Mul model Simula on of Water Flow in a Field Soil Using Pedotransfer Func ons

1 S fl ow simulations have a multitude of applications at various scales, from irrigation scheduling to obtaining global climate change estimates. Modeling of fl ow in variably saturated soils requires water retention and hydraulic conductivity parameters that are impractical to measure for largescale projects. As opposed to saturated fl ow, the nonlinearity of the unsaturated fl ow constitutive (hydraulic) properties seriously complicates calibration of the variably saturated fl ow models against fi eld monitoring data. Pedotransfer functions are routinely used to relate the hydraulic parameters to readily available data on soil properties that can be found on soil maps or extracted from soil survey reports. Since pedotransfer functions are empirical regressiontype relationships, their accuracy outside of their development region is essentially unknown. A wealth of pedotransfer information has recently accumulated in nearly all parts of the world. Unfortunately, no good method currently exists to decide which pedotransfer function model should be used for a specifi c site or application (Pachepsky and Rawls, 2004). Climate predictions faced similar uncertainties in model selection in the 1980s. To deal with these uncertainties, multimodel prediction emerged as a popular technique in climate prediction (Barnston et al., 2003; Palmer et al., 2000, 2004; Shukla et al., 2000). Th e objective of multimodel prediction is to reduce modeling errors by combining forecasts of various independent models. Since its introduction by Bates and Granger (1969), multimodel prediction has been subject to much debate that can be summarized into questions: (i) is a multimodel prediction better than the single best forecast, and (ii) what is the best approach to weigh predictions obtained with the diff erent models? Regarding the fi rst question, several studies (Kharin and Zwiers, 2002; Krishnamurti et al., 2000; Palmer et al., 2004) have reported systematic improvements in the forecasts with multimodel ensemble prediction, while others have argued that the benefi t of multimodel prediction in seasonal predictions is marginal compared with that of using the best single model (Doblas-Reyes et al., 2000; Graham et al., 2000; Peng et al., 2002). Regarding the second question, some studies have demonstrated that proper weight selection leads to relatively better forecasts compared with simple averaging of the predictions from Mul model Simula on of Water Flow in a Field Soil Using Pedotransfer Func ons