Assistant Technical Representative: John E. Janowiak Contributors: Robert J. Joyce Period of Performance: November, 2001 1. WORK PERFORMED DURING THE PAST MONTH A. Precipitation derived from the AMSU-B sensor aboard both the NOAA-15 and NOAA-16 satellites has a limb problem which is most often characterized as "blocky" precipitation overestimation yielding strong discontinuities parallel to the track of the satellite, reflecting strong satellite zenith angle gradients at the edge of the swath. After careful inspection of the AMSU-B limb problem, a method was developed to trim the edges of both the NOAA-15 and NOAA-16 swaths. For the NOAA-15, the original swath width is 19 degrees of longitudinal coverage at the equator, and increases at ~ 1/cos(lat) toward the poles. A 130 km (~1.2 deg longitude at equator) trim from each side of the NOAA-15 swath boundary is used to remove the discontinuities, (seen in the below mentions eps files) this means more pixels at the mid-latitudes than at the equator. For the NOAA-16 which has a original swath width of 22 degrees of longitudinal coverage at the equator, the limb problem extends a little more earth distance into the swath, thus a 150 km (~1.35 deg lon at equator) trim is used. About ~85-88% of the original earth coverage of the AMSU-B derived precipitation remains. Three eps files were uploaded to the ftp.ncep.noaa.gov, in the data/rjoyce/amsu-b_trimmed directory. Two show the before and after of the AMSU-B trim correction at different latitudes, the other eps file is a larger coverage view giving a good indication of the remaining coverage. B. The next task is to tackle the problem of requiring the geostationary satellite ID (and IR information to be used) is the same for each 5x5 latitutde /longitude regions within each pair of successive IR images to be used for spatial lag correlation. To begin this process, code has been added which obtains geostationary satellite ID from the 0.5 degree IR products along with the full res IR images in the software used to derive the zonal and meridional half-hourly advection vectors from IR and then advects microwave derived precipitation. Next we need to set criteria to determine geostationary satellite advection zones, logistically how to derive half-hourly advection vectors from one-hourly imagery of the GMS, and how to deal with temporal gaps in geostationary satellite data for purposes of deriving advection vectors. C. Created software enabling development of daily, global, most frequently used geostationary satellite ID, 5 degree latitude/longitude resolution templates using CPC's half hourly merged IR images. Along with the most frequent satellite ID is the number of images that the satellite was the dominant satellite in the region for that day. Although for any image time, two satellites may contribute information within a particular 5 degree region in overlapping coverage, the satellite with the greater coverage within the region is determined as dominant. Resulting templates for 6, 7, and 8 August 2001 yielded over 45 images per day for the METEOSAT-5 and 7 domains as well as northern hemispheric GOES-8 and GOES-10 northern hemispheric sector images, 35-43 in the southern hemispheres. The division of the most frequent satellite used within these regions is and should be the midpoint between sub-satellite positions. However since GMS-5 scans only hourly images on a regular basis, and less frequently in the southern hemisphere, MET-5 and GOES-10 (conic sector region only) are the most used satellites in their southern hemispheric overlap coverage with GMS-5 which averages only 14-16 images in the area. D. Inspection of satellite ID in sequential half hourly merged IR images the within southern hemispheric region of the GMS domain reveals the irregular temporal sampling of the GMS-5 scanner. Typical image times of the GMS-5 southern hemispheric sector would be : 2:30, 4:30, 5:00, 5:30, 8:00, 10:30, 11:00, 11:30, 14:00, 17:00, 17:30, 18:00, 20:00, 22:30, 23:00, and 23:30 GMT. Unfortunately GMS-5 is the most dominant satellite in the southern hemispheric region between MET-5 and GOES-10 coverage, with these infrequent, irregular scans. This deficiency leads toward studying spatial and temporal interpolation of half hourly advection vectors derived from successive half hourly images (and hourly GMS images). E. Set up software to average 1 to 6 half hourly advection vector arrays to then correlate with a future vector array. Among other problems with this approach, the result is one number per correlation, thus regional properties are not shown. However this work is not lost because the software needed to compare the spatial and temporal interpolation properties of the advection vector fields (#2 below) will be modified from this attempt. F. Developed software enabling the temporal interpolation of half hourly zonal and meridional IR derived advection vectors (sections 2.A and 2.C in the Work Planned section of last week's report). The method computes the temporal interpolation of both zonal and meridional advection vector components for a controlled spatial domain (individual satellites) at hourly interpolation increments (half hourly past increment combined with half hourly future increment) to be correlated with the original advection vectors for each half hourly advection frame. Correlations from successive half hourly advection periods are averaged to smooth values for each hourly interpolation interval. G. Tested temporal advection software for the 6-8 August 2001 period for the GOES-10, METEOSAT-7, and METEOSAT-5 satellites for hourly temporal interpolation intervals ranging from 1- 48 hours. Since the total data domain period is 72 hours only 48 half hourly vector frames (00:00 GMT - 23:30 GMT 7 August) were interpolated and correlated to the original vectors. Results of correlations are broken down to zonal and meridional components for the following individual satellites: Zonal Meridional Time Interval (hours) GOES-10 MET-7 MET-5 GOES-10 MET-7 MET-5 1 .974 .948 .970 .931 .848 .911 2 .968 .927 .966 .940 .801 .898 3 .959 .918 .955 .925 .742 .863 4 .955 .917 .955 .933 .737 .866 5 .952 .909 .947 .920 .690 .842 6 .950 .905 .947 .911 .702 .848 The results indicate high correlations for the temporal interpolation of zonal advection vectors for the six hourly intervals shown. Values for the GOES-10 domain are possibly higher because of a smaller spatial domain used, thus fewer 5x5 degree