SEARCH

SEARCH BY CITATION

Keywords:

  • catchment hydrology;
  • distributed hydrologic model;
  • downscaling;
  • remote sensing;
  • soil moisture

Abstract

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations and Modeling
  5. 3. Soil Moisture Downscaling
  6. 4. Hydrologic Evaluation and Data Assimilation
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[1] Demonstrating the utility of satellite-based soil moisture (θ) products for hydrologic modeling at high resolution is a critical component of mission design. In this study, we utilize aircraft and ground θdata collected during the SMEX04 experiment in Sonora (Mexico) to compare two downscaling frameworks using C-band and L-band sensors. We show that the L-band framework, which mimics the disaggregation of SMAP products, has considerably better performance than the C-band framework simulating the downscaling of AMSR-E. Disaggregated L-bandθ fields are able to characterize with reasonable accuracy the θvariability at multiple extent scales, including the SMAP footprint and the catchment scale, and along an elevation transect. We then test the utility of coarse and downscaled C- and L-bandθestimates for hydrologic simulations through data assimilation experiments using a distributed hydrologic model. Results reveal that the model prognostic capability is significantly enhanced when using L-bandθfields at the SMAP scale and, to a greater extent, when downscaled L-bandθfields are assimilated. L-band data assimilation leads to higher model fidelity relative to ground data as well as more realistic soil moisture patterns at the catchment scale. This study indicates the potential value of satellite-based L-band sensors for hydrologic modeling when coupled with a statistical downscaling algorithm.

1. Introduction

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations and Modeling
  5. 3. Soil Moisture Downscaling
  6. 4. Hydrologic Evaluation and Data Assimilation
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[2] The availability of soil moisture (θ) is fundamental for enhancing predictions of climate [Ni-Meister et al., 2006], weather [Scipal et al., 2008] and hydrologic [Bindlish et al., 2009] models. Satellite-based microwave sensors have the potential to provideθglobally in the top soil layer. Existing sensors, such as the Advanced Microwave Scanning Radiometer (AMSR-E) [Njoku et al., 2003] at C-band (7.32 GHz), however, have limited accuracy as vegetation increases [Schmugge, 1998]. As a result, L-band (1.4 GHz) sensors have been selected for the Soil Moisture and Ocean Salinity [Kerr et al., 2001] and the forthcoming Soil Moisture Active and Passive (SMAP) [Das et al., 2011] missions. The use of L-band sensors should produce more accurateθ estimates, as observations at these wavelengths are less sensitive to vegetation, soil roughness and radio interferences [Calvet et al., 2011].

[3] Intensive field campaigns are used to test the ability of satellite sensors to observe soil moisture through comparisons with ground and aircraft θ measurements [e.g., Bindlish et al., 2008]. Field campaign data have also been utilized in the development of data assimilation schemes [e.g., Houser et al., 1998] and to characterize conditions prior to flood predictions [Brocca et al., 2010]. In addition, the fine resolution of ground and aircraft θ measurements have been instrumental in the development of downscaling models that characterize the spatial variability of soil moisture within a coarse satellite pixel using statistical [e.g., Kim and Barros, 2002; Mascaro et al., 2010] or deterministic methods [e.g., Merlin et al., 2010].

[4] Here, we demonstrate the value of coarse and downscaled θproducts from L-band sensors, relative to products derived at C-band, for hydrologic modeling at the catchment scale in a semiarid region. We utilize ground and aircraftθdata collected in Sonora, Mexico, during the SMEX04 campaign in an area with complex topography and moderate vegetation cover. First, we compare the performance of a soil moisture downscaling algorithm applied to the disaggregation of coarse AMSR-E (C-band) and SMAP (L-band)θestimates. We then evaluate the utility of C- and L-bandθ products at coarse and fine (downscaled) resolutions within the SMEX04 region, a watershed and along a topographic transect. Finally, using a set of data assimilation experiments, we show the utility of coarse and downscaled θ products for enhancing the prognostic ability and spatial representation of soil moisture in a catchment.

