Article | Open | Published:

An early warning system to forecast the close of the spring burning window from satellite-observed greenness

Scientific Reportsvolume 7, Article number: 14190 (2017) | Download Citation


Spring represents the peak of human-caused wildfire events in populated boreal forests, resulting in catastrophic loss of property and human life. Human-caused wildfire risk is anticipated to increase in northern forests as fuels become drier, on average, under warming climate scenarios and as population density increases within formerly remote regions. We investigated springtime human-caused wildfire risk derived from satellite-observed vegetation greenness in the early part of the growing season, a period of increased ignition and wildfire spread potential from snow melt to vegetation green-up with the aim of developing an early warning wildfire risk system. The initial system was developed for 392,856 km2 of forested lands with satellite observations available prior to the start of the official wildfire season and predicted peak human-caused wildfire activity with 10-day accuracy for 76% of wildfire-protected lands by March 22. The early warning system could have significant utility as a cost-effective solution for wildfire managers to prioritize the deployment of wildfire protection resources in wildfire-prone landscapes across boreal-dominated ecosystems of North America, Europe, and Russia using open access Earth observations.


The importance of changing wildfire regimes under a changing climate has become a major focus of research and management efforts globally1. In the boreal forests of North America it is anticipated that climate warming will result in increases in fuel availability and wildfire severity2 with concomitant increases in wildfire danger and wildfire season length3, particularly for the circumboreal forest4. The effects of increasing human activity such as urbanization and resource extraction within boreal forests remain poorly understood with respect to potential human-caused wildfire ignitions and wildfire risk5,6. Humans ignited approximately 57% of all wildfires in Alberta, accounting for 63% of annual area burned (~200,000 ha) between 1996 and 20147. More than two-thirds of human-caused wildfires in Alberta occur during what is recognized across Canada as the spring burning window8, a period of increased ignition and wildfire spread potential spanning the time between the disappearance of snow and vegetation green-up. The spring burning window is most pronounced in Alberta, which ends when understory and overstory vegetation produce fresh leaves that are high in moisture content, reducing the potential of wildfire ignition and spread8. Alberta is a model jurisdiction for investigating human-caused wildfire risk during the springtime due to the confluence of human ignition sources and a seasonal change in fuel moisture modulated by climate and leaf phenology. For these reasons, May is the most active wildfire month resulting in the highest monthly area burned in Alberta7, and springtime burning is common to circumboreal ecosystems across North America, Europe, and Russia9,10,11.

Previous research has shown that the moisture content in fine fuels is a critical risk factor of human-caused wildfire activity12,13. Fine fuels with diameters smaller than 2.5 cm tend to respond rapidly to abrupt changes in weather conditions (i.e., temperature, wind speed, relative humidity, and precipitation) due to a high surface area to volume ratio that increases rates of evapotranspiration compared with larger fuels14. Current models of fuel moisture, such as the Canadian Forest Fire Danger Rating System, integrate wildfire weather data with empirical knowledge of evapotranspiration rates for different forest fuel types to estimate daily fuel availability and potential wildfire behavior14. These models are ledgers of fuel moisture that capture moisture content of fine fuels and large coarse fuels including deep soils and duff, accounting for daily wetting and drying as well as seasonal drying factors15. The Canadian system has been broadly adopted around the world, including across Europe, however similar systems of wildfire danger rating exist such as the National Fire Danger Rating System in the United States16 and the MacArthur Forest Fire Danger Index in Australia17. Although these systems are widely used around the world to estimate wildfire danger and wildfire behavior, the weather-based inputs are not direct measurements of fuel moisture, and the Canadian system makes broad assumptions about the homogeneity of fuel types. As a result, there remains unaccounted spatial variability in the estimate of wildfire behavior within and between wildfires from these indirect fuel moisture models18.

Satellite observations of vegetation greenness have been successfully used to estimate wildfire danger codes and fuel moisture codes for northern boreal forests19,20. Satellite observations from the Moderate Resolution Imaging Spectroradiometer (MODIS) provide a unique opportunity to observe and monitor fuel moisture from the flush of new vegetation, since healthy leaf area has long been shown to have a strong relationship with spectral greenness indices and leaf moisture21. Since leaf area is a proxy for foliar fine fuel moisture8 and is readily measured directly from satellite-derived greenness indices, fewer assumptions are required about site-specific fuel types when assessing wildfire spread potential. Moreover, using cloud-free composites of MODIS imagery, surface reflectance observations are available nearly every eight days at a spatial resolution of 250 m for the entire planet22. Thus, the spectral, spatial, and temporal resolutions of the satellite observations are ideal for monitoring the spring burning window as expressed by vegetation greenness for large areas.

