Principal Component Analysis of Hydrologic Data

Abstract

Analysis and modeling of the characteristics and regionalization of hydrologic data in which both the spatial and the tempora1 components are simultaneously considered has recently received increased attention. Two problems related to the analysis of spatial-temporal data are discussed in this paper. The first of these relates to the predictability of spatial-temporal data over a region. Although the predictability of a process measured at a location has received considerable attention in time series analysis literature, the predictability of a process over a region has not received commensurate attention. Methods to estimate regional predictability are discussed and some examples are presented. These examples deal with rainfall, runoff and droughts over a region. The time scales considered are months and years. The seoond problem considered is that of representation of spatial cause-etfect data. Principal component analysis may be used to develop principal daIum and estimated patterns whim bring out the causa1 connectiom between two sets of spatial data. The theory of computation of estimated pattems is discussed. Examples of the use of this method to investigate the causa1 connections between spatial rainfiill-runoff and rainfiill-drought data are presented.