Improving carbon monitoring and reporting in forests using spatially-explicit information
© The Author(s) 2016
Received: 21 July 2016
Accepted: 21 October 2016
Published: 26 October 2016
Understanding and quantifying carbon (C) exchanges between the biosphere and the atmosphere—specifically the process of C removal from the atmosphere, and how this process is changing—is the basis for developing appropriate adaptation and mitigation strategies for climate change. Monitoring forest systems and reporting on greenhouse gas (GHG) emissions and removals are now required components of international efforts aimed at mitigating rising atmospheric GHG. Spatially-explicit information about forests can improve the estimates of GHG emissions and removals. However, at present, remotely-sensed information on forest change is not commonly integrated into GHG reporting systems. New, detailed (30-m spatial resolution) forest change products derived from satellite time series informing on location, magnitude, and type of change, at an annual time step, have recently become available. Here we estimate the forest GHG balance using these new Landsat-based change data, a spatial forest inventory, and develop yield curves as inputs to the Carbon Budget Model of the Canadian Forest Sector (CBM-CFS3) to estimate GHG emissions and removals at a 30 m resolution for a 13 Mha pilot area in Saskatchewan, Canada.
Our results depict the forests as cumulative C sink (17.98 Tg C or 0.64 Tg C year−1) between 1984 and 2012 with an average C density of 206.5 (±0.6) Mg C ha−1. Comparisons between our estimates and estimates from Canada’s National Forest Carbon Monitoring, Accounting and Reporting System (NFCMARS) were possible only on a subset of our study area. In our simulations the area was a C sink, while the official reporting simulations, it was a C source. Forest area and overall C stock estimates also differ between the two simulated estimates.
Both estimates have similar uncertainties, but the spatially-explicit results we present here better quantify the potential improvement brought on by spatially-explicit modelling. We discuss the source of the differences between these estimates. This study represents an important first step towards the integration of spatially-explicit information into Canada’s NFCMARS.
KeywordsForest Emissions reporting Carbon Greenhouse gas Landsat CBM-CFS3 Terrestrial ecosystem modelling
Current levels of atmospheric carbon dioxide (CO2) are unprecedented in the last 20 million years . These levels of CO2 and other greenhouse gases (GHGs) have caused changes in climate that in turn have had widespread impacts on human and natural systems . The latest Intergovernmental Panel on Climate Change (IPCC) report (2014) confirms that the earth has warmed about 1 °C in response to rising GHG concentrations and that further warming is highly likely. The rate of CO2 increase in the atmosphere can be reduced by taking advantage of the process by which atmospheric CO2 accumulates as carbon (C) in vegetation and soils in terrestrial ecosystems and in harvested wood products. Forest systems are the largest terrestrial C sink, removing about one quarter of annual anthropogenic CO2 emissions [3, 4]. Understanding C exchanges between the biosphere and the atmosphere, specifically the process of C removal from the atmosphere, and how this process is changing or might change, is the basis for developing appropriate adaptation and mitigation strategies. Presently, boreal forest systems are estimated to be C sinks [3, 5, 6]. However, projected changes in environmental conditions may result not only in a reduction in the proportion of anthropogenic CO2 forests remove from the atmosphere, but also in forests becoming a source of GHG to the atmosphere (example, [6–8]). Accurate representation of the emissions and removals from forest systems is a key component of monitoring and predicting changes in the global C cycle.
Historically, forest information and statistics have largely relied on plot-based field measurements . However, with recent technological advancements, and increasingly refined reporting requirements, other sources of information provide new opportunities. Acquiring and using data from satellites or aircraft has notably increased the information base for describing forests in the last decades; for example, trends in phenology at northern latitudes have been remotely observed , remotely-sensed observations have provided support for renewable energy decisions , and aboveground biomass has been estimated remotely . Observations from different types of sensors (i.e., optical and radar) are being explored to improve forest monitoring (example, ). Large international initiatives, such as Group on Earth Observations (GEO), are coordinating efforts to build a Global Earth Observation System of Systems, or GEOSS. GEO’s Global Forest Observations Initiative (GFOI), aims to foster the sustained availability of observations for national forest monitoring systems, exploiting the growing potential of remote observations to support decision making. GFOI works with national governments that report to international forest assessments [such as the global forest resources assessment (FRA) of the Food and Agriculture Organization, FAO] and the national GHG inventories reported to the UN Framework Convention on Climate Change (UNFCCC) using methods of the IPCC.
Robust national forest monitoring based on objective observations is widely accepted as a pre-requisite for countries to participate in international forest C agreements. University and government research activities, as well as initiatives such as GFOI, have fostered the development of methods and applications to produce remotely-sensed information products of relevance to monitoring and reporting programs. However, as of yet few of these products have been integrated into the reporting structures they aim to support, such as the national GHG inventories reported to the UNFCCC. To address these considerations, we fully integrate a 30 m spatial resolution change detection product derived from remotely-sensed data (Landsat, ), and a spatial forest inventory , into the Carbon Budget Model of the Canadian Forest Sector (CBM-CFS3—) the core model of Canada’s National Forest Carbon Monitoring, Accounting and Reporting System (NFCMARS—). CBM-CFS3 has a community of users around the world with applications in many forest systems outside Canada (examples, [18, 19]) making our integration advances relevant to other regions.