In this paper we propose that satellite-observed greenness trends will be useful for estimating the peak of the spring burning window for any location, thus forming an early warning system for forecasting human-caused wildfire activity, which has wide application in forested ecosystems beyond the focus site. To do so, we investigated the relationship between fine temporal changes in satellite-derived greenness, and the commencement of the spring burning window as quantified by human wildfire ignitions in Alberta’s Forest Protection Area, the forested areas where wildfires are actively managed using provincial resources. To facilitate the development of the early warning system, we provide an assessment of the model error (in days) as a function of day-of-year of observation acquisition and show ground conditions of modelled predictions using a ground network of phenology cameras. The developed relationships of the spring burning window can augment existing decision support systems by providing near real-time proxy information of fuel moisture to wildfire managers for allocating suppression resources and planning prescribed burns.


Study area

The province of Alberta comprises 661,848 km2 of land and freshwater ecosystems23, of which approximately 42% is considered to be forested24. The provincial government actively manages a wildfire protection area totaling 392,856 km2 or approximately 59% of the province25. The wildfire protection area spans 16 natural sub-regions with varying continental and interior climates (Fig. 1a). Growing degree days >5 °C ranges from 317 in the Alpine to 1,388 in the Foothills Fescue and the mean date of last spring frost ranges from May 27 (day-of-year [DOY] 147) in the Athabasca Plain and Peace-Athabasca Delta to July 13 (DOY 194) in the Alpine26. Topography of the region is highly variable with elevations ranging from 150 m in the Northern Mixedwoods and Kazan Uplands to 3,650 m in the Alpine26. Mean annual precipitation ranges from 377 mm in the Peace-Athabasca Delta to 989 mm in the Alpine, of which, 68% and 48% falls during the growing season, respectively26. Approximately 74% of the wildfire protection area is considered boreal forest, dominated by pure and mixedwood stands of Populus tremuloides, Populus balsamifera, Picea mariana, Picea glauca, Pinus contorta, Pinus banksiana, and Larix laricina. Distribution of various vegetation types are determined climatically by precipitation and minimum annual temperature (i.e., cold hardiness), which vary along the large latitude and elevation gradients that dominate the province.

Figure 1
Figure 1

(a) Map of natural sub-regions within Alberta’s wildfire protection area: (1) Alpine; (2) Athabasca Plain; (3) Boreal Subarctic; (4) Central Mixedwood; (5) Dry Mixedwood; (6) Foothills Fescue; (7) Foothills Parkland; (8) Kazan Uplands; (9) Lower Boreal Highlands; (10) Lower Foothills; (11) Montane; (12) Northern Mixedwood; (13) Peace-Athabasca Delta; (14) Subalpine; (15) Upper Boreal Highlands; (16) Upper Foothills. (b) Mean maximum greenness as expressed by the Normalized Vegetation Differenced Index (NDVI) for years 2000 to 2016. Maps created using ArcMap 10.5.1. (

Human-caused wildfire ignitions data

Wildfire ignitions for Alberta were retrieved from the historical wildfire database for the years 2000 to 20147. The historical wildfire database contains wildfire ignitions observed from a variety of sources including lookout observers, forest officers, the public (310-FIRE), wildfire patrols, aircraft, wildfire crews, infrared scans, and satellite imagery. Attributes are recorded for each wildfire ignition including the estimated date of ignition, the cause of ignition, and wildfire size after being extinguished. Small wildfire ignitions (<0.1 ha) likely represented illegal campfires and were removed from the dataset. Ignitions caused by lightning or without sufficient information on the date of occurrence were also removed. The remaining human-caused wildfire ignitions between 2000 and 2014 totaled N = 5,325 and accounted for approximately 45% of the total area burned represented in the database and approximately 92% of human-caused area burned. The number of human-caused ignitions that occurred between February 26 (DOY 57) and July 15 (DOY 196) totaled n = 4,017 or approximately 75% of all human-caused wildfire ignitions across all calendar years.

