Estimating hill country rainfall without full data sets for the Manawatu River catchment : a thesis presented in partial fulfillment of the requirements for the degree of Master of Resource and Environmental Planning at Massey University, Turitea, New Zealand
Nowadays, people anticipate floods using flood warning systems, and building stop banks and flood ways in place that use flood models generated with hydrological information in their design. Nevertheless, various regions in the world are still hit by floods with catastrophic effects to urban areas, because of a lack of local hydrological knowledge, especially of upstream areas in their catchments. This lack of hydrological knowledge is a result of difficult accessible highly elevated upstream areas, which makes monitoring of hydrological variables difficult or impossible.
This thesis examines models for determining montane rainfall using spatial estimation methods and data sets. The distribution and quantity of montane rainfall were assessed by applying five appropriated spatial estimation methods, data of historical and current rain gauges, and a performance measurement.
The methodology applied to gain more knowledge about montane rainfall was established with the results of a literature analysis of 40 articles about montane rainfall. This literature analysis revealed that ordinary kriging is the most frequently applied spatial estimation method for montane rainfall, with regression and regression kriging completing the top three of the most applied methods. Also, two other spatial estimation methods, empirical Bayesian kriging and geostatistical simulation, performed well with rainfall data. The same literature analysis disclosed that the root mean square error was predominantly used as a performance measure of spatial estimation methods.
The literature analysis revealed a number of data gap-filling techniques, with the inverse distance weighting method and the coefficient of correlation weighting method as the two most suitable techniques. These techniques were applied to complete historical rainfall data sets and their performance was compared within this research. The result showed that the coefficient of correlation weighting method outperformed the inverse distance weighting method in 74% of all data gap estimations, and the coefficient of correlation weighting method was 22% more accurate (based on the overall performance) than the inverse distance weighting method. The most accurate data gap-filling technique, the coefficient of correlation weighting method, was used to complete the historical rain gauges data.
The overall ranking of the spatial estimation methods revealed that Gaussian geostatistical simulation performed the best. Regression kriging was the second best spatial estimation method, but there was no significant difference with Gaussian geostatistical simulation. At the same time, the results showed that the best performance of the spatial estimations was accomplished without the maximum number of rain gauges. However, better visual
representation of the distinct pattern of rainfall was generated with the historical rain gauges in the second and third experiment of the spatial estimations.
Finally, this research discussed the factors that can impact the performance of the spatial estimations. Two of these factors were the removal of ?bad data? and the the strategic placing of rain gauges. The results of this research clearly defined that the removal of ?bad data? increased the accuracy of estimation, while a more even and strategic distribution of rain gauges was suggested to increase the accuracy of the spatial estimation of rainfall.