Predicting leaf area index from scaling principles: corroboration and consequences.

Abstract

Leaf area index (LAI) is a key biophysical variable in most process-based forest-ecosystem models. However, most such models require LAI as an input, typically obtained from empirical observations. We tested whether scaling principles based on trade-offs between single leaf and canopy properties could be effectively used to model LAI, thereby obviating the need for empirical observations. To do so, we used the process-oriented model, PnET, configured to estimate LAI from these same scaling principles. We derived biologically based LAI predictions (LAIPnET) for the Harvard Forest (Massachusetts, USA) eddy covariance tower site, a predominately mixed deciduous hardwood forest, using PnET, and compared these with a locally observed phenology record and with LAI estimates from both local (ground-based) photosynthetically active radiation transmittance (LAITRANS) and normalized difference vegetation index satellite data (LAINDVI). We generated the LAIPnET trajectory by running the PnET model with meteorological observations from the flux tower as model drivers. We derived LAITRANS from measurements of above- and below-canopy photosynthetically active radiation at the flux tower, and LAINDVI from observations from the Advanced Very High Resolution Radiometer (AVHRR) satellite-borne sensor of surface greenness for the 1 km2 cell containing the flux tower. Over a 5-year period, LAIPnET and LAITRANS values were comparable intra- and interannually, with maximum values differing by less than 0.1 to 0.2 LAI units (m2 m(-2)). Values of LAINDVI were similar to LAIPnET and LAITRANS in midsummer, but higher LAI values were predicted in the early and late portions of the growing season. In addition, we used the three alternative LAI trajectories in a modified version of the PnET model and compared the resulting outputs of gross primary production (GPP) with GPP estimates from the flux tower for 5 continuous years. The LAIPnET and LAITRANS inputs resulted in a difference of less than 3% in mean annual GPP from 1995 to 1999, and these were within 7 and 9%, respectively, of the annual eddy flux-based estimates over the same time period. The results indicate that biologically based LAI scaling approaches can closely track temporal changes in a deciduous forest and have potential for spatial and temporal scaling of LAI.