Canada’s forests return following disturbance

In a recently published paper, Joanne White (Research Scientist, Canadian Forest Service) and co-authors communicate their findings characterizing 25 years of forest disturbance and recovery for Canada using satellite data. Using open data from the Landsat series of satellites and high performance computing opportunities, White and co-authors use tens-of-thousands of images representing billions of pixels, to map forest wildfire and harvesting on an annual basis and to then characterize the return of forest vegetation following the disturbance. Forest disturbance is captured and mapped to type (such as wildfire or harvest). Using the measurements from the post-disturbance satellite data, it is found that given time forests return following disturbance. As Joanne indicates,

“The capacity to distinguish recovery trends by disturbance type provides insights on forest dynamics that are both spatially detailed and national in scope.”

What are the major findings of this research?

  • 57.5 Mha or 10.75% of Canada’s net forested ecosystems (exclusive of water) were disturbed by either wildfire or harvest, representing an annual rate of disturbance of approximately 0.43% per year.
  • Wildfire accounted for 2.5 times more area disturbed than harvest. On average, wildfire impacted 1.5 Mha annually, compared to 0.65 Mha impacted by harvesting. Moreover, the amount of area impacted by wildfire was much more variable annually (σ = 1.5 Mha) when compared to harvest (σ= 0.1 Mha).
  • Regionally, boreal ecozones had the greatest absolute area of disturbance from both wildfire (Boreal Shield West) and harvest (Boreal Shield East).
  • Spectral recovery metrics indicate that vegetation returns following disturbance.
  • Areas impacted by harvest exhibited more rapid spectral recovery rates.
  • 78.6% of areas impacted by harvest having a mean recovery time ≤ 10 years, compared to 35.5% of areas impacted by wildfire
  • Overall, less than 1% of the areas disturbed by wildfire and harvest were identified as non-recovering by all three spectral measures of recovery used in our analysis.

Citation and open access:

White, J.C., Wulder, M.A., Hermosilla, T., Coops, N.C., Hobart, G.W. 2017. A nationwide annual characterization of 25 years of forest disturbance and recovery for Canada using Landsat time series. Remote Sensing of Environment, 194: 303–321. DOI: 10.1016/j.rse.2017.03.035

Get it here:

Best Available Pixel (BAP) proxy image composite of Canada in 2010 (Landsat TM/ETM+bands 5,4,3) superimposed with the boundaries of the forested ecozones analyzed in this study (A). Area disturbed by wildfire and harvest (1985–2010), as identified using the Composite to Change (C2C) approach (B).

Vegetation recovery following disturbance is a process that varies greatly in both time and space. A 5-year window, as measured by our absolute or relative short-term spectral recovery metrics provides an initial assessment of forest conditions post disturbance. To evaluate recovery over a longer temporal window, we calculated how many years it took for a pixel to reach 80% of its pre-disturbance Normalized Burn Ratio (NBR) value (Y2R). The pre-disturbance NBR value is determined using the average NBR value for the two years prior to disturbance.

Every paper has a key figure that really brings home the key findings. In this paper, it is this figure showing the years to recovery following disturbance (by type), with the recovery trends also stratified by epochs.

So, Joanne, what does this figure show?
National summary of Years to Recovery (Y2R) values for wildfire and harvest. The cumulative proportion of area disturbed by Y2R is plotted for all disturbances (1985–2005), for disturbances that occurred prior to 1990, and for disturbances that occurred between 1990 and 2000
“In the context of our definition of recovery, and given the emphasis that we place on characterizing relative rates of recovery, meeting or exceeding 80% of the pre-disturbance NBR value relates a positive trend in the return of vegetation, but does not necessarily indicate a return to the same forest conditions that existed at a site prior to disturbance.”

Further, Joanne offers, “Our results indicated that there was a difference in long-term recovery between areas disturbed by wildfire and harvest. We found that approximately, 68.4% of wildfire areas and 92.5% of harvest areas attained the 80% benchmark by the end of the time series in 2010. It should be noted that this result includes all disturbances that occurred between 1985 and 2005, so for some disturbances, there is only a 5-year period post-disturbance available for analysis. Therefore, we specifically considered only those disturbances in our time series that have had the longest time period of recovery (that is, those disturbances that occurred prior to 1990). When we focused on this subset of disturbances, we found that 86.3% of wildfire areas, and 98.4% of harvest areas attained an NBR value that was 80% of their respective pre-disturbance NBR.”

What are the unique contributions of this research?

• Remote sensing-informed assessment of 25 years of wildfire and harvest in Canada’s forested ecosystem (not done previously from Landsat time series at 30 m spatial resolution);

• National, spatially-explicit record of harvesting (no such record exists in Canada);

• Characterize national trends in recovery following wildfire and harvest, which has previously only been achieved on a sample-basis or by using AVHRR data;

• Characterize trends in Canada’s forest recovery separately by disturbance type (wildfire and harvest), which has likewise not been previously undertaken.


The objective of this @NRCan research, accelerated by support from the Canadian Space Agency (@CSA_ASC), and through collaboration with the University of British Columbia (Prof Nicholas Coops, @IRSS_UBC and Dr. Txomin Hermosilla, @txominhermos) was to characterize national trends in stand replacing forest disturbance caused by wildfire and harvest, and subsequent recovery, for the period 1985–2010 for Canada’s forested ecosystems (~650 Mha), using information derived from Landsat time series data. These results represent a major information outcome from the Composite-2-Change (C2C) approach of generating best-available-pixel (BAP) composites from Landsat data, and identifying and attributing forest change on an annual basis (as presented in Hermosilla et al. 2016). This publication is important because it is national in scope, and represents the first wall-to-wall characterization of wildfire and harvest and subsequent vegetation recovery in Canada at a spatial resolution commensurate with human impacts. The information outcomes represent 25 years of change in Canada’s forests, derived from a single, consistent spatially-explicit data source, derived in a fully automated manner. This demonstrated capacity to characterize forests at a resolution that captures human impacts is key to establishing a baseline for current and future climate change monitoring. Moreover, characterizing both the depletion (disturbance) and accrual (regrowth) of vegetation provides a more holistic understanding forest change in the context of long-term forest monitoring and carbon accounting. The high performance computing environment of @WestGrid, a part of @ComputeCanada, made this research and collaboration possible. Further, the free and open access data policy of the United States Geological Survey (@USGSLandsat) provided the data (from the @NASA_Landsat series of Landsat satellites) in a robust, analysis ready form, promoting our scientific analyses.

