Harvard Forest >

Harvard Forest Symposium Abstract 2013

• Title: Mapping and Modeling of Vegetation Structure in the northeastern US
• Primary Author: Paul Siqueira (University of Massachusetts - Amherst )
• Additional Authors: Caitlin Dickinson (University of Massachusetts - Amherst ); Yang Lei (University of Massachusetts - Amherst )
• Abstract:

  The characterization of vegetation structure has a variety of applications in carbon accounting, habitat identification, resource tracking and estimation of disturbance history. Because of the need for establishing these applications over large geographic regions, remote sensing and forest modeling are important tools for mapping these characteristics over regional and continental scales. The transition from local measures of forest structure to continental scales is a difficult task, requiring observational and modeling resources that are imperfect and serving as sources of uncertainty.
  
  
  
  
  The current state of the art technology for the use of remote sensing for characterizing vegetation structure is primarily through the use of active sensors, lidar and radar, which have a basic sensitivity to the desired parameters of interest (the density and distribution of vegetative scatters in a volume which can be related to habitat identification, biomass and terrestrial carbon storage). To date, many of the relationships between these remote sensing observations and their desired quantities of interest are determined through empirical relationships. The functional forms of the relationships are created through a basic physical modeling of the remote sensing signal’s interaction with the vegetated target and a comparison with measured ground validation. In this scenario, model coefficients can be adjusted to better fit remote sensing observations to ground validation data with the resulting model applied to the wider area of data collected by the remote sensing instrument.
  
  
  
  
  While there has been a considerable degree of success in using these approaches, a limit is now being reached where such approaches are difficult to extend beyond the region where the empirical relationships were immediately formed. As a result, the global application of such techniques is severely limited, thus making it difficult (or impossible) to design reliable spaceborne systems necessary to meet the Earth and Carbon Cycle science communities’ needs for forest monitoring and mensuration.
  
  
  
  
  The missing component in the application of remote sensing for the characterization of vegetation structure is a frame of reference of these estimates based on physical and environmental constraints (e.g. water resources, light availability and soil type) that are inherent to ecosystems. Models that take these types of constraints into account do exist, ranging from regional, climate-type models (e.g. the ED2 model) to individual tree-based models, known as IBM’s (e.g. the SORTIE-ND model).
  
  
  
  
  What has not been accomplished thus far, to a high degree of resolution, is the use of remote sensing and IBM’s in a combined probabilistic model. Such a model would simulate possible trajectories of the forest environment, and, using an electromagnetic interaction model, create the manifestation of these simulations into an expectation of what the remote sensing system (optical or microwave) would observe. The actual remote sensing observations and instrument error model would then be used to determine which trajectory of the ecosystem was most likely for a particular region. Once accomplished, forest properties of interest (biomass, vertical structure, species diversity, etc.) would be calculated from the IBM and ecologically consistent estimates of these properties would be produced. Further, by taking into account uncertainties in the IBM and observational models, with the use of Bayes’ theorem, the probability density functions (pdf’s) of these desired quantities can be determined, and methods such as the Maximum Likelihood Estimator or metrics such as mean, mode and variance, produced in a statistically and environmentally meaningful way. This model will serve as an important link between remote sensing observations and the larger-scale models, such as ED2, which determines biophysical forest characteristics important for regional assessments and projections.

• Research Category: Ecological Informatics and Modelling
  Regional Studies