Modelling human-caused wildfire ignitions

Recent studies have used lightning ignition data to estimate wildfire risk during the wildfire season27,28. However, wildfire risk is known to vary according to ignition source and season29. For example, it has been postulated that deciduous leaf phenology in the western Canadian interior forest provides a major climate feedback on seasonal patterns of temperature and precipitation, with the possible conclusion that evapotranspiration from the leaf-on forest is at least partly responsible for the development of thunderstorms in the region30. Thus, lightning ignitions are not a suitable proxy for human-caused wildfire activity during leaf-off conditions, and local environmental conditions such as rapid springtime greening of vegetation, time since last snow cover (or some other suitable representation of the moisture content of deep, compact organic layers), and human-caused wildfire ignitions also need to be considered. The timing of vegetation greening shifts inter-annually, however vegetation greening has been shown to be relatively stable in terms of amplitude31. By contrast, it is challenging to know a priori when and where humans ignite wildfires or what weather conditions will be prevalent beyond several weeks. Therefore, our approach was first to relate the timing of human-caused ignitions to vegetation green-up and then use the satellite remote sensing data to predict green-up dates as an indicator of peak springtime human-caused wildfire risk.

Human-caused wildfire risk was modelled by converting wildfire ignition counts to wildfire days, defined as any day of the calendar year when one or more human-caused wildfire ignitions are reported. A wildfire day is a meaningful concept to wildfire managers who require a prediction of whether a given location will require wildfire protection resources on a given day and provides a well-established, robust statistical random variable for binomial modelling32. The probability of a human-caused wildfire day during a given year may be estimated as a function of a Poisson distribution, which approximates the binomial process, where the number of wildfire starts occurring on any given DOY for all available years 2000–2014 from the historical wildfire database is the rate parameter λ12.

Satellite observations

The MODIS sensor onboard the Aqua and Terra satellites acquires imagery in the red and near infrared spectral wavelengths for most northern latitudes on a daily basis at 250 m spatial resolution. We retrieved eight-day surface reflectance image composites generated for the MOD09Q1 Version 6 product between February 26 (DOY 57) and July 12 (DOY 193) for years 2000 to 201622. The eight-day MODIS composites represented a compromise between obtaining sufficient observations to track vegetation phenology over a given season while minimizing data gaps caused by clouds and atmospheric conditions22. For every year, a total of 18 eight-day composite MODIS images were available. The MODIS images were downloaded directly from the United States Geological Survey22, converted to a 10 degree North American Datum 1983 Transverse Mercator projection, and mosaicked to cover the province of Alberta. Wildfire managers in Alberta use broad ecological natural sub-regions to manage wildfire protection resources, so we adopted those regions for summarizing trends of satellite-derived greenness across the wildfire protection area of the province26.

Modelling vegetation green-up

Greenness was estimated for each eight-day MODIS composite using the normalized difference vegetation index (NDVI)21,33. We first excluded low-quality observations using the quality control data distributed with the MOD09Q1 product. The annual NDVI curves were averaged to derive a mean greenness normal for each DOY and then each season between 2000 and 2016 was screened for noise by converting the annual NDVI observations to a z-score. Observations were flagged as noise if they exceeded the one-sided upper tail critical value at α = 0.10. Assessing the upper tail ensured that observations with significantly lower greenness than expected from the 17-year average were flagged as noise since these observations were likely related to error in atmospheric correction procedure. For each year from 2000 to 2016, a one-dimensional cubic spline was fit34 to the seasonal NDVI values for each MODIS pixel to temporally interpolate the 8-day MODIS composites to daily estimated values of NDVI. Green-up was defined as the date when the temporally interpolated NDVI curve reached the mean of the minimum and maximum NDVI values for the season, also known as the half-maximum35. This procedure resulted in a spatially-explicit estimation of vegetation green-up across the region. A mask was applied to extract pixels occurring within the wildfire protection area of Alberta and green-up statistics were further summarized by natural sub-regions (Fig. 1a). Vegetation productivity varies considerably across the wildfire protection area and the mean maximum vegetation greenness as expressed by NDVI is shown in Fig. 1b for the years 2000 to 2016. Green-up was not estimated for seasonal NDVI pixel series with more than five noisy observations since this degree of missing data can result in inaccurate curve fitting. Additionally, our initial analyses showed that the maximum NDVI value is most likely observed in the last two dates of our seasonal series, so if those dates were noisy then we did not estimate green-up.

Forecasting vegetation green-up

We tested the ability of the half-maximum greenness system to accurately forecast the peak spring burning window by simulating real-time availability of satellite observations throughout each annual season. This was achieved by sequentially replacing the 2000–2016 mean NDVI values with actual satellite observations for any given day-of-year and then recalculating the green-up date using the new greenness curve. The greenness curves that emerged from concatenating actual greenness observations with the mean greenness normal were then used to estimate the green-up date for the cumulative set of satellite observations {NDVIi=0, … NDVIn} in a year, where i indexes the observation position in the year out of n = 18 total observations in any year. The mean absolute error (in days) between the observed and the forecasted green-up DOY from the cubic spline model was calculated for each pixel and for each satellite observation replaced sequentially in the season. This process was repeated for all years and days-of-year, up to and including July 4 (DOY 185).

