Prediction versus Understanding (The 2006 Ven Te Chow Lecture)

Abstract

flow hydrographs and peak flows on the time scales of individual rainfall-runoff events. This in contrast to the statistical approach, which was spatial. The two approaches were developed to solve practical engineering problems. The regional statistical analysis of annual flood frequencies was designed to predict floods of a given return period for highway, bridge, and spillway design at ungauged locations, and rainfall-runoff models were developed for planning and project design. The dichotomy in the two approaches to hydrologic analyses continues until today in hydrologic research, education, and practice. It was also recognized in the early 1960s that the two approaches must be closely connected because physical rainfallrunoff mechanisms generating floods were the basis for the flood values used in the statistical analysis. For example, federal agencies other than the USGS have often used deterministic rainfallrunoff analyses to determine flood frequencies. An intercomparison of use of the various models for the prediction of flood frequencies was undertaken, and it was shown that the USGS model predicted with less bias and less variance than any of the other models. Thus, once again, better understanding did not lead to better prediction. However, a general theoretical framework for connecting the two, i.e., how the physics of the system could be used to predict spatial statistical parameters in quantile analysis was not known, and tools had not been developed to address this fundamental hydrologic question. The parameters used in the regression models were surrogates for the physics underpinning the prediction. However, it was not recognized at the time that the problem of connecting the two approaches was a nonlinear geophysical problem belonging to hydrological sciences rather than a practical engineering problem belonging to hydrologic engineering and that it required the development of new ideas and new tools.