Quantifying post-fire recovery of forest canopy structure and its environmental drivers using satellite image time-series

Abstract

Fire is a recurring disturbance in most of Australia's forests. About one quarter of Australia's total 149 Mha forests is located in south-east Australia. These forests are largely dominated by eucalypts and support several key ecological services including carbon sequestration, biodiversity conservation, and water supply for major cities. Eucalypts are generally considered well adapted to fire impact by having special adaptive features such as the ability to quickly rebuild their canopy after fire through epicormic shoots. However, projected changes in fire regimes under future climate conditions are expected to cause severe but poorly resolved impacts on Australia's eucalypt forest ecosystems. This thesis focuses on fire impacts on the canopy structure of eucalypt forests and the associated reduction in carbon uptake during the post-fire recovery. Depending on fire severity, impacts on forest canopies vary from light scorching to complete defoliation, with related variation in the magnitude and duration of post-fire gas exchange by that canopy. Regional estimates of fire impacts on forest canopy structure and post-fire carbon uptake for south-eastern Australia's forests do not exist. Further, the contribution of environmental factors such as climate and fire history to spatiotemporal variation in the duration of the post-fire recovery has not been quantified for the major vegetation types in the region. These knowledge gaps can only be practically addressed using remotely sensed imagery and will require an approach to identify fire impacts on forest canopies of high heterogeneity in structure and species composition across environments of differing climate, terrain and fire regimes. Vegetation Indices (VIs) computed by transformation of spectral bands were developed to monitor vegetation parameters including canopy structure and photosynthetic activity (Huete et al. 2002). VIs are almost linearly related to the fraction of photosynthetically active radiation (FPAR) absorbed by a plant canopy and are thus demonstrated as useful tools in monitoring photosynthesis up to canopy scale (Glenn et al. 2008). Further, improvement in quantification of post fire recovery duration of forest canopy may be attained by including other VIs from the MODIS especially the more water sensitive ones. Thus, the main aims of this study are a) to evaluate the applicability of the Moderate Resolution Imaging Spectroradiometer (MODIS) data products to identify the post fire canopy recovery period, b) to develop, test and apply a new approach to quantify the duration of the post-fire recovery period from MODIS image time series, and c) to understand the spatial pattern of post-fire canopy recovery and quantify the role of environmental controls across forests in south-east Australia. In this study I used 8-day composite measurements of MODIS FPAR to characterise forest canopies before and after fire and to compare burnt and unburnt sites. FPAR is a key biophysical canopy variable and primary input for estimating Gross Primary Productivity (GPP). A new method is proposed to determine the duration of post-fire recovery from MODIS-FPAR time-series. Here, I assume that the post-fire recovery of a burnt site continues until the local FPAR is no longer distinguishable from that of similar unburnt sites. The method involves a spatial-mode principal component analysis on full FPAR time series followed by a k-means clustering of components to group pixels based on similarity in temporal FPAR patterns. This classification provides populations of unburnt pixels and identifies FPAR time series that provide the information to predict the hypothetical FPAR of burnt pixels (of the same class or cluster) in the absence of fire. The difference between observed and predicted FPAR of burnt pixels provides both a quantification of fire impact or severity and an objective criterion to identify the post-fire recovery period. Using fire history data, time series of FPAR for burnt and unburnt pixels in each cluster were then compared to quantify the duration of the post-fire recovery period, which ranged from less than 1 to 8 years. Validation of the approach indicated reasonable accuracy based on correspondence with field based severity classes as well as performance of clustering large spatio-temporal FPAR dataset. The proposed approach can be readily applied to other forest environments and/or other disturbances affecting forest canopy structure, FPAR and carbon uptake (Chapter 2). The duration of the post-fire recovery period was thus quantified by modelling the response observed in MODIS FPAR time series. In this study post-fire recovery is defined in terms of a convergence of the FPAR in burnt and unburnt pixels. The duration of this process can be assumed to be a function of the severity of initial fire impacts on the canopy and the rate at which foliage, branches and possibly stems, can be replaced. Hence, the duration of post-fire recovery can be expected to be controlled by environmental factors affecting plant growth and primary productivity. In temperate forest ecosystems of south-east Australia, water availability, fire history and topographic position are potentially important factors. Thus, we produced several predictors related to canopy condition, fire history and environmental controls were produced for forest areas burnt in the period 2000-2010 and applied generalized additive modelling (GAM) to investigate their contribution to the duration of the post-fire recovery period. The optimum GAM had seven significant predictors and explained 45.6% of the deviance. From the categories of predictors, a GAM based on predictors related to canopy condition as indicated by pre-fire and maximum FPAR explained 33.3% of deviance in the duration of the recovery period suggesting that pre-fire canopy condition as well as site productivity exerts important control over canopy recovery. Climatic control based on indicator of rainfall condition (average annual rainfall and daily soil water during recovery period) alone explained 19.7% of deviance. This finding demonstrated the important role of water for canopy foliage growth across diverse vegetation types in the region. Predictors based on local fire history, which included fuel age, the number of times burnt previously and the last burn severity, explained 10% of the duration in post-fire recovery while topographic predictors (elevation and slope) explained just 2% of deviance. The performance of the best model was verified using a bootstrapped comparison between predicted and observed durations of post-fire recovery, yielding a spearman's correlation between 0.56 and 0.76 (Chapter 3).

Date of Award	2014
Original language	English