Validating vegetation green-up

A network of 16 phenological cameras were installed in situ in the Foothills, Subalpine, and Alpine natural sub-regions during the growing season between April and October in 200936,37,38. The cameras acquired five red-green-blue photos daily over an hour at noon local standard time to compensate for varying shadow and illumination effects in the scene. From these sets of images, we show some examples of the forest condition on or around the dates that we predict high probability of a wildfire day based on the half-maximum green-up model.

Data availability

Alberta wildfire ignitions are publicly available from The 8-day surface reflectance MODIS products (MOD09Q1) are publicly available from All processed data and results including estimations of vegetation green-up and forecast simulations are available from the corresponding author upon reasonable request.


Human-caused wildfire activity during the years 2000 and 2014 peaked in May (Fig. 2a). Days on which the probability of a human-caused wildfire day exceeded 20% ranged from May 1 to May 17 (Fig. 2b). Human-caused wildfires burned an average of 74,872 ha per year between 2000 and 2014, and approximately 95% of total area burned by human-caused wildfires across all years burned during the month of May.

Figure 2
Figure 2

(a) Total number of human-caused wildfire starts by date between 2000 and 2014 and (b) the probability of a human-caused wildfire day modelled as a binomial process by eight-day intervals corresponding to the frequency of satellite observations of greenness.

Vegetation green-up varied from year-to-year within each natural sub-region (Fig. 3). The Alpine and Subalpine natural sub-regions consistently showed the latest green-up dates across all years and natural sub-regions. The earliest observed green-up occurred on April 7, 2015 (day-of-year [DOY] 97) in the Montane natural sub-region, April 7, 2006 in the Upper Foothills natural sub-region, and April 7, 2010 in the Athabasca Plain natural sub-region. The latest observed green-up was on June 13, 2002 (DOY 164) in the Alpine natural sub-region. Much of the wildfire protection area was greener earlier in the season for the years 2005, 2006, 2010, 2015, and 2016 (Fig. 3). The years 2002, 2008, and 2013 showed that vegetation green-up was later in the season (Fig. 3).

Figure 3
Figure 3

Green-up date for each natural sub-region modelled from half-maximum greenness for all years. Earlier dates represent earlier than average green-up and later dates represent later than average green-up relative to the entire wildfire protection area for all years (April 29). Maps created using ArcMap 10.5.1. (

Springtime human-caused wildfire activity was concomitant with satellite-observed green-up. Human-caused wildfire ignitions had a wider range in the season compared with green-up in the wildfire protection area. The interquartile ranges for vegetation green-up and human-caused wildfire ignitions overlapped completely for all years, confirming that wildfire activity peaked shortly following vegetation green-up during the spring burning window (Fig. 4). The median green-up date for all years was May 3 (DOY 123) compared with May 8 (DOY 128) for human-caused wildfire ignitions.

Figure 4
Figure 4

Barplot showing significant overlap in the interquartile ranges of the green-up predictions derived from normal greenness curves and human-caused wildfire starts during the spring burning window. The median green-up date was May 3 (day-of-year [DOY] 123) compared with May 8 (DOY 128) for human-caused wildfire ignitions.

The mean absolute error (days) for the forecasted green-up date decreased in all natural sub-regions as more observations were made available in the season (Figs 5 and 6). By March 22 (DOY 81), vegetation green-up was forecasted to be within 10 days of the actual green-up date in any given year for 76% of the wildfire protection area (Fig. 5). A 10-day window of accuracy was not achieved until later in the season for the non-boreal natural sub-regions (Fig. 6): Alpine (May 25); Lower Foothills (April 7); Montane (April 23); Subalpine (May 25); and Upper Foothills (April 23).

Figure 5
Figure 5

Forecasting the green-up day-of-year (DOY) was simulated for each year by successively adding greenness observations to the mean greenness curve and calculating the mean absolute error (MAE) within each natural sub-region. The natural sub-regions shown represent 76% of the wildfire protection area and the green-up DOY was forecasted to be within 10 days of the actual green-up DOY by March 22 (DOY 81).

Figure 6
Figure 6

Forecasting the green-up day-of-year (DOY) was simulated for each year by successively adding greenness observations to the mean greenness curve and calculating the mean absolute error (MAE) within each natural sub-region. The natural sub-regions shown represent 24% of the wildfire protection area and the green-up DOY was forecasted to be more than 10 days from the actual green-up DOY by March 22 (DOY 81).

