Comparing predictions of aboveground woody biomass of PnET-II, LINKAGES v2.2, and ED2 with decadal field data at plot and landscape scales

Background/Question/Methods

Forest ecosystem models are a primary approach to predict aboveground woody biomass (AWB) at various temporal and spatial scales.  These models vary substantially in their complexities and can be classified into three types: simple mechanistic; hybrid empirical and mechanistic; and complex mechanistic models.  Many studies that compare forest ecosystem model predictions with field data are conducted at small temporal and spatial scales, and a key finding is that complex mechanistic models provide the best predictions.  However, whether this holds true at large temporal and small spatial scales, as well as at large temporal and spatial scales remains untested.  We compared model predictions of AWB density using PnET-II (simple mechanistic model), LINKAGES v2.2 (hybrid empirical and mechanistic model), and ED2 (complex mechanistic model) with decadal field data at plot and landscape scales.  We used three sites in the Central Hardwood forest region for comparisons at the plot scale: Sinkin Experimental Forest (30 years of data), Vinton-Furnace Experimental Forest (33 years), and Kaskaskia Experimental Forest (78 years).  For comparisons at a landscape scale, we used the three ecological subsections in which each forest site was located: Current River Hills (21 years), Western Hocking Plateau (19 years), and Lesser Shawnee Hills (25 years).    

Results/Conclusions

The ED2 model was the most concordant with field data at the three forest sites (all p-values>0.1) for the plot scale.  The average percent mean bias of AWB density predictions were greatest for PnET-II (-2.4% and -10.3%), followed by LINKAGES v2.2 (2.3%) and ED2 (1.4%).  These results suggest that the expectation that complex mechanistic models provide the best predictions at plot and small temporal scales may also hold true at decadal scale. 

At landscape scale, the average percent mean bias of AWB density predictions were greatest for PnET-II (11.7% and -12.5%), followed by ED2 (-6.7%) and LINKAGES v2.2 (3.0%).  This suggests that models employing both empirical and mechanistic processes may provide better AWB density predictions at decadal and landscape scales.  ED2 did not provide the best prediction due to fine scale physiological process such as photosynthesis; albeit the relationships between such processes and population dynamics are not fully understood at decadal and landscape scales.  The average percent mean bias of AWB density predictions was greater for all models at the landscape scale than that at the plot scale, possibly because these models do not simulate landscape processes.