Flow prediction in data scarce catchments : a case study of Northern Thailand

• Main Author: Visessri, Supattra
• Other Authors: Maksimovic, Cedo; McIntyre, Neil
• Published: Imperial College London 2014
• Subjects: 624
• Online Access: http://ethos.bl.uk/OrderDetails.do?uin=uk.bl.ethos.656699

Flow time-series data are crucial for water resources and floods management. In catchments where flow observations are not available or data are of poor quality, a method for modelling flow time-series is needed. The overall objective of this study is to assess the applicability of recent regionalisation methods to predict flows in tropical monsoon-dominated catchments where hydrological response is particularly variable over the seasons. This PhD thesis addresses six primary challenges in the context of such catchments: 1) data quality, 2) rainfall estimation in mountainous catchments, 3) using regression for regionalising rainfall-flow response indices, 4) catchment non-stationarity, 5) conditioning rainfall-runoff models and 6) uncertainty analysis. The main novel contributions are: developing a data quality scoring system and exploring its effects on modelling; comparing a customised technique for rain gauge interpolation and using satellite products for spatial rainfall estimation; demonstrating practical difficulties in predicting land use change impacts; assessing the performance of recent conditioning methods and estimating prediction uncertainty in monsoonal areas. The main practical outcomes are: 1) the most in-depth study yet published of methods for predicting flows in northern Thailand for water resources planning; 2) recommendations towards improving data support for water resources estimation in Thailand. Using data from 44 gauged sub-catchments of the upper Ping catchment in northern Thailand from the period 1995-2006, three relevant flow response indices (runoff coefficient, base flow index and seasonal flow elasticity) were regionalised by regression against 14 available catchment properties. The runoff coefficient was the most successfully regionalised, followed by base flow index and lastly the seasonal elasticity of flow. The non-stationarity (represented by the differences between two six-year sub-periods) was significant both in the flow response indices and in land use indices; however relationships between the two sets of indices were weak. The regression equations were not helpful in predicting the non-stationarity in the flow indices except somewhat for the runoff coefficient. Rainfall estimation errors from two different estimation methods were large and believed to significantly contribute to uncertainty in regionalised flow response indices and modelled flow time-series. The three regionalised flow response indices were used individually and in combination to condition the IHACRES rainfall-runoff model using a Bayesian approach. The runoff coefficient was the most informative index. This is followed by the base flow index and lastly the seasonal flow elasticity. Using the variance of the regression coefficients and of the regression residuals had limited success in estimating the flow uncertainty intervals because uncertainty from the IHACRES model structure is not sufficiently represented by the variance of the regression. The regionalised model was considered to be too imprecise at the daily time scale but offers good support to water resources planning at the monthly and seasonal time scales. A partly subjective data quality scoring system showed the clear influence of rainfall and flow data quality on regionalisation uncertainty. Recommendations include developing more relevant soils databases, improved records of abstractions and investment in the gauge network.