From phenological camera observations within the Foothills and Subalpine natural sub-regions, it is apparent that leaf-off conditions in mixedwood stands were concomitant with the timing of peak human-caused wildfire activity in early May (Figs 7 and 8). In a coniferous site in the Alpine natural sub-region, changes in greenness were more subtle and consequently fine fuel moisture content was less obvious (Fig. 9). Snow cover was evident in all of the photos around the predicted green-up date, indicating possible interference in the signal-to-noise ratio for NDVI and may explain the high variability in predictions from year-to-year in these regions.

Figure 7
Figure 7

Leaf-off trembling aspen (Populus tremuloides) shown on clear dates (both photos taken on April 29, 2009) for two mixedwood sites around the mean predicted green-up date (April 27, 2009) for the Lower Foothills natural sub-region using the half-maximum greenness method. Photo by C.W. Bater.

Figure 8
Figure 8

Mixedwood site in the Subalpine natural sub-region of Alberta. Leaf-off condition of balsam poplar (Populus balsamifera) shown on a clear day (May 7, 2009) with recent snow cover around the mean predicted green-up date (May 15, 2009) for the Subalpine natural sub-region using the half-maximum greenness method. Photo by C.W. Bater.

Figure 9
Figure 9

Coniferous site shown on a clear day (April 29, 2009) with recent snow cover around the mean predicted green-up date (April 28, 2009) for the Upper Foothills natural sub-region using the half-maximum greenness method. Photo by C.W. Bater.


Globally, northern boreal forests are seeing an increase in industrial activity, including the establishment of road networks, oil and gas development, timber harvesting, and mining39. These activities have transformed historically remote boreal forests into anthropogenically-modified landscapes that pose increased risks for human-caused wildfire ignitions. As human uses of boreal forests increases in the near term, we can expect that the legacy of wildfire suppression from the beginning of the 20th century may exacerbate persistent wildfire hazards around settlements and critical infrastructures.

The economic costs associated with wildfire suppression are recognized to be very high in the early months of the wildfire season and the variability of wildfire-fighting costs from year-to-year increases uncertainty for wildfire managers40. Additionally, increased human-caused wildfire activity can be expected to increase social costs such as displacement from evacuations and increased smoke concentrations near settlements and over large areas41. The wildfire that devastated the town of Fort McMurray, Alberta, in May 2016 was the costliest natural disaster in Canadian history and the second costliest wildfire globally, resulting in an estimated C$3.58 billion insurable damages42 and the mandatory evacuation of approximately 88,000 residents43. Additionally, the proximity of the Fort McMurray wildfire to one of the largest oil mining operations globally resulted in an estimated loss of 800,000 barrels per day due to production stoppages44. Thus, an economical early warning system is critical for forecasting wildfire spread potential and mitigating these socioeconomic costs in the wildland-urban interface from human-caused wildfires40,45. With the rapid increase in wildfire suppression activities in recent decades, the wildfire statistics in Alberta demonstrate that the spring burning window remains the most dangerous part of the year with respect to loss of property and infrastructure40.

The ability of wildfire ignitions to quickly spread in early spring is a well-known phenomenon attributed to a dip in foliar moisture and the drying of fine fuels8. These vegetation conditions contribute to increased wildfire frequency, area burned, and wildfire size during springtime. Historical and contemporary wildfire records show that the spring burning window results in increased wildfire activity between May and June, which coincide with observable rapid changes in the vegetation phenology using remote sensing observations from this study. The increasing popularity and use of inexpensive visible spectrum digital cameras offers notable potential for the monitoring and measurement of phenological events such as spring green-up37,46. Repeat photography allows for very fine temporal sampling, often at daily or hourly intervals, for monitoring vegetation. By mounting these camera systems on towers, data can be acquired at an intermediate scale of observation, providing a link between field-based observation methods and satellite-derived estimates. In addition, the cameras provide a visual link between satellite-observed greenness over larger spatial scales to field-based measurements and observations from local experts. Our results demonstrate that the phenological camera network utilized in this study provides evidence of leaf-off conditions in mixedwood sites during peak human-caused wildfire activity in early May. To be effectively integrated into a near-realtime monitoring and forecasting system, however, these remote cameras would require additional investment in wireless technology for downlinking the imagery.

Some challenges remain for accurately forecasting vegetation green-up in higher elevation areas with shorter growing seasons and less pronounced deciduous vegetation. In Alberta, these areas account for only 24% of the wildfire protection area (Fig. 6) and are sparsely populated (i.e., lower overall human-caused wildfire risk). The phenological camera observations demonstrated that green-up was reliably detected within mixedwood forests (Figs 7 and 8), underscoring the reliability of NDVI for observations with low signal-to-noise ratio. NDVI is quite sensitive to initial greening while leaf area index is below 247, which is ideal for studying vegetation green-up. In sparser and coniferous forests occurring at higher elevations, the sensitivity of NDVI may outperform other vegetation indices that are less sensitive to lower leaf area index values. Additionally, MODIS is limited to the number of vegetation indices that may be calculated at 250 m spatial resolution because there are only two bands that are acquired at this scale. The signal-to-noise ratio may increase for NDVI with larger pixel sizes (e.g., 500 m, 1000 m) and this could possibly degrade the accuracy of a forecast especially for the higher elevation forests.

Simulation of the greenness curves from the end of winter to the end of vegetation green-up indicate that the vegetation green-up period was forecasted with 10-day accuracy for the majority of wildfire protected lands in Alberta by March 22 (DOY 81). This provides key evidence that vegetation greenness may be used as a reliable proxy for fine fuel moisture content. The eight-day composites limit the temporal precision available from the satellite data and introduce some uncertainty in the forecast. There is a fundamental trade-off between acquiring cloud-free satellite observations from MODIS and having sufficient temporal resolution to undertake the forecast on a near-weekly basis. Although in some cases a wildfire start date may be estimated, we report the probability of a wildfire day in terms of the 8-day MODIS intervals (Fig. 2). This procedure ensures that any uncertainty in the wildfire start date is minimized. For example, it would be very unlikely that a wildfire start date is off by more than 8 days because initial attack protocols are very successful and rely on timely reporting of wildfire starts. Despite this limitation, the early warning system was able to achieve 10-day accuracy very early in the wildfire season and the difference in timing of green-up and human-caused wildfire activity was only five days at the median. Moreover, satellite-observed greenness data are standardized and processed on a continual basis as observations become available. The early-warning system developed here offers significant benefits to wildfire managers for prioritizing the deployment of wildfire protection resources.

Additional information

Publisher's note: Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.


  1. 1.

    Bowman, D. M. J. S. et al. Fire in the Earth System. Science (5926). 324, 481–484 (2009).

  2. 2.

    Barichivich, J. Responses of boreal vegetation to recent climate change. (University of East Anglia, 2014).

  3. 3.

    Jolly, W. M. et al. Climate-induced variations in global wildfire danger from 1979 to 2013. Nat. Commun. 6, 7537 (2015).

  4. 4.

    Flannigan, M., Stocks, B., Turetsky, M. & Wotton, M. Impacts of climate change on fire activity and fire management in the circumboreal forest. Glob. Chang. Biol. 15, 549–560 (2009).

  5. 5.

    Smith, A. M. S. et al. The Science of firescapes: achieving fire-resilient communities. Bioscience 66, 130–146 (2016).

  6. 6.

    Moritz, M. A. et al. Learning to coexist with wildfire. Nature 515, 58–66 (2014).

  7. 7.

    Alberta Agriculture and Forestry. Alberta Historical Wildfire Database: 1996–2014. Available at: (Accessed: 29th September 2017) (2015).

  8. 8.

    Jolly, W. M., Hintz, J., Kropp, R. C. & Conrad, E. T. Physiological drivers of the live foliar moisture content ‘spring dip’ in Pinus resinosa and Pinus banksiana and their relationship to foliar flammability. in VII International Conference on Forest Fire Research (ed. Viegas, D. X.) 1–9 (2014).

  9. 9.

    Shvidenko, A. & Nilsson, S. Extent, distribution, and ecological role of fire in Russian forests. in Fire, Climate Change, and Carbon Cycling in the Boreal Forest (eds Kasischke, E. & Stocks, B.) 132–150 (Springer-Verlag, 2000).

  10. 10.

    de Groot, W. J. D. et al. A comparison of Canadian and Russian boreal forest fire regimes. For. Ecol. Manage. 294, 23–34 (2013).

  11. 11.

    Johnson, E. Fire and Vegetation Dynamics: Studies from the North American Boreal Forest. (Cambridge University Press, 1992).

  12. 12.

    Cunningham, A. A. & Martell, D. L. A stoachastic model for the occurrence of man-caused forest fires. Can. J. For. Res. 3, 282–287 (1973).

  13. 13.

