Soil Moisture Remote Sensing: State-of-the-Science

As remote sensing platforms are becoming more strategic for global monitoring of Earth resources, various ground-based invasive and noninvasive soil moisture measurement techniques and their monitoring networks have been utilized for their validation around the globe. Bogena et al. (2015) provided a review of recent developments of various noninvasive soil moisture measurement techniques and monitoring networks. In particular, cosmic ray neutron probes (CRNPs) with Cosmic-ray Soil Moisture Interaction Code (COSMIC) as well as global navigation satellite system reflectometry have shown good promise to capture soil moisture with a support area of a few hundred to a few thousand square meters. In addition, several soil moisture testbeds encompassing different in situ sensors with different accuracy and precision have been developed (e.g., TERestrial ENvironmental Observatories [TERENO] in Germany, Marena Oklahoma In Situ Sensor Testbed [MOISST] in Oklahoma, and the Texas Soil Observation Network [TxSON]) to evaluate the satellite-based soil moisture products on a time-continuous basis. Furthermore, various sparse soil moisture networks (e.g., NOAA’s US Climate Reference Network and the USDA’s Soil Climate Analysis Network) as well as time-limited field campaigns (e.g., SMAPVEX) including airborne and ground-based sampling have been used for validating the satellite soil moisture products. In addition to the spatial extent of soil moisture networks with a global distribution, these networks need a long-term perspective to evaluate long time series such as provided by the European Space Agency Climate Change Initiative (CCI) soil moisture product (Liu et al., 2012; Dorigo et al., 2015).

The first real-time comparison of satellite soil moisture products with in situ data was achieved with the launch of the SMAP mission (Entekhabi et al., 2010). The first large-scale soil moisture satellite validation program for the Advanced Microwave Scanning Radiometer (AMSR-E) on the Aqua satellite helped to establish several moderate-resolution watershed networks, resulting in a specially designed suite of landscapes for validation (Jackson et al., 2010). These watershed networks provide a diversity of conditions among them, but within a watershed there is limited heterogeneity to limit additional sources of error in the process. For the SMOS mission, a follow-up study was conducted with comparisons of these same watershed networks with SMOS, in addition to newer networks that came online more recently (Jackson et al., 2012; Panciera et al., 2009). Other satellites had validation programs along the way, including Aquarius (Bindlish et al., 2015) and ASCAT (Albergel et al., 2009). With SMAP, however, there was a regular automated ingestion of in situ data for weekly comparisons with in situ resources. These resources included both pixel-scale watershed networks and larger sparse networks to provide a spectrum of land covers and moisture conditions each week. Error estimates were able to be generated for a given site as soon as enough data points were collected to provide confidence in the statistics. These results were documented by Chan et al. (2016), who detailed meeting the success criteria of a validated 0.04 m³/m³ RMSE for the Level 2 soil moisture product. The new satellites that have launched or are on schedule, such as the Global Change Observation Mission–Water (GCOM-W) mission or the Sentinel missions, will benefit from the increased performance of in situ resources that have matured in preparation for the SMAP mission. Watersheds such as the Little Washita (Cosh et al., 2014) have a long history of soil moisture monitoring, with established accuracies (Cosh et al., 2006). Newer watershed networks, such as the Carman network, near Winnipeg, MB, Canada, have only recently begun to document their accuracy as well (Adams et al., 2015). The variety of networks and technologies has necessitated the development of an in situ sensor testbed to provide some estimate of the interoperability and accuracy of these different sensors. The MOISST (Cosh et al., 2016) network was established in 2010 to provide a long-term data series of diverse soil moisture technologies to address what impact these technology selections have on calibration and validation efforts. Most networks are available through the International Soil Moisture Network (Dorigo et al., 2011) and have been used for satellite soil moisture validation (e.g., de Jeu et al., 2014; Paulik et al., 2014; van der Schalie et al., 2016; Wu et al., 2016). Figure 1 shows the partial list of in situ soil moisture networks used for global calibration and validation activities.

