Data-intensive modeling of forested ecosystems

Olga Rumyantseva

doi:10.7273/000005485

In the present work, we have performed various applied mathematics techniques to model forested ecosystems in North America using Quebec and USA forest inven- tory data. We start with autoregressive models which are analytically tractable and operate with continuous state space. We perform time series statistical analysis of Quebec forest data recorded in 1970–2007. We have obtained that geometric random walk with normal increments adequately describes dynamics of forest biomass yearly averages. For individual forest locations, the best fit also turns out to be geometric random walk, however, the normality tests for residuals fail. We fulfilled the same analysis at the level of USA ecological regions, where we noticed the same pattern in the absolute majority of ecoregions. The exception was California Coastal Province, where geometric random walk with normal increments adequately describes dynam- ics of both biomass yearly averages and biomass on individual forest plots. Using Bayesian approach, we have generated comparable USA forest growth rate diagram. The other direction of my research was to model the change of spatial distribution of species under climatic changes. We investigated how various combinations of bio- climatic characteristics affect the potential distribution of Pitch Pine tree. We in- troduced two novel data-intensive models (VIMM and VNM) and calculated Shapley scores which reveal the most important climatic factors for Pitch Pine spatial distri- bution. In continuation of climate modeling, we investigated how forests in the USA are affected by various climatic characteristics. We performed various dimensionality reduction techniques (stepwise regressions, principal component analysis and random forest algorithm) in order to reveal the most influential climatic variables for forest biomass in the USA. We have obtained that precipitation related factors are the most essential for forests in the majority of USA ecoregions. The distinct area of my research was to use analytical methods (integral equations) to understand which tree crown shapes show stable behavior under such perturbations as forest harvesting, hurricanes and etc. As a result, we have found that the tree species with the most ’classical’ tree shapes show stable behaviour.

Data-intensive modeling of forested ecosystems

Files and links (2)

Abstract

Metrics

Details