No abstract

Variable sensitivity and error variance of six evaporation and evapotranspiration models were calculated. Their sensitivity values were computed by approximating the partial derivatives by finite differences. To compare the models, relative sensitivity •p was computed by •Pt = (100/E•)(AE/Axt), where i = 1, ,.., n, with sensitivity defined as the percent change in evaporation E per unit change in the input variable x,. A second relative sensitivity •R = [(xt -Xto)/E•](/XE//Xxt) was calculated to compare the t. relative importance of the different input variables. Error variance was analyzed to find the variance of error due to instrument inaccuracies by E[Var (Z)] = •".t=•'•(AE/Axt) •' Vat (xt), where Vat (x,) is the instrument variance estimated from the manufacturer's statements. Although no tests were conducted to determine the bias or prediction accuracy of the model, a technique was proposed to show how the instrument error variance could be added to the prediction error variance to determine the overall system variance. In model development the best model will be one whose sum of these two variances is a minimum if there is no prediction bias.

No abstract

Do quasi-physically based models with more detail perform better than regression or other empirical models? This is a question that was raised many years ago and still remains. In an effort to respond to this question, the author reviews the needs and concerns of users and then divides the large number of models into three classes: (1) screening, (2) research, and (3) planning, monitoring, and assessment. Empirical and causal (physically based) models are contrasted and the advantages and disadvantages of each described. Sources of model uncertainty (properties of data bases, model structure, parameter estimation methods, algorithmic implementation, verification and validation, and future users) that lead to skepticism about models' performance are investigated. Simulation scale and spatial variability are also important considerations. The leading question is then discussed from the perspective of screening, research, and planning, monitoring, and regulatory models.

No abstract