latitude regional vector pairs used in the correlations. Correlations for meridional vectors are somewhat smaller but not substantially. Generally meridional vector amplitudes are considerably smaller than zonal and spatial patterns less distinct, so the fact that the temporal correlations are smaller is not surprising. What is strange and a bit of concern is the values for the 4 hour interval compared to 3 hours and 6 hour interval compared to the 5 hour interval is nearly the same (or in some cases higher) for all satellites, this needs a little investigating, after which, GrADS plots need to be created. This study also needs repeating for a different time period, possibly when we get both the even and odd satellite high resolution IR imagery needed in order to compare collocated vector fields derived from separate satellites. H. Refined software temporally interpolating half hourly zonal and meridional IR derived advection vectors by adding logic to resolve interpolations ending in .5 increments. Since advection vectors are discrete, either rounding up or down previously biased the interpolated vectors, affecting correlation with original vectors. The biased was eliminated and temporal correlations slightly increased by choosing to round the interpolated vector towards the amplitude of the vector (of the two used in the temporal interpolation) with the higher spatial lag correlation. Since the temporal distance from both the past and future value used in interpolation is the same (currently fixed) it makes sense to use the vector with higher spatial lag correlation to resolve the rounding issue. I. Tried to determine why correlations of 4 hour temporal interpolations are nearly the same as those of 3 hours, and correlation of 6 hour interpolations as high as those 5 hour interpolations (check last week's report tables). The strange thing is that correlations at the 1 hour interval (both zonal and meridional) are always the highest, followed by the 2 hour interval. A 1 hour interval means that for missing 12:00-12:30 GMT advection vectors, interpolation is performed between the 11:30-12:00 GMT and the 12:30-13:00 advection vectors. Note that the intervals (even hours) with slightly higher than expected correlations are temporally interpolated from vectors derived from IR imagery scanned at the same minutes of the hour as the advection period for which the interpolation is being performed, i.e. correlation of 10:00-10:30 GMT and 14:00-14:30 GMT interpolated vectors with the original 12:00-12:30 GMT vectors is about the same as the relationship of 10:30-11:00 GMT and 13:30-14:00 GMT interpolated vectors with original 12:00-12:30 GMT vectors, both zonally and meridionally. There is no difference in these relationships if the interpolations are separated for beginning of the hour and end of the hour advection vectors, either in advection vector amplitude, correlation amplitude, or correlation dominance relative to interval. J. Developed software spatially interpolating zonal and meridional IR derived advection vectors for a specified region within a particular satellite's domain. By using the temporal interpolation software as the beginning template, convenient inter-comparisons of temporal and spatial interpolation can be made over collocated domains. The interpolated values were then categorized by the distance of the nearest value used in the interpolation process by ~550 km intervals, the resolution of vectors at the equator (~5 degrees latitude/ longitude). This work will falls under the 2.B and 2.D sections below. Resulting correlations of spatial interpolation show good correlation (~0.70 - 0.9) in the first few distance intervals for zonal advections vectors for the GOES West, METEOSAT-7, and METEOSAT-5 satellites, but not as high as temporal interpolation correlations for the first few hourly intervals (~0.90-0.98). Spatial interpolation correlation values for meridional advection vectors (~0.4 - 0.7) in the first few distance intervals were considerably smaller than temporal interpolation (0.75-0.93). For both zonal and meridional IR derived advection vectors, correlation drops off much quicker with spatial interpolation relative to temporal interpolation, leaving usefulness of spatial interpolation to cases when a geostationary satellite is inoperative for a substantial period of time. 2. WORK PLANNED FOR NEXT MONTH For the next week, even and odd satellite configuration, full resolution IR imagery will be archived for determination the relationship of collocated advection vectors produced by different satellites. The even and odd satellite configuration preserves the full coverage of each satellite allowing simultaneous overlap between neighboring satellites. Software will then be modified in order to produce advection vectors from more than one satellite and for comparison. This work will fall under the 2nd section below. A. The overall goal is to get the best half hourly global advection field, at all locations. 1. The assumption that this must begin with preforming spatial lag correlations at lag of only a half hour between successive images (or hourly only in GMS domain) must be proved. Show the decay in correlation by increasing the temporal lag between successive images. The resulting spatial lag correlation at each location should be averaged for each temporal lag increment. Also, smoothed maps of temporally lagged correlation are produced by averaging several maps of correlation at the same lag for a view of regional dependency. 2. Compare collocated vector fields derived from separate satellites. In order to do this parallax and zenith angle corrected full resolution IR from both the even and odd geostationary satellite configuration must be written to disk. This is necessary in order to perform spatial lag correlation separately for both geostationary satellites in their overlap regions for the same period of time. Comparisons of these vector fields derived from separate satellites will determine if an additional satellite will improve the vector field, in an overlap region, if the dominant satellite displays large temporal gaps in imagery. These results should be compared to spatial and temporal interpolation properties of single satellite derived vector fields (2B&C) to determine usefulness. 3. Repeat procedure tallying all IR images per day at 5.0 degree latitude/longitude region for the dominant satellite by including all satellites observing the region.