Begin Main Content

Home | Soil Moisture | Skin Temperature | Snow | References

Soil Moisture Assimilation

The GMAO land assimilation system is based on the Ensemble Kalman filter ( Reichle et al., 2002b) and was successfully tested with a historic record of soil moisture retrievals, including the development of bias reduction methods. Bias removal is an integral part of any assimilation system and typically relies on knowledge of the correct climatology of the variable of interest. Yet for soil moisture there is no universally accepted climatology ( Reichle et al ., 2004). Many factors contribute to this lack, including the absence of a global in situ observation network and the large uncertainties associated with modeling and retrieving soil moisture. The situation is further complicated by the fact that the data record of AMSR-E . currently the only operational sensor that produces reliable soil moisture estimates . spans only a few years . This means that even the climatology of soil moisture retrievals from the sensor is difficult to obtain, whether or not it correctly describes the true soil moisture statistics. Nevertheless, we were able to show how biases can still be addressed even with short satellite records through the ergodic substitution of variability in space for variability in time (Figure 1, Reichle and Koster , 2004).

(click image for larger view)

Figure 1: Difference in the 1979-1987 (top) mean and (bottom) standard deviation of satellite and model soil moisture (left) before and (right) after scaling. Units are volumetric moisture percent (absolute soil moisture typically ranges between 0 and 0.5). Scaling of the satellite soil moisture is based only on data from a single year (1979) using the ergodic substitution of variability in space for variability in time. From ( Reichle and Koster , 2004).

In a retrospective analysis ( Reichle and Koster, 2005) , bias-corrected, global retrievals of surface soil moisture from the Scanning Multichannel Microwave Radiometer (SMMR; 1979-87) ( Owe et al. , 2001 ) were assimilated into the NASA Catchment land surface model as it is driven with surface meteorological data derived from observations ( Berg et al. , 2003 ) . Validation against ground-based measurements from the Global Soil Moisture Data Bank ( Robock et al ., 2000 ) in Eurasia and North America demonstrated a long assumed (but rarely proven) property of soil moisture fields derived from data assimilation . that the assimilation product is superior to either satellite data or model data alone (Table 1).

(click image for larger view)

Table 1: The table demonstrates that the assimilation product is superior to either satellite data or model data alone and shows average time series correlation coefficients R with in situ surface and root zone soil moisture for satellite (SMMR), model, and assimilation estimates with 95% confidence intervals. N denotes the number of locations with sufficient data in North America and Eurasia that contributed to the average. Also shown are confidence levels that R for assimilation estimates is higher than R for satellite or model data alone. From ( Reichle and Koster , 2005).