Next Article in Journal
Genetic Species Identification Using ycf1b, rbcL, and trnH-psbA in the Genus Pinus as a Complementary Method for Anatomical Wood Species Identification
Previous Article in Journal
Predicting Soil Saturated Water Conductivity Using Pedo-Transfer Functions for Rocky Mountain Forests in Northern China
 
 
Search for Articles:
Title / Keyword
Author / Affiliation / Email
Journal
Article Type
 
 
Advanced Search
 
Section
Special Issue
Volume
Issue
Number
Page
 
Journals
Forests
Volume 14
Issue 6
10.3390/f14061098
forests-logo
Submit to this Journal Review for this Journal Propose a Special Issue
► Article Menu

Article Menu

share Share announcement Help format_quote Cite question_answer Discuss in SciProfiles thumb_up
...
Endorse textsms
...
Comment
first_page
settings
Order Article Reprints
Font Type:
Arial Georgia Verdana
Font Size:
Aa Aa Aa
Line Spacing:
Column Width:
Background:
Article

Exploring the Influence of Forest Tenure and Protection Status on Post-Fire Recovery in Southeast Australia

by
Sven Huettermann
*,
Simon Jones
,
Mariela Soto-Berelov
and
Samuel Hislop
Remote Sensing Centre, Geospatial Science, School of Science, RMIT University, 402 Swanston Street, Melbourne 3000, Australia
*
Author to whom correspondence should be addressed.
Forests 2023, 14(6), 1098; https://doi.org/10.3390/f14061098
Submission received: 29 March 2023 / Revised: 18 May 2023 / Accepted: 23 May 2023 / Published: 25 May 2023
(This article belongs to the Topic Application of Remote Sensing in Forest Fire)
Browse Figures
Versions Notes

Abstract

:
Research Highlights: We used Landsat time series data to investigate the role forest tenure and protection status play in the recovery of a forest after a fire. Background and Objectives: Changing fire regimes put forests in southeast Australia under increasing pressure. Our investigation aimed to explore the impact of different forest management structures on a forest’s resilience to fire by looking at the post-fire recovery duration. Materials and Methods: The analysis included a total of 60.6 Mha of land containing 25.4 Mha of forest in southeast Australia. Multispectral time series data from Landsat satellites and a local reference dataset were used to model attributes of disturbance and recovery over a period of 33 years. Results: Protected public forest spectrally recovered 0.4 years faster than protected private forest. No other significant effects in relation to different tenure and protection status were found. Climatic and topographic variables were found to have much greater influence on post-fire spectral recovery. Conclusions: Protected area status in public forests resulted in slightly faster recovery, compared with the private protected forest estate. However, factors outside the control of land managers and policy makers, i.e., climatic and topographic variables, appear to have a much greater impact on post-fire recovery.

1. Introduction

In Australian forests, disturbances caused by wildfire have occurred regularly for millions of years. However, although many local species have adapted to regular fire through mechanisms such as epicormic resprouting and fire-resistant canopy-stored seeds [1,2], it has been suggested that changing fire regimes with higher burn frequencies and severities might pose a threat to some of Australia’s unique forest ecosystems [3]. Recent studies have indicated that current forest management strategies in combination with climate change might lead to irreversible changes in some forested landscapes [4,5,6]. The fire season of summer 2019/20 can be regarded as an indication of changing fire regimes, as it was unprecedented in scale [7] and affected highly vulnerable and diverse forest ecosystems that are not typically exposed to regular fire disturbance [8]. Despite the scale and severity of the 2019/20 fire events, a relatively fast recovery has been documented, which has been attributed to above-average rainfall in the months after the fires [9]. However, other studies that examined native Australian forests have indicated that post-fire recovery usually takes several years [10,11].
Forest management policies in Australia vary with tenure and protection status. The management of protected areas under the National Reserve System (NRS) requires adherence to strict guidelines, usually not permitting timber extraction, and fire mitigation measures such as prescribed burning only under strict conditions [12]. Most protected areas can be found on public land; however, some are managed privately. On the other hand, most of the Australian forest is not under NRS protection, which permits private owners and public land stakeholders to have more invasive management strategies, including timber extraction and fire prevention measures.
To ensure Australian forests retain their ecological key functions in the future it is necessary to evaluate how different tenure and protection status might impact forest resilience to fire. Of particular interest is recovery duration, as some forested ecosystems rely on a certain period of recovery before re-developing the ability to reproduce [13]. This study aims to explore the relationship between different forest management regimes and the length of post-fire recovery in southeast Australia.
Forest recovery is a complex process that includes a broad range of ecological and structural variables [10,14]. However, collecting detailed terrestrial information on these variables is cost intensive and often not feasible across large areas. Satellite-based remote sensing, on the other hand, offers wall-to-wall coverage at regular temporal intervals. For example, the Landsat archive provides multispectral satellite imagery reaching back to the 1970s, and has proven particularly valuable for forest change detection, as its spatial and temporal resolution is well suited to map forest-related phenomena [15,16,17]. Although multispectral remote sensing is not capable of capturing the full complexity of structural forest change, many studies have demonstrated that the spectral signals of disturbance and recovery are sufficiently correlated with the structural changes to address a broad range of monitoring questions [14,18]. The Normalised Burn Ratio (NBR) spectral index, in particular, has been found to be capable of capturing forest change [19,20], as it exploits the sensor’s near infrared and short-wave infrared bands, which are more sensitive to structural complexity within the forest canopy [21].
To leverage data from the Landsat archive, a range of tools have been developed. A frequently used algorithm for processing Landsat time series data is LandTrendr [20,22], which fits linear trend segments to the spectral time series of each pixel, filtering noise and deriving generalised trends [21]. For this study, LandTrendr segmentation based on the NBR index was used to extract spectral metrics of disturbance and recovery for the study area in southeast Australia. The disturbance information obtained from LandTrendr was used in conjunction with further predictor variables to model higher-level disturbance attributes with the Random Forests algorithm, including fire severity and causal agent. To describe forest recovery, we adopted the Years to Recovery (Y2R) approach introduced by Pickell et al. [23], defined as the years it takes a pixel to return to 80 percent of its pre-disturbance spectral value (Y2R80) [14]. Subsequently, the Y2R80 metric was used to assess whether forest tenure and protection status had a significant correlation with recovery length. The key objective was to investigate whether land management and protection status influence a forest’s recovery from wildfire, and how these factors compare with the impact of other environmental variables (e.g., climate and topography) in post-fire recovery.

2. Materials and Methods

To investigate how land tenure and protection status relate to post-fire recovery, a number of processing steps were undertaken. Firstly, disturbance and recovery maps were created for the study area using Landsat time series data from 1988 to 2021 and the segmentation algorithm LandTrendr [21,24]. Based on these maps, a random forest (RF) classifier was used to predict disturbance severity on the pixel level [25]. Disturbance patches were identified and attributed a disturbance agent using a second RF model [17,26], retaining only disturbances caused by fire. The remaining disturbance events were then used to examine how forest recovery behaviour varies across areas of different protection status and tenure in southeast Australia. The methods are discussed in detail below.

2.1. Study Area

The study area covered 25.4 Mha of forested land in southeast Australia. Its boundaries were defined as the extent of all 11 biogeographic regions (Interim Biogeographic Regionalisation for Australia, or IBRA) that can be found in the state of Victoria, some of which stretch across the border into the neighbouring states New South Wales and South Australia, as shown in Figure 1. Each IBRA region features distinct climatic, geological and ecological properties [27].
Sclerophyll forests account for a large percentage of all woody vegetation in the study area [28]. Many eucalypt species found in the study area have adapted to regular fire disturbances through mechanisms such as epicormic resprouting [1,10].

2.2. Restricting the Area of Interest

We restricted our analysis to forested areas within the study area, defined in Australia as areas with greater than 20 percent canopy cover with the potential to reach more than 2 m of height, over an area of more than 0.2 ha [29]. To create the corresponding forest mask, we used the National Forest and Sparse Woody Vegetation Data product of the Department of the Environment and Energy. The dataset provides a total of 23 near-annual layers of forest extent for Australia between 1988 and 2018, based on a classification of Landsat imagery [29]. Using these maps, a composite forest mask was produced to restrict the analysis to areas that were continuously forested between 1988 and 2021, whilst allowing for periods of disturbance and regrowth. Sparse woody vegetation, defined as areas with a canopy cover between 5 and 20 percent, was considered non-forest for the purpose of our analysis.
To investigate the correlation of forest tenure and protection status with recovery duration, we used tenure information from the Department of Agriculture and Water Resources Australia for the year 2013 [30] and data on protection status from the Commonwealth of Australia for the year 2016 [31].

2.3. Forest Change Maps

We used Landsat L1T Surface Reflectance imagery [32] from the sensors ETM, ETM+ and OLI to create annual image stacks for the years 1988 to 2021. Only imagery from between February 1 and May 1 was considered, as this time window allowed most disturbance events happening during the Australian fire season to be detected in the same year as they occurred. For each year, a medoid composite was created [33,34,35]. Clouds and cloud shadow were filtered out using the FMASK algorithm [36]. All processing was carried out on the Google Earth Engine platform [37].
The medoid composites were then used to create forest change maps with LandTrendr [21,38], an algorithm widely acknowledged for its suitability for spectral segmentation of Landsat time series [15,16]. The LandTrendr algorithm uses the values of a given spectral index to fit trend segments to the time series of each pixel. We chose the Normalised Burned Ratio (NBR) index, as it has been shown to be well suited to the monitoring of fire disturbance [19,39] and is sensitive to structural forest change [14].
We chose parameters for the LandTrendr algorithm that were similar to the approach of Kennedy et al. [40]. Areas that were affected by a secondary disturbance were excluded from further analysis, given the objective of this study was to analyse uninterrupted recovery from fire.
In addition to the change maps derived from the LandTrendr analysis of the primary index NBR, the algorithm allows for the creation of change maps from additional spectral indices. The spectral values of the chosen indices are fitted to the trend segments obtained from the segmentation of the primary index [41]. For our analysis, we exported additional change map products created using the Tasseled Cap indices Brightness (TCB), Greenness (TCG) and Wetness (TCW). These indices have been found to respond to disturbance and recovery in different ways [20,22,42], and thus were expected to broaden the spectrum of forest changes that we could distinguish, such as different disturbance severities and causal disturbance agents.
The final outputs from LandTrendr included maps of the pre-disturbance value as well as the magnitude and relative magnitude of disturbance. Furthermore, a map of the disturbance onset year was generated. To exclude very small disturbance events, we applied a minimum mapping unit to the output, restricting the size of a change area to a minimum of 11 pixels (1 ha). Lastly, pixels that experienced an NBR change of less than 50 were also filtered out.

2.4. Predicting Disturbance Severity at a Pixel Level

The year of detection (YOD) of a pixel’s primary disturbance event was directly obtained from LandTrendr; however, the disturbance severity was assigned through a RF classifier at the pixel level. RF is an ensemble classification technique that uses a large number of individual decision trees, each trained with a random subset of the reference data [43]. For our model, we set the number of trees to 200. Widely used in remote sensing applications, RF has been found to be a reliable classification tool even when trained on non-Gaussian distributed data [44]. In our RF model, we used a range of predictor variables that included spectral, climatic and topographic variables.
To train the RF classifier we relied on a reference dataset. The Victorian Forest Monitoring Program (VFMP) features 786 plots, with locations systematically stratified according to biogeographic region and public land tenure, namely multiple-use forests and nature conservation reserves [45]. Based on the VFMP framework, Soto-Berelov et al. [46] created an extensive reference dataset for forest disturbance in Victoria. In a 4 km2 area surrounding each of the 786 VFMP plots, ten Landsat pixels were randomly chosen and attributed by human interpreters to determine disturbance events and associated severity.
To develop the RF model, we used disturbance severity information from the VFMP-based reference dataset. It contains 7823 reference plots with associated disturbance history, including severity as indicated by a number between 0 and 10 [46]. For the analysis, the values were reclassified into 3 categories: low (1–3), medium (4–7) and high severity (8–10). To ensure that the reference dataset and the LandTrendr-derived raster datasets were describing the same disturbance event for each reference pixel, only pixels where the onset years coincided (+/−1 year tolerance) were considered for training purposes. Undisturbed pixels or pixels that experienced a secondary disturbance were excluded, leaving a total of 613 disturbed training pixels. Predictor variables were extracted from LandTrendr disturbance segments and included NBR, TCB, TCG and TCW information on the pre-disturbance value (start value), as well as the spectral magnitude and relative spectral magnitude. Furthermore, climatic and topographic variables were included (mean temperature, annual rainfall, elevation, slope and aspect).
A total of 75 percent of the reference pixels were used to train the RF model, while 25 percent were held back for the assessment of model performance using the accuracy score metric of sklearn in Python [47]. Furthermore, we evaluated the relative importance of predictor variables using permutation importance [47].

