Abstract:
Rainfall is a hydrological phenomenon that varies in magnitude in space as well as in time and requires suitable tools to predict values in space and time. The identification of spatial rainfall pattern is an essential task for hydrologists, climatologists as well as regional and local planners and managers. This is due to the variability of both the temporal spatial distribution of rainfall. Multi-day rainfall events are an important cause of recent severe flooding in the India in general and in Andhra Pradesh in particular and are required for the design of structures such as dams, urban drainage systems and flood defences and cause failures to occur. Daily rainfall data from a network of 51 one degree gird stations for the period 1951-
2003 has been in used for the study. On the basis of rainfall data for the heaviest storms that occurred in different part of Andhra Pradesh during the period 1951-2003, estimates of one-day, two-day, three-day, five-day and seven-day rainfalls were made. The main objectives of this work are: (1) to analyze and model the annual maximum rainfalls of various durations (2) to analyze and model the spatial variability of rainfall, (3) to interpolate kriging maps for different durations, and (4) to compare prediction errors and prediction variances with those of kriging methods for different durations. L-moment ratio diagrams have been used to identify candidate regional distribution of the data. Generalized Extreme Value distribution found to be the representative distribution. Parameters were estimated using maximum likelihood method. Return period quantiles were estimated using the fitted distribution for each station, Using these rainfall estimates, a geo-statistical analysis was performed. Rainfall surfaces have been predicted using ordinary kriging method. It was observed that the rainfall data is skewed and Box-cox transformation has been used for converting the skewed data to normal. It was found that one-day peak rainfall over the region varied from 25 mm to 360 mm. It is observed that the trend is present in all the cases, the first order polynomial fits well for all durations No-directional effects were observed in the region. The Spherical model fits well for higher order return periods of 50, 100 and 200 year return periods for one day and two day durations and for all return periods for five day and seven
duration storms, where as the Gaussian model fits lower return periods up to 20 year for one-day, two-day duration storms. Fitted model resulted in a Mean Error (ME) varied in the range of —1.85 to —0.07, (which is very near to zero), Mean Square Error (MSE) altered in the range of 22.2 to 90.8 , (which is very low as compared to the variance of the data), Kriged Reduced Mean Square Error (KRMSE) of changed from 0.9547 to 1.09 , (which is very near to 1) and a Kriged Reduced Mean Error (KRME) varied in the range of —0.165 to —0.0153, (which is near to zero) for one-day duration events. The exploratory data analysis, variogram model fitting, and generation of prediction map through kriging were accomplished by using ESRI'S ArcGIS and geostatistical analyst extension.
