compared to observations (m3 m−3; black) and CaPA (cm day−1; blue) during the summer of 2014. (a) Comparison of simulated domainwide average with observations and (b) comparison of the spatial average of grid points corresponding to stations with observations. (c),(d) Stations where SPS-simulated values are closest and farthest from observations in terms of bias, respectively. The MAE, BIAS, STDE, and CORR between simulated and observed data are indicated.
Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0070.1
On average, simulated soil moisture is very close to observations, as shown by the mean absolute error (MAE) and the mean bias (BIAS) in Fig. 3. It is interesting to see that, in this case, the simulated domainwide average is closer to observations than is the simulated stations-only average, with an MAE of 0.026 m3 m−3 and a standard deviation of the error (STDE) of 0.030 m3 m−3. It is important to note that the crop fractional cover used in this study is from the AAFC crop inventory with target accuracies of 85%. When compared to field observations from 2014 (data not shown), it shows that the AAFC crop inventory is 76% accurate over the domain. Thus, this source of error has to be taken into account since differences in crop type can lead to differences in soil moisture as shown in Fig. 2.
SPS responds well to precipitation events (Fig. 3), whether it be spatial averages or single grid points when compared to observations. However, the model seems to have difficulties in periods of dry-down, such as the first half of June 2014, during which the undergrown crops allow too much evaporation, and the first half of July 2014, during which crops are mature and the soil retains too much soil moisture. This suggests that SPS-simulated evaporation from top soil layers is greatly affected by vegetation cover. At the station represented in Fig. 3c, the simulated soil moisture follows the observed data closely, although it tends to underestimate it slightly. On the contrary, at the station in Fig. 3d, SPS overestimates soil moisture in a constant manner throughout the summer. Since both the simulated and observed crop type is canola, we can assume that such large differences do not come from vegetation but rather from soil composition. For example, a larger proportion of sand in the soil would reduce water retention compared to simulated results.
Figure 4 is set up similarly to Fig. 3 but compares simulated soil temperature to observations. As seen in Figs. 4a and 4b, SPS does very well when looking at evolution in time compared to observations. Nonetheless, most of the time in Figs. 4a and 4b, the magnitude of temperature extremes in SPS is overestimated, particularly the minima, which suggests that soil conductivity could be overestimated, thus inducing excessive heat gain and loss. In this case, the simulated stations-only average is closest to observations. Figure 4c is very similar to Figs. 4a and 4b, but Fig. 4d shows an underestimation of soil temperature from the beginning of July onward. Since the simulated soil water content is greater than observed at this station (Fig. 3d), the difference in temperature could be related to an increased evaporation rate, thus leading to a greater cooling of the surface in the simulated data.
As in Fig. 3, but for soil temperature (°C).
Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0070.1
To evaluate the performance of SPS with respect to simulated variability of soil moisture and soil temperature compared to observations, Figs. 5a and 5b show the spatial standard deviation evolution in time, while Figs. 5c and 5d show the probability density functions, for both simulated stations-only variables compared to observations. As indicated in Figs. 5c and 5d, on average over time and space, SPS simulates soil temperature 0.51°C lower and soil moisture 0.008 m3 m−3 higher than observations. SPS simulates a much weaker variability in space (Figs. 5a,b) throughout the summer of 2014 but a slightly higher variability in time (Figs. 5c,d) in both soil moisture and soil temperature. Although the simulated time variability is relatively good, the much weaker spatial variability in SPS could be the result of uniform soil composition, lack of small-scale orographic features, and predefined growth cycle for each crop type. Moreover, the weak variability in soil temperature is mostly the result of moisture variability. However, it is important to note that, in some sense, a network site cannot represent the variance of a grid if one considers the scale of support. Indeed, where the sensor is measuring an area of a few 100 cm3, the model is averaging an area of 100 m × 100 m. It would take hundreds of sensor observations to match this scale and the variance would decrease because of the larger number of samples.
Soil moisture and soil temperature (a),(b) spatial std dev evolution in time and (c),(d) probability density function of simulated (red) and observed (black) data, with indications of spatial and time averages (mean) and spatial mean std dev in (c) and (d). The observed data are compared to simulated data from grid points corresponding to observation stations in terms of location.