2. Observations and Modeling

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations and Modeling
  5. 3. Soil Moisture Downscaling
  6. 4. Hydrologic Evaluation and Data Assimilation
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[5] The study region is a semiarid area in Sonora (Figure 1a) characterized by complex terrain and vegetation that greens during the North American monsoon [Vivoni et al., 2007]. Land cover includes (from low to high elevation) desert scrub, subtropical scrubland and oak savanna, with a few agricultural sectors and rural towns. SMEX04 datasets consist of ground data collected at 30 sites during 3–14 August, 2004, across a transect spanning a range of soil and vegetation types [Vivoni et al., 2007, 2008]. The transect is almost entirely located (23 out of 30 sites) in the Sierra Los Locos (SLL) basin (92.5 km2), where ground, aircraft and numerical modeling estimates of θ have been used by Vivoni et al. [2010] to quantify the spatiotemporal evolution of soil moisture and the controls exerted by landscape conditions.

image

Figure 1. (a) Soil moisture field in the 75 × 50 km2Soil Moisture Experiment 2004 (SMEX04) area on 25 August from 2-D Synthetic Aperture Radiometer (2D-STAR, L-AIR-θ). The boundary of the Sierra Los Locos (SLL) basin is shown. (b) Polarimetric Scanning Radiometer (PSR) θfield (C-AIR-θ) on 25 August over the coarse-scale domainL × L (with L= 25.6 km) used in the C-band framework to mimic the AMSR-E footprint (dashed black square). (c) L-AIR-θon 25 August over two coarse-scale domainsL × L (L= 12.8 km) used in the L-band framework to mimic SMAP footprint (gray squares). The SLL boundary, river network and ground sites are shown.

Download figure to PowerPoint

[6] During SMEX04, two aircraft sensors collected brightness temperature over a 75 by 50 km2 area (Figure 1a), which was converted into θmaps at 800-m resolution. The sensors include: (a) the C-band Polarimetric Scanning Radiometer (PSR), which acquired images during 10 days from 5 to 26 August [Bindlish et al., 2008], and (b) the L-band 2-D Synthetic Aperture Radiometer (2D-STAR), with data observed on 5 days within the same period [Ryu et al., 2010]. Figure 1ashows an example of a 2D-STAR field on 25 August.Ryu et al. [2010]demonstrated that 2D-STARθ data captured ground observations with higher accuracy than the PSR θ estimates. As an example, the θestimates in an area over the SLL basin retrieved on 25 August from PSR and 2D-STAR sensors are compared inFigures 1b and 1c. Aircraft data were used to calibrate a downscaling algorithm [Mascaro et al., 2010, 2011] able to reproduce the multifractal properties of soil moisture fields across a range of climate settings. Here, the algorithm was calibrated using two approaches: (a) the C-band framework that mimics the disaggregation of coarse AMSR-E (∼25 km) products using PSR data at 800-m resolution, and (b) the L-band framework that simulates downscaling of coarse SMAP (∼10 km) products based on 2D-STAR data.

[7] The two frameworks are used to carry out soil moisture data assimilation experiments with the Triangulated Irregular Network (TIN)-based Real-time Integrated Basin Simulator (tRIBS) model [Ivanov et al., 2004]. tRIBS is a fully-distributed model where infiltration is simulated using a modified Green-Ampt approach in a sloped, heterogeneous soil column. A range of possible soil moisture states is obtained due to the interaction of the infiltration front with antecedent conditions and the water table position, and by lateral water movement in the saturated and unsaturated zones.Vivoni et al. [2010] applied tRIBS at an eddy covariance tower and in the SLL basin, revealing good agreement with the tower water and energy fluxes and ground θ observations in the basin. Here, we utilize the simulation conditions of Vivoni et al. [2010] from 23 July to 30 September to conduct a set of data assimilation experiments with coarse and downscaled soil moisture products. Table 1 summarizes the characteristics of the θ data and experiments used in the study, including the name conventions adopted.