The second largest forest biome after tropical systems, the boreal forests spans the higher latitudes through Canada, Alaska, Siberia, China, and Scandinavia [3, 20]. Canada represents about one-third of the total boreal forest . The sheer size of these forests, coupled with the fact that boreal forests are expected to experience the greatest warming of any forest biome as global temperatures rise (IPCC 2014), means that climate-related changes here have the capacity to significantly impact the global C cycle . Recent studies have also shown that these systems are demonstrating altered environmental conditions via changes in forest growth [23–26] and mortality rates . However, the C balance of boreal forests is also affected by changes in disturbance regimes [28–32] and at present, it is not clear how these many changes will affect the net C balance of boreal forests . Unlike other forest systems where direct human-induced disturbance such as deforestation and forest degradation dominate the GHG balance (example, in Mexico, see ), the GHG emissions and removals from the atmosphere from boreal forest systems seem to be driven by natural disturbances . Data from 1959 to 1997 show that an average of ~2 million ha burned annually in Canada with high inter annual variability . With rising temperatures, the frequency and intensity of natural disturbances, specifically fires and especially in boreal forests, are also expected to increase [36, 37]. Information on the location and extent of forest disturbances are therefore essential inputs for accurate estimates of the boreal forest C balance . For the Canadian portion of the managed boreal forest that is under reporting obligation , the dominant disturbances are fire and harvest . As disturbance from fire and harvesting have vastly different C consequences, the ability to distinguish the cause or type of forest disturbance occurring on the landscape through space and time is especially important .
Assembling consistent (e.g., scale, coverage, vintage) and standardized (e.g., categories, estimating protocols) information products for a large and multi-jurisdictional nation, such as Canada, is difficult. Collaborative efforts to assemble the best available data have already resulted in an operational, science-based reporting system for the 2.3 million square kilometers of managed forests of Canada  that directly supports reporting and policy development (examples, [40, 41]), with completed reviews from UNFCCC expert review teams. For such a vast expanse of land (>230 Mkm2), field-based information can only be part of the inputs required for GHG estimation procedures. The current model, CBM-CFS3, allocates disturbance events from a variety of spatial and aspatial sources to deplete forest stand C from a forest inventory stand list in each of about 540–634 spatial units (depending on reporting year) defined by forest management units and jurisdiction (example, ).
Some remotely-sensed data and processing methods allow for forest-change detection that is consistent with the resolution of forest land management (~30 m)  in a cost effective way over large areas . As noted, fire and harvest are the most common disturbances in the boreal forests of Canada, and have the greatest impact on the emission estimates [6, 31, 34]. As stand-replacing disturbances, fire and harvest are also the most reliably distinguished changes using remotely sensed data and analysis techniques . From a spatial monitoring and modelling perspective, models of forest growth are mature and capture well biomass increment, but do not account for forest depletions, which can result in large changes in biomass and impact the ability of forests to exchange gasses with the atmosphere. Therefore, stand-replacing disturbances need to be well captured by monitoring systems. By incorporating change-detection products into regional GHG accounting and monitoring programs, it will also be possible to improve estimates of pre-disturbance forest conditions and it is anticipated that the reliability of present estimates will be improved, thereby enhancing our knowledge and understanding of the status of the large C stores contained in, and emissions from, these northern forests.
Historically modelling and computational constraints had limited the spatial resolution at which C accounting was performed in most jurisdictions, including Canada. Efforts to improve many facets of monitoring and reporting systems such as NFCMARS and CBM-CFS3 (CBM-CFS3—) are starting to alleviate those constraints, by enabling the ingestion of spatially-explicit data for modelling C estimates for an extremely large number of records. Here we incorporate a newly available pan-Canadian forest change product derived from time series of Landsat data  with change attributed to disturbance type  into the CBM-CFS3, for a test area within the boreal forests of Saskatchewan, Canada under reporting obligations (~13 Mha). We aim to demonstrate the capacity for increased inclusion of detailed data (from spatial, temporal, and categorical perspectives) into a regional GHG modelling approach and hypothesize that incorporating the refined disturbance estimates will result in reduced uncertainties in GHG balance estimates for boreal forests.
For our study, we have modified the input to CBM-CFS3, the core model to the Canadian reporting system (NFCMARS), to enable the use of spatially-explicit 30 m pixels instead of spatially-referenced forest stands, as the base unit for modelling. The CBM-CFS3 uses a combination of statistical modelling and process modelling to simulate forest C dynamics. Yield information compiled from forest inventory data and allometric equations are used to grow live biomass components on a yearly basis and process modelling is used to simulate dead organic matter pool dynamics. Biomass C includes all C in above- and below-ground living matter. Dead organic matter pools include hardwood and softwood snags and branches, all litter and organic horizons, as well as organic C in the mineral soil. More information on the model and its input requirements for use in NFCMARS are available in Stinson et al.  and the model is described in Kurz et al. .
CBM-CFS3 simulations require information specific to each forest stand, such as age, dominant species, productivity level, and growth information. Simulations also require ecological parameters such as decomposition and turnover rates, general climate information as well as the timing, location, and type of disturbances.
Dead wood, litter, and soil organic matter C dynamics are explicitly simulated in CBM-CFS3, from the creation of snags to the decay of litter and dead wood and the eventual transfer of C into humified soil organic matter pools and the atmosphere. Annual turnover rates are specified for each of the above- and below-ground biomass pools tracked by the model, and when yield tables indicate declining biomass, the biomass C is transferred to the appropriate dead organic matter pools. Dead wood, litter, and soil organic matter decomposition rates are sensitive to mean annual temperature (using climate inputs developed after ), but no other climatic sensitivity is accounted for in this version of the model. We used the same turnover and decay rates that were developed for GHG balance reporting for the forests of Saskatchewan (see—[16, 34]).
