Estimating Confidence Bands for Soil‐Crop Simulation Models

Abstract

AbstractA general method is described that provides confidence bands for the ouput from soil‐crop simulation models. Stochastic data are generated for model inputs and model coefficients. These data reproduce the statistical characteristics of the climate, soil properties, and internal model coefficients including distributions, temporal, spacial and variable cross correlations, means, and standard deviations. Model output based on these inputs is then analyzed to produce means, standard deviations, and confidence bands. A typical application of the method is illustrated using a stochastic model for soils input data together with the nitrogen‐tillage‐residue management (NTRM) simulation model to generate confidence bands for soil NO3‐N profiles in Houston County, MN. Results indicate that 380 to 1800+ computer runs may be required to provide confidence bands for soil NO3‐N concentrations that are within 10% of the means, and that use of singular mean input data sets in a nonlinear model such as NTRM can produce deviations in the simulated results ranging from 0.7 to 101% of the mean output obtained from many runs.