Soil Moisture Modelling and Data Assimilation

The Ouémé catchment is located in the central part of Benin in West Africa. Its climate, extremely dry in winter and heavy rains in summer, makes the water cycle a true challenge to model, especially for the water management in the agricultural areas. DHSVM (Distributed Hydrology Soil Vegetation Model) is a physically based and distributed hydrological model that solves the energy and water balance. This model simulates the soil moisture at each soil layer, the snow quantity, the evapotranspiration, the runoff and the streamflow. It was used with an hourly time step at a 1 km resolution. Model parameters have been calibrated using 2010 in situ streamflow and soil moisture measurements along with in situ precipitations.

Because there are not enough rain gauges in most of the cases, satellite precipitation products are used as inputs in the hydrological models. In this study, three products are tested: PERSIANN, TRMM, and CMORPH. Simulations are run for 3 years, from 2010 to 2012, Unfortunately, the streamflow simulations are no longer in agreement with the in situ measurements when satellite precipitation products are used because they all bring too much water and sometimes not at the correct time (too much water in early spring).

The goal of this work is to assimilate the SMOS soil moisture product (Level 3 product available at a 25 km resolution with a 3-day global coverage) into the hydrological model DHSVM using a Kalman filter for a better constraint on the water cycle model when using satellite precipitation observations. SMOS acts on the system by adding or removing some water when satellite observations are either under or overestimating the precipitations. Besides the choice of the assimilation method, the difference in spatial resolutions is also being taken care of by using an influence circle area.

The results show an improvement on the soil moisture simulations (significant decrease of the bias at the surface but also at deeper layers) and on the streamflow simulations (peaks are smoothed and Nash model efficiency increases significantly going from negative to positive values for all precipitation products) proving that coarse observations from space can still provide useful information into a finer hydrological model. Furthermore, a comparison between the water table depth simulations the in situ water levels measured from a deep well clearly shows the benefit of the SMOS soil moisture assimilation.

It is also possible to correct the precipitations directly using the SMOS soil moisture instead of assimilating them. A comparison between the two methods is also presented here.

Plant-available soil moisture is a key element which affects plant properties in their ecosystems.  Thus, monitoring of vegetation responses during different soil moisture deficit steps are directly applicable to agriculture/food security research, ecology, water resources engineering,  rainfall-runoff modeling, land and disaster management. This study shows Poa pratensis -a species of grass- responses to soil moisture deficit during an artificial drought episode in a greenhouse experiment. We used radiative transfer model inversion to monitor the gradual manifestation of soil moisture deficit effects on vegetation in a laboratory setting. Plots of 21 cm x 14.5 cm surface area with Poa pratensis plants that formed a closed canopy were subjected to water stress for 50 days. In a regular weekly schedule, canopy reflectance was measured. The 1-D bidirectional canopy reflectance model SAIL, coupled with the leaf optical properties model PROSPECT, was inverted using hyperspectral measurements by means of an iterative optimization method to retrieve vegetation biophysical and biochemical parameters (mainly; LAI, Cab, Cw  and Cs). The relationships between these retrieved parameters with soil moisture content were established in two separated groups; stress and non-stressed. To differentiate the water stress condition from the non-stressed condition, a threshold was defined that was based on the laboratory produced Soil Water Characteristic (SWC) curve. All parameters retrieved by model inversion using canopy spectral data showed good correlation with soil moisture content in the drought condition. These parameters co-varied with soil moisture content under the stress condition (Chl: R²= 0.91, Cw: R²= 0.97, Cs: R²= 0.88 and LAI: R²=0.48) at the canopy level. we concluded that analysis of the canopy reflectance spectrum in the VIS-NIR-MIR region is a promising tool for monitoring soil moisture deficit effects on vegetation properties.

