The use of entropy optimization principles in parameter estimation: Applications to global water demand modeling.

摘要

A major challenge in global water demand forecasts is the lack of detailed and consistent databases. The IWMI-IFPRI project follows a two-track strategy: one part is the continuing process of collecting and compiling more data from alternative sources (national statistics, grey literature, expert knowledge). The other facet is dealing with data scarcity and uncertainty by extracting more information from existing data, i.e. data mining. This dissertation focuses on the data mining aspect. It explores the use of entropy optimization principles in data analysis and applies it to three aspects of global modeling.; The first application addresses the lack of reliable crop yield data disaggeegated into irrigated and rain fed production modes. Entropy principles are used to estimate agricultural production functions for Indian agriculture and trends in irrigated and rain fed crop yields. The second estimation problem deals with the disaggregation of total traded quantities into bilateral trade flows. This is essential in order to determine the impact of food trade and virtual water flows on global water use. The third estimation problem relates to reliable parameter estimation with uncertain data. This is relevant for the calibration of the hydrologic component of the global model, where both model input (climate and water use data) and model output (river flow measurement) may be corrupted by measurement errors.; Entropy based estimation methods are relatively new and not as well established as conventional regression techniques, such as Least Squares. However, when data are scarce and/or uncertain, conventional regression techniques require restrictive assumptions, which, as this dissertation shows, may be violated and, therefore, yield unreliable estimates. Entropy estimation offers a promising alternative.