ABSTRACT
The processes underlying year-to-year variability in almost any ecological variable are typically numerous and nonlinear. Because of the relatively short length of most instrumental records – commonly several decades or less – the data do not readily divulge these mechanisms. Here, we outline a simple practical procedure for analyzing such records. The procedure has evolved during the course of data analyses for several lake and estuarine sites in California. It consists of (1) decomposing spatial-temporal series into individual modes of variability using a special application of rotated principal component analysis, (2) exploring the qualitative relations between individual variability modes and plausible causal factors using general additive models, and (3) parameterizing the relations for hypothesis testing. The parameterization is seen primarily as the final step in identifying dominant mechanisms for inclusion in a predictive model, whether numerical or statistical, not as a predictive tool in itself. Nonetheless, it might be useful at times to a1Tive at a simple predictive statistical tool by (4) choosing from among plausible parameterizations 286identified in Item 3 the one that minimizes the prediction error. These steps are briefly illustrated by specific applications from Lake Tahoe, Castle Lake, and the northern San Francisco Estuary.
