- Approaches to assessing risk:
- General: Analysis of remotely sensed active/burned area data to derive historical fire series
- Enhancing remotely sensed fire data with fire weather indices to reconstruct longer-term historic fire records
- Quantification of biomass lost during forest fires (fire severity) via remote sensing
- Remote sensing of fire risk enhanced with vegetation models
- Wildfire ensembles modelling – worst case scenarios for investments
- Analysis of forest carbon losses post fire for pan-tropic regions
- Analysis of the size distribution of fires and their impact
- Fire behaviour modelling and practical application
- Self-propagating mega-fires in the Mediterranean
Primarily UK focused work:
- Mapping spatial and temporal wildfire risk in the UK using: satellite & IRS data
- Using radar data to detect UK fires
- Forest fire danger ratings/indices and associated risk management systems – especially application in the UK
- Enhanced calculation and communication of UK wildfire severity forecasts
- Wildfire Threat Analysis for the UK
- Developing a risk of ignition and wildfire hazard map for Britain using Wildfire Threat Analysis
- Prediction of future UK fires under climate change scenarios (moorland approach that could be adapted to forests)
- UK Climate Change Risk Assessment & Fire
Fire is one of the largest causes of forest loss in the world. E.g. across Europe between 1950 and 2000, for example, fire is estimated to have accounted for 16% of the total wood damaged by disturbances (Schelhaas et al., 2003). Standardised global information on fire losses in forests is poor. FAO data supplied by national authorities suffers from inconsistencies in definition and quality; however, it shows that less than 1% of forest is affected by fire each year in most regions. Fires are most prevalent in Africa, particularly Western and Central areas, where 12% of land was affected. Many of these fires will not, however, be the tall plantation forests normally invested in by investors. Oceania was the only other region where fires affected more than 1% of land at 2.4% (FAO, 2010). Where fires do occur at the national and local level, they can of course be devastating.
The likelihood of fire is primarily driven by the presence of an ignition source and the availability of dry fuel to burn. The spread and intensity of a fire depends on a number of other factors including forest type, management regime (e.g. fire-fighting capability), topography, and meteorological conditions – especially temperature and humidity.
The most promising development in terms of monitoring fire risk is the development of appropriate remote sensing technologies. Remotely sensed data allows the monitoring of active fires as well as the historical analysis of the scale and extent of past fires through the identification of burned areas using specially developed algorithms. Many of these datasets are now available in a widely usable form e.g. The FAO’s Global Fire Information Management System (Link), and the European Commission’s European Forest Fire Information System (EFFIS) (Link), provide fire information from remote sensors in a simple format for a wide range of stakeholders. The information can provide basic trend information for countries and regions where there is an absence of quality historical data.
There are complex interactions and feedbacks between fire, vegetation type, land use and climate. Researchers have been developing approaches to modelling these interactions however it is early days for this sort of modelling and the ability to predict future risk is compounded by uncertainties in future climate projections of key variables such as precipitation and drought. It does, however, have high potential.
A number of fire danger indices exist which indicate the scale of the current threat to forests. Many national authorities publish daily indices for different areas of the country from these.
We present below a summary of the research being undertaken by leading academics in this field which seek to enhance and combine these approaches to derive new means of estimating historic and future fire risk.
Global historical fire data is now readily available from sensors on board satellites. Two different methods are used: either mapping the intense heat and light triggered by active fires, giving the location of fires occurring as the satellites passed over, or looking at changes in land surface colour after fires have passed over, producing maps of the area burned. The resolution and historical availability varies by sensor. The most commonly used sensors to provide historical fire data are MODIS and Landsat. MODIS fire information is available globally at 1km resolution for active fires, and 500m resolution for burned areas from 2000 (link). Landsat scenes are available at a far higher resolution than MODIS, and for a far longer time period: from 1972 at a 60 m resolution, and from 1982 at 30m resolution, but are not routinely processed for fire data so significant post-processing is necessary. The latest Landsat satellites (Landsat 7 & 8, from 2000 onwards) provide 15m resolution data in their panchromatic bands, theoretically allowing very small burn scars to be detected: though in all likelihood in practice fires would need to be at least a hectare in size (100 x 100 m or similar) to be easily detectable even from the latest Landsat data. Datasets from all these sensors are free to download and use, even for commercial purposes. The non-technical end user can view active fires from the last 24/48hours and fire history information for a particular location from MODIS using the online FIRMS tool (link). Academics working with GIS can provide tailored history and statistical analyses by downloading the historic datasets and processing them. As an example, processing historic MODIS data from 2000 for a country such as Cameroon could take half a day and around a day for a medium sized country. Coarser resolution data is available using the Meteosat satellites, which have low spatial resolution but very high temporal resolution (15 minutes), as these are geostationary satellites looking at the same area continuously. These satellites also allow an idea of fire intensity, rather than just extent, though their Fire Radiative Power products.