Linearizing Hoerl’s function (Eq. 1b) simplifies the processes of finding the best parameters for our dataset. Permanent sample plots are re-measured through time and therefore violate the basic assumption of independence necessary for basic statistical fitting methods. Hence, to account for these data dependencies, we determined the long-term means of change in volume and its components with plot identity as a random factor to account for temporally autocorrelated plot-specific site condition and species composition. We estimated parameters with random effects (one intercept by plot—1|plot) and fixed effects using maximum likelihood by maximizing the joint density of the parameters and the random component. Adding the random effects in our model fitting lends structure to the random variation in volume measurements, accounting for plot re-measurements. Effectively, we split the error term in two. We assumed a normal distribution for both error terms with a covariance matrix equal to 0. The data were also stratified into nine strata defined by leading species and categorized by soil moisture regime from our spatial inventory information. The R-package lme4  was used to fit this mixed-effect model.
Using the plentiful Landsat Thematic Mapper and Enhanced Thematic Mapper image archive for Canada , a best-available pixel image compositing approach was developed  that was in-turn used as an input to a time series-based change-detection algorithm . A conceptual framework was developed followed by application of a disturbance attribution algorithm over the forested portion of the province of Saskatchewan . Land-cover changes over the test area are detected with high overall accuracy (92.2%), with the majority of changes labeled to the correct occurrence year (91.1%) or within ±1 year (98.7%). Fire and harvesting events, the most important disturbances for C balance in these forest types, are the most successfully attributed (commission error <10%).
Our simulations estimate C stocks in our study area to have increased by 17.98 Tg C between 1984 and 2012, starting with an average C density of 204.85 Mg C ha−1 in 1984, and ending with 207.86 Mg C ha−1 in 2012. Over the course of the simulations, C increased in total biomass, soil C remained almost the same with a small gain, while the deadwood and snag C pools experienced losses, and the litter C pool increased.
When compared to the spatially-referenced simulations used for UNFCCC reporting, the landscape-level C balance (NBP in Fig. 8) shows the same general patterns (synchronized peaks and valleys) in the reported values as our spatially-explicit results. However, the spatially-referenced simulations depict the area as a C source to the atmosphere (i.e., it is mostly below the 0 line), while the spatially-explicit simulations show the area as a sink. Thus, despite the forest area in our spatially-explicit simulations having fewer forested hectares (14% less) than the spatially-referenced simulations, the large fluxes (Rh and NPP) are in the same range, with the difference that the overall productivity (NEP) in our spatially-explicit simulations is greater than respiration. More productive forests tend to store more C in boreal systems and recently disturbed forests areas tend to be C sources for 10–20 years post disturbances , which both contribute to the differences in absolute C leaving the system. The causes of our spatial simulations having higher NPP over Rh could result from differences in disturbance and growth rates, as well as age-structure, all of which differ between the simulations. We explore these differences in the following text.
The area disturbed by year is compared in Fig. 9 for (a) the total area disturbed, and the two most important disturbances, fire (b) and harvest (c). For comparison purposes the larger number of disturbance types in the spatially-referenced simulations were lumped into four categories according to the four disturbance types used in our spatially-explicit simulations.
The cumulative difference in hectares disturbed between the spatially-referenced and our spatially-explicit simulations for the period 1990–2011 is 924,924 ha. Of that, 56% is due to differences in area burned (fire), and 29% is due to a difference in area harvested. In both cases, the spatially-referenced simulations disturbed more hectares overall than our spatially-explicit simulations, contributing to a large proportion of the differences in the C balance. The number of hectares disturbed in the spatially-referenced simulations is based on provincial statistics and national compilations for both fire and harvest (see ). Fires are the predominant disturbance on the landscape, and years of large fires are detected in both simulations. For fires, these burned-area compilations use both coarse-resolution (e.g., SPOT-VGT) and medium-resolution remotely-sensed data sources which include Landsat information, as well as other sources, including aerial GPS surveys and manual delineation from air photos. In some cases, unburned islands or water bodies found within fire perimeters are not removed, but rather included in area burned totals. This would contribute to differences in the total area burned. The Landsat change product used in our spatially-explicit simulations detected, in most years, less area burned than the data used in the spatially-referenced simulations, primarily as a result of differences in the data, spatial resolution, and methods used for fire detection and mapping. With increasing spatial resolution, greater detail is captured with less generalization of disturbance edges  and improved capture of within-fire boundary variability, such as unburned islands . For harvest, much less area was estimated as harvested by the Landsat change product relative to the harvest data used in the spatially-referenced simulation. For the latter, harvest area is allocated from aspatial provincial summaries. All of these differences contribute to the differences we see in our comparison: less area disturbed contributes to lower emissions and the maintenance of higher C stocks on the landscapes, more recently-disturbed stands act as C sources for 10–20 years post disturbance.