Further in situ observatories to support satellite soil moisture validation in recent years are based on CRNPs. The networks COSMOS (Zreda et al., 2012), TERENO (Baatz et al., 2014), and COSMOS-UK (Evans et al., 2016) are able to provide footprint-scale reference data instead of typical point-scale measurements. According to Köhli et al. (2015), the horizontal cosmic ray probe footprint radius ranges from 130 to 240 m depending on air humidity, soil moisture, and vegetation. For short-period campaigns, the utilization of mobile CRNP rovers is possible (Dong et al., 2014). A first comparison of SMAP Level 3 soil moisture data with CRNPs in the Rur catchment, Germany (Montzka et al., 2016), is shown in Fig. 2. The two curves agree well, but SMAP follows a higher dynamic range, which can be explained by the lower penetration depth compared with CRNPs (Montzka et al., unpublished data, 2016). Further research is needed to consider the different soil volumes observed, e.g., by the data assimilation method published by Rosolem et al. (2014). Moreover, hydrogen stored in vegetation affects the counts of registered neutrons at the sensor. Therefore, a vegetation correction procedure has to be applied in regions with high vegetation dynamics (Baatz et al., 2015). More details and examples of new techniques in large-scale soil moisture monitoring were given by Ochsner et al. (2013).

Besides satellite- and ground-based measurements, different land surface models using climatology and atmospheric forcing have been used for estimating soil moisture at different space and time scales. Statistical tools such as triple collocation techniques using multiple satellite products and their signal/noise ratio along with land surface model products have been used as alternative tools for validation of near-surface soil moisture (Gruber et al., 2013, 2016; McColl et al., 2014; Pan et al., 2015). While cross-validation of different satellite data using these novel schemes has been found promising, debate continues about their accuracy and bias at different scales, heterogeneities, and hydroclimates. In addition to these statistical techniques for validation of remote sensing data, other efforts including rules for scaling up or down based on various land surface physical controls are being used to decipher soil moisture dynamics at different scales.

To accomplish this, studies are being conducted to understand the influence and sensitivity of different physical attributes (e.g., soil type, roughness, vegetation water content, and albedo) in different hydroclimates. In addition, it is necessary to determine how these attributes impact the observed brightness temperature, backscatter, and retrieved soil moisture across various space and time scales.

Recent studies have suggested that the performance of available radiative transfer models (RTMs) used for satellite (e.g., SMOS and SMAP) soil moisture algorithms are not optimum under highly heterogeneous landscape conditions and vary with hydroclimate. Using global sensitivity analysis for parameter interaction in a zero-order radiative transfer model for data collected during field campaigns SMEX02 (Iowa) and SMAPVEX12 (Winnipeg, Canada), Neelam and Mohanty (2015) found that: (i) four parameters (soil moisture, vegetation water content [VWC], surface roughness height, and its correlation length in Iowa, whereas one parameter, soil moisture (SM), in Winnipeg, Canada, were sensitive to brightness temperature, (ii) attenuation of soil emissions by vegetation (VWC and vegetation structure) can be significant in structured plants (e.g., corn [Zea mays L.]) and this attenuation or scattering increases with roughness and SM conditions, (iii) for similar surface roughness conditions, sensitivity to roughness parameters is higher in wet soils than dry soils, and (iv) SM derived from brightness temperature pixels representative of a wide range of SM conditions are more accurate than pixels representative of a narrow (and higher) range of SM conditions because SM retrieval accuracy will be compromised due to higher sensitivity to other parameters at narrow SM ranges. Based on these findings of Neelam and Mohanty (2015), it can be inferred that a uniform RTM approach for soil moisture retrieval across the globe may have limitations and there exist opportunities to further improve soil moisture retrieval accuracy and associated soil water fluxes in the vadose zone using various current and future microwave remote sensing platforms. Specifically, because the most RTM sensitivity originates from vegetation water content and surface roughness attributes, more carefully designed high-resolution field studies are warranted to gain a deeper understanding of their magnitude and seasonality in different hydroclimates.

Besides near-surface soil moisture, root-zone (the top 1 m below the land surface) soil moisture dynamics are a key factor in management of water resources and agricultural water, rainfall–runoff processes, and ecosystem health and dynamics. Using near-surface soil moisture from satellites in conjunction with soil hydrologic models, efforts are being made to produce regional (Das and Mohanty, 2006; Das et al., 2010; Ridler et al., 2014; Dumedah et al., 2015; González-Zamora et al., 2016; De Lannoy and Reichle, 2016) and global (Muñoz-Sabater, 2015) scale root-zone soil moisture. The ASCAT root zone soil moisture products have been publicly available for several years (Wagner et al., 1999; Brocca et al., 2010), operational SMOS Level 4 root-zone products can currently be delivered to the scientific community only (Mecklenburg et al., 2016). Besides SMAP Level 2 and 3 products near the land surface, the SMAP Level 4 product provides top 1-m root-zone soil moisture estimates at 9-km resolution on a weekly basis by assimilating passive near-surface soil moisture products with a GEOS-5 catchment land surface model (Entekhabi et al., 2014). Encompassing various soil types (e.g., texture), topographic features (e.g., slope), vegetation and land cover, and climatic conditions, SMAP Level 4 data provide a perfect baseline to better understand the role of root-zone soil moisture “memory” in hydrologic, agricultural, weather, and seasonal climate predictions. This will lead to a closing of the water budget at the field, catchment, watershed, and regional scales.

