From Point to Pixel Scale: An Upscaling Approach for In Situ Soil Moisture Measurements

Abstract

Given the high spatial variability of soil moisture content (SMC), direct comparison and integration of observations from different sources and measurement scales is becoming a major challenge. We have developed a spatial upscaling method for SMC that enables the direct combination of in situ measurements and remotely sensed data. The approach is based on the fact that spatial soil moisture patterns are related to ancillary fea-tures like topography, land cover, and soil type. This study used in situ data from a well-equipped research site in the northern Italian Alps. One of the main goals was to enable the use of these data for the validation of the NASA Soil Moisture Active Passive (SMAP) products. Dealing with medium-to coarse-resolution satellite imagery, especially in mountain areas, requires compensating for different measurement scales. The study approach was assessed based on Envisat advanced synthetic aperture radar (ASAR) data, which were resampled to reproduce the spatial scale of the SMAP data. Results show that the representativeness of in situ data, with respect to the 3-by 3-km SMAP pixel scale, can be improved significantly—direct correlation between SMC and satellite backscatter was improved from R = 0.05 to 0.28; furthermore, the error of the estimated SMC was improved from RMSE = 0.12 to 0.03 m 3 m −3 . This leads to more accurate reference data, which can help to improve the retrieval of SMC from remotely sensed imagery. Abbreviations: ASAR, advanced synthetic aperture radar; SAR, synthetic aperture radar; SMAP, Soil Moisture Active Passive; SMC, soil moisture content; SVR, support vector regression. Soil moisture content (SMC) is a key element in the global cycles of water, energy, and carbon. Knowledge of the spatial and temporal distribution of this parameter is, therefore, essential for a number of hydrological applications, as well as other geosciences like meteorol-ogy or climatology (Heathman et al., 2003). Soil moisture content can be measured in situ at the field or catchment scales, using, e.g., time-domain reflectometry or capacity sensors. For the continuous sensing of wider areas, satellite remote sensing is widely used (Wagner et al., 1999; Barrett et al., 2009; Bindlish et al., 2015). Among the different methodologies, active and passive microwave systems, like radiometers, scatterometers, or synthetic aperture radar (SAR) (Pathe et al., 2009), have proved to be promising technologies for the retrieval of SMC. In areas with complex terrains, such as in the Alps, the relationship between the measured signal and SMC is complicated and ill posed. Therefore, retrieval is possible only after consideration of further parameters. Several approaches have been proposed to address this issue. In one method, introduced by Pasolli et al. (2011), machine learning is used to relate SMC with SAR data and several auxiliary features, like topography, land cover, and vegetation parameters. This method allows the robust estimation of SMC in mountain areas; nevertheless, it depends on the availability of in situ data and a reliable integration of data from different sensors (with different spatial resolutions). Otherwise, alpine areas have been considered only marginally, and only pioneer studies can be found in the literature (Brocca et al., 2013; Bertoldi et al., 2014). One of the main open issues is the validation of spatially dis-tributed data using point measurements. If the variability of SMC is not taken into account correctly, it can lead to biased performances. This is particularly important in mountainous Core Ideas