ON THE SYNERGY OF SMOS AND TERRA/AQUA MODIS: HIGH RESOLUTION SOIL MOISTURE MAPS IN NEAR REAL-TIME M.Piles1,4 ,M.Vall-llossera1,4 ,A.Camps1,4 ,N.S´anchez2,J.Mart´ınez-Fern´andez2,J.Mart´ınez3,4 ,V.Gonz´alez-Gambau3,4,R.Riera5 1
Universitat Polit`ecnica de Catalunya, IEEC/UPC, Jordi Girona 1-3, E-08034 Barcelona, Spain 2 Universidad de Salamanca/CIALE, Duero 12, E-37185 Villamayor, Spain 3 Institut de Ci`encies del Mar/CSIC, Pg. Mar´ıtim Barceloneta 37-49, E-08003 Barcelona, Spain 4 SMOS Barcelona Expert Centre, Pg. Mar´ıtim Barceloneta 37-49, E-08003 Barcelona, Spain 5 Diputaci´o de Barcelona, Rambla de Catalunya, 126, E-08008 Barcelona, Spain E-mail:
[email protected]
ABSTRACT An innovative downscaling approach to obtain fine-scale soil moisture estimates from 40 km SMOS observations has been developed. It optimally blends SMOS multi-angular and fullpolarimetric information with MODIS visible/data into high resolution soil moisture maps. The core of the algorithm is a model that links microwave/optical sensitivity to soil moisture and linearly relates the two instruments across spatial scales. This algorithm has been implemented at SMOS-BEC facilities and near real-time maps of disaggregated soil moisture over the Iberian Peninsula are being distributed. In this work, the temporal and spatial variability of these maps is evaluated through comparison with ground-based mesurements acquired at the REMEDHUS soil moisture network, in the central part of the Duero basin, Spain. Results from a two-year time-series comparison show that downscaled soil moisture maps compare well with in situ data and nicely reproduce soil moisture dynamics at a 1 km spatial scale.
(a) Morning overpass
1. INTRODUCTION Soil moisture is a key state variable that links the Earth’s water, energy and carbon cycles, and its variations affect the evolution of weather and climate over continental regions. The ESA’s Soil Moisture and Ocean Salinity (SMOS) is the first satellite mission ever designed to measuring this variable, and its accurate (0.04 m3 /m3 ) and frequent (< 3 days) observations of soil moisture are helping to improve our understanding of water and energy fluxes interactions between the atmosphere, the soil surface and subsurface at a global scale. However, due to practical constraints on antenna size and altitude of low Earth orbits, its spatial resolution is limited to ∼40 km. The work presented on this paper was supported by the Spanish Ministry of Science and Innovation under the projects AYA2010-22062-C05 and AYA2012-39356-C05.
(b) Afternoon overpass
Fig. 1: Images of 1 km disaggregated SMOS SM maps over Catalonia included in the Barcelona’s fire risk prevention service daily bulletin (from July 8, 2012). Empty areas in the image correspond to clouds masking MODIS observations.
Within this context, an innovative microwave/optical downscaling approach for SMOS has been developed that permits resolving soil moisture dynamics within 1 to 10 km catchments [1]. Knowledge of soil moisture at these scales
Fig. 2: Location of the REMEDHUS network and layout of permanent soil moisture stations within the area. Left: SMOS disaggregated soil moisture map at 1 km spatial resolution from January 24, 2012. Right: orthophoto showing terrain features. is important for numerous applications, including water resources management, agricultural productivity estimation, ecological modeling and fast reaction to natural hazards (e.g. floods, landslides)[2]. This algorithm has been implemented at SMOS-BEC facilities and 1 km spatial resolution soil moisture maps over the Iberian Peninsula are being freely distributed (cp34-bec.cmima.csic.es/NRT): maps from the first three years of SMOS in-orbit are available and two near real-time (NRT) maps are daily generated corresponding to ascending and descending overpasses with a delay of less than 12 hours. In its current form, the algorithm only requires SMOS and MODIS remotely sensed data for implementation. In NRT mode, MODIS/Terra LST are used as baseline and MODIS/Aqua LST are employed in case of cloud masking Terra measurements. As a prime NRT application, SMOS disaggregated soil moisture maps are being used by Barcelona’s fire prevention services to detect extremely dry soil and vegetation conditions posing a risk of fire. Figure 1 shows two 1 km disaggregated soil moisture images as they appear in the Barcelona’s fire prevention service daily bulletin from July 8, 2012. They reflect a clear increase in surface soil moisture from SMOS morning to SMOS afternoon overpass due to a rain event. Taking advantage of the first-ever availability of high resolution soil moisture maps over the Iberian Peninsula, this work is focused on evaluating this data set. An overview of the downscaling strategy is presented in Section 2. Results of long-term dedicated validation studies at the soil moisture measurement network (REMEDHUS), central part of the Duero basin, Spain, are reported in Section 3. Conclusions and perspectives are presented in Section 4. 