Flaviana D. L. Hilario
Point estimates of hourly, 3-hr. 6-hr, 12-hr and 24-hr rainfall from tropical cyclones were computed using 3-hr interval of infrared and visible images from the Geostationary Meteorological Satellite (GMS) of Japan. Multiple regression models involving IR, VIS and combination of the two parameters were developed to estimate short duration rainfall Independent data were used to validate the models.
For very short duration rainfall estimation (hourly and 3-hr), the correlation between the observed and the estimated rainfall from IR only models is very low. However, it was observed that high rainfall values are always associated with high IR brightness value but high IR brightness value is not always associated with high rainfall values. On the other hand, the visible parameters were found to be highly correlated with very short duration rainfall. This could be attributed to the presence of cirrus anvil which is more sensitive to the infrared than the visible measurements.
The total time period of the rainfall estimation appears to be an important factor in the accuracy of the calculations particularly with models involving infrared parameters only. The scatter of hourly and 3-hr rainfall estimates are relatively large, however, the scatter decreases as estimates are accumulated over longer period of time. For models with visible parameters, the scatter also decreases as the period of accumulation increases.
In general, the estimated rainfall for the periods hourly, 3-hr, 6-hr, 12-hr and 24-hr showed overestimation for lower rainfall values and underestimation for higher rainfall values. The results also showed that as the period of accumulation of rainfall increased, the correlation between the observed and the estimated rainfall also increased.
Validation of the five models (with IR parameters only) using independent data sets (T.C. Akang) yielded slightly lower correlation coefficients for 3-hr to 12- hr rainfall estimation. The 24-hr estimated rainfall has much lower correlation coefficient which can attribute to the limited number of samples used in the model development. Models with visible brightness as parameter adequately estimated 3-hr, 6-hr and 9-hr rainfall of T.C. Akang. Validation of the 3rd order polynomial of IR for 6-hr and 12-hr rainfall showed relatively poor performance of the model. The observed rainfall from the synoptic stations are much higher than the rainfall used in model development which could account for the low skill of the model.