Time Series Analysis of Forest Dynamics at the Ecoregion Level

Olga Rumyantseva,Nikolay Strigul,Andrey Sarantsev

doi:10.3390/forecast2030020

Olga Rumyantseva, Nikolay Strigul + Show 1 more

Open Access

https://doi.org/10.3390/forecast2030020

Copy DOI

Abstract

Forecasting of forest dynamics at a large scale is essential for land use management, global climate change and biogeochemistry modeling. We develop time series models of the forest dynamics in the conterminous United States based on forest inventory data collected by the US Forest Service over several decades. We fulfilled autoregressive analysis of the basal forest area at the level of US ecological regions. In each USA ecological region, we modeled basal area dynamics on individual forest inventory pots and performed analysis of its yearly averages. The last task involved Bayesian techniques to treat irregular data. In the absolute majority of ecological regions, basal area yearly averages behave as geometric random walk with normal increments. In California Coastal Province, geometric random walk with normal increments adequately describes dynamics of both basal area yearly averages and basal area on individual forest plots. Regarding all the rest of the USA’s ecological regions, basal areas on individual forest patches behave as random walks with heavy tails. The Bayesian approach allowed us to evaluate forest growth rate within each USA ecological region. We have also implemented time series ARIMA models for annual averages basal area in every USA ecological region. The developed models account for stochastic effects of environmental disturbances and allow one to forecast forest dynamics.

Highlights

IntroductionForested ecosystems demonstrate self-organization at different spatial and temporal scales
Time series modeling traditionally employed in financial mathematics, meteorology and remote sensing is a promising tool for the forecasting of forest dynamics at large scales
We model aggregated dynamics of individual trees growing on a certain forest patch

Summary

Introduction

Forested ecosystems demonstrate self-organization at different spatial and temporal scales Current modeling approaches such as the Markov chain model [2,3,4,5], individual-based models [6,7,8,9,10,11] and differential equations models [12] are powerful predictive tools, especially when applied to forest dynamics at local and intermediate spatial scales; they have particular limitations in capturing continuous large scale forest dynamics. Autoregressive models are computationally efficient and analytically tractable, and, are especially suitable for large scale carbon cycle and biogeochemistry forecasting models, which do not require species-level predictions of forest dynamics [17]

Methods

Results

Discussion

Conclusion