Example: Dr Thomas Smith processed GIS fire data for ForestRe Ltd (insurance). He analysed MODIS burned area data on China for the purposes of quantifying premiums for forest fire insurance. He provided datasets on monthly and annual burned areas, for each of the 22 provinces covering the whole of China, from 1997-2011. This took 7 days to complete.
Academics in our network with expertise in processing fire data via GIS for forestry include:
– Balzter, Prof. Heiko
– Mitchard, Dr Edward. (use of processed fire data for analysis)
– McMorrow, Julia – (particular expertise in the UK)
– Smith, Dr Thomas
– Wooster, Prof. Martin – (particular expertise in the development of new algorithms to process satellite data)
For additional areas of expertise see sections below.
Enhancing remotely sensed fire data with fire weather indices to reconstruct longer-term historic fire records
Prof. Balzter has developed a new approach of analysing remotely sensed burned area datasets from satellites over periods of around 20 years, and comparing them to local fire danger weather indices to evaluate and calibrate the influence of climate feedbacks on the fire regime. Using this dynamic, fire histories can be estimated over longer periods than that for which remotely sensed data is available, using longer term weather records. He has worked extensively on such an approach in Siberia as part of the EU project SIBERIA-2. He compared two burned area datasets from satellite observations: the L3JRC daily global burned area dataset derived from SPOT-VEGETATION and the FFID burned area dataset from MODIS and compared them to the Russian fire danger index – the Nesterov index (Balzter et al., 2010). The index uses daily temperature and dew point measurements. Regression analysis of the results showed that the number of days in which the index exceeded 4,300 explained about half of the interannual variability in the burned area assessed (R2 = 0.49). Such an approach takes around 3 months to develop. Contact: Prof. Heiko Balzter
Prof. Wooster is developing remote sensing methods to quantify the emissions and impacts of forest and grassland fires – through the analysis of thermal imagery from aircraft and satellites. He has developed an algorithm which estimates the rate of fuel consumption from the rate of radiative heat output of burning carbon and converts this to estimates of carbon emissions. Vegetation is typically around 50% ±5% carbon and the heat yield from the carbon burned is relatively constant regardless of the type of vegetation burned. The algorithms are usually tested on prescribed fires being conducted for management purposes, where comparison data are available, and then can be applied to unplanned ‘wildfires’ whose impacts we want to assess. Maps of carbon emitted per unit area can be created by dividing the tons of carbon emitted (estimated from the thermal methods) by the area burned. This information could have a particular application in loss adjusting by determining how much material was burned, not just the area burned. Because the thermal sensors are very sensitive to the strong thermal emissions from fires the satellites can detect fires covering only 0.01% of the pixel area. So for example a 1 km resolution sensor can potentially detect the heat output coming from a 100 square meter fire. This is far more sensitive that mapping of burned area, but the fire cannot be seen through cloud cover. Prof. Wooster’s group is now working with a new satellite sensor that provides 10x the sensitivity than this. The quality and temporal range of information available is dependent on current satellites. MODIS burned area data is available back to 1998 at 250-500 m resolution, whereas Landsat data is available back to the 1970s at higher resolution. However, the radiative heat information required has only been available from MODIS since 1998. There is a dataset already processed that provides fuel consumption data at coarse (10 km) resolution. Geostationary satellites provide continuously streamed information every 15 – 60 minutes, and can thus estimate fire intensity and carbon emissions at very high temporal resolution, but they may miss some of the smaller fires due to the coarser resolution (around 3 km). There are currently geostationary satellites over Europe (1), the Americas (2) and Asia (1). A new generation of geostationary satellites is in the process of being launched including a third generation of Meteosat satellites which Prof. Wooster has been involved in and which will provide data over Europe every 2.5 mins at 1km. Contact: Prof. Martin Wooster
In the EU project CARBOAFRICA, Prof. Balzter studied the African fire regime, using MODIS burned area data at 500 m scale and the dynamic fire model SPITFIRE, coupled to the vegetation model LPJ-GUESS. This work has led to improved greenhouse gas emission estimates from fire by using a combined fire/dynamic vegetation model forced with satellite-derived burned area data (Lehsten et al., 2009). CARBOAFRICA was awarded the Italian WWF (World Wildlife Fund for Nature) prize for scientific research in 2009. He has also investigated fire and vegetation structure in Kruger National Park (Khalefa et al., 2013). Contact: Prof. Heiko Balzter
Dr. Smith has developