EVALUATION OF MODEL PERFORMANCE WHEN THE OBSERVED DATA ARE SUBJECT TO ERROR
EVALUATION OF MODEL PERFORMANCE WHEN THE OBSERVED DATA ARE SUBJECT TO ERROR
复制标题
观测数据存在误差时模型性能的评估
DOI：
10.1080/02723646.1990.10642414
复制
发表时间：
1990
影响因子：
1.6
通讯作者：
J. D. Rowland
中科院分区：
文献类型：
--
作者：
R. Moore;J. D. Rowland
In many cases of model evaluation in physical geography, the observed data to which model predictions are compared may not be error free. This paper addresses the effect of observational errors on the mean squared error, the mean bias error and the mean absolute deviation through the derivation of a statistical framework and Monte Carlo simulation. The effect of bias in the observed values may either decrease or increase the expected values of the mean squared error and mean bias error, depending on whether model and observational biases have the same or opposite signs, respectively. Random errors in observed data tend to inflate the mean squared error and the mean absolute deviation, and also increase the variability of all the error indices considered here. The statistical framework is applied to a real example, in which sampling variability of the observed data appears to account for most of the difference between observed and predicted values. Examination of scaled differences between modelled and obs...