2. DOWNSCALING STRATEGY FOR SMOS The downscaling algorithm is based on a linking model that optimally combines the high radiometric resolution of SMOS data with the high spatial resolution of MODIS Terra/Aqua data into 1 km soil moisture maps. This model is a regression formula that relates a soil moisture reference to MODIS Land
Surface Temperature (LST), MODIS Normalized Difference Vegetation Index (NDVI) and SMOS brightness temperatures (TB) [1]. SMOS L2 data is used as baseline soil moisture reference; the prospect use of alternative references –such as the European Centre for Medium-range Weather Forecast product– is under consideration. The model is first applied at low resolution (40 km) to determine fitting parameters. Then, it is applied at high resolution (1 km) to derive the fine-scale soil moisture estimates. In [3], an in-depth analysis of the linking model formulation was performed using collocated SMOS and MODIS observations under different climatic contexts and a robust formulation for the linking model was stablished. This linking model exploits SMOS multi-angular and full-polarimetric capabilites, is shown to be stable over time and reduces the algorithm minimization error to 0.010.03 m3 /m3 . From the range of available optical sensors that could be used in the algorithm to disaggregate SMOS observations, we selected 1 km MODIS data to meet the temporal requirements of land hydrology applications (∼ 3 days). Yet the algorithm can potentially provide soil moisture maps at finer spatial resolutions than 1 km, given higher spatial resolution optical data is used (e.g. 90-m, 16-day revisit ASTER). However, the transferability of results when using other optical platforms will require further validation. Also, it is important to note that this downscaling methodology is dynamic and can be transported to any region of interest. 3. LONG-TERM VALIDATION RESULTS Two years of SMOS disaggregated observations over the Iberian Peninsula have been selected for this study, from March 1, 2010 to February 29, 2012. They are obtained using MODIS/Terra as fine-scale auxiliary information in the algorithm and are available in netCDF format at SMOS-BEC data distribution services (cp34-bec.cmima.csic.es/NRT). For validation purposes, in situ soil moisture observations from REMEDHUS stations as well as SMOS L2 v. 500 soil moisture data during this period have also been collected. The REMEDHUS network has previously been used in validation and applications activities with passive microwave sensors [4], active [5], or both of them [6], all devoted to soil moisture retrieval. The layout of 19 permanent soil moisture stations within REMEDHUS and a sample NRT SMOS disaggregated map from January 24, 2012 are shown in Fig.2. A comparison at 40 km between SMOS L2, SMOS downscaled and ground-based SM has been performed to ensure SMOS sensitivity is preserved with the proposed technique. The temporal evolution of these three soil moisture timeseries over REMEDHUS are shown in Fig.3. It shows that ground-based soil moisture temporal dynamics are captured by the downscaled soil moisture estimates at a regional scale. Areal-averaged downscaled estimates match well in situ data and also compensate from the dry bias present on SMOS L2
Fig. 3: Temporal evolution of surface soil moisture time-series over REMEDHUS: ground-based mean (green solid line) and standard deviation (green shaded areas), SMOS L2 (black stars), and 40-km aggregated SMOS-Terra downscaled (red triangles). Both SMOS ascending and descending orbits are included. Daily mean rainfall on top.
Correlation coefficient R
1
performed to analyze spatio-temporal correlation of disaggregated maps. Statistics of the comparison at each REMEDHUS station are reported in Fig.4. Bar graphs show spatial patterns are captured at 1 km, with a median R of 0.6 and a median ubRMSD of 0.05 m3 /m3 .
0.8 0.6 0.4 0.2 0
E10 F6 F11 H7
H9 H13 I6
J3 J12 K4
K9 K10 K13 L3
L7
M5 M9 M13 N9
RMSD ubRMSD
RMS differences
0.2
4. CONCLUSION
0.15 0.1 0.05 0
E10 F6 F11 H7
H9 H13 I6
J3 J12 K4
K9 K10 K13 L3
L7
M5 M9 M13 N9
Fig. 4: Correlation coefficient, RMSD and unbiased RMSD from the comparison of 1 km disaggregated SMOS SM vs. ground-based SM at each REMEDHUS station.