    Martell, D. L., Bevilacqua, E. & Stocks, B. J. Modelling seasonal variation in daily people-caused forest fire occurrence. Can. J. For. Res. 19, 1555–1563 (1989).

  14. 14.

    Van Wagner, C. Development and structure of the Canadian Forest Fire Weather Index System. Forestry Technical Report 35 (Canadian Forestry Service, 1987).

  15. 15.

    Turner, J. A. & Lawson, B. D. Weather in the Canadian forest fire danger rating system: A user guide to national standards and practices. Rep. BC-X-177 (Environment Canada, 1978).

  16. 16.

    Bradshaw, L. S., Deeming, J. E., Burgan, R. E. & Cohen, J. D. The National Fire-Danger Rating System–1978. (USDA Forest Service, 1983).

  17. 17.

    Noble, I. R., Gill, A. M. & Bary, G. A. V. McArthur’s fire‐danger meters expressed as equations. Aust. J. Ecol. 5, 201–203 (1980).

  18. 18.

    Ferster, C. J., Eskelson, B. N. I., Andison, D. W. & Lemay, V. M. Vegetation mortality within natural wildfire events in the Western Canadian boreal forest: what burns and why? Forests 7, 1–20 (2016).

  19. 19.

    Leblon, B., García, P. A. F., Oldford, S., Maclean, D. A. & Flannigan, M. Using cumulative NOAA-AVHRR spectral indices for estimating fire danger codes in northern boreal forests. Int. J. Appl. Earth Obs. Geoinf. 9, 335–342 (2007).

  20. 20.

    Oldford, S., Leblon, B., Maclean, D. & Flannigan, M. Predicting slow‐drying fire weather index fuel moisture codes with NOAA-AVHRR images in Canada’s northern boreal forests. Int. J. Remote Sens. 27, 3881–3902 (2006).

  21. 21.

    Tucker, C. J. Red and photographic infrared linear combinations for monitoring vegetation. Remote Sens. Environ. 8, 127–150 (1979).

  22. 22.

    NASA LP DAAC. MOD09Q1: Surface Reflectance 8-Day L3 Global 250m. Version 6. NASA EOSDIS Land Processes DAAC, USGS Earth Resources Observation and Science (EROS) Center, Sioux Falls, South Dakota (, Available at: (Accessed: 29th September 2017) (2016).

  23. 23.

    Statistics Canada. Land and freshwater area, by province and territory. Available at: (Accessed: 29th September 2017) (2005).

  24. 24.

    Statistics Canada. Forest land by province and territory. Available at: (Accessed: 29th September 2017) (2011).

  25. 25.

    Government of Alberta. Forest Protection Area. Available at: (Accessed: 29th September 2017) (2017).

  26. 26.

    Natural Regions Committee. Natural Regions and Subregions of Alberta. Pub. No. T/852 (2006).

  27. 27.

    Albert-Green, A., Dean, C. B., Martell, D. L. & Woolford, D. G. A methodology for investigating trends in changes in the timing of the fire season with applications to lightning-caused forest fires in Alberta and Ontario, Canada. Can. J. For. Res. 43, 39–45 (2013).

  28. 28.

    Blouin, K. D., Flannigan, M. D., Wang, X. & Kochtubajda, B. Ensemble lightning prediction models for the province of Alberta, Canada. Int. J. Wildl. Fire 25, 421–432 (2016).

  29. 29.

    Taylor, S. W., Woolford, D. G., Dean, C. B. & Martell, D. L. Wildfire Prediction to Inform Fire Management: Statistical Science Challenges. Stat. Sci. 28, 586–615 (2013).

  30. 30.

    Hogg, E. H., Price, D. T. & Black, T. A. Postulated feedbacks of deciduous forest phenology on seasonal climate patterns in the Western Canadian interior. J. Clim. 13, 4229–4243 (2000).

  31. 31.

    Coops, N. C., Wulder, M. A., Duro, D., Han, T. & Berry, S. L. The development of a Canadian dynamic habitat index using multi-temporal satellite estimates of canopy light absorbance. Ecol. Indic. 8, 754–766 (2008).

  32. 32.

    Martell, D. L., Otukol, S. & Stocks, B. J. A logistic model for predicting daily people-caused forest fire occurrence in Ontario. Can. J. For. Res. 17, 394–401 (1987).

  33. 33.

    Zhang, K. et al. Vegetation greening and climate change promote multidecadal rises of global land evapotranspiration. Nat. Publ. Gr. 15956, (2015).

  34. 34.

    Flannigan, M. D. & Wotton, B. A study of interpolation methods for forest fire danger rating in Canada. Can. J. For. Res. 19, 1059–1066 (1989).