2.5. Predicting Disturbance Agent at a Patch Level

Since our study was focused on recovery after fire disturbances, we proceeded to remove other types of disturbances (e.g., logging and drought) by classifying disturbance agents at the patch level. We defined a disturbance patch as a collection of adjacent pixels that were disturbed in the same year and with the same severity [20,41] by intersecting the YOD layer obtained from LandTrendr with the disturbance severity layer obtained from the first RF model described above.
To infer the causal disturbance agent for individual patches, we assigned a range of predictor variables to each patch, namely the mean and standard deviation of all contained pixels, computed from the same raster datasets as used in the pixel-level classification described in Section 2.4. Additionally, the area size and the circumference of each patch were determined and added to the range of predictor variables.
We used the same reference dataset as described in Section 2.4 to train the RF classifier for the patch-level classification of disturbance agents [46]. The original description of a disturbance agent in the reference data comprised a total of 10 different types of disturbance; however, we simplified these classes for our analysis and regrouped the values into three categories that covered the two most relevant disturbance agents within the study area: fire, logging and other. Every patch that contained one or more reference pixels was considered a candidate for the training dataset.
To ensure the integrity of the training data, a range of consistency checks were performed. If prospective training patches contained two or more reference pixels, which suggested different disturbance agents or onset years, the patch was excluded from training purposes. Furthermore, for each training patch, the YOD value as indicated by the reference dataset was checked against the value suggested by LandTrendr. If the two differed by more than one year, we assumed that different disturbance events were depicted, and removed the patch from the training dataset. A total of 216 patches passed the consistency checks, of which 75 percent were used to develop the RF model for the prediction of the disturbance agent on the patch level. The resulting RF model was assessed using a held-back portion of the patches (25 percent), and the relative importance of predictor variables was investigated using permutation importance [47].

2.6. Assessing Forest Recovery

For the recovery analysis, we used the annual NBR values of the time series fitted to the segments provided by the LandTrendr algorithm. The advantage of this approach is that, for each pixel, it yields a stabilised time series that is less affected by year-to-year noise resulting from sun angle, atmospheric effects, phenology or other factors [41].
To assess the impact of land tenure and protection status on the recovery process of disturbed forest, we defined recovery duration as the time it takes a pixel to return to 80 percent of its pre-disturbance NBR value [23]. The Years to Recovery 80 percent metric (Y2R80) has been previously tested by White et al. [14] for describing forest regrowth after clearfell logging, and found to be a good indicator of the state of actual structural recovery. Other studies also found the Y2R80 metric to be a suitable descriptor of recovery after fire disturbance [23,48].
To be eligible for the final analysis of recovery duration, a range of conditions were placed on Y2R80 pixels. The mean NBR value of the six years prior to the onset of a disturbance event was used to establish a pre-disturbance reference value. Hence, pixels disturbed before the year 1994 were excluded from the analysis due to a lack of pre-disturbance years. Furthermore, no Y2R80 pixels burnt after the year 2011 were considered, leaving pixels disturbed in more recent years a maximum of 10 years to recover. Consequently, we imposed this maximum as a rule on all disturbed pixels for consistency. Thus, every pixel that needed longer than 10 years to recover to 80 percent of its pre-disturbance NBR value was considered non-recovered.

2.7. Examining the Effects of Tenure and Protection on Recovery

As a means of evaluating the relative importance of forest tenure and protection status for recovery, we used a random subset (n = 100,000) of all Y2R80 pixels in a RF regression model to predict recovery length. A range of other predictor variables were also included in the model, including disturbance severity, elevation, annual rainfall and mean temperature.
Additionally, violin plots were used to assess post-fire recovery across various forest tenure and protection status categories. To examine the findings further, a series of one-way Analyses of Variance (ANOVA) tests were conducted.

3. Results

3.1. Forest Change Maps

The change mapping using the LandTrendr algorithm resulted in a total area of 7.3 Mha of disturbed forest, including pixels that experienced a secondary disturbance. Figure 2 shows an example of resulting disturbance information (namely, year of disturbance) for a selected part of the study area. Investigation of the temporal agreement (year of disturbance) between the VFMP-based reference data and the LandTrendr output yielded an overall agreement of 63.5 percent. This comparison allowed for a tolerance of +/− one year, which accounted for disturbance events happening outside the annual Landsat observation window from 1 February to 1 May.

3.2. Classification of Disturbance Severity

For pixel-level modelling of disturbance severity, the most important predictor variables were the start value and relative change magnitude of NBR, the relative magnitude of TCW, the year of disturbance and temperature (Table 1). The accuracy score of the RF-driven severity classification was 0.65.

3.3. Classification of Causal Agent of Disturbance

The RF classifier used to predict the disturbance causal agent was based on a total of 38 predictor variables. The most important predictor variables were the patch-based mean and standard deviation of the TCW relative magnitude of disturbance, as well as the mean TCW pre-disturbance value. Furthermore, the year of disturbance, mean slope and standard deviation of elevation also demonstrated a high predictive importance. The model had an accuracy score of 0.82. Figure 3 shows examples of classified disturbance patches with the associated disturbance agent.

3.4. Impact of Tenure and Protection on Recovery

In total, 14.5 percent of all Y2R80 pixels did not spectrally recover within 10 years, and thus were considered unrecovered and excluded from the subsequent analysis (Table 2). The remaining eligible Y2R80 pixels (Figure 4) in our analysis together covered an area of 942,629 ha.
Central to the analysis was to investigate the correlation of forest tenure and protection status with the recovery duration. Firstly, a random subsample (n = 100,000) of the recovery dataset was used to predict Y2R80 using RF regression [47]. Table 3 shows the relative importance of all examined predictor variables. The RF model ranked protection status as the least important predictor variable, while the forest tenure had a slightly higher importance. Both factors, however, were of minor importance when compared with environmental predictor variables and fire severity.
Subsequently, we created violin plots of the Y2R80 distribution across forest tenure and protection status categories and conducted ANOVA tests using randomly subsampled groups (n = 1000 in each). Figure 5 demonstrates that the mean values of all categories are similar. ANOVA tests and subsequent pairwise comparisons yielded no significant differences between most groups (p > 0.05), except for those between the protected public and protected private group (p < 0.005) which on average took 4.5 and 4.9 years to recover to pre-disturbance spectral levels, respectively.
However, the described differences between tenure and protection categories were eclipsed by the impact of other factors that we identified as strong predictor variables of Y2R80, such as fire severity, as demonstrated by the violin plots of Figure 6. According to ANOVA tests, the fire severity categories low, medium and high all differed significantly. Each of the severity categories showed a unique distribution of Y2R80 values. Most pixels disturbed by low-severity fires recovered within less than 2 years, while pixels disturbed by medium-severity fires displayed a wider range of recovery responses. High-severity fires caused a bi-modal distribution of Y2R80 values, with a statistical mode of 7 years.