While satellite platforms provide the coarse resolution information, most agricultural, hydrological, meteorological, and land-use and change applications need representation of small-scale spatial heterogeneity of land surface as hydrological processes manifest at scales ranging from centimeters to kilometers. Although remote sensing platforms provide large-scale soil moisture dynamics, scale discrepancy between the observation scale (e.g., approximately several kilometers for satellite observations and point scales for in situ observations) and the modeling scale (e.g., a few hundred meters to kilometers) leads to uncertainties in land surface hydrologic model performance. To improve hydrologic and other applications, satellite-based soil moisture should be downscaled to an appropriate level. The spatiotemporal variability of root-zone soil moisture influences runoff at the soil surface and subsurface, evapotranspiration and atmospheric feedback, and groundwater recharge. Root-zone soil moisture content also plays a pivotal role in ecological processes at individual plant to system scales. Ongoing efforts to downscale soil moisture products at these application scales (e.g., Shin et al., 2013; Ines et al., 2013) have been undertaken using different scaling techniques as well as to improve the capability to predict root-zone soil hydraulics and associated hydrologic fluxes at different spatial and temporal scales.

At a particular point in time, root-zone soil moisture content is influenced by (i) precipitation history, (ii) the texture of the soil, which determines the water-holding capacity, (iii) the slope of the land surface, which affects runoff and infiltration, and (iv) vegetation and land cover, which influences evapotranspiration and deep percolation. Traditionally, soil moisture spatial variability studies using ground-based, point-scale measurements are limited to small fields with uniform soil characteristics, topographic features, and vegetation conditions. On the other hand, satellite footprint-scale soil moisture measurements may be coarse, and errors at the local scales of field, catchment, and watershed may propagate into inaccurate regional-scale hydrologic fluxes important for water resource assessment, agriculture, and hydroclimatic predictions. To develop more efficient scaling tools for utilizing satellite soil moisture (SMAP and SMOS) products, studies investigating the scale dependence of the dominant physical controls on soil moisture distribution have been undertaken. The temporal and spatial distribution of soil moisture and the resultant fluxes to different hydrologic reservoirs are affected by interactions between soil, topography, vegetation, and climate (Joshi and Mohanty, 2010; Gaur and Mohanty, 2013). Differences in these dominant controlling mechanisms were evaluated using data from various past airborne remote sensing field campaigns (e.g., SGP97, SGP99, SMEX02, SMEX03, SMEX05, CLASIC07, and SMAPVEX12). Using wavelet analysis with various airborne remote sensing data, Gaur and Mohanty (2016) observed the dominance of various physical controls on the soil moisture status evolved in different hydroclimates (see Fig. 3).

There have been attempts to downscale soil moisture from the satellite footprint scale to various Earth system application scales. A few methods have exploited sensor characteristics and raw data records, e.g., Lindell and Long (2016) as well as Lindsley and Long (2016) modified the ASCAT retrieval algorithm to generate 15- to 20-km soil moisture products. However, most of these methods take into account supplemental land surface attributes such as soil, vegetation, evapotranspiration, surface temperature, and topography at higher spatial resolutions. Some of these include the use of vegetation temperature (Peng et al., 2016), surface temperature (Peng et al., 2015; Im et al., 2016), vegetation (Zhao and Li, 2015), and vegetation, topography, and soil texture (Ranney et al., 2015). Higher resolution microwave measurements at higher frequencies (Ka band) were used by de Jeu et al. (2014) to downscale AMSR-E C-band data. Merlin et al. (2013) improved the evaporation-based disaggregation of SMOS data down to scales of 3 km and 100 m. Further downscaling approaches have been developed for SMOS data, such as methods similarly exploiting SMOS–optical sensor relationships (Piles et al., 2014, 2016; Sánchez-Ruiz et al., 2014), copulas (Verhoest et al., 2015), other deterministic approaches (Shin and Mohanty, 2013; Shin et al., 2013), statistical unmixing (Ines et al., 2013), and machine learning (Srivastava et al., 2013). Rüdiger et al. (2016) disaggregated L-band SMOS data by C-band Envisat ASAR data, a dual-frequency concept currently evaluated also for SMAP downscaling by Sentinel-1. In addition to estimating downscaled soil moisture, some of these techniques also estimate effective soil hydraulic properties of the vadose zone at matching resolution (Ines et al., 2013; Shin and Mohanty, 2013; Lee et al., 2014). Mohanty (2013) provided an extended review of available methods in this regard.