Table 1. Summary of Soil Moisture Datasets and Their Properties
SourceData AvailabilitySpatial CoverageUsageResolutionName
PSR5, 8–10, 12–14, 24–26 August75 × 50 km2Original aircraft800 mC-AIR-θ
Aggregated at AMSR-E25.6 kmC-SAT-θ
Downscaled from AMSR-E800 mC-DIS-θ
2D-STAR7, 8, 24–26 August75 × 50 km2Original aircraft800 mL-AIR-θ
Aggregated at SMAP12.8 kmL-SAT-θ
Downscaled from SMAP800 mL-DIS-θ
Ground3–13, AugustTransect with 23 sitesAverage of 5 measures in a 1 m × 1 m plot at each locationPointGR-θ
tRIBS23 July to 30 SeptemberSLL basinNo assimilationVariable, 7 m to 230 mSIM-θ
Assimilation of C-SAT-θSIM-C-SAT-θ
Assimilation of C-DIS-θSIM-C-DIS-θ
Assimilation of L-SAT-θSIM-L-SAT-θ
Assimilation of L-DIS-θSIM-L-DIS-θ

3. Soil Moisture Downscaling

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations and Modeling
  5. 3. Soil Moisture Downscaling
  6. 4. Hydrologic Evaluation and Data Assimilation
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[8] The downscaling model reproduces observed multifractal properties by means of a log-Poisson generator based on two parameters,c and β, which are estimated through scale-invariance and multifractal analysis between a fine (l) and a coarse (L) scale [Deidda, 2000]. The model is applied by determining calibration relations that link the parameters to a few predictors computed in the coarse domain (L × L). Hence, starting from coarse-scaleθdata, an ensemble of fields reproducing the probability distribution of the observed small-scaleθ field is generated by stochastic multifractal cascades. Mascaro et al. [2010, 2011] applied the model from L = 25.6 km to l = 0.8 km by selecting coarse domains in multiple campaign datasets (SGP97, SGP99, SMEX02, SMEX04). βwas found to be a constant, region-specific value, andc was linked with the spatial mean soil moisture 〈θ〉 as follows for SMEX04:

  • display math

with parameters c, a and γ. In a calibration approach named ancillary (ANC), the parameters were linked to landscapes factors (terrain, soil, vegetation) affecting soil moisture dynamics for different 〈θ〉. Additional details can be found in Mascaro et al. [2010, 2011].

[9] Here, in the C-band framework, we used the ANC calibration relation estimated byMascaro et al. [2011]in SMEX04 using C-AIR-θ (Table 1). As an example, the boundaries of a domain with L = 25.6 km are shown in Figure 1b (domain A), while the corresponding calibration line (1) is shown in Figure 2a with the cvalues estimated on the C-AIR-θ fields in the domain. A constant β= 0.71 was found to be appropriate for this framework. We also tested the use of operational AMSR-E products for 〈θ〉. However, comparison between satellite estimates and aggregated PSR values revealed poor agreement (not shown), likely due to the degraded retrieval performance in this region with moderate vegetation [Jackson et al., 2009]. As a result and for consistency with the L-band framework, the coarse mean soil moisture 〈θ〉 (C-SAT-θ) was obtained by aggregating the 800-m C-AIR-θ up to 25.6 km.

image

Figure 2. (a) ANC calibration relation in the C-band framework andc estimates from observed θfields for domain A (squares). (b) REG calibration relation in the L-band framework andc estimates from observed θfields of domains B and C (circles and asterisks). (c) ECDFs of C-AIR-θon 25 August, compared against the 90% confidence intervals derived by 50 C-DIS-θfields (dashed gray lines). Inset is a scatterplot between the STD of C-AIR-θversus the ensemble mean of the STDs of 50 C-DIS-θfields for all domains. (d) Same as Figure 2c, but for L-AIR-θand L-DIS-θ in domain B with similar inset.

Download figure to PowerPoint

[10] In the L-band framework, we calibrated the model using L-AIR-θ in the range from L = 12.8 km to l = 0.8 km. For this aim, we selected 65 coarse domains with a criterion similar to Mascaro et al. [2011] and estimated c and β in each domain. As an example, two domains, B and C, are shown in Figure 1c. As in the first framework, we found β to assume a constant value of 0.71 and a relation between c and 〈θ〉 described by (1). However, we found similar values of c, a and γacross all domains, suggesting that a single, regional parameter set can be assumed. A number of reasons are possible for the low variability among domains in this framework, including: (a) a shorter range of scale invariance (0.8 to 12.8 km); (b) different features of 2D-STAR, and (c) a lower number of available images. The calibration relation (labeled regional, REG) is shown inFigure 2c with the c estimates in domains B and C.