4. Discussion

In this study, we examined the impact of forest tenure, protection status and environmental variables on spectral recovery duration (Y2R80) of forest after fire. The results showed that forest tenure and protection status were mostly unrelated to Y2R80. Almost no significant differences were found between the recovery of private and public tenure forests, and between protected and unprotected forests. Our analysis furthermore confirmed that, compared with other factors such as climatic and topographic variables, forest tenure and protection status were very weak predictors of recovery length after fire events. Considering these two factors are the only variables investigated in this study that land managers and policy makers can influence, the results of this study suggest that forest resilience to fire might not depend on forest-management-related factors as much as other studies have suggested [4]. On the other hand, mean annual temperatures in southeast Australia have exceeded the average temperature of the reference period (1961 to 1990) in every year between 2000 and 2022 [49]. The results of this study suggest that this development might have far greater implications for post-fire forest recovery than forest tenure and protection status. However, our analysis found one significant difference, showing that protected forests under public tenure spectrally recovered 0.4 years faster than privately held forest under protection status. This finding was backed by the low number of unrecovered pixels (Y2R80 > 10 years) in public protected forests, where only 6.2 percent of the disturbed forest had not spectrally recovered within 10 years of the fire, compared with 21.3 percent in private protected forests. Although reviewing different protection policies across Australia was out of the scope of this study, this result might be of interest to land managers and should be further explored in future studies.
One limitation of our analysis was its focus on pixels that experienced only one disturbance during the whole study period (1988 to 2021). While this ensured a spectrally stable pre-disturbance period for each pixel and prevented recovery periods from being interrupted by a secondary disturbance, it also might have concealed some effects of Australian fire management policies. Prescribed burns are a common feature of fire mitigation strategies, aiming at reducing the fuel load, but can vary in their utilisation depending on forest tenure and protection status [50]. However, as prescribed burns can be registered by LandTrendr as a secondary disturbance signal, it can be assumed that some pixels experiencing such provisions were masked out from our analysis. Thus, we expected our recovery analysis to be partly insensitive to possible effects of these fire management strategies on forest and its recovery.
The use of Landsat imagery in conjunction with an algorithm for time series segmentation such as LandTrendr is widely used for forest disturbance mapping [15,35,41]. In our study, we combined LandTrendr output with a local reference dataset of forest disturbance to model attributes of forest disturbance and recovery. The temporal agreement between disturbance maps derived from LandTrendr and the VFMP-based reference dataset was 63.5 percent. It can be assumed that the NBR magnitude threshold of LandTrendr affected this level of agreement, which allowed only disturbances with an NBR loss of greater than 50 to be detected. This threshold value might not have been sensitive enough to detect all disturbance events registered using the multiple sources of evidence training data [46]. However, it can be assumed that the effect of this discrepancy on the RF models was limited since only reference pixels in agreement with the LandTrendr output were used for training purposes.
Our first RF model predicted disturbance severity at a pixel level and achieved an OOB score of 0.61. In comparison, the authors of the VFMP-based training dataset reported an overall accuracy of 68 percent for a similar RF-based modelling approach [46]. One source of confusion named by the authors was the high rate of disagreement between interpreters for pixels of low disturbance severity, which might have affected the results of our study as well. Our second RF classifier for modelling the causal disturbance agent on patch level achieved an OOB score of 0.90. Our findings on the importance of predictor variables align with earlier studies conducted in the area of southeast Australia, confirming spectral magnitude and relative magnitude as valuable predictors of disturbance agent [20]. One notable difference, however, was the relatively low importance that disturbance patch shape metrics had in our model.
In the second part of our study, we examined recovery after fire disturbance for forested pixels. The Y2R80 metric relies on a stable pre-disturbance period and an uninterrupted recovery period, so we decided to examine only the recovery of disturbance events that occurred between 1994 and 2011. This left at least six years of pre-disturbance history for each pixel from the beginning of our Landsat time series in 1988, and a maximum of ten years for pixels disturbed in 2011 to recover before the end of the time series. Consequently, we decided to label all pixels that did not recover within 10 years as unrecovered, to avoid bias relating to events occurring late in the time series [23,51] and make the comparison of Y2R80 pixels more coherent. This resulted in the discarding of 14.5 percent of otherwise eligible pixels from the analysis. Lowering the recovery threshold (e.g., using a 60 percent instead of an 80 percent threshold) would have yielded a higher percentage of pixels being eligible for the final analysis. However, White et al. [14] observed that the Y2R60 metric overestimated recovery and showed a narrower distribution of values, indicating a decrease in sensitivity.
Lastly, it must be emphasised that spectral recovery, as measured in this study, does not adequately reflect all elements of structural recovery [10,18,52]. Vertical forest structure is a key factor in biomass monitoring [53,54] and fire modelling [55], as well as habitat mapping [10,56]. Despite the sensitivity of the NBR index to some elements of vegetation structure [14,39], passive remote sensing data generally have limits in capturing vertical forest structure in detail. Thus, it is increasingly used in conjunction with active remote sensing technologies such as spaceborne lidar [18,57,58,59] or synthetic aperture radar [60], as well as airborne lidar [55,61,62,63,64]. Future investigations into the role of forest tenure and protection policies in post-fire recovery would benefit from taking advantage of such approaches.

5. Conclusions

We used Landsat imagery from 1988 to 2021 and the LandTrendr algorithm to create disturbance and recovery maps for southeast Australia across 11 bioregions. Applying random forest models and a local reference dataset, disturbance severity and causal agent were modelled. We then conducted an analysis of recovery (Y2R80) for a total of 942,629 ha of forested pixels affected by fire events between 1994 and 2011, excluding areas that experienced a secondary disturbance. The findings suggest that the variables of interest, forest tenure and protection status, were mostly unrelated to recovery duration. The only significant difference was found for protected forests, where public forest recovered 0.4 years faster than its private equivalent. However, other factors such as climatic and topographic variables were much more strongly associated with post-fire recovery duration.

Author Contributions

Conceptualization, S.H. (Sven Huettermann), S.J., M.S.-B. and S.H. (Samuel Hislop); data curation, S.H. (Sven Huettermann); formal analysis, S.H. (Sven Huettermann), S.J., M.S.-B. and S.H. (Smuel Hislop); funding acquisition, S.H. (Sven Huettermann); investigation, S.H. (Sven Huettermann); methodology, S.H. (Sven Huettermann), S.J., M.S.-B. and S.H. (Samuel Hislop); project administration, S.H. (Sven Huettermann); resources, S.H. (Sven Huettermann); software, S.H. (Sven Huettermann); supervision, S.J., M.S.-B. and S.H. (Samuel Hislop); validation, S.H. (Sven Huettermann); visualization, S.H. (Sven Huettermann); writing—original draft, S.H. (Sven Huettermann); writing—review & editing, S.J., M.S.-B. and S.H. (Samuel Hislop). All authors have read and agreed to the published version of the manuscript.

Funding

This research received no external funding.

