Open Collections

A hybrid modeling approach to simulating past-century understory solar irradiation in Alberta, Canada

University of British Columbia

In western Canada, the effects of warming and increasing human activity may alter the structure, composition, and function of forests, producing quantitatively and qualitatively different understory light conditions. While difficult to measure directly, process-based models may facilitate inference of historical forest states. Yet, existing models are limited in the dynamics they represent. A promising new approach in hybrid modeling, first demonstrated here, is the fusion of machine learning and process-based models to simulate pattern-based processes. The objective of this dissertation was to simulate the effects of past-century climate and fire conditions on understory global solar irradiation trajectories across a 25.2 million ha landscape in Alberta, Canada. The LANDIS-II forest landscape model was applied to simulate past-century changes to competition, fire, and regeneration. Simulated tree species and age maps were classified into landcover types. A regression model of canopy light transmission as a function of landcover and site index showed good fit with field observations (R2 = 0.94) and was applied to a classification of LANDIS-II outputs. Canopy light transmission was multiplied by mean annual bare-earth global solar irradiation to produce understory light maps. Empirical and semi- mechanistic fire models were also applied. A variant of stochastic gradient descent was applied for parameter optimization, improving fire model performance (R² = 0.96; ΔR²= +0.14). Simulations showed a mild decline in forested area across the 1923-2012 period, attributable to a velocity of warming three times faster than migration. Migration was primarily controlled by fire and secondarily by regeneration. Simulated understory light levels declined across the period due to reduced mortality rates, preceding a likely long-term increase in light attributable to reduced regeneration rates. The key innovations of this work are as follows: characterization of human-dominated fire regimes in western Alberta (Chapter 4); advancement of the TACA-GEM regeneration model (Chapter 5); development of an algorithm for fire model parameter optimization (Chapter 6); development of new LiDAR models of canopy light transmission (Chapter 7); demonstration of a new hybrid modeling approach to simulating pattern-based processes, applied to understory light (Chapter 8); demonstration of long-term climatic regulation of understory solar irradiation through forest regeneration (Chapter 8).

Text

2017-08-21

Attribution-NonCommercial-NoDerivatives 4.0 International

http://hdl.handle.net/2429/62748

Forestry

University of British Columbia

UBCV

http://creativecommons.org/licenses/by-nc-nd/4.0/