The aim of the proposed research is to analyze and evaluate several Artificial Neural Networks approaches to the interpolation of daily precipitation over space and time. The work will attend to the following objectives:
- Consideration of appropriate date, weather classification and terrain variables and neighboring observations in both time and space as predictors of rainfall
- The design of an appropriate back propagation network structure for interpolating daily precipitation that accounts for trend and covariance within the data
- Independent testing of results using withheld data, particularly in relation to the constant drizzle problem affecting traditional local interpolation methods
- Comparison of results with those of advanced mathematical interpolators
- Investigation of a combined network for the estimation of daily temperatures in conjunction with precipitation. The research will focus on the development of network models to estimate precipitation as a function of both trend and covariance. A formal comparison of ANN versus advanced mathematical interpolation techniques will be done for daily precipitation. Rigorous testing of the method against state-of-the-art traditional interpolation methodologies, using independent data and uncertainty metrics, is intended to provide quality assurance.