How do the C2C outputs compare to other data sources?

In the paper, remotely sensed outputs are compared to statistics for wildfire fire and harvest from the National Forestry Database and Canadian National Fire Database. Trends are similar among data sources, although absolute estimates of area differ. In the case of harvests, much of the earlier data used to report national trends in harvesting over time were aspatial. Methods for recording harvest information have evolved over time as GIS technology has become ubiquitous in natural resource management agencies. Despite this, there is currently no national, spatially-explicit record of forest harvesting in Canada. In the case of wildfire, the mapped outcomes are expected to differ, based on known differences in the methods applied and data used, especially as related to differences in source data spatial resolution, definitions of minimum mapable units, and treatment of unburned islands and waterbodies within wildfire boundaries. We also acknowledge that the C2C outputs have a latency issue in detecting disturbance as a function of the image compositing window used (i.e. August 1 ± 30 days); however a national independent accuracy assessment indicated that 97.7% of changes were labelled to within ± 1 year (Hermosilla et al. 2016).

What about other disturbance types (e.g. insects)?

While the focus of this analysis was on stand replacing disturbance, other forms of disturbance (e.g. insects, water stress) can also play an important role in the dynamics of forested ecosystems and can have significant impacts on applications such as carbon accounting. While some insect disturbances manifest in ways that are relatively abrupt and readily detectable (e.g. bark beetles) others are more ephemeral (e.g. defoliators) and although separating the two has been demonstrated, it it is challenging to do so in a national context, where there are many potential and region-specific (and often overlapping) insects and pathogens present. The capacity to use Landsat time series analysis to characterize these various non-stand replacing disturbances is emerging; however, spectral measures of recovery from these non-stand replacing changes have yet to be explored and will require further consideration and research.

How do these satellite measures relate to ground (plot) measurements?

Information from remotely sensed data can provide a useful framework for assessing relative rates and changes in spectral recovery that enables the national assessment of trends. These trends can be related to available ground observations, and can form the basis for additional sampling and investigations to further relate spectral measures of recovery to ecological and silvicultural understanding of the recovery process. Such national assessments are not possible using existing ground plot measurements alone, as demonstrated in Bartels et al. (2016).

Key takeaways

By leveraging recent advances in image compositing capability and the holdings of the Landsat archive, we generated a synoptic, consistent national baseline of stand-replacing forest disturbance and recovery and characterized important regional variations in the disturbance and recovery trends observed, as well as variations related to the disturbance type (i.e. wildfire, harvest). These baseline data provide unprecedented reference information against which present and future trends in disturbance and recovery can be assessed. Efforts are underway to share the data in an open access form. To view the change data online, see:

NRCan Information Links:

Tracking forest change and recovery in Canada

How disturbances shape Canada’s forests

References (peer review):

Multi-decade high-resolution forest disturbance and recovery for Canada:
White, J.C., Wulder, M.A., Hermosilla, T., Coops, N.C., Hobart, G.W. 2017. A nationwide annual characterization of 25 years of forest disturbance and recovery for Canada using Landsat time series. Remote Sensing of Environment, 194: 303–321. DOI: 10.1016/j.rse.2017.03.035
Get it here, #OpenAccess:

Data used:

White, J. C., & Wulder, M. A. (2014). The Landsat observation record of Canada: 1972–2012. Canadian Journal of Remote Sensing, 39(6), 455–467. Link

Project rationale and information on image compositing:

White, J. C., Wulder, M. A., Hobart, G. W., Luther, J. E., Hermosilla, T., Griffiths, P., … & Guindon, L. (2014). Pixel-based image compositing for large-area dense time series applications and science. Canadian Journal of Remote Sensing, 40(3), 192–212. Link

Generation of gap-free surface reflectance composites:

Hermosilla, T., Wulder, M. A., White, J. C., Coops, N. C., & Hobart, G. W. (2015a). An integrated Landsat time series protocol for change detection and generation of annual gap-free surface reflectance composites. Remote Sensing of Environment, 158, 220–234. Link

Labeling of change type using spatial, spectral, temporal, and contextual information:

Hermosilla, T., Wulder, M. A., White, J. C., Coops, N. C., & Hobart, G. W. (2015b). Regional detection, characterization, and attribution of annual forest change from 1984 to 2012 using Landsat-derived time-series metrics. Remote Sensing of Environment, 170, 121–132. Link

Overall methods and accuracy assessment:

Hermosilla, T., Wulder, M. A., White, J. C., Coops, N. C., Hobart, G. W., & Campbell, L. B. (2016). Mass data processing of time series Landsat imagery: pixels to data products for forest monitoring. International Journal of Digital Earth, 9(11), 1035–1054. Link

Description of plot based insights on forest recovery in Canada:

Bartels, S. F., Chen, H. Y., Wulder, M. A., & White, J. C. (2016). Trends in post-disturbance recovery rates of Canada’s forests following wildfire and harvest. Forest Ecology and Management, 361, 194–207. Link.