Short Duration Rainfall Estimation Using GMS IR and VIS Images
Rainfall Data: Rainfall data from PAGASA telemetering stations were used in this study. These stations normally record hourly rainfall data. Due to unavoidable circumstances such as power failure, there were times when some stations were unable to record hourly rainfall. During T.C. with complete 24 hr hourly rainfall data.
Discussion of Results
The following discussion tackles the probability of rain-no rain occurrence using infrared and visible parameters half and two & half hours after the satellite pass. This will be the basis of the choice of threshold values for both parameters. The main focus of the discussion, however, is the results of the regression analysis to determine the model/s for hourly, 3-hr, 6-hr and 12-hr rainfall estimation.
Rain-No Rain Delineation
Table 1 shows the rainfall probability half and 2.5 hour after the satellite pass using infrared data during TC Ditang(19 July 1992) and T.C. Akang (14 June 1990) at 3 hourly interval. For IR brightness values of less than or equal to 190, the probability that it is going to rain within the next half hour is very low. On the otherhand, for very high IR brightness values(236-240), the rainfall probability is 87% that indicates a very high probability of rain half-hour after the satellite pass. In general, there is an increasing probability of rain from column 1 to column 7. However, the results for columns` 2,3 and 4 cannot be considered as a trend due to limited number of cases. The results for 2.5 hr after the satellite pass are somewhat similar to that of the results for 0.5hr.
The results showed increasing rainfall probability as the temperature brightness decrease (IR brightness value increases). The same relationship was found by Lethbridge (1967) on the relationship between precipitation probability and satellite data from TIROS IV. The actual probability values, however, are not the same. This may be attributed to the differences in geographical locations of areas for which estimates were made (Martin and Howland, 1986).
| IR Brightness value | <191 | 191-200 | 201-210 | 211-220 | 221-230 | 231-235 | 236-240 |
| Total no. of Observations | 70 | 23 | 24 | 30 | 84 | 65 | 70 |
| Rainfall Prob'ty 1/2 hr after sat. pass | 4.3 | 26.1 | 37.5 | 46.7 | 50.0 | 53.8 | 87.1 |
| Rainfall Prob'ty 2.5 hr after sat.pass | 15.3 | 39.1 | 75.0 | 73.3 | 63.1 | 67.7 | 97.1 |
Table 1. Rainfall probability (in %) based on infrared brightness values.
The rainfall probability half and two and half hours after satellite pass using visible brightness of T.C. Ditang and T.c. Akang are shown in Table 2. The rainfall probability for both cases increases as the visible brightness values increases. However, the results are not conclusive since the number of cases for visible brightness value greater than 160 are very limited. Most of the visible brightness values are less than 160 where the rain fall probability is only 14.5% for half hour and 23.6% for 2.5 hour after satellite pass.
| Visible brightness | <160 | 160-170 | 171-180 | 181-190 | 191-200 | 201-210 | >210 |
| ½ hr after sat pass | 14.5 | 28.6 | 33.3 | 80.0 | 87.5 | 76.9 | 100.0 |
| 2& ½ after sat pass | 23.6 | 42.9 | 50.0 | 80.0 | 100.0 | 100.0 | 100.0 |
| No. of observation | 55 | 7 | 6 | 5 | 8 | 13 | 9 |
Table 2. Rainfall probability (in %) based on visible brightness value.
Model Development
The following discussion centered on the development of models for short duration rainfall estimation using infrared and visible satellite date of T.C. Ditang for twelve hours (July 19, 1992) at 3 hour's interval. This involves four IR and four VIS satellite pictures taken at the following periods; 0031, 0331, 0631 and 0931.
The correlation analysis of average visible brightness (AveVis) with hourly and 3-houly rainfall is much higher than the correlation between IR(AveIR) and rainfall from the 3x3 to 21x21 pixel grid. For the variance of IR and VIS at different grids, the correlation coefficients are negative for hourly as well as 3-houly. This means that as the variance increases (or the less smooth the image is), the rainfall decreases.