Carbon dioxide (CO2) is the most important anthropogenic greenhouse gas contributing to about half of the total anthropogenic change in the Earth's radiation budget. About half of the anthropogenic CO2 emissions stay in the atmosphere, the remainder is taken up by land and oceans. In the context of climate change it is of paramount importance to understand CO2 sources and sinks and their spatio-temporal distribution. This information is needed to improve the projections of future trends in carbon sinks and sources, and thus the potential magnitude of climate change. In the present study we employ a Carbon Cycle Data Assimilation System to assimilate atmospheric CO2 flask samples with and without the SMOS L3 soil moisture product over two years into a process-based model of the terrestrial carbon cycle to constrain land surface CO2 exchange fluxes. We find that the assimilation of SMOS data improves the agreement with independent soil moisture data from ASCAT. Also the validation against in situ data provided by Fluxnet shows considerable improvement when SMOS data is assimilated jointly with CO2, compared to assimilating CO2 only. In both cases the assimilation also improves the fit of modelled atmospheric CO2 to the observations at flask sampling sites, which have not been used in the assimilation. Reduction of uncertainty relative to the prior is generally high for both regional NEP and NPP and considerably higher than for assimilating CO2 only, which quantifies the added value of SMOS observations as a constraint on the terrestrial carbon cycle. The study demonstrates a high potential for a SMOS L4 carbon flux product.

In 2014, the European Space Agency initiated the "Passive Microwave Soil Moisture Data Fusion Study" with the goal to provide guidelines for the integration of soil moisture retrievals from the Soil Moisture and Ocean Salinity satellite mission (SMOS) into the long term soil moisture record from the Climate Change Initiative soil moisture project (CCI; http://www.esa-soilmoisture-cci.org/). Soil moisture retrievals that use observations from the Advanced Scanning Microwave Radiometer – EOS (AMSRE) currently form the basis of the passive microwave part within the CCI Soil Moisture project. Therefore, three different approaches were designed and tested to fuse soil moisture retrievals from AMSRE and SMOS over the period between 2003 and 2014, which are:

A Neural Network Fusion approach (Rodríguez-Fernández et al., 2015), which uses SMOS Level 3 soil moisture (L3) as a reference to train a neural network with AMSRE brightness temperatures as input to retrieve soil moisture.
 

 A regression approach (Wigneron et al., 2004; Al-Yaari et al., submitted), which investigates the use of physically based statistical regressions with AMSR-E brightness temperatures using the SMOS L3 as training data for the retrieval of soil moisture.
 

A radiative transfer based approach (Van der Schalie et al., 2015), which is based on applying the current CCI baseline algorithm for passive microwave observations, the Land Parameter Retrieval Model (LPRM; Owe et al., 2008), to SMOS observations and updating the AMSR-E LPRM to best match the SMOS LPRM soil moisture retrievals.

In this particular study, the soil moisture retrievals from the three approaches are cross compared and tested for their skill against multiple datasets, which include MERRA-Land, ERA-Interim/Land, the current CCI Soil Moisture dataset v2.2 and in situ measurements from the International Soil Moisture Network. The skill of these three approaches are based on the use of common statistics, like the correlation, unbiased root mean square error and bias. In addition, an advanced version of the triple collocation  (e.g. Gruber et al., 2015) is applied to determine the global uncertainties of each developed product. With a direct comparison of the three approaches to the current CCI soil moisture product, which also includes soil moisture retrievals from active microwave observations, this study can directly reveal the added value of SMOS soil moisture retrievals within the current framework of the CCI Soil Moisture data record. Based on all these results, a recommendation will be presented that will contribute to a successful integration of SMOS derived soil moisture into the CCI soil moisture climate record. 

References

Al-Yaari, A., Wigneron, J.-P., Kerr, Y.H., Jeu, R.A.M., Rodriguez-Fernandez, N.J., Van der Schalie, R., Bitar, A.A., Mialon, A., Richaume, P., Dolman, A.J., and Ducharne, A., “Testing regression equations to derive long-term global soil moisture datasets from passive microwave observations”, Submitted to: RSE

Gruber, A., Su, C.-H., Zwieback, S., Crowd, W., Dorigo, W., and Wagner, W. (2015), “Recent advances in (soil moisture) triple collocation analysis”, Int. J. Appl. Earth Observ. Geoinf, doi: http://dx.doi.org/10.1016/j.jag.2015.09.002.