Further, in the same overlap area, the Landsat change product identified disturbances occurring between 1990 and 2011 for areas that were not modeled due to a lack of corresponding forest inventory data for those areas. Notwithstanding that the spatial forest inventory used in our spatially-explicit simulations could likely be further refined, since photo-interpreted inventories have error that is estimated to be in the range of 20–30% for individual attributes , it would be difficult to determine if those disturbed, but not modelled hectares contribute to the 14% difference in forested area (i.e., hectares considered forested under the spatially-referenced simulations that had no corresponding data in our spatial inventory), or if they are part of the mix between wetland/peatland and forests that may have been disturbed but is not classified as forests. These areas will therefore contribute to a reduced C sink in the spatially-referenced simulations. In the non-modelled category of disturbed ha (0.07%), the majority falls in the non-stand-replacing condition (42%) and another portion in the unclassified disturbance (20%); in this non-modelled category there is very little harvest or roads (1.25%), with some fire (29%). This may indicate that these areas represent less productive forests. Also, >60% of hectares in this category would have been modelled as a 20% mortality disturbance, with low impact on the short-term C-balance.
We are aware that assigning a 20% mortality C-redistribution scenario to the low disturbance condition and to the unclassified disturbance is a simplification of the forest dynamics in boreal forests. This simplification results in 26% of the disturbed area being assigned a 20% mortality scenario, more than any individual disturbance, apart from fire or harvest, would have in the forests of Saskatchewan. This is a consequence of the relatively coarse attribution of forest change to disturbance types. The origin of much of the non-stand replacing disturbance remains to be determined. The proximity of much of this class to surficial water may suggest that it represents the ecotone between peatlands/wetlands and forests that is pervasive in much of the boreal forests . Considering the low impact on the C-balance of this category, the relatively small area impacted by this (0.06% of the modelled forested hectares), and that further refining this category is not presently possible, we assess the consequences on our estimated C-balance of this generalization to be minimal. It may also be the case that these types of changes are difficult to specify at a 30 m resolution . Deforestation events in general (other than roads) are rare in this part of Saskatchewan and would have little influence on the differences in C dynamics that we are exploring. Further, the Landsat disturbance product uses an agricultural mask in 2011, and would therefore not track swaps between forested lands and agricultural lands before that date, something the spatially-referenced simulations are designed to track. The spatially-referenced simulations simulated 44,849 ha of deforestation in the comparison area between 1990 and 2012.
Disturbance types used in UNFCCC reporting spatially-referenced simulations were more diverse than those depicted in our spatially-explicit simulations, as the Landsat change product only identifies four types of disturbances. The initial change/no-change attribution of stand replacing disturbance is however, of a high accuracy (92.2%). It is also worth noting that of these four types that could be reliably labeled, two types represented >72% of the area disturbed, had a reported area level accuracy of 98%, and are the dominant drivers of C balance in boreal forests. While refinement of the Landsat-based attribution could improve the C-balance estimate further, the attribution task is not a simple one [14, 33, 59]. Some understanding of the disturbance types in a given region will dictate the importance of capturing particular classes. The high accuracy of stand replacing change detection, as well as the harvest and wildfire categories, allows for additional focused investigation to allocate types to remaining changes, such as for the attribution of deforestation and the related land-use and land-cover transitions involved. Insect infestations are harder to detect  yet are known cover large areas of the Canadian forests and impact the C balance (—mountain pine beetle, —spruce bud worm), while in other jurisdictions, human activities dominate the C balance [19, 33]. Hence, improvements in the attribution of the change detected via remote sensing are desirable, and research in this domain is ongoing [62, 63].
Another contribution to the differences in C-balance between the spatially-referenced and spatially-explicit simulations (for the overlap area) is growth rates. Growth estimates that drive our spatial CBM-CFS3 simulations were directly developed from permanent sample field plots which are the best available data on growth , and using mixed-effect models that take into account the data structure, hence, proven statistical methods . The growth information used for the GHG reporting spatially-referenced simulations in this region relied on 15-year old compilations of fewer plots, not specifically designed to monitor growth, with poor fitting methods (see, ), and hence, were less specific to the target area. A cursory comparison of the two sets of growth estimates showed that the growth rates depicted by the field plots (Fig. 4) used in our spatial simulations are higher than those used for the spatially-referenced simulations. More productive forests, those sites that are able to carry more C per ha, tend to remove more CO2 from the atmosphere, further contributing to differences in productivity. Further, in the CBM-CFS3 simulations, growth curves are used to initialise C pools, hence, from the starting point, our spatial simulations had more C in the system, or at least more C concentrated on the hectares that were modelled, than the reported GHG balance for the area. One could argue that selecting the most productive sites for estimating growth across the landscape, as permanent sample field plots do, may not reflect the reality of forest growth conditions across the landscape, and this is an area of active research that we are exploring.
The goals of our study were to advance C-balance modelling towards spatially-explicit systems and to evaluate the impact of these new approaches on the estimates of the C balance of a boreal forest landscape. Our simulations revealed that a 30 m resolution brings into focus the heterogeneity of the landscape, and that although the general fluxes have similar patterns (valleys and peaks) to those reported, C density and sink estimates are higher than those reported. Mascorro, Coops  analysed the impacts of four remote sensing products on GHG balance estimates and also found that using different remotely-sensed data for determining disturbances yielded different estimates of C fluxes over their study area in Mexico.
Using spatially-explicit disturbance information in combination with a spatial forest inventory will improve the characterization of pre-disturbance forest conditions (age, volume, fuel loading) and should therefore help to reduce uncertainties in GHG emissions and removals estimates that are inherent in the current spatially-referenced approaches. Another benefit of spatially-explicit modelling is the potential to better inform post-disturbance productivity information based on pre-existing stand-specific conditions. There are still, however, many aspects of C balance estimation that can be improved.
