Pretzsch and Dursky (2001), for example, found a temporal trend with an overestimation in the first half of the century and an underestimation in the last half of the century. Also, hypothesized climate change recommends a test for temporal bias (Sterba and Monserud, 1997). Ideally, Akt inhibitor models should be based on data that can be regarded as the climatic mean for the evaluation period. Previous studies

showed that temporal bias is smallest in the period that overlaps with the parameterization period (Sterba and Monserud, 1997). Temporal bias can be exceedingly high if the evaluation period is shorter than 10–15 years (Pretzsch, 2002). Inferring from the data used for model fitting, temporal bias should be very small for the growth this website models Silva and BWIN, which were fit from long term research plots. Growth rates in

these models can be interpreted as the long term climatic mean. In contrast, Prognaus was fit from a relatively short period, and temporal bias could be prevalent. The evaluation period of this study of 15–30 years should be sufficiently long to avoid excessive temporal bias. Spatial bias also frequently occurs (Sterba and Monserud, 1997, Schmid et al., 2006 and Froese and Robinson, 2007). Deviations are caused by site-specific variation not captured in the model (Sterba and Monserud, 1997). For example, this can be due to regionally variable trends between elevation and prediction accuracy or different ownership not accounted for by the model (Froese and Robinson, 2007). Spatial bias is an important problem, where the data used for model fitting are not spatially representative. It is the strength of inventory data to be spatially representative for a study area because national inventories are usually systematic samples covering the full range of conditions. Spatial bias is expected to be high for growth models fit from permanent research plots, because permanent research plots are often clustered

at lower elevations on good sites; they rarely are representative of the site variation across a region. Spatial bias should therefore be relatively small for Prognaus, but higher for BWIN, Moses and Silva. This seems to be confirmed by evaluation results by Schmid et al. (2006). They found that Silva correctly predicted see more growth within the range of the parameterization data up to an elevation of about 1000 m, whereas at higher elevations there were notable deviations. In addition to temporal and spatial deviations, other trends can be found in the evaluation data set. Often deviations with respect to size are found. In agreement with our results, most frequently there is an over-prediction for small trees and an under-prediction for larger trees (Sterba et al., 2001, Schmid et al., 2006, Froese and Robinson, 2007 and Mette et al., 2009).