[11] The performance of C- and L-band frameworks is illustrated inFigures 2c and 2d, where the Empirical Cumulative Density Function (ECDF) of the 800-mθfield observed on 25 August is compared against the 90% confidence intervals from an ensemble of 50 downscaled fields. Inspection of the ECDFs in all domains revealed that the downscaling algorithm in the L-band framework captured better theθ distributions within the satellite footprint, especially in the left tail. In fact, in the retrieval algorithm of PSR, θ tends to be artificially set to a lower bound of 2% in many pixels (arrow in Figure 2c), thus affecting the multifractal analysis and, in turn, degrading the downscaling performance [Mascaro and Vivoni, 2012]. This effect is not present in 2D-STAR (Figure 2d), likely due to the higher sensitivity of the L-band sensor to soil moisture variations, even in dry conditions. Despite this difference, both frameworks capture well the small-scale standard deviation (STD), as shown inFigures 2c and 2d(insets), with root mean square errors (RMSE) of 0.67% and 0.39% for the C- and L-band frameworks, respectively.

4. Hydrologic Evaluation and Data Assimilation

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations and Modeling
  5. 3. Soil Moisture Downscaling
  6. 4. Hydrologic Evaluation and Data Assimilation
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

4.1. Variability Within the Watershed and Along a Topographic Transect

[12] Vivoni et al. [2008, 2010]compared the variability of GR-θ, C-AIR-θand SIM-θin the SLL basin. Here, we extend this analysis with L-AIR-θ, C-DIS-θand L-DIS-θ. Figure 3a reports the relation between the spatial standard deviation (σθ) of θ and the mean soil water content (μθ) in the SLL basin for all products for each day of observation. The downscaled results are averages of an ensemble of 50 synthetic fields. σθ increases with μθ in a fairly linear fashion for all products, consistent with studies in other semiarid areas [e.g., Martinez-Fernandez and Ceballos, 2003]. The robustness of the parameters of the regression line estimated with all products (dashed line) was assessed through two sets of bootstrap tests, summarized in the caption of Figure 3. Note that the downscaled data are able to capture well the pattern of the corresponding aircraft products, except C-DIS-θ that slightly overestimates σθof C-AIR-θ. This comparison shows the capability of the downscaling model to capture the θ variability over a wide range of extent scales, including the size of a medium catchment (∼100 km2). If the products are compared directly in concomitant days (not shown), the variability of μθ and σθcomputed for C-AIR-θand C-DIS-θis always much wider than that of ground and simulated data, since C-band retrievals overestimate soil moisture in wetter days [Vivoni et al., 2008, 2010]. In contrast, L-AIR-θand L-DIS-θexhibit patterns very close to those of GR-θand SIM-θ, implying that L-bandθproducts, observed from aircraft or disaggregated from the SMAP footprint, are able to capture the soil moisture mean and variability at the catchment scale with much higher accuracy than estimates derived from C-band products.

image

Figure 3. (a) Relation between the spatial standard deviation σθ and mean soil moisture μθ in the SLL basin for each day of observation of available products (Table 1). The dashed line is the linear regression σθ = mμθ + q line for all points, with expected value and STD (parentheses) of m and q of 0.297 (0.017) and 0.689 (0.163), respectively. Two bootstrapping techniques provide similar values: (i) 0.299 (0.013) and 0.676 (0.082) with a total of 6 samples, each obtained by eliminating the (μθ, σθ) points of one soil moisture product; and (iii) 0.297 (0.013) and 0.687 (0.097) with a total of 1000 samples, each created by randomly sampling with replacement of 70% of the (μθ, σθ) points. (b)–(c) Comparison of soil moisture products as a function of elevation for (b) 8 and (c) 24 August. GR-θare plotted as error bars with average and standard deviation of 5 measurements per plot. C-AIR-θand L-AIR-θare values in colocated pixels with ground data. SIM-θ are shown as average of θin colocated and neighbor polygons (continuous line) with ±STD (dashed lines). Horizontal lines are the mean ± STD of 50 values provided by C-DIS-θand L-DIS-θ in the colocated pixels.