The sparsity of field data is one of the main impediments to many possible improvements to C balance estimates. In the Canadian case, spatial forest inventories are developed by provincial/territorial jurisdictional resource management agencies, with the extents, attribution content, and time periods represented intended to meet strategic information needs . As a result, the data are of variable vintage, spatial coverage, scales, and information content. The current version of CASFRI which covers most of the managed boreal forests of Canada and is the input spatial inventory for our study region, comprises 36 source datasets and contains 25,319,505 polygons covering a total area of 3,635,970 km2. The minimum mapping unit ranges from 1 to 8 ha, comparable with the resolution of many remotely-sensed land-cover products. In most inventories, smaller identifiable features such as small patches of a distinct tree species are incorporated into the description of the larger unit in which they lie. Refinement and update of spatial inventory inputs with a more direct link to current field inventories will be essential in the production of reliable GHG balance estimates for forests.
Field and remote sensing data could also inform the post-disturbance productivity or stand dynamics. CBM-CFS3 can accommodate such information but data were not available at the appropriate scale to inform post-disturbance transition in our simulations, nor were they available for the current GHG reporting simulations. Refining the post-disturbance succession trajectories and productivity would greatly improve the accuracy of the C balance, and may permit the incorporation of the climate change effects on species dynamics. Remote sensing is well poised to contribute information to these efforts [65–68], although field data would still be necessary to support remote sensing observations.
Present reporting requirements use the IPCC “managed land proxy” and limit reporting to anthropogenic emissions, which are defined as those occurring within the managed portion of the forested systems. However, for scientific reasons and to understand the contribution of Canada’s forests to the global C cycle, Canada also plans to estimate the stocks and fluxes of all its forests. Efforts are presently underway in Canada and elsewhere to expand GHG balance estimates to include managed and unmanaged forests.
This study is a step towards comprehensive spatially-explicit forest C estimation and reporting, with insights relevant for practices implemented in Canada and elsewhere. The full implementation of a second generation National Forest Carbon Monitoring, Accounting and Reporting System will require further developments of the modelling environment to enable the processing of the entire spatially-explicit time series of disturbances that is now available from Landsat satellites and, in the future, could be augmented by measurements from other operational satellites . However, building models and tools that enable explicit links between earth observation products and estimates of GHG emissions is a much needed effort that will enhance the capabilities of future monitoring, reporting and verification (MRV) systems for the land use, land-use change and forestry sector, that can also be used in support of efforts to reduce emissions from deforestation and degradation (REDD) in developing countries.
Carbon Budget Model of the Canadian Forest Sector
- CO2 :
Intergovernmental Panel on Climate Change
Group on Earth Observations
Global Earth Observation System of Systems
Global Forest Observations Initiative
UN Framework Convention on Climate Change
Food and Agriculture Organization
forest resources assessment
Canada’s National Forest Carbon Monitoring, Accounting and Reporting System
balsam fir (Abies balsamea)
balsam poplar (Populus balsamifera)
black spruce (Picea mariana)
jack pine (Pinus banksiana)
trembling aspen (Populus tremuloides)
white birch (Betula papyrifera)
white spruce (Picea glauca)
Canada’s Forest Resource Inventories
black spruce medium productivity
black spruce good productivity
white spruce medium productivity
white spruce good productivity
net primary productivity
heterotrophic respiration (decomposition)
net ecosystem productivity
net biome productivity
net ecosystem exchange
monitoring, reporting and verification
reduce emissions from deforestation and degradation
National Terrestrial Ecosystem Monitoring System
Canadian Space Agency
Government Related Initiatives Program
Canadian Forest Service
CB designed the study, completed the analyses, and wrote the manuscript. BPS completed all the supporting spatial data manipulations and contributed to study refinement. JW provided the Landsat-based disturbance data, provided support for the use of these data and also acted as the main editor for this manuscript. WK participated in study design and conceptualization, study refinement and provided a thorough review of the paper. MW provided support for the use of the Landsat-based data, information about all aspects of remote-sensing information used in this study and provided a thorough review of the paper. All authors read and approved the final manuscript.
We would like to thank all supporting members of the Canadian Forest Service Carbon Accounting Team and the team of remote sensing experts within the Canadian Forest Service for continued efforts with special thanks to Max Fellows and Scott Morken for their invaluable support for spatial CBM-CFS3 simulations. We also thank Dr. Steve Cumming and Mélina Houle of Laval University for contributing spatially-explicit forest inventory data.
The authors declare that they have no competing interests.
This study was supported by funding from Natural Resources Canada. The remote sensing products used in this project were generated via research undertaken as part of the “National Terrestrial Ecosystem Monitoring System (NTEMS): Timely and detailed national cross-sector monitoring for Canada” project jointly funded by the Canadian Space Agency (CSA) Government Related Initiatives Program (GRIP) and the Canadian Forest Service (CFS) of Natural Resources Canada.
Open AccessThis article is distributed under the terms of the Creative Commons Attribution 4.0 International License (http://creativecommons.org/licenses/by/4.0/), which permits unrestricted use, distribution, and reproduction in any medium, provided you give appropriate credit to the original author(s) and the source, provide a link to the Creative Commons license, and indicate if changes were made.
