Modeling Forest Productivity in a Complex Forested Landscape Using High Spectral Resolution Remote Sensing and Extensive Field Data

Spatial estimates of forest carbon fluxes are needed for a growing number of environmental issues that range from local to global scales. Although integration of ecosystem models and remote sensing data has become a relatively common approach, prediction accuracy is often limited by a lack of detailed input data for important physiological variables such as plant photosynthetic capacity, as well as errors within the models themselves. However, understanding the various sources of prediction error can be challenging in areas that lack extensive field data for rigorous model validation. Previous work with high spectral resolution remote sensing has suggested that the ability to remotely-detect biochemical properties of a plant canopy, most notably leaf N concentrations, may improve our ability to derive high-quality productivity estimates. In this study, we evaluated the effectiveness of this approach in a topographically-complex forested landscape in New Hampshire's White Mountain National Forest. Spatial estimates of net primary productivity were generated by combining an ecosystem model (PnET) with a canopy nitrogen coverage derived using NASA's Airborne Visible/Infrared Imaging Spectrometer (AVIRIS). Model predictions were compared with field-measured productivity data from an extensive network of forest inventory plots, as well as with predictions derived using directly-measured canopy nitrogen. Results showed that model predictions using remotely-sensed canopy N were improved substantially over those generated using input data that were aggregated by cover type. However, remaining prediction error was mostly related to error in the remote sensing coverage, indicating that model accuracy was largely a function of input data quality and that future improvements can be achieved through improved methods for canopy N detection.