Figure 4 depicts a downscaling algorithm based on a published method outlined by Fang et al. (2013). It uses MODIS-derived normalized difference vegetation index (NDVI) and surface temperature (T_s) as conditioning attributes. These two variables are derived for each 1-km subpixel in the 36-km SMAP footprint soil moisture data. Fang et al. (2013) constructed lookup curves between daily surface temperature difference and daily average soil moisture from the Global Land Data Assimilation System (GLDAS) (Rodell et al., 2004) for different NDVI values (derived from the Advanced Very High Resolution Radiometer [AVHRR]) for a 30-yr period (1979–2008). Based on the collocated MODIS-based NDVI and T_s, and the established lookup curves, they downscaled SMAP soil moisture to 1-km subpixels (as shown in Fig. 4). Downscaling of satellite soil moisture is very important because the spatial resolution of the soil moisture retrievals are on the order of kilometers. With the absence of the radar on SMAP, there is further imperative to use the readily available visible–near infrared and thermal observations corresponding to vegetation and surface temperature as the constraining parameters.

The ASCAT data have been available at near real time over the Satellite Application Facilities (http://hsaf.meteoam.it) for several years (Hahn et al., 2012). Therefore, operational utilization is very advanced, especially in numerical weather prediction (de Rosnay et al., 2013). Further examples of ASCAT applications are, e.g., the improvement of rainfall (Ciabatta et al., 2016; Wanders et al., 2015) and discharge estimates (Laiolo et al., 2016), the identification of potential flash-flood areas (Bangira et al., 2015) and the prediction of floods (Alvarez-Garreton et al., 2015), and identification of vegetation drought (Schroeder et al., 2016).

Both SMAP and SMOS data have been used by different early adopters for various Earth system operational purposes, ranging from agriculture, to weather forecasting, to hazard mitigation or management. In March 2016, the ESA released the new SMOS Level 2 soil moisture product resulting from processor version V620 for both reprocessed and operational products. However, for the improvement of numerical weather forecasting and drought and flood prediction, fast soil moisture product dissemination gains more importance. To overcome the time lag limitations of the operational Level 2 product, a neural net approach after Rodriguez-Fernandez et al. (2015) was applied to rapidly provide near-real-time SMOS soil moisture data within a time lag of 4 h after sensing. Because the neural net was trained by historical time series of the operational SMOS product, the two data sets follow the same global climatology.

Although several uncertainties related to retrieval, validation, and climate-specific bias persist, microwave remote sensing of soil moisture has matured in the past decade. Developing adaptive scaling, data assimilation, and modeling schemes, these near-surface-sensing global products can be used for many societal applications in the coming decades. Among various applications, some of the prominent ones include (i) agricultural water management (irrigation scheduling) by including remotely sensed soil moisture status as a boundary condition in soil hydrology and crop growth models at various spatiotemporal scales, (ii) weather and climate forecasting by assimilating regional- and global-scale soil moisture into numerical weather and climate models, (iii) hydrologic flux (evapotranspiration, groundwater recharge, surface runoff, and base flow) estimation by linking fluxes to surface and root-zone soil moisture, resulting in improved drought or flood forecasting, (iv) battle ground trafficability and landslide potential determination by relating root-zone soil moisture to soil tensile strength, (v) nutrient and contaminant transport potential of the soil where soil moisture status reflects the flow and transport attributes, and (vi) large-scale effective soil hydraulic property estimation by inverting time series soil moisture and other hydrologic fluxes. Although remotely sensed data provides these unprecedented advantages, some scientific challenges in terms of coarse space and time resolution, shallow penetration depth, and mismatched governing hydrologic principles persist. In addition, to minimize uncertainty and improve accuracy of the soil moisture products and their applications, new environment-specific RTMs, up- and downscaling functions, and data assimilation schemes need to be developed.