Assimilation of Land Surface Data

Abstract

(1969) first suggested combining current and past data in an explicit dynamical model, using the model’s prognostic equations to provide time continuity and dynamic coupling amongst the fields (Figure 1). This concept has evolved into a family of techniques known as four-dimensional data assimilation. “Assimilation is the process of finding the model representation which is most consistent with the observations” (Lorenc, 1995). In essence, data assimilation merges a range of diverse data fields with a model prediction to provide that model with the best estimate of the current state of the natural environment so that it can then make more accurate predictions. The application of data assimilation in hydrology has been limited to a few one-dimensional, largely theoretical studies (i.e. Entekhabi et al., 1994; Milly, 1986), primarily due to the lack of sufficient spatially-distributed hydrologic observations (McLaughlin, 1995). However, the feasibility of synthesizing distributed fields of soil moisture by the novel application of four-dimensional data assimilation applied in a hydrological model was demonstrated by (1998). Six Push Broom Microwave Radiometer images gathered over the United States Department of Agriculture, Agricultural Research Service Walnut Gulch Experimental Watershed in southeast Arizona were assimilated into a land surface model using several alternative assimilation procedures. Modification of traditional assimilation methods was required to use these high-density Push Broom Microwave Radiometer observations. The images were found to contain horizontal correlations with length scales of several tens of kilometres, thus allowing information to be advected beyond the area of the image. Information on surface soil moisture was also assimilated into the subsurface using knowledge of the surface-subsurface correlation. Newtonian nudging assimilation procedures were found to be preferable to other techniques because they nearly preserve the observed patterns within the sampled region, but also yield plausible patterns in unmeasured regions, and allow information to be advected in time.