- Tripati AK, Roberts CD, Eagle RA. Coupling of CO2 and ice sheet stability over major climate transitions of the last 20 million years. Science. 2009;326(5958):1394–7.View ArticleGoogle Scholar
- IPCC. Climate change 2014: synthesis report. Contribution of working groups I, II and III to the fifth assessment report of the intergovernmental panel on climate change. Geneva: IPCC; 2014. p. 2014.Google Scholar
- Pan Y, Birdsey RA, Fang J, Houghton R, Kauppi PE, Kurz WA, et al. A large and persistent carbon sink in the world’s forests. Science. 2011;333(6045):988–93.View ArticleGoogle Scholar
- Le Quéré C, Moriarty R, Andrew RM, Canadell JG, Sitch S, Korsbakken JI, et al. Global carbon budget 2015. Earth Syst Sci Data. 2015;7(2):349–96.View ArticleGoogle Scholar
- Federici S, Tubiello FN, Salvatore M, Jacobs H, Schmidhuber J. New estimates of CO2 forest emissions and removals: 1990–2015. For Ecol Manag. 2015;352:89–98.View ArticleGoogle Scholar
- Kurz WA, Shaw CH, Boisvenue C, Stinson G, Metsaranta J, Leckie D, et al. Carbon in Canada’s boreal forest—a synthesis 1. Environ Rev. 2013;21(4):260–92.View ArticleGoogle Scholar
- Metsaranta JM, Dymond CC, Kurz WA, Spittlehouse DL. Uncertainty of 21st century growing stocks and GHG balance of forests in British Columbia, Canada resulting from potential climate change impacts on ecosystem processes. For Ecol Manag. 2011;262(5):827–37.View ArticleGoogle Scholar
- Boisvenue C, Running S. Simulations show decreasing carbon stocks and potential for carbon emissions in Rocky Mountain forests over the next century. Ecol Appl. 2010;20(5):1302–19.View ArticleGoogle Scholar
- Wiersum KF. 200 years of sustainability in forestry: lessons from history. Environ Manag. 1995;19(3):321–9.View ArticleGoogle Scholar
- Jeganathan C, Dash J, Atkinson PM. Remotely sensed trends in the phenology of northern high latitude terrestrial vegetation, controlling for land cover change and vegetation type. Remote Sens Environ. 2014;143:154–70.View ArticleGoogle Scholar
- Kumar L, Sinha P, Taylor S, Alqurashi AF. Review of the use of remote sensing for biomass estimation to support renewable energy generation. APPRES. 2015;9(1):097696.View ArticleGoogle Scholar
- Liu Q, Yang L, Liu Q, Li J. Review of forest above ground biomass inversion methods based on remote sensing technology. J Remote Sens. 2015;19(1):62–74.Google Scholar
- Lucas RM, Mitchell AL, Armston J. Measurement of forest above-ground biomass using active and passive remote sensing at large (subnational to global) scales. Curr For Rep. 2015;1(3):162–77.Google Scholar
- Hermosilla T, Wulder MA, White JC, Coops NC, Hobart GW. Regional detection, characterization, and attribution of annual forest change from 1984 to 2012 using Landsat-derived time-series metrics. Remote Sens Environ. 2015;170:121–32.View ArticleGoogle Scholar
- Cumming SG, Drever CR, Houle M, Cosco J, Racine P, Bayne E, et al. A gap analysis of tree species representation in the protected areas of the Canadian boreal forest: applying a new assemblage of digital forest resource inventory data. Can J For Res. 2014;45(2):163–73.View ArticleGoogle Scholar
- Kurz WA, Dymond CC, White TM, Stinson G, Shaw CH, Rampley GJ, et al. CBM-CFS3: a model of carbon-dynamics in forestry and land-use change implementing IPCC standards. Ecol Model. 2009;220(4):480–504.View ArticleGoogle Scholar
- Kurz WA, Apps MJ. Developing Canada’s national forest carbon monitoring, accounting and reporting system to meet the reporting requirements of the Kyoto protocol. Mitig Adapt Strateg Glob Change. 2006;11(1):33–43.View ArticleGoogle Scholar
- Pilli R, Grassi G, Kurz WA, Smyth CE, Blujdea V. Application of the CBM-CFS3 model to estimate Italy’s forest carbon budget, 1995–2020. Ecol Model. 2013;266:144–71.View ArticleGoogle Scholar
- Mascorro V, Coops NC, Kurz WA, Olguin M. Choice of satellite imagery and attribution of changes to disturbance type strongly affects forest carbon balance estimates. Carbon Balance Manag. 2015;10(30):1–15.Google Scholar
- FAO. Agriculture, forestry and other land use emissions by sources and removals by sinks: 1990–2011 analysis. Rome: UN FAO; 2014. http://www.fao.org/docrep/019/i3671e/i3671e.pdf. Accessed 14 Mar 2016.