Citation: Journal of Hydrometeorology 17, 1; 10.1175/JHM-D-15-0070.1
4. Conclusions
This study shows that with quality initial conditions and surface condition forcings based on observations, EC’s Surface Prediction System (SPS) can simulate soil moisture and soil temperature evolution in time with high accuracy (see stats in Figs. 3, 4). However, our simulation was performed over a small domain, thus limiting the weight of our conclusions, and with a uniform soil texture. Improvements could be brought upon the initialization of soil texture since this variable greatly influences soil capacity to retain water and heat. We further show that this factor, as well as the predefined crop growth cycle and probably small-scale orographic features, limits the simulated spatial variability in soil conditions. As a result, although the variability in time of both soil moisture and soil temperature is well simulated, the spatial variability is underestimated when compared to observations, even with SPS running at hyperresolution.
Thus, although our study provides a good foundation for future cal–val efforts in the context of SMAP, SPS in its present state is not able to adequately simulate spatial variability in soil moisture, which undermines the utility of the model at the moment. Nonetheless, both simulated domainwide and stations-only spatial averages of soil conditions compare very well with observations over time. In turn, this shows that observed data at stations are able to represent the entire domain through spatial averages, which suggests that both in situ observations and simulated data from SPS can be used in future comparisons with, and validation of, SMAP satellite data. Furthermore, undergoing research shows that SPS-simulated soil moisture spatial variability can be increased by applying high-resolution observed soil texture to the model, and other tests will be performed with increased forcing data resolution (i.e., precipitation and radiation).
Once a new version of SPS gives satisfactory results with respect to spatial variability, the model will be used at hyperresolution in triple-collocation calculations to compare coarse-scale satellite soil moisture retrievals and as a tool to link finescale in situ data to a satellite-scale mean. SPS will also be used outside the scope of SMAP activities as a mean to downscale soil moisture analyses for atmospheric models to scales of a few hundred meters.
Acknowledgments
The authors thank Maria Abrahamowicz and Dorothy Durnford for their technical assistance, and Erika Tetlock for her assistance with the Brightwater Creek network data. This work was funded by the Canadian Space Agency through the Government Related Initiative Program (GRIP).
REFERENCES
Bélair, S., Crevier L.-P. , Mailhot J. , Bilodeau B. , and Delage Y. , 2003a: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part I: Warm season results. J. Hydrometeor., 4, 352–370, doi:10.1175/1525-7541(2003)4<352:OIOTIL>2.0.CO;2.
Bélair, S., Brown R. , Mailhot J. , Bilodeau B. , and Crevier L.-P. , 2003b: Operational implementation of the ISBA land surface scheme in the Canadian regional weather forecast model. Part II: Cold season results. J. Hydrometeor., 4, 371–386, doi:10.1175/1525-7541(2003)4<371:OIOTIL>2.0.CO;2.
Berg, A., Lintner B. R. , Findell K. L. , Malyshev S. , Loikith P. C. , and Gentine P. , 2014: Impact of soil moisture–atmosphere interactions on surface temperature distribution. J. Climate, 27, 7976–7993, doi:10.1175/JCLI-D-13-00591.1.
Bernier, N., Bélair S. , Bilodeau B. , and Tong L. , 2014: Assimilation and high resolution forecasts of surface and near surface conditions for the 2010 Vancouver winter Olympic and Paralympic Games. Pure Appl. Geophys., 171, 243–256, doi:10.1007/s00024-012-0542-0.
Bonan, G., 2008: Ecological Climatology: Concepts and Applications. 2nd ed., Cambridge University Press, 563 pp.
Carrera, M., Bélair S. , Fortin V. , Bilodeau B. , Charpentier D. , and Doré I. , 2010: Evaluation of snowpack simulations over the Canadian Rockies with an experimental hydrometeorological modeling system. J. Hydrometeor., 11, 1123–1140, doi:10.1175/2010JHM1274.1.
