Extreme Rainfall Estimation at Ungauged Locations: Information That Needs to be Included in Low-Lying Monsoon Climate Regions

Design rainfall is associated with the estimation of rainfall depth for a particular duration in terms of return period. The estimation of such at high return periods, often referred to as frequency analysis, is required in hydrological modelling which subsequently helps in managing flood risk.

Annual maximum series which satisfies the IID (independent identically distributed) concept is generally used to perform frequency analysis. However, problems arise when an estimate to be made at a target site where the data is unavailable or the site is ungauged.

This study considers two approaches that are commonly used in ungauged estimation: the regional ROI (region-of-influence) and the interpolation method, and investigates what information does it require to have reliable estimate in both cases. A comparison is also carried out between those methods.

Both approaches incorporate generalized extreme value (GEV) based index-flood estimation procedure in which the growth factor is used as the means of comparison.

Bangladesh, a tropical monsoon climate region, is used as a case study where annual maximum daily rainfall data at 34 stations have been employed to assess the performance.

Several ROI schemes which differ in their association with different site descriptors (e.g. elevation, location, annual average monsoon rainfall, annual average rainfall) are assessed for its suitability in ungauged cases whereas popular interpolation techniques: inverse distance weighting (IDW) and kriging are examined to find an appropriate model for the same purpose.

The geographical location based ROI scheme which is suitable in ungauged conditions shows promising results. The scheme has successfully passed the homogeneity test and showed the unbounded characteristics of frequency model which is required in practical applications. The ordinary kriging (OK) is found to be superior to the IDW method and the kriging with external drift that can take information other than location, in terms of cross validation error measures.

Regarding comparison between OK and ROI, both methods show a similar performance, indicating that both can be used for ungauged estimation. The overall results suggest that the spatial information about rainfall is an important factor in terms of formation of governing character of extreme rainfall in a low-lying region like Bangladesh.