- Brandt JP, Flannigan MD, Maynard DG, Thompson ID, Volney WJA. An introduction to Canada’s boreal zone: ecosystem processes, health, sustainability, and environmental issues. Environ Rev. 2013;21(4):207–26.View ArticleGoogle Scholar
- Price DT, Alfaro RI, Brown KJ, Flannigan MD, Fleming RA, Hogg EH, et al. Anticipating the consequences of climate change for Canada’s boreal forest ecosystems. Environ Rev. 2013;21(4):322–65.View ArticleGoogle Scholar
- Boisvenue C, Running SW. Impacts of climate change on natural forest productivity—evidence since the middle of the 20th century. Glob Change Biol. 2006;12:862–82.View ArticleGoogle Scholar
- Hember RA, Kurz WA, Metsaranta JM, Black TA, Guy RD, Coops NC. Accelerating regrowth of temperate-maritime forests due to environmental change. Glob Change Biol. 2012;18(6):2026–40.View ArticleGoogle Scholar
- Chen HYH, Luo Y. Net aboveground biomass declines of four major forest types with forest ageing and climate change in western Canada’s boreal forests. Glob Change Biol. 2015;21(10):3675–84.View ArticleGoogle Scholar
- Ma Z, Peng C, Zhu Q, Liu J, Xu X, Zhou X. Long-term changes in tree basal area across the boreal zone, Canada. Ecoscience. 2014;21(3–4):232–41.View ArticleGoogle Scholar
- Luo Y, Chen HYH. Observations from old forests underestimate climate change effects on tree mortality. Nat Commun. 2013;4:1655.View ArticleGoogle Scholar
- Bond-Lamberty B, Rocha AV, Calvin K, Holmes B, Wang C, Goulden ML. Disturbance legacies and climate jointly drive tree growth and mortality in an intensively studied boreal forest. Glob Change Biol. 2014;20(1):216–27.View ArticleGoogle Scholar
- Rogers BM, Veraverbeke S, Azzari G, Czimczik CI, Holden SR, Mouteva GO, et al. Quantifying fire-wide carbon emissions in interior Alaska using field measurements and Landsat imagery. J Geophys Res Biogeosci. 2014;119(8):1608–29.View ArticleGoogle Scholar
- Metsaranta JM, Kurz WA, Neilson ET, Stinson G. Implications of future disturbance regimes on the carbon balance of Canada’s managed forest (2010–2100). Tellus B. 2010;62(5):719–28.View ArticleGoogle Scholar
- Kurz WA, Stinson G, Rampley GJ, Dymond CC, Neilson ET. Risk of natural disturbances makes future contribution of Canada’s forests to the global carbon cycle highly uncertain. Proc Natl Acad Sci. 2008;105(5):1551–5.View ArticleGoogle Scholar
- Bond-Lamberty B, Peckham SD, Ahl DE, Gower ST. Fire as the dominant driver of central Canadian boreal forest carbon balance. Nature. 2007;450(7166):89.View ArticleGoogle Scholar
- Mascorro V, Coops N, Kurz W, Olguín M. Attributing changes in land cover using independent disturbance datasets: a case study of the Yucatan peninsula, Mexico. Reg Environ Change. 2014;16:213–28.View ArticleGoogle Scholar
- Stinson G, Kurz WA, Smyth CE, Neilson ET, Dymond CC, Metsaranta JM, et al. An inventory-based analysis of Canada’s managed forest carbon dynamics, 1990 to 2008. Glob Change Biol. 2011;17(6):2227–44.View ArticleGoogle Scholar
- Stocks BJ, Mason JA, Todd JB, Bosch EM, Wotton BM, Amiro BD, et al. Large forest fires in Canada, 1959–1997. J Geophys Res Atmos. 2002;107(D1):FFR 5-1–-12.Google Scholar
- Gauthier S, Bernier PY, Boulanger Y, Guo J, Guindon L, Beaudoin A, et al. Vulnerability of timber supply to projected changes in fire regime in Canada’s managed forests. Can J For Res. 2015;45:1439–47.View ArticleGoogle Scholar
- Allen CD, Breshears DD, McDowell NG. On underestimation of global vulnerability to tree mortality and forest die-off from hotter drought in the Anthropocene. Ecosphere. 2015;6(8):1–55.View ArticleGoogle Scholar
- Schroeder TA, Wulder MA, Healey SP, Moisen GG. Mapping wildfire and clearcut harvest disturbances in boreal forests with Landsat time series data. Remote Sens Environ. 2011;115(6):1421–33.View ArticleGoogle Scholar
- Wulder MA, Kurz WA, Gillis M. National level forest monitoring and modelling in Canada. Prog Plan. 2004;61:365–81.View ArticleGoogle Scholar
- Smyth CE, Stinson G, Neilson E, Lemprière TC, Hafer M, Rampley GJ, et al. Quantifying the biophysical climate change mitigation potential of Canada’s forest sector. Biogeosciences. 2014;11(13):3515–29.View ArticleGoogle Scholar
- Environment Canada E. National inventory report—greenhouse gas sources and sinks in CANADA. The Canadian government’s submission to the UN framework convention on climate change. Ottawa: Her Majesty the Queen’s press; 2015.Google Scholar
- Cohen WB, Yang Z, Kennedy R. Detecting trends in forest disturbance and recovery using yearly Landsat time series: 2. TimeSync—tools for calibration and validation. Remote Sens Environ. 2010;114(12):2911–24.View ArticleGoogle Scholar
- Gillanders SN, Coops NC, Wulder MA, Gergel SE, Nelson T. Multitemporal remote sensing of landscape dynamics and pattern change: describing natural and anthropogenic trends. Prog Phys Geogr. 2008;32(5):503–28.View ArticleGoogle Scholar
- Wulder MA, White JC, Coops NC. Identifying and describing forest disturbance and spatial pattern. Understanding forest disturbance and spatial pattern. Boca Raton: CRC Press; 2006. p. 31–61.View ArticleGoogle Scholar
- Hermosilla T, Wulder MA, White JC, Coops NC, Hobart GW. An integrated Landsat time series protocol for change detection and generation of annual gap-free surface reflectance composites. Remote Sens Environ. 2015;158:220–34.View ArticleGoogle Scholar
- Group ESW. A national ecological framework for Canada. Ottawa: Agriculture and Agri-Foods Canada, Research Branch, Centre for land and Biological Resources Research and Environment Canada; 1996. Report No.: and map at scale 1:7.5 million.