  35. 35.

    Schwartz, M. D., Reed, B. C. & White, M. A. Assesing satellite-derived start-of-season measures in the conterminous USA. Int. J. Climatol. 22, 1793–1805 (2002).

  36. 36.

    Bater, C. W., Wulder, M. A., Road, B., Nielsen, S. E. & Mcdermid, G. Design and installation of a camera network across an elevation gradient for habitat assessment. Instrum. Sci. Technol. 39, 231–247 (2011).

  37. 37.

    Bater, C. W. et al. Using digital time-lapse cameras to monitor species-specific understorey and overstorey phenology in support of wildlife habitat assessment. Environ. Monit. Assess. 180, 1–13 (2011).

  38. 38.

    Nijland, W., Coops, N. C., Nielsen, S. E. & Stenhouse, G. International Journal of Applied Earth Observation and Geoinformation Integrating optical satellite data and airborne laser scanning in habitat classification for wildlife management. Int. J. Appl. Earth Obs. Geoinf. 38, 242–250 (2015).

  39. 39.

    Gauthier, S., Bernier, P., Kuuluvainen, T., Shvidenko, A. Z. & Schepaschenko, D. G. Boreal forest health and global change. Science (6250). 349, 819–822 (2015).

  40. 40.

    Stocks, B. J. & Martell, D. L. Forest fire management expenditures in Canada: 1970–2013. For. Chron. 92, 298–306 (2016).

  41. 41.

    Beverly, J. L., Flannigan, M. D., Stocks, B. J. & Bothwell, P. The association between Northern Hemisphere climate patterns and interannual variability in Canadian wildfire activity. Can. J. For. Res. 41, 2193–2201 (2011).

  42. 42.

    Insurance Bureau of Canada. Northern Alberta Wildfire Costliest Insured Natural Disaster in Canadian History - Estimate of insured losses: $3.58 billion. Available at: (Accessed: 29th September 2017) (2016).

  43. 43.

    Government of Alberta. Final Update 39: 2016 Wildfires (June 10 at 4:30 p.m.). (2016). Available at: (Accessed: 29th September 2017)

  44. 44.

    U.S. Energy Information Administration. Fires near Fort McMurray are reducing Canada’s oil sands production. Available at: (Accessed: 29th September 2017) (2016).

  45. 45.

    Parisien, M.-A. Science can map a solution to a fast-burning problem. Nature 534, 297 (2016).

  46. 46.

    Graham, E., Riordan, E. C., Yuen, E. M., Estrin, D. & Rundel, P. W. Public Internet-connected cameras used as a monitoring system. Glob. Chang. Biol. 16, 3014–3023 (2010).

  47. 47.

    Gitelson, A. A. Wide Dynamic Range Vegetation Index for remote quantification of biophysical characteristics of vegetation. J. Plant Physiol. 161, 165–173 (2004).

Download references


This work was funded by the Western Partnership for Wildland Fire Science, Government of Alberta, and a Natural Sciences and Engineering Research Council (NSERC) Discovery grant to N. Coops.

Author information


  1. Department of Forest Resources Management, University of British Columbia, Vancouver, Canada

    • Paul D. Pickell
    • , Nicholas C. Coops
    •  & Colin J. Ferster
  2. Alberta Agriculture and Forestry, Edmonton, Canada

    • Christopher W. Bater
    •  & Jinkai Zhang
  3. Department of Renewable Resources, University of Alberta, Edmonton, Canada

    • Karen D. Blouin
    •  & Mike D. Flannigan


  1. Search for Paul D. Pickell in:

  2. Search for Nicholas C. Coops in:

  3. Search for Colin J. Ferster in:

  4. Search for Christopher W. Bater in:

  5. Search for Karen D. Blouin in:

  6. Search for Mike D. Flannigan in:

  7. Search for Jinkai Zhang in:


C.W.B., C.J.F., K.D.B. and N.C.C. conceived the study. P.D.P., N.C.C., and C.J.F. carried out the analyses of satellite data and wildfire ignition statistics. C.W.B. provided the phenological camera observations. P.D.P., N.C.C., C.J.F., C.W.B., K.D.B., M.D.F., and J.Z. wrote the manuscript.

Competing Interests

The authors declare that they have no competing interests.

Corresponding author

Correspondence to Paul D. Pickell.

About this article

Publication history






By submitting a comment you agree to abide by our Terms and Community Guidelines. If you find something abusive or that does not comply with our terms or guidelines please flag it as inappropriate.