NON-LINEAR, MULTIVARIATE FORECASTING OF HYDROLOGIC AND ANTHROPOGENIC RESPONSES TO METEOROLOGICAL FORCING

摘要

Water resource managers often need to forecast natural system conditions for optimal resource allocation. Even though detailed meteorological forecasting over weeks and months is impractical, hydrologic behaviors such as groundwater cycling can transpire over months and years. Alternatively, man-made system behaviors both lag and lead causal forcing, e.g., seasonal weather changes. This paper compares forecasting the behaviors of two systems - the upper Klamath Basin in Oregon where resource managers allocate water among hydropower, farming, and fisheries, and a South Carolina water utility whose demand varies significantly with seasonal irrigation. A similar approach was used to model both systems. Meteorological, basin streamflow, and consumer demand signals were decomposed into components that differentiated standard (seasonally periodic) behaviors from non-standard chaotic behaviors. Empirical process models then were synthesized using multi-layer perceptraon artificial neural networks (ANN), a non-linear, multivariate curve fitting technique. The ANNs predicted chaotic output behaviors (basin streamflow or consumer demand) from chaotic meteorological inputs. Finally, prediction sensitivity to shifting the output forward in time relative was determined. The results for the purely natural Klamath and the anthropogenic water demand systems were found to be instructively dissimilar.