- Diffenbaugh NS, Field CB. Changes in ecologically critical terrestrial climate conditions. Science. 2013;341(6145):486–92.View ArticleGoogle Scholar
- Scheffer M, Hirota M, Holmgren M, Van Nes EH, Chapin FS. Thresholds for boreal biome transitions. Proc Natl Acad Sci. 2012;109(52):21384–9.View ArticleGoogle Scholar
- McKenney DW, Hutchinson MF, Kesteven JL, Venier LA. Canada’s plant hardiness zones revisited using modern climate interpolation techniques. Can J Plant Sci. 2001;81(1):129–43.View ArticleGoogle Scholar
- Leckie DG, Gillis MD. Forest inventory in Canada with emphasis on map production. For Chron. 1995;71(1):74–8.View ArticleGoogle Scholar
- Sharma T, Kurz WA, Stinson G, Pellatt MG, Li Q. A 100-year conservation experiment: impacts on forest carbon stocks and fluxes. For Ecol Manag. 2013;310:242–55.View ArticleGoogle Scholar
- Bates D, Maechler M, Bolker B, Walker S. Fitting linear mixed-effects models using lme4. J Stat Softw. 2015;61(1):1–48.Google Scholar
- White JC, Wulder MA. The Landsat observation record of Canada: 1972–2012. Can J Remote Sens. 2013;39(06):455–67.View ArticleGoogle Scholar
- White JC, Wulder MA, Hobart GW, Luther JE, Hermosilla T, Griffiths P, et al. Pixel-based image compositing for large-area dense time series applications and science. Can J Remote Sens. 2014;40(3):192–212.View ArticleGoogle Scholar
- IPCC. IPCC meeting on current scientific understanding of the processes affecting terrestrial carbon stocks and human influences upon them. Geneva: Intergovernmental Panel on Climate Change (IPCC) Working Group I Technical Support Unit—National Oceanic and Atmospheric Administration (NOAA); 2003.Google Scholar
- Wulder M, White J, Gillis M, Walsworth N, Hansen M, Potapov P. Multiscale satellite and spatial information and analysis framework in support of a large-area forest monitoring and inventory update. Environ Monit Assess. 2010;170(1):417–33.View ArticleGoogle Scholar
- Wieder RK, Vitt DH, editors. Boreal peatland ecosystems. Berlin: Springer; 2006.Google Scholar
- Stewart BP, Wulder MA, McDermid GJ, Nelson T. Disturbance capture and attribution through the integration of Landsat and IRS-1C imagery. Can J Remote Sens. 2009;35(6):523–33.View ArticleGoogle Scholar
- Liang L, Hawbaker TJ, Chen Y, Zhu Z, Gong P. Characterizing recent and projecting future potential patterns of mountain pine beetle outbreaks in the Southern Rocky Mountains. Appl Geogr. 2014;55:165–75.View ArticleGoogle Scholar
- Kurz WA, Dymond CC, Stinson G, Rampley GJ, Neilson ET, Carroll AL, et al. Mountain pine beetle and forest carbon feedback to climate change. Nature. 2008;452(7190):987–90.View ArticleGoogle Scholar
- Dymond C, Neilson E, Stinson G, Porter K, MacLean D, Gray D, et al. Future spruce budworm outbreak may create a carbon source in eastern Canadian forests. Ecosystems. 2010;13:917–31.View ArticleGoogle Scholar
- Goodwin NR, Magnussen S, Coops NC, Wulder MA. Curve fitting of time-series Landsat imagery for characterizing a mountain pine beetle infestation. Int J Remote Sens. 2010;31(12):3263–71.View ArticleGoogle Scholar
- Schroeder TA, Healey SP, Moisen GG, Frescino TS, Cohen WB, Huang C, et al. Improving estimates of forest disturbance by combining observations from Landsat time series with US forest service forest inventory and analysis data. Remote Sens Environ. 2014;154:61–73.View ArticleGoogle Scholar
- Wulder MA, Campbell C, White JC, Flannigan M, Campbell ID. National circumstances in the international circumboreal community. For Chron. 2007;83(4):539–56.View ArticleGoogle Scholar
- Madoui A, Gauthier S, Leduc A, Bergeron Y, Valeria O. Monitoring forest recovery following wildfire and harvest in boreal forests using satellite imagery. Forests. 2015;6(11):4105.View ArticleGoogle Scholar
- Bartels SF, Chen HYH, Wulder MA, White JC. Trends in post-disturbance recovery rates of Canada’s forests following wildfire and harvest. For Ecol Manag. 2016;361:194–207.View ArticleGoogle Scholar
- Wulder MA, Skakun RS, Kurz WA, White JC. Estimating time since forest harvest using segmented Landsat ETM+ imagery. Remote Sens Environ. 2004;93(1–2):179–87.View ArticleGoogle Scholar
- Pickell PD, Hermosilla T, Frazier RJ, Coops NC, Wulder MA. Forest recovery trends derived from Landsat time series for North American boreal forests. Int J Remote Sens. 2016;37(1):138–49.View ArticleGoogle Scholar
- Wulder MA, Hilker T, White JC, Coops NC, Masek JG, Pflugmacher D, et al. Virtual constellations for global terrestrial monitoring. Remote Sens Environ. 2015;170:62–76.View ArticleGoogle Scholar