Owe, M., De Jeu, R.A.M., and Holmes, T.R.H. (2008), "Multi-Sensor Historical Climatology of Satellite-Derived Global Land Surface Moisture", J. Geophys. Res., vol. 113, F01002, doi: 1029/2007JF000769.

Rodríguez-Fernández, N.J., Aires, F., Richaume, P., Kerr, Y.H., Prigent, C., Kolassa, J., Cabot, F., Jiménez, C., Mahmoodi, A., and Drusch, M. (2015), “Soil Moisture Retrieval Using Neural Networks: Application to SMOS”, IEEE Trans. on Geosc. and Remote Sens., vol. 53, no. 11, doi: 10.1109/TGRS.2015.2430845.

Van der Schalie, R., Kerr, Y.H., Wigneron, J.P., Rodriguez-Fernandez, N.J., Al-Yaari, A., and De Jeu, R.A.M. (2015), “Global SMOS Soil Moisture Retrievals from The Land Parameter Retrieval Model”, Int. J. Appl. Earth Observ. Geoinf, doi:  http://dx.doi.org/10.1016/j.jag.2015.08.005.

Wigneron J.-P., Calvet, J.-C., De Rosnay, P., Kerr, Y., Waldteufel, P., Saleh, K., Escorihuela, M.J. and Kruszewski, A. (2004), “Soil Moisture Retrievals from Bi-Angular L-band Passive Microwave Observations”, IEEE Trans. Geosc. Remote Sens. Let., vol. 1, no. 4, pp. 277-281.

Assimilation of satellite data is a key component of current Numerical Weather Prediction (NWP) systems. For land surfaces, initialisation of soil moisture is particularly important to ensure accurate forecast of surface and near surface weather conditions at short range, medium range and seasonal range. Satellite data from active and passive microwave sensors, such as ASCAT, SMOS and SMAP provide relevant information on soil moisture at global scale. They are being increasingly used for data assimilation and monitoring in NWP systems, and scatterometer soil moisture data is already assimilated in operational NWP systems at the UK MetOffice and at the European Centre for Medium-Range Weather Forecasts (ECMWF). A critical prerequisite for NWP applications is the availability of the observations in Near-Real-Time (NRT), which is preferably within three hours after sensing. The ASCAT soil moisture product and the ESA SMOS brightness temperature products have been available in NRT since several years via EUMETCAST and ESA, respectively. More recently a new SMOS NRT soil moisture product, based on a neural network approach was developed by the Centre d'Etudes Spatiales de la Biosphère (CESBIO) and it is being implemented by ECMWF and ESA. It will ensure the NRT availability of a high quality SMOS soil moisture product, for the scientific community as well as for the operational hydrological and NWP communities.

This paper gives a review of the large range of activities related to the use of satellite data for soil moisture analysis and reanalyses at ECMWF. Firstly, the status of the operational ECMWF soil moisture data assimilation is described. ASCAT soil moisture is assimilated together with in situ SYNOP observations of air temperature and relative humidity. We show the complementarity between these different types of observations and the impact of the soil analysis on NWP. Secondly, developments of SMOS brightness temperature data assimilation is presented and SMOS impact on NWP is assessed. Furthermore results of recent investigations on a SMOS soil moisture neural network product data assimilation are presented. The different data assimilation approaches using either SMOS brightness temperature or soil moisture are compared. Finally, this paper presents applications involving longer time scale processes, to produce and to use soil moisture data records in the context of H-SAF, ESA-CCI and future reanalysis activities.  Perspectives to use data from future missions, such as SWOT, to consistently constrain NWP and flood forecast systems are also discussed.