Table 1: Comparisons of 2 years of areal daily averages between SMOS L2 and 1 km disaggregated soil moisture with groundbased soil moisture measurements over REMEDHUS: correlation (R), slope of linear regression, mean difference (bias), mean standard deviation (stdev), root mean square difference (RMSD), unbiased RMSD (ubRMSD), skewness (S) and kurtosis (K). R
slope
bias stdev RMSD ubRMSD
SMOS L2 0.55 0.97 -0.06 0.04 Downscaled 0.49 1.13 -0.03 0.05
0.07 0.06
0.03 0.04
S
K
1.12 6.78 0.73 4.20
data. Statistics on the comparison between SMOS L2 and areal-averaged SMOS disaggregated with soil moisture measurements are reported in Table 1. Scores confirm that SMOS sensitivity is preserved, with correlation coefficients of ∼0.5 and unbiased Root Mean Square Differences (ubRMSD) of ∼0.04 m3 /m3 . Measures of skewness and kurtosis evidence that the downscaled estimates are closer to a normal distribution than SMOS L2 data. A spatial comparison at high resolution of downscaled soil moisture estimates with in situ measurements has been
Accurate knowledge of the soil moisture status at fine scales (∼1 km) is essential to know how to manage and utilise soil water -one of the Earth’s scarcest and most valuable natural resource- to its maximum potential. A downscaling strategy to combine SMOS and MODIS observations into high resolution soil moisture fields has been presented as a clear step in this direction. Two years of SMOS disaggregated soil moisture estimates at 1 km spatial resolution have been shown to compare well with in situ soil moisture measurements acquired at REMEDHUS stations. SMOS sensitivity at 40 km is preserved and temporal dynamics of ground-based measurements are captured at high resolution with a median correlation coefficient of 0.6 and ubRMSE of 0.05 m3 /m3 . Near real-time SMOS disaggregated soil moisture maps over the Iberian Peninsula are freely distributed through SMOS-BEC data services as a path opener to new water management and fast reaction applications. These maps have been already integrated in the Barcelona local fire prevention service to detect extremely dry soil and vegetation conditions posing a risk of fire. An overview of the downscaling strategy and its key research results will be presented at the conference. 5. REFERENCES [1] M. Piles, A. Camps, M. Vall.llossera, I. Corbella, R. Panciera, C. R¨udiger, Y. Kerr, and J. Walker, “Downscaling SMOS-Derived Soil Moisture Using MODIS Visible/Infrared Data,” IEEE Transactions on Geoscience and Remote Sensing, vol. 49, pp. 3156–3166, 2011. [2] W.F. Krajewski, M.C. Anderson, W.E. Eichinger, D. En-
tekhabi, B.K. Hornbuckle, P.R. Houser, G.G. Katul, W.P. Kustas, J.M. Norman, C. Peters-Lidard, and E.F. Wood, “A remote sensing observatory for hydrologic sciences: A genesis for scaling to continental hydrology,” Journal of Water Resources Research, vol. 42, 2006. [3] M. Piles, M. Vall.llossera, L. Laguna, and A. Camps, “A downscaling approach to combine SMOS multi-angular and full-polarimetric observations with MODIS VIS/IR data into high resolution soil moisture maps,” in Proc. IEEE Geoscience and Remote Sensing Symposium, 2012. [4] N. S´anchez, J. Mart´ınez-Fern´andez, A. Scaini, and C. P´erez-Guti´errez, “Validation of the SMOS L2 soil moisture data in the REMEDHUS Network (Spain),” IEEE Transactions on Geoscience and Remote Sensing, vol. 50, pp. 1602–1611, 2012. [5] A. Ceballos, K. Scipal, W. Wagner, and J. MartinezFernandez, “Validation of ERS scatterometer-derived soil moisture data in the central part of the Duero basin, Spain,” Hydrological Processes, vol. 19, pp. 1549–1566, 2005. [6] L. Brocca, S. Hasenauer, T. Lacava, F. Melone, T. Moramarco, W. Wagner, W. Dorigo, P. Matgen, J. MartnezFernndez, P. Llorens, J. Latron, C. Martin, and M. Bittelli, “Soil moisture estimation through {ASCAT} and amsr-e sensors: An intercomparison and validation study across europe,” Remote Sensing of Environment, vol. 115, no. 12, pp. 3390 – 3408, 2011.