Download figure to PowerPoint

[13] To extend the spatial analysis, the products are compared in Figures 3b and 3c as a function of transect elevation and wetness condition (wet day of 8 August and dry day of 24 August). In the case of the disaggregated θ, a deterministic comparison is meaningless, as the multifractal model produces spatially homogeneous fields, so that the ensemble statistics do not vary with location and elevation within a coarse pixel. Hence, the comparison was made by plotting, for each downscaled product, two horizontal lines representing the mean ± STD of 50 synthetic fields. In the wet day, C-AIR-θslightly decreases from low to high elevation, with large discrepancies with other products. Patterns of L-AIR-θ, SIM-θand GR-θ are consistent with each other. The lines of downscaled θdata envelop the corresponding aircraft data, with mean ± STD equal to 15.0 ± 6.4% for C-DIS-θ, and to 7.4 ± 3.5% and 9.9 ± 1.8% for L-DIS-θat elevations lower and higher than 1100 m, respectively (due to the two coarse pixels covering the SLL basin for 2D-STAR). Note that L-DIS-θ are in reasonable agreement with ground and model data. In the dry period, all products lack significant variations with elevation, except for the model that has slightly higher θat lower sites. The C-AIR-θare always close to the lower bound of 2%, while L-AIR-θ show some variability between low and high elevation. For the dry day, the ensemble mean of the downscaled θ fields is 3.3% in both frameworks, with a STD of 1.7% (shown as single set of horizontal lines in Figure 3c). Overall, this confirms the higher accuracy of the L-band framework, as disaggregatedθestimates compare better than C-band against ground data and simulations from a distributed model along a transect spanning a wide range of elevation, vegetation and soil conditions.

4.2. Utility of Coarse and Downscaled Soil Moisture for Hydrologic Modeling

[14] Assimilation of remotely-sensed soil moisture data into distributed hydrologic models can improve their capability to simulate land surface fluxes and runoff generation [e.g.,Crow and Ryu, 2009]. We carried out experiments by ingesting coarse and downscaled products from each framework on both the wet day (8 August) and dry day (24 August). A total of 103 simulations were conducted with identical model parameterizations to Vivoni et al. [2010]: no assimilation (SIM-θ), coarse assimilation (SIM-C-SAT-θ, SIM-L-SAT-θ) and downscaled assimilation (50 downscaled fields for each framework as SIM-C-DIS-θand SIM-L-DIS-θ). For the coarse products, direct insertion was used in the model to assimilate soil moisture in the top 0-5 cm layer [Houser et al., 1998]. This approach assumes that the ingested observation is correct, thus leading to a modification of the simulated spatial pattern. For the downscaled products, a statistical correction assimilation method was applied such that simulated θ values in each model element are varied, according to the criteria proposed by Houser et al. [1998], to match the μθ and σθ provided by the downscaled field. This scheme assumes the statistics of the observation to be correct and preserves the simulated spatial pattern. The modification of the 0–5 cm θrequired in both schemes was accomplished in tRIBS by updating the states of the modified Green-Ampt infiltration model [Ivanov et al., 2004], including the position of top and wetting fronts, and the water table, while preserving the overall water balance in each model element. Future work will be devoted to refine the assimilation schemes, including the development of a method where θ in each element is modified to match the entire statistical distribution of the disaggregated field.

[15] Time series of the basin-averaged (μθ) soil moisture within the SLL are presented in Figures 4a and 4bfrom the assimilation of coarse and downscaled products at C- and L-band, respectively. These are compared against basin-averaged SIM-θand GR-θ, whose RMSE is 1.2%. Note that μθfor SIM-C-SAT-θconsiderably overestimates GR-θand SIM-θ during the wet day, while it sharply decreases to ∼2% by the dry day (Figure 4a). For coarse C-band, the RMSE with GR-θrises to 2.3%. Performance improves when assimilating downscaled C-band data, with skill increasing in time, as shown by the approach of SIM-C-DIS-θto SIM-θand a decrease in ensemble spread. Results indicate that when C-bandθproducts are ingested on the wet day, an artificially high water content is introduced, leading to unrealistic fluxes during the dry down. In contrast, the assimilation of coarse L-band data enhances the model prognostic capability as compared to the original model (SIM-θ). The μθfor SIM-L-SAT-θcaptures better GR-θafter the wet day, and no sharp change is introduced on the dry day, due to the higher sensitivity of the L-band sensor in dry conditions. As a result, the RMSE with GR-θis lowered to 1.0%. While the ensemble spread of SIM-L-DIS-θis much narrower than SIM-C-DIS-θ, enhancements in capturing μθare more limited when ingesting downscaled L-band data as compared to C-band. Overall, assimilating accurate coarse and downscaled L-bandθ products leads to a more reliable simulation of the states and fluxes in the basin.