Data Availability Statement

Publicly available datasets were analysed in this study. These data can be found here: https://developers.google.com/earth-engine/datasets/catalog/landsat (Landsat time series data; Access date: 19 September 2022), https://data.gov.au/data/dataset/d734c65e-0e7b-4190-9aa5-ddbb5844e86d (National Forest and Sparse Woody Vegetation Data used for forest mask; Access date: 24 August 2020), https://data.gov.au/data/dataset/2afe071b-f050-43cf-af91-6afa368330cb (Tenure of Australia’s forests 2013; Access date: 12 January 2021), http://www.environment.gov.au/fed/catalog/search/resource/details.page?uuid=%7B57645456-C5D3-483C-89F4-51A0EC6070EC%7D (Collaborative Australian Protected Areas Database 2016; Access date: 14 May 2020). The VFMP-based reference dataset used in this study is not publicly available due to license restrictions.

Acknowledgments

We gratefully acknowledge the support of an RMIT GOstralia Stipend Scholar scholarship from RMIT University Australia. Furthermore, we would like to acknowledge the science and development teams behind Landsat, Google Earth Engine and LandTrendr for providing the data and tools used in this study. Likewise, we thank the authors of the VFMP-based reference dataset (LandFor project team) for supplying the foundation of our training datasets. Lastly, we would like to thank the RMIT RACE hub (specifically Robert Shen and Mohammad Kazemi Beydokhti) for providing the computational resources we partly relied on for this study.

Conflicts of Interest

The authors declare no conflict of interest.