Champagne, C., Berg A. , Belanger J. , McNairn H. , and Jeu R. D. , 2010: Evaluation of soil moisture derived from passive microwave remote sensing over agricultural sites in Canada using ground-based soil moisture monitoring networks. Int. J. Remote Sens., 31, 3669–3690, doi:10.1080/01431161.2010.483485.
Chaney, N., Roundy J. , Herrera-Estrada J. , and Wood E. , 2015: High-resolution modeling of the spatial heterogeneity of soil moisture: Applications in network design. Water Resour. Res., 51, 619–638, doi:10.1002/2013WR014964.
Crow, W., Ryu D. , and Famiglietti J. , 2005: Upscaling of field-scale soil moisture measurements using distributed land surface modeling. Adv. Water Resour., 28, 1–14, doi:10.1016/j.advwatres.2004.10.004.
Cutforth, H. W., and Coauthors, 2013: Comparing rooting characteristics and soil water withdrawal patterns of wheat with alternative oilseed and pulse crops grown in the semiarid Canadian Prairie. Can. J. Soil Sci., 93, 147–160, doi:10.4141/cjss2012-081.
Djamai, N., Magagi R. , Goita K. , Hosseini M. , Cosh M. , Berg A. , and Toth B. , 2015: Evaluation of SMOS soil moisture products over the CanEx-SM10 area. J. Hydrol., 520, 254–267, doi:10.1016/j.jhydrol.2014.11.026.
Dorigo, W., Scipal K. , Parinussa R. , Liu Y. , Wagner W. , de Jeu R. , and Naeimi V. , 2010: Error characterisation of global active and passive microwave soil moisture datasets. Hydrol. Earth Syst. Sci., 14, 2605–2616, doi:10.5194/hess-14-2605-2010.
Leroyer, S., Bélair S. , Mailhot J. , and Strachan I. B. , 2011: Microscale numerical prediction over Montreal with the Canadian external urban modeling system. J. Appl. Meteor. Climatol., 50, 2410–2428, doi:10.1175/JAMC-D-11-013.1.
Mahfouf, J.-F., Brasnett B. , and Gagnon S. , 2007: A Canadian Precipitation Analysis (CaPA) project: Description and preliminary results. Atmos.–Ocean, 45, 1–17, doi:10.3137/ao.450101.
Mailhot, J., and Coauthors, 2006: The 15-km version of the Canadian regional forecast system. Atmos.–Ocean, 44, 133–149, doi:10.3137/ao.440202.
Miralles, D., Crow W. , and Cosh M. , 2010: Estimating spatial sampling errors in coarse-scale soil moisture estimates derived from point-scale observations. J. Hydrometeor., 11, 1423–1429, doi:10.1175/2010JHM1285.1.
Noilhan, J., and Planton S. , 1989: A simple parameterization of land surface processes for meteorological models. Mon. Wea. Rev., 117, 536–549, doi:10.1175/1520-0493(1989)117<0536:ASPOLS>2.0.CO;2.
Rochoux, M., Bélair S. , Abrahamowicz M. , and Pellerin P. , 2015: Subgrid-scale variability for thermodynamic variables in an offline land surface prediction system. J. Hydrometeor., doi:10.1175/JHM-D-15-0016.1, in press.
Scipal, K., Holmes T. , de Jeu R. , Naeimi V. , and Wagner W. , 2008: A possible solution for the problem of estimating the error structure of global soil moisture data sets. Geophys. Res. Lett., 35, L24403, doi:10.1029/2008GL035599.
Seneviratne, S. I., Corti T. , Davin E. L. , Hirschi M. , Lehner E. B. J. I. , Orlowsky B. , and Teuling A. J. , 2010: Investigating soil moisture–climate interactions in a changing climate: A review. Earth Sci. Rev., 99, 125–161, doi:10.1016/j.earscirev.2010.02.004.
Stanhill, G., 1957: The effect of differences in soil-moisture status on plant growth: A review and analysis of soil-moisture experiments. Soil Sci., 84, 205–214, doi:10.1097/00010694-195709000-00003.
Teuling, A., and Coauthors, 2010: Contrasting response of European forest and grassland energy exchange to heatwaves. Nat. Geosci., 3, 722–727, doi:10.1038/ngeo950.
