Geosdas Scm Study

The current SCM (Successive Correction Method) routine in GEOSDAS seems to leave room for improvement. This report first shows the problem with the current SCM smoothing solution, and then introduces an iterative algorithm and a Sigmoid function to improve the solution. Parameters in the iterative algorithm are studied in this report. The overall performance and accuracy of the iterative method exceeds the current SCM routine. The Current Result The precipitation simulation in the GEOSDAS employs a single column model as a simulation base. After the analysis is done, the analysis increment is in the form of vertical profile (column); it requires a mathematical routine to smooth the increment solution on the horizontal grid. This is currently done by a routine called SCM (Successive Correction Method). In this report, an offline SCM routine with a given data set is studied. Figure 1 shows the input increment data and the SCM results at the 55 computational level, which is near the ground level (1000mb). The computational resolution is 1x1.25 degrees. It is clear that there are four bands of increment data (input of SCM) in Figure 1a. Between the four bands of data there is a gap in which there are no increment data. The SCM result is shown in Figure 1b. Here, four distinct bands are seen, as well, after the smoothing. This Figure 1. Input satellite observation data and the smoothing. (a) (b) indicates that the current SCM scheme may have difficulty in the region where no input data are available. Zooming into the region near the gap (see Figure 2) it is obvious that the current method may not give accurate results as it should be. Figure 2 overlays the increment data and the SCM solution within a smaller region. The isolated “red” data (see the arrow) has a value of 0.006, but the SCM has a value of 0.002 in that location. In addition, the clifflike feature indicates that there may be problems in the SCM routine when input data are insufficient. SCM Introduction The algorithm of the current SCM routine is as follows: Given a set of input data for each computational grid point, i r r , define a maximum radius of influence (R). Within the radius of R, there are Ki numbers of input data. Let ) ( k o r f r be the input data at location k r r . Then, the smoothing solution at i r r is given by