You are here

Home

Harvard Forest >

Harvard Forest REU Symposium Abstract 2016

  • Title: Quantifying uncertainties in forest landscape model outputs using a variance-based global sensitivity analysis
  • Author: Patrick F McKenzie (University of Tennessee at Knoxville)
  • Abstract:

    LANDIS-II is a popular forest landscape model that simulates tree cohort colonization, photosynthesis, and succession in response to climatic conditions. It integrates these processes with stochastic disturbances to make spatiotemporal predictions of forest biomass and composition. The model is structured as a core program with a series of extensions that can be activated to simulate different succession and disturbance processes. The new PnET-Succession extension is the most mechanistic of the LANDIS-II succession extensions; it uses equations from the PnET-II physiological model to predict changes in species biomass and spatial distribution as the result of competition for photosynthetic resources. The extension integrates many parameters, so knowing how parameters influence model outputs is necessary for interpretation of model results. I applied the Fourier Amplitude Sensitivity Test (FAST), a popular global sensitivity analysis method, to describe which PnET-Succession parameters contribute most to variation in LANDIS-II outputs. Using indices generated by FAST, sixteen PnET-Succession input parameters were ranked by their contributions to uncertainty in LANDIS outputs. Total biomass outputs of PnET-Succession are best explained by variation in precipitation loss fraction, maintenance respiration, and climate parameters. In contrast, the biomass of an individual species is influenced primarily by foliar nitrogen for that species. Understanding the relative importance of these parameters in determining model outputs will improve interpretation of model outputs and prioritization of data collection. The analysis also identifies a challenge in that the most influential parameters are difficult to measure empirically. Future steps involve using the extended FAST method to examine interactions between parameters.

  • Research Category: Regional Studies; Physiological Ecology, Population Dynamics, and Species Interactions; Ecological Informatics and Modelling; Biodiversity Studies