image

Figure 4. Time series of simulated basin-averaged soil moisture (0-5 cm)μθwith assimilation of coarse and downscaled (a) C-band and (b) L-bandθproducts at two times (8 and 24 August indicated by arrows) and comparison to hypsometric averages of GR-θ in basin from Vivoni et al. [2008], along with the mean areal precipitation (MAP). Spatial distributions of soil moisture (0–5 cm) for (c) the C-band coarse (SIM-C-SAT-θ) and (d) downscaled (SIM-C-DIS-θ) and (e) the L-band coarse (SIM-L-SAT-θ) and (f) downscaled (SIM-L-DIS-θ) products on 9 August. The downscaled fields correspond to averages of an ensemble of 50 simulations. The insets are the frequency distributions of θ within the basin. The marked zones in soil moisture shown in Figure 4d represent Thiessen polygon areas from the rain gauge forcing.

Download figure to PowerPoint

[16] The ingestion of downscaled fields is preferable to coarse products since the model is updated while accounting for the spatial variability of soil moisture within the basin. This is shown in Figures 4c–4f through the spatial distribution of surface (0 to 5 cm) θobtained 24 hours after the first assimilation date, selected to allow soil moisture redistribution processes to take place. The assimilation of coarse C- and L-band products leads to low spatial variability sinceθis mainly distributed around one (SIM-C-SAT-θ, Figure 4c) or two (SIM-L-SAT-θ, Figure 4e) values (note the frequency distributions for each map). In contrast, the spatial variability of θis much higher when assimilating downscaled data from C-band (SIM-C-DIS-θ, Figure 4d) and L-band (SIM-L-DIS-θ, Figure 4f). Note that spatial patterns in SIM-L-DIS-θ show finer details related to terrain, soil, vegetation and rainfall fields [Vivoni et al., 2010]. Thus, accounting for the spatial heterogeneity of θthrough assimilation of a downscaled L-band product leads to more realistic soil moisture patterns at the catchment scale. Ultimately, this impact the generation of surface fluxes, such as evapotranspiration and runoff, in a nonlinear fashion [Crow and Wood, 2002; Vivoni et al., 2010]. Furthermore, assimilation of an ensemble of downscaled products provides the opportunity to issue probabilistic hydrologic predictions.

5. Conclusions

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations and Modeling
  5. 3. Soil Moisture Downscaling
  6. 4. Hydrologic Evaluation and Data Assimilation
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

[17] In this study, we compared the utility of coarse and downscaled θproducts derived from C- and L-band sensors for hydrologic applications at the catchment scale. We found an improved performance of a downscaling algorithm with L-band data derived from a coarse scale similar to the forthcoming SMAP for the semiarid region of moderate vegetation and complex terrain. Disaggregation of L-bandθdata allowed reproducing with higher accuracy: (a) the sub-footprintθ distribution across the entire region, (b) the θvariability in a watershed located within a set of coarse pixels, and (c) the variation of soil moisture along a topographic transect in the basin during wet and dry conditions. The comparison between C- and L-band products was then extended to include a set of data assimilation experiments with coarse and downscaled data ingested in a distributed hydrologic model. Results reveal that more accurate L-bandθproducts enhanced the model prognostic capability as compared to no assimilation, while the assimilation of coarse C-band data degraded performance. The downscaled L-band assimilation also improved the spatial representation of soil moisture within the basin. This holds promise for the utility of L-band soil moisture products from current and forthcoming satellite missions in hydrologic modeling applications at high resolution.

References

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations and Modeling
  5. 3. Soil Moisture Downscaling
  6. 4. Hydrologic Evaluation and Data Assimilation
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information

Supporting Information

  1. Top of page
  2. Abstract
  3. 1. Introduction
  4. 2. Observations and Modeling
  5. 3. Soil Moisture Downscaling
  6. 4. Hydrologic Evaluation and Data Assimilation
  7. 5. Conclusions
  8. Acknowledgments
  9. References
  10. Supporting Information
FilenameFormatSizeDescription
grl29224-sup-0001-t01.txtplain text document1KTab-delimited Table 1.

Please note: Wiley Blackwell is not responsible for the content or functionality of any supporting information supplied by the authors. Any queries (other than missing content) should be directed to the corresponding author for the article.