Application of the metabolic scaling theory and water–energy balance equation to model large‐scale patterns of maximum forest canopy height

Abstract

AbstractAimForest height, an important biophysical property, underlies the distribution of carbon stocks across scales. Because in situ observations are labour intensive and thus impractical for large‐scale mapping and monitoring of forest heights, most previous studies adopted statistical approaches to help alleviate measured data discontinuity in space and time. Here, we document an improved modelling approach which links metabolic scaling theory and the water–energy balance equation with actual observations in order to produce large‐scale patterns of forest heights.MethodsOur model, called allometric scaling and resource limitations (ASRL), accounts for the size‐dependent metabolism of trees whose maximum growth is constrained by local resource availability. Geospatial predictors used in the model are altitude and monthly precipitation, solar radiation, temperature, vapour pressure and wind speed. Disturbance history (i.e. stand age) is also incorporated to estimate contemporary forest heights.ResultsThis study provides a baseline map (c. 2005; 1‐km2 grids) of forest heights over the contiguous United States. The Pacific Northwest/California is predicted as the most favourable region for hosting large trees (c. 100 m) because of sufficient annual precipitation (> 1400 mm), moderate solar radiation (c. 330 W m−2) and temperature (c. 14 °C). Our results at sub‐regional level are generally in good and statistically significant (P‐value < 0.001) agreement with independent reference datasets: field measurements [mean absolute error (MAE) = 4.0 m], airborne/spaceborne lidar (MAE = 7.0 m) and an existing global forest height product (MAE = 4.9 m). Model uncertainties at county level are also discussed in this study.Main conclusionsWe improved the metabolic scaling theory to address variations in vertical forest structure due to ecoregion and plant functional type. A clear mechanistic understanding embedded within the model allowed synergistic combinations between actual observations and multiple geopredictors in forest height mapping. This approach shows potential for prognostic applications, unlike previous statistical approaches.