References

  1. Pausas, J.G.; Keeley, J.E. Epicormic Resprouting in Fire-Prone Ecosystems. Trends Plant Sci. 2017, 22, 1008–1015. [Google Scholar] [CrossRef]
  2. Lamont, B.B.; Le Maitre, D.C.; Cowling, R.M.; Enright, N.J. Canopy seed storage in woody plants. Bot. Rev. 1991, 57, 277–317. [Google Scholar] [CrossRef]
  3. Canadell, J.G.; Meyer, C.P.M.; Cook, G.D.; Dowdy, A.; Briggs, P.R.; Knauer, J.; Pepler, A.; Haverd, V. Multi-decadal increase of forest burned area in Australia is linked to climate change. Nat. Commun. 2021, 12, 6921. [Google Scholar] [CrossRef] [PubMed]
  4. Lindenmayer, D.B.; Taylor, C. New spatial analyses of Australian wildfires highlight the need for new fire, resource, and conservation policies. Proc. Natl. Acad. Sci. USA 2020, 117, 12481–12485. [Google Scholar] [CrossRef]
  5. Bowman, D.M.; Murphy, B.P.; Boer, M.M.; Bradstock, R.A.; Cary, G.J.; Cochrane, M.A.; Fensham, R.J.; Krawchuk, M.A.; Price, O.F.; Williams, R.J. Forest fire management, climate change, and the risk of catastrophic carbon losses. Front. Ecol. Environ. 2013, 11, 66–67. [Google Scholar] [CrossRef]
  6. Collins, L. Eucalypt forests dominated by epicormic resprouters are resilient to repeated canopy fires. J. Ecol. 2020, 108, 310–324. [Google Scholar] [CrossRef]
  7. Boer, M.M.; Resco de Dios, V.; Bradstock, R.A. Unprecedented burn area of Australian mega forest fires. Nat. Clim. Chang. 2020, 10, 171–172. [Google Scholar] [CrossRef]
  8. Godfree, R.C.; Knerr, N.; Encinas-Viso, F.; Albrecht, D.; Bush, D.; Christine Cargill, D.; Clements, M.; Gueidan, C.; Guja, L.K.; Harwood, T.; et al. Implications of the 2019–2020 megafires for the biogeography and conservation of Australian vegetation. Nat. Commun. 2021, 12, 1023. [Google Scholar] [CrossRef]
  9. Qin, Y.; Xiao, X.; Wigneron, J.-P.; Ciais, P.; Canadell, J.G.; Brandt, M.; Li, X.; Fan, L.; Wu, X.; Tang, H.; et al. Large loss and rapid recovery of vegetation cover and aboveground biomass over forest areas in Australia during 2019–2020. Remote Sens. Environ. 2022, 278, 113087. [Google Scholar] [CrossRef]
  10. Karna, Y.K.; Penman, T.D.; Aponte, C.; Hinko-Najera, N.; Bennett, L.T. Persistent changes in the horizontal and vertical canopy structure of fire-tolerant forests after severe fire as quantified using multi-temporal airborne lidar data. For. Ecol. Manag. 2020, 472, 118255. [Google Scholar] [CrossRef]
  11. Hislop, S.; Jones, S.; Soto-Berelov, M.; Skidmore, A.; Haywood, A.; Nguyen, T.H.; Pettorelli, N.; Disney, M. High fire disturbance in forests leads to longer recovery, but varies by forest type. Remote Sens. Ecol. Conserv. 2019, 294, 376–388. [Google Scholar] [CrossRef]
  12. Natural Resource Management Ministerial Council Australia. Directions for the National Reserve System: A Partnership Approach; Natural Resource Management Ministerial Council: Canberra, Australia, 2005; ISBN 0642551367.
  13. Bassett, O.D.; Prior, L.D.; Slijkerman, C.M.; Jamieson, D.; Bowman, D.M.J.S. Aerial sowing stopped the loss of alpine ash (Eucalyptus delegatensis) forests burnt by three short-interval fires in the Alpine National Park, Victoria, Australia. For. Ecol. Manag. 2015, 342, 39–48. [Google Scholar] [CrossRef]
  14. White, J.C.; Saarinen, N.; Kankare, V.; Wulder, M.A.; Hermosilla, T.; Coops, N.C.; Pickell, P.D.; Holopainen, M.; Hyyppä, J.; Vastaranta, M. Confirmation of post-harvest spectral recovery from Landsat time series using measures of forest cover and height derived from airborne laser scanning data. Remote Sens. Environ. 2018, 216, 262–275. [Google Scholar] [CrossRef]
  15. Zhu, Z. Change detection using landsat time series: A review of frequencies, preprocessing, algorithms, and applications. ISPRS J. Photogramm. Remote Sens. 2017, 130, 370–384. [Google Scholar] [CrossRef]
  16. DeVries, B.; Decuyper, M.; Verbesselt, J.; Zeileis, A.; Herold, M.; Joseph, S. Tracking disturbance-regrowth dynamics in tropical forests using structural change detection and Landsat time series. Remote Sens. Environ. 2015, 169, 320–334. [Google Scholar] [CrossRef]
  17. Schroeder, T.A.; Schleeweis, K.G.; Moisen, G.G.; Toney, C.; Cohen, W.B.; Freeman, E.A.; Yang, Z.; Huang, C. Testing a Landsat-based approach for mapping disturbance causality in U.S. forests. Remote Sens. Environ. 2017, 195, 230–243. [Google Scholar] [CrossRef]
  18. Healey, S.P.; Yang, Z.; Gorelick, N.; Ilyushchenko, S. Highly Local Model Calibration with a New GEDI LiDAR Asset on Google Earth Engine Reduces Landsat Forest Height Signal Saturation. Remote Sens. 2020, 12, 2840. [Google Scholar] [CrossRef]
  19. Bolton, D.K.; Coops, N.C.; Wulder, M.A. Characterizing residual structure and forest recovery following high-severity fire in the western boreal of Canada using Landsat time-series and airborne lidar data. Remote Sens. Environ. 2015, 163, 48–60. [Google Scholar] [CrossRef]
  20. Nguyen, T.H.; Jones, S.D.; Soto-Berelov, M.; Haywood, A.; Hislop, S. A spatial and temporal analysis of forest dynamics using Landsat time-series. Remote Sens. Environ. 2018, 217, 461–475. [Google Scholar] [CrossRef]
  21. Kennedy, R.E.; Yang, Z.; Cohen, W.B. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 1. LandTrendr–Temporal segmentation algorithms. Remote Sens. Environ. 2010, 114, 2897–2910. [Google Scholar] [CrossRef]
  22. Frazier, R.J.; Coops, N.C.; Wulder, M.A. Boreal Shield forest disturbance and recovery trends using Landsat time series. Remote Sens. Environ. 2015, 170, 317–327. [Google Scholar] [CrossRef]
  23. Pickell, P.D.; Hermosilla, T.; Frazier, R.J.; Coops, N.C.; Wulder, M.A. Forest recovery trends derived from Landsat time series for North American boreal forests. Int. J. Remote Sens. 2016, 37, 138–149. [Google Scholar] [CrossRef]
  24. Gafvert, U.; Kirschbaum, A. Landsat-Based Monitoring of Landscape Dynamics at Indiana Dunes National Lakeshore, 2007–2012; U.S. Department of the Interior, National Park Service, Natural Resource Stewardship and Science: Fort Collins, CO, USA, 2015.
  25. Haywood, A.; Verbesselt, J.; Baker, P. Mapping Disturbance Dynamics in Wet Sclerophyll Forests Using Time Series Landsat. Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci. 2016, XLI-B8, 633–641. [Google Scholar] [CrossRef]
  26. Senf, C.; Pflugmacher, D.; Wulder, M.A.; Hostert, P. Characterizing spectral–temporal patterns of defoliator and bark beetle disturbances using Landsat time series. Remote Sens. Environ. 2015, 170, 166–177. [Google Scholar] [CrossRef]
  27. Thackway, R.; Cresswell, I. An Interim Biogeographic Regionalisation for Australia: A Framework for Setting Priorities in the National Reserves System Cooperative Program; Australian Nature Conservation Agency, Reserve Systems Unit: Canberra, Australia, 1995.
  28. Montreal Process Implementation Group for Australia & National Forest Inventory Steering Committee. Australia’s State of the Forests Report 2018: Five-Yearly Report; Department of Agriculture, ABARES: Canberra, Australia, 2018; ISBN 9781743234075.
  29. Department of the Environment and Energy. National Forest and Sparse Woody Vegetation Data, 3rd ed.; Commonwealth of Australia: Canberra, Australia, 2018.
