Soil Moisture Assimilation Research


An alternative methodology for diagnosing soil moisture (SM) exploits observations of land surface temperature (LST) in the thermal infrared (TIR) atmospheric window (10 to 12 microns). It has been shown that the evolution of LST in the morning hours is strongly dependent on current SM conditions, as wet soil or well-watered vegetation will heat up more slowly while dry soil or stressed vegetation will heat up more rapidly [Anderson et al. 2007a; Hain et al. 2009]. A number of past studies have attempted to develop strategies to exploit TIR LST for monitoring SM [Carlson et al. 1981; Price 1983; Carlson 1986; Taconet et al. 1986; Carlson et al. 1994; McNider et al. 1994; Gilles and Carlson 1995; van den Hurk et al. 1995; Jones et al. 1998]. Although, several of these studies showed some success in the diagnosis of SM, very few of these techniques have transitioned to either an “operational framework” or have been attempted on a large spatial (regional to global) or temporal scale. One of the more promising TIR SM monitoring techniques exploits surface flux predictions retrieved from the Atmosphere Land Exchange Inverse (ALEXI) model, a two-source energy balance algorithm [Anderson et al. 1997; Anderson et al. 2007; Hain et al. 2009]. ALEXI uses a time-differential measurement of morning LST rise to diagnose the partitioning of net radiation into fluxes of HLE and soil heat flux (G). As stated above, surface flux partitioning of H and LE heat from ALEXI contains meaningful information about SM as shown by Anderson et al. [2007a; 2010] and Hain et al. [2009]. The TIR-based methodology is advantageous it provides because surface SM (surface stress) information over sparsely vegetated pixels and, importantly, root-zone SM information over pixels with moderate to dense vegetation through the detection of vegetation stress. Additionally, geostationary-based TIR sensors operate at spatial resolutions that are significantly higher than that of current microwave sensors, (on the order of 3 km [Meteosat Second Generation (MSG)] to 10 km [GOES Sounder]).

Main Results:

The two-source energy balance (TSEB) component of ALEXI partitions the total system LE flux into soil evaporation (LEs) and canopy transpiration (LEc) sub-components. These fluxes in turn are largely controlled by SM in the surface layer and the root-zone layer, respectively. In general, wet SM conditions lead to increased LE (decreased H) and a depressed morning surface temperature amplitude, while dry SM conditions lead to decreased LE (increased H) and an increased morning surface temperature amplitude. Anderson et al. [2007a] and Hain et al. [2009] outline a technique for simulating the effects of SM on LE estimates from ALEXI using a SM stress function, relating the value of the fraction of actual to potential evapotranspiration (fPET) to the fraction of available water (fAW) in the soil profile. In many prognostic-modeling frameworks, a semi-empirical linear or non-linear relationship is defined between fPET and fAW, to account for effects of SM depletion on the surface evaporative fluxes [Anderson et al. 2007a; Hain et al., 2009.] Here, we assume a linear relationship (Hain et al. 2011), and in the computation of standardized anomalies in θALEXI, soil texture-specific values ofθfc and θwpare not necessary as standardized anomalies in θALEXI are equivalent to standardized anomalies in fPET. This retrieval of SM from diagnosed evaporative fluxes (the inverse of the prognostic approach) should be reasonable when SM is between field capacity and the permanent wilting point, but will lose sensitivity as SM exceeds θfcand approaches saturation.

For ALEXI-retrieved SM, it is assumed that the contributions from the surface and root-zone SM layers are related to the observed fraction of green (e.g., actively transpiring) vegetation (fc). In general, over bare soil, ALEXI LE is dominated by direct soil evaporation and reflects SM conditions in only the first few centimeters of the profile, similar to an effective sensing depth of microwave sensors [Crow and Zhan, 2007]. However, over dense to full vegetation cover (fc greater than 75%), ALEXI LE is predominantly partitioned to canopy transpiration, and soil evaporation becomes negligible. In this case, fPET is governed by moisture conditions in the plant root zone. Between these two extremes (bare soil to full vegetation cover), ALEXI provides a composite of both surface and root-zone SM information, with relative influence related to fc. Therefore θALEXI values represent a composite of surface and root-zone SM conditions depending on surface vegetation conditions (see Hain et al. 2011), although in this case θALEXIsfc(surface, 0-5 cm SM) and θALEXIrz (5 cm to 2 m SM) are not being retrieved independently. Hain et al. [2009] evaluated ALEXI fAW retrievals in comparison soil moisture observations over a multi-year period (2002-2004) from the Oklahoma Mesonet and found reasonable temporal and spatial agreement (RMSE values around 20% of mean observed SM).

Remotely-sensed SM studies have mainly focused on retrievals using active and passive microwaves sensors whose measurements can directly be related to SM. Microwave retrievals have obvious advantages: they are physically based, can retrieve through non-precipitating cloud cover, and typically have short repeat cycles. However, microwave sensors exhibit reduced retrieval accuracy over moderate to dense vegetation, and have spatial resolutions that are too coarse for many kinds of applications. One potential avenue toward filling these informational gaps is to exploit the retrieval of SM information from TIR observations [Hain et al. 2009; Anderson et al. 2007a; 2007b, 2011], which can provide SM information under dense vegetation cover and at resolutions down to meter scales. The physical basis for such a diagnosis is that wet soil conditions lead to increased LE and reduced surface temperature amplitude, while dry SM conditions lead to decreased LE flux and increased surface temperature amplitude.
A series of analyses has been used to study relationships between SM products from a passive microwave retrieval (AMSR-E), a TIR based model (ALEXI), and a land surface model (Noah). An analysis of spatial anomaly correlations on a seasonal time scale showed that seasonal composites were spatially consistent between the three datasets, except for the degradation in AMSR-E anomaly detection over moderate to dense vegetation, mainly occurring over the eastern half of the CONUS. The ALEXI SM maps showed better spatial correspondence with Noah SM than did the microwave SM retrievals over the period studied here (2003-2008). This may have ramifications regarding the utility of TIR vs. MW-based surface moisture products used in drought monitoring, which requires spatially consistent measures of drought conditions across the monitoring domain.

A temporal correlation analysis showed that Noah and AMSR-E time-series were better correlated in sparsely vegetated (fc < 0.5) regions (western and central CONUS), while ALEXI was better correlated with the Noah reference under moderate to dense vegetation (fc > 0.5), primarily in eastern CONUS. Analysis of variations with ALEXI / Noah time series anomaly correlation with repeat cycle suggest that sampling errors may play a larger role in degrading ALEXI retrieval composites.

Hain et al. (2011) shows that a triple collocation error estimation technique, which estimates relative SM error among independent datasets, yielded average TC error assessments of 0.0211 m3 m-3 for Noah, 0.0245 m3 m-3 for AMSR-E, and 0.0261 m3 m-3 for ALEXI. The Noah TC error was smaller than ALEXI and AMSR-E over the entire observed range of fc, while AMSR-E performed better than ALEXI over low to moderate vegetation and ALEXI performed better over moderate to dense vegetation.  The TC analysis provides absolute error values from ALEXI and AMSR-E SM proxies (after re-scaling to the Noah model SM climatology). Such error values are required as input for the assimilation of either SM proxy into a land surface model.

The intercomparison of TIR-based and MW-based SM (Hain et al. 2011) has not been thoroughly investigated in the literature. The two datasets provide complementary information about the current SM state. TIR methods can provide SM information over dense vegetation, a large gap in current MW methods, while serving as an additional independent source of information over low to moderate vegetation. The complementary nature of the two retrieval methods leads to the potential of integration within an advanced data assimilation system, provided that accurate representation of relative errors are available for each method.