Search

Rainfall features of the Baiu season estimated from GMS IR images are investigated for the Japanese islands and their adjacent ocean for June 1989. The total monthly rainfall for the period is distributed in a band-like shape extending from west to east, showing its maximum on 35°N in the latitudinal section and a sudden decrease to the west of 130°E in the longitudinal section. The latitudinal distributions of rainfall relative to the Baiu front show almost the same shape with the maximum at 1.25° north of the front at the surface over both land and ocean, but the peak amount is larger over the island than that over the ocean. The temporal changes of rainfall are dominated by a period of synoptic scale for all regions over both land and ocean, and secondly, by a peak corresponding to daily variation observed only over the continent. The rainfall region moves eastward along the Baiu front with almost same phase speed of about 7.5° longitudes per 12 hours over almost all regions.

The method developed for estimating rainfall over several sample areas over the main islands of Japan from IR imagery data of GMS is extended to estimate rainfall over a wider region including the Japanese islands and their adjacent ocean. In this method, rainfall estimation over a grid of 1.25° latitudes × 1.25° longitudes is made by assuming linear relations between hourly rainfall and cold cloud fractional coverage (FC) for the 3 rain cloud types (A: Cumulus; B: Cumulonimbus and C: Middle Clouds) and assuming rain free in cases of cloud amount less than 70% as well as cloud type D (Cirrus). The threshold temperatures to define FC for the 3 rain cloud types are kept the same as those used in Xie, that is, 245 K, 235 K and 255 K for each cloud type A, B and C, respectively. The proportional constants between hourly rainfall and FC are assumed to be functions of latitude and are calculated by using the rainfall data obtained by radar-AMeDAS composite and FC data of the grids over the Japanese islands. The resultant relative errors of rainfall estimation over the Japanese islands are 73% and 17% for daily and monthly estimation, respectively, which satisfy the useful level of the accuracy of rainfall data in climatic use.

A new method is proposed for estimation of rainfall in mid-latitudinal areas from GMS infrared imagery data. This method was developed by using the satellite and gage data in 5 areas of about 10^4 km^2 in Japan, taking account of the different features of rainfall from various types of clouds. In this method, clouds are classified into six types, S(Clear Sky), F(Fine), A(Cumulus), B(Cumulonimbus), C(Middle Clouds) and D(High Clouds), by using the discriminate analysis technique from IR imagery data. The cloud types S, F and D are assumed to be no-rain cases and in the cases of A, B and C, rainfall is assumed to be proportional to the cold cloud fractional coverage (FC). A proportional constant for each cloud type is determined from the regression equation between FC and 3-hr rainfall, where the threshold T_BB used to define FC is determined to give the highest correlation for each cloud type. This method gives a good rainfall estimation for 24 hr cumulative rainfall with the correlation coefficient of 0.651 and relative root mean square error (RRE) of 0.968. To apply this method to the Hokkaido area, the results should be adjusted by an empirically determined factor, because weather conditions are different in this area.

Techniques estimating cloud amount and classifying cloud type from GMS IR and/or VIS imagery data have been developed and evaluated. Cloud amount is estimated by a Two-Threshold-Method (TTM) which takes account of the partially cloud-covered pixels. Cloud type is classified by a discriminant analysis technique. Fairly good agreement is obtained between the satellite estimated and the surface observed cloud amounts. The comparison between the satellite classified and ground observed cloud types shows correctness ratios of 52.9% and 47.7% for the daytime and nighttime respectively, but the distinction between cumulus and middle clouds seems insufficient. The techniques were applied to produce the time series of cloud amounts and types during the period from 3 to 11 Sept. 1980, and the results were acceptable.