  30. Department of Agriculture and Water Resources Australia. Tenure of Australia’s Forests (2013); Department of Agriculture and Water Resources Australia: Canberra, Australia, 2015.
  31. Commonwealth of Australia. Collaborative Australian Protected Areas Database (CAPAD) 2016; Commonwealth of Australia: Canberra, Australia, 2016.
  32. Masek, J.G.; Vermote, E.F.; Saleous, N.E.; Wolfe, R.; Hall, F.G.; Huemmrich, K.F.; Gao, F.; Kutler, J.; Lim, T.-K. A Landsat Surface Reflectance Dataset for North America, 1990–2000. IEEE Geosci. Remote Sens. Lett. 2006, 3, 68–72. [Google Scholar] [CrossRef]
  33. Saarela, S.; Holm, S.; Healey, S.; Andersen, H.-E.; Petersson, H.; Prentius, W.; Patterson, P.; Næsset, E.; Gregoire, T.; Ståhl, G. Generalized Hierarchical Model-Based Estimation for Aboveground Biomass Assessment Using GEDI and Landsat Data. Remote Sens. 2018, 10, 1832. [Google Scholar] [CrossRef]
  34. Flood, N. Seasonal Composite Landsat TM/ETM+ Images Using the Medoid (a Multi-Dimensional Median). Remote Sens. 2013, 5, 6481–6500. [Google Scholar] [CrossRef]
  35. Bright, B.C.; Hudak, A.T.; Kennedy, R.E.; Braaten, J.D.; Henareh Khalyani, A. Examining post-fire vegetation recovery with Landsat time series analysis in three western North American forest types. Fire Ecol. 2019, 15, 8. [Google Scholar] [CrossRef]
  36. Zhu, Z.; Woodcock, C.E. Continuous change detection and classification of land cover using all available Landsat data. Remote Sens. Environ. 2014, 144, 152–171. [Google Scholar] [CrossRef]
  37. Gorelick, N.; Hancher, M.; Dixon, M.; Ilyushchenko, S.; Thau, D.; Moore, R. Google Earth Engine: Planetary-scale geospatial analysis for everyone. Remote Sens. Environ. 2017, 202, 18–27. [Google Scholar] [CrossRef]
  38. Kennedy, R.; Yang, Z.; Gorelick, N.; Braaten, J.; Cavalcante, L.; Cohen, W.; Healey, S. Implementation of the LandTrendr Algorithm on Google Earth Engine. Remote Sens. 2018, 10, 691. [Google Scholar] [CrossRef]
  39. Hislop, S.; Jones, S.; Soto-Berelov, M.; Skidmore, A.; Haywood, A.; Nguyen, T.H. Using Landsat Spectral Indices in Time-Series to Assess Wildfire Disturbance and Recovery. Remote Sens. 2018, 10, 460. [Google Scholar] [CrossRef]
  40. Kennedy, R.E.; Yang, Z.; Cohen, W.B.; Pfaff, E.; Braaten, J.; Nelson, P. Spatial and temporal patterns of forest disturbance and regrowth within the area of the Northwest Forest Plan. Remote Sens. Environ. 2012, 122, 117–133. [Google Scholar] [CrossRef]
  41. Kennedy, R.E.; Yang, Z.; Braaten, J.; Copass, C.; Antonova, N.; Jordan, C.; Nelson, P. Attribution of disturbance change agent from Landsat time-series in support of habitat monitoring in the Puget Sound region, USA. Remote Sens. Environ. 2015, 166, 271–285. [Google Scholar] [CrossRef]
  42. Huang, C.; Wylie, B.; Yang, L.; Homer, C.; Zylstra, G. Derivation of a tasselled cap transformation based on Landsat 7 at-satellite reflectance. Int. J. Remote Sens. 2002, 23, 1741–1748. [Google Scholar] [CrossRef]
  43. Wilkes, P.; Jones, S.; Suarez, L.; Mellor, A.; Woodgate, W.; Soto-Berelov, M.; Haywood, A.; Skidmore, A. Mapping Forest Canopy Height Across Large Areas by Upscaling ALS Estimates with Freely Available Satellite Data. Remote Sens. 2015, 7, 12563–12587. [Google Scholar] [CrossRef]
  44. Mellor, A.; Boukir, S.; Haywood, A.; Jones, S. Exploring issues of training data imbalance and mislabelling on random forest performance for large area land cover classification using the ensemble margin. ISPRS J. Photogramm. Remote Sens. 2015, 105, 155–168. [Google Scholar] [CrossRef]
  45. Haywood, A.; Mellor, A.; Stone, C. A strategic forest inventory for public land in Victoria, Australia. For. Ecol. Manag. 2016, 367, 86–96. [Google Scholar] [CrossRef]
  46. Soto-Berelov, M.; Haywood, A.; Jones, S.; Hislop, S.; Nguyen, T. Creating robust reference (training) datasets for large area time series disturbance attribution. In Remote Sensing Time Series Image Processing; Weng, Q., Ed.; CRC Press: Boca Raton, FL, USA, 2018; ISBN 9781351680554. [Google Scholar]
  47. Pedregosa, F.; Varoquaux, G.; Gramfort, A.; Michel, V.; Thirion, B.; Grisel, O.; Blondel, M.; Prettenhofer, P.; Weiss, R.; Dubourg, V.; et al. Scikit-learn: Machine Learning in Python. J. Mach. Learn. Res. 2011, 12, 2825–2830. [Google Scholar]
  48. White, J.C.; Wulder, M.A.; Hermosilla, T.; Coops, N.C.; Hobart, G.W. A nationwide annual characterization of 25years of forest disturbance and recovery for Canada using Landsat time series. Remote Sens. Environ. 2017, 194, 303–321. [Google Scholar] [CrossRef]
  49. Australian Bureau of Meteorology. Australian Climate Variability and Change–Time Series Graphs. Time Series Graphs. 2023. Available online: http://www.bom.gov.au/climate/change/index.shtml#tabs=Tracker&tracker=timeseries (accessed on 17 May 2023).
  50. Country Fire Authority Victoria. Fire Ecology: Guide to Environmentally Sustainable Bushfire Management in Rural Victoria; CFA: Burwood East, Victoria, Australia, 2011. [Google Scholar]
  51. Nguyen, H.T.; Soto-Berelov, M.; Jones, S.D.; Haywood, A.; Hislop, S. Mapping forest disturbance and recovery for forest dynamics over large areas using Landsat time-series remote sensing. In Remote Sensing for Agriculture, Ecosystems, and Hydrology XIX, Proceedings of the Remote Sensing for Agriculture, Ecosystems, and Hydrology, Warsaw, Poland, 11–14 September 2017; Neale, C.M., Maltese, A., Eds.; SPIE Press: Warsaw, Poland, 2018; p. 35. ISBN 9781510613065. [Google Scholar]
  52. Bolton, D.K.; White, J.C.; Wulder, M.A.; Coops, N.C.; Hermosilla, T.; Yuan, X. Updating stand-level forest inventories using airborne laser scanning and Landsat time series data. Int. J. Appl. Earth Obs. Geoinf. 2018, 66, 174–183. [Google Scholar] [CrossRef]
  53. Nguyen, T.H.; Jones, S.D.; Soto-Berelov, M.; Haywood, A.; Hislop, S. Monitoring aboveground forest biomass dynamics over three decades using Landsat time-series and single-date inventory data. Int. J. Appl. Earth Obs. Geoinf. 2020, 84, 101952. [Google Scholar] [CrossRef]
  54. Margolis, H.A.; Nelson, R.F.; Montesano, P.M.; Beaudoin, A.; Sun, G.; Andersen, H.-E.; Wulder, M.A. Combining satellite lidar, airborne lidar, and ground plots to estimate the amount and distribution of aboveground biomass in the boreal forest of North America. Can. J. For. Res. 2015, 45, 838–855. [Google Scholar] [CrossRef]
  55. Bright, B.; Hudak, A.; Meddens, A.; Hawbaker, T.; Briggs, J.; Kennedy, R. Prediction of Forest Canopy and Surface Fuels from Lidar and Satellite Time Series Data in a Bark Beetle-Affected Forest. Forests 2017, 8, 322. [Google Scholar] [CrossRef]
  56. Schneider, F.D.; Ferraz, A.; Hancock, S.; Duncanson, L.I.; Dubayah, R.O.; Pavlick, R.P.; Schimel, D.S. Towards mapping the diversity of canopy structure from space with GEDI. Environ. Res. Lett. 2020, 15, 115006. [Google Scholar] [CrossRef]
  57. Campbell, M.J.; Dennison, P.E.; Kerr, K.L.; Brewer, S.C.; Anderegg, W.R.L. Scaled biomass estimation in woodland ecosystems: Testing the individual and combined capacities of satellite multispectral and lidar data. Remote Sens. Environ. 2021, 262, 112511. [Google Scholar] [CrossRef]
  58. Kacic, P.; Hirner, A.; Da Ponte, E. Fusing Sentinel-1 and -2 to Model GEDI-Derived Vegetation Structure Characteristics in GEE for the Paraguayan Chaco. Remote Sens. 2021, 13, 5105. [Google Scholar] [CrossRef]
