Adaptation of a technique to estimate rainfall from satellite data in Bangladesh

In order to estimate rainfall from satellite data GMS (Geostationary Meteorological Satellite) Precipitation Index (GPI) was calculated for north-eastern part (Sylhet) of Bangladesh. GPI was calibrated with ground-based raingauge rainfall. As was found, GPI underestimated i


INTRODUCTION • \|
Flood is a common phenomenon in Bangladesh. Every year about one fifth of the \ country gets flooded '. The main component of flood is rainfall. Usually raingauge and radar are used to estimate rainfall. There are some problems associated with raingauges; the shape of the container, its exposure, the wind and evaporation between measurements. Also a dense network of raingauges is required to produce accurate estimates of areole averaged precipitation. Such a network is impossible over ocean and inaccessible areas. Despite problems such as variation in the reflectivity rainfall relation varying droplet size spectra and beam attenuation among others good estimates of areole averaged rainfall can be obtained using suitably calibrated digital radars 2 . However radar ranges are rather small (100-500 km approximately) and its deployment is impracticable over the ocean and it would be prohibitively expensive to have a large radar network on land. So the solution to overcome the difficulties of land-based equipment is to make use of satellite based remote sensing devices. Satellite provides data round the clock and they can monitor very large areas. Therefore meteorological satellite data are the only realistic means to monitor the spatial and temporal distribution of precipitation. Though there is inherent indirectness of satellite observable quantities (e.g., cloud top reflectance or thermal radiance) as measures of surface precipitation intensity; but cloud infrared area is highly crenulated with rain area 3 ' 4 . So these data become useful when averaged over large space and or time scales and then only when carefully calibrated for the region and monsoons in question 5 .

DATA AND PROCEDURE
We used three hourly satellite infrared (IR) data, provided by the Institute of Flood We compared satellite estimated rainfall (GPI) with raingauge rainfall in the north eastern region of Bangladesh to calibrate the GMS Precipitation Index (GPI) for this region. To get GPI the steps are given below.
Step 1. The whole region was divided into five boxes each having an area of 0.5°x0.5°.
We assigned each location a name; they are Jaflong, Chatak, Amalshid, Habiganj and Hakaluki. Each box contains 30 pixels, while each pixel was 11 kmx9 km.
Step 2. The fractional coverage of cloud 'F c ' was calculated for each 0.5°x0.5° grid cell. "F c " was defined as for each 0.5° x 0.5° cell covered by clouds whose cloud top temperatures are colder than 253K. The threshold so chosen to include all kind of precipitating clouds.
According to the definitiontotal number of cloudy pixels

(2-1) total pixels
Step 3. Estimation of precipitation is carried out using a linear a) . The rainfall estimates are referred to as GMS-5-precipitation index (GPI).
The Lineaiform of GPI is,

Where 'Fc' is the fractional coverage of cloud and 'K' is a constant related to rain rate and is taken as 3 mm/h as by Arkin and Meissner 7 . And T' is the length of averaging period in hour.
Steps J to Step 3 were repeated for each 8 images of the day. Then the total number of pixels for the day was calculated by simple addition. The daily GPI values were averaged for three-days and seven-days to produce 3-day and 7-day average GPI for each 0.5°x0.5°grid cell.
Step 4. We also calculated raingauge rainfall for the time resolution of 1-day, 3-day and 7-day period. GPI and rainfall for 0.5°x0.5° boxes were then compared with each other.
Step 5. AGPI was also calculated for the time resolution of 1-day, 3-day and 7-day period. In calculation of AGPI, we used 4 mm/h instead of 3 mm/h for 'K' in equation 2.2. The value of 'K' was found as the best fit for this study.
Step 6. AGPI and rainfall for 0.5°x0.5° boxes were then compared with each other.

Figure 3 shows the accumulation of 3-day variation in GPI and RAIN in Sylhet.
Among five stations we again found that two stations Chatak and SUST show higher raingauge rainfall than GPI.

Adaptation of GPI as AGPI
In order to get acceptable rain estimates from satellite data we may readjust constant rain rate in calculating GPI. We tried to find a better constant rain rate in calculating GPI from GMS-5 for this region. In this work we used the constant rain rate 4 mm/hour instead of 3 mm/hour used by Arkin and Meissner 7 . This new rain rate better enabled us to calculate rain from satellite IR data. In this study the new adjusted GPI is termed as AGPI which is described in this sub-section. and Hobigang were 106%, 94%, 54%, 70% and 24% of AGPI respectively. The mean percentage of RAIN was 69.6% of AGPI. This value fully satisfies our expectation. Figure 6 shows the 3-day accumulated raingauge rainfall (RAIN) and Adjusted GMS Precipitation Index (AGPI) at Sylhet. All stations except Chatak show that RAIN was smaller than AGPI, which is expected. At. Chatak we see that the 3-day accumulated RAIN was 9 mm higher than AGPI, hence RAIN was 109% of AGPI. This is a bit underestimation by satellite. The reason is discussed for daily average in Figure 5. Without this enormous rain in a day AGPI might be higher than RAIN as expected. The  and 67.2% for 3-day and 7-day respectively. The average RAIN was 68% of AGPI.
Hence AGPI gives very satisfactory rainfall from satellite data.

CONCLUSIONS
From this study we found that the percentage of raingauge rainfall was 67.8%, 68.7% and 67.2% for 1-day, 3-day and 7-day average respectively of AGPI. Hence, we may conclude that AGPI is good enough to estimate rainfall from satellite data. AGPI was applied successfully at different stations in Sylhet. In future we may expand the analysis area for whole of the country to apply this method of estimation of rainfall from satellite data. This may be our next task in near future.