  59. Potapov, P.; Li, X.; Hernandez-Serna, A.; Tyukavina, A.; Hansen, M.C.; Kommareddy, A.; Pickens, A.; Turubanova, S.; Tang, H.; Silva, C.E.; et al. Mapping global forest canopy height through integration of GEDI and Landsat data. Remote Sens. Environ. 2020, 253, 112165. [Google Scholar] [CrossRef]
  60. Fagua, J.C.; Jantz, P.; Rodriguez-Buritica, S.; Duncanson, L.; Goetz, S.J. Integrating LiDAR, multispectral and SAR data to estimate and map canopy height in tropical forests. Remote Sens. 2019, 11, 2697. [Google Scholar] [CrossRef]
  61. Francini, S.; D’Amico, G.; Vangi, E.; Borghi, C.; Chirici, G. Integrating GEDI and Landsat: Spaceborne Lidar and Four Decades of Optical Imagery for the Analysis of Forest Disturbances and Biomass Changes in Italy. Sensors 2022, 22, 2015. [Google Scholar] [CrossRef]
  62. Kane, V.R.; Lutz, J.A.; Roberts, S.L.; Smith, D.F.; McGaughey, R.J.; Povak, N.A.; Brooks, M.L. Landscape-scale effects of fire severity on mixed-conifer and red fir forest structure in Yosemite National Park. For. Ecol. Manag. 2013, 287, 17–31. [Google Scholar] [CrossRef]
  63. Matasci, G.; Hermosilla, T.; Wulder, M.A.; White, J.C.; Coops, N.C.; Hobart, G.W.; Zald, H.S.J. Large-area mapping of Canadian boreal forest cover, height, biomass and other structural attributes using Landsat composites and lidar plots. Remote Sens. Environ. 2018, 209, 90–106. [Google Scholar] [CrossRef]
  64. Matasci, G.; Hermosilla, T.; Wulder, M.A.; White, J.C.; Coops, N.C.; Hobart, G.W.; Bolton, D.K.; Tompalski, P.; Bater, C.W. Three decades of forest structural dynamics over Canada’s forested ecosystems using Landsat time-series and lidar plots. Remote Sens. Environ. 2018, 216, 697–714. [Google Scholar] [CrossRef]
Forests 14 01098 g001 550
Figure 1. The study area, stretching across three states in southeast Australia. VFMP = Victorian Forest Monitoring Program.
Figure 1. The study area, stretching across three states in southeast Australia. VFMP = Victorian Forest Monitoring Program.
Forests 14 01098 g001
Forests 14 01098 g002 550
Figure 2. Year of detection (YOD) of disturbance mapped for a selected part of the study area. The YOD information was derived from a Landsat time series using the LandTrendr segmentation algorithm. Non-forested areas and pixels that experienced a secondary disturbance were masked out.
Figure 2. Year of detection (YOD) of disturbance mapped for a selected part of the study area. The YOD information was derived from a Landsat time series using the LandTrendr segmentation algorithm. Non-forested areas and pixels that experienced a secondary disturbance were masked out.
Forests 14 01098 g002
Forests 14 01098 g003 550
Figure 3. Random forest-based predictions of causal disturbance agent for disturbance patches in a selected part of the study area. Non-forested areas and pixels that experienced a secondary disturbance were masked out.
Figure 3. Random forest-based predictions of causal disturbance agent for disturbance patches in a selected part of the study area. Non-forested areas and pixels that experienced a secondary disturbance were masked out.
Forests 14 01098 g003
Forests 14 01098 g004 550
Figure 4. Spectral recovery after fire as measured by the Years to Recovery 80 percent (Y2R80) metric, for a selected part of the study area.
Figure 4. Spectral recovery after fire as measured by the Years to Recovery 80 percent (Y2R80) metric, for a selected part of the study area.
Forests 14 01098 g004
Forests 14 01098 g005 550
Figure 5. Violin plots of the distribution of the Years to Recovery (Y2R80) values across combined tenure and protection categories. Black lines indicate the mean. MUF = multiple-use forest.
Figure 5. Violin plots of the distribution of the Years to Recovery (Y2R80) values across combined tenure and protection categories. Black lines indicate the mean. MUF = multiple-use forest.
Forests 14 01098 g005
Forests 14 01098 g006 550
Figure 6. Violin plots of the distribution of the Years to Recovery (Y2R80) values across the three fire severity categories. Black lines indicate the mean value.
Figure 6. Violin plots of the distribution of the Years to Recovery (Y2R80) values across the three fire severity categories. Black lines indicate the mean value.
Forests 14 01098 g006
Table
Table 1. Feature importance (permutation importance) of predictor variables in a random forest model predicting pixel-level disturbance severity.
Table 1. Feature importance (permutation importance) of predictor variables in a random forest model predicting pixel-level disturbance severity.
PredictorPermutation Importance
Mean temperature0.021
TCW relative magnitude0.014
Year of disturbance0.012
NBR relative magnitude0.009
NBR start value0.008
NBR magnitude0.007
TCB relative magnitude0.005
TCB magnitude0.003
TCW magnitude0.002
TCG relative magnitude−0.001
Elevation−0.003
TCG start value−0.003
Slope−0.003
TCB start value−0.005
TCG magnitude−0.005
Aspect−0.005
Annual rainfall−0.009
TCW start value−0.01
Table
Table 2. Total area examined in the Years to Recovery (Y2R80) analysis per tenure and protection category, and percentage of unrecovered pixels.
Table 2. Total area examined in the Years to Recovery (Y2R80) analysis per tenure and protection category, and percentage of unrecovered pixels.
TenureProtection StatusY2R80 Analysis Area (ha)Unrecovered Pixels (%)
PublicProtected27,6406.2
PublicUnprotected641,30511.5
PrivateProtected23,03921.3
PrivateUnprotected250,64621.4
Total 942,63014.5
Table
Table 3. Feature importance (permutation importance) of predictor variables in a random forest model predicting the Y2R80 metric.
Table 3. Feature importance (permutation importance) of predictor variables in a random forest model predicting the Y2R80 metric.
PredictorPermutation Importance
Rainfall0.26
Fire severity0.161
Mean temperature0.108
Elevation0.097
Forest tenure0.012
Protection status0.001
Disclaimer/Publisher’s Note: The statements, opinions and data contained in all publications are solely those of the individual author(s) and contributor(s) and not of MDPI and/or the editor(s). MDPI and/or the editor(s) disclaim responsibility for any injury to people or property resulting from any ideas, methods, instructions or products referred to in the content.

Share and Cite

MDPI and ACS Style

Huettermann, S.; Jones, S.; Soto-Berelov, M.; Hislop, S. Exploring the Influence of Forest Tenure and Protection Status on Post-Fire Recovery in Southeast Australia. Forests 2023, 14, 1098. https://doi.org/10.3390/f14061098

AMA Style

Huettermann S, Jones S, Soto-Berelov M, Hislop S. Exploring the Influence of Forest Tenure and Protection Status on Post-Fire Recovery in Southeast Australia. Forests. 2023; 14(6):1098. https://doi.org/10.3390/f14061098

Chicago/Turabian Style

Huettermann, Sven, Simon Jones, Mariela Soto-Berelov, and Samuel Hislop. 2023. "Exploring the Influence of Forest Tenure and Protection Status on Post-Fire Recovery in Southeast Australia" Forests 14, no. 6: 1098. https://doi.org/10.3390/f14061098

Note that from the first issue of 2016, this journal uses article numbers instead of page numbers. See further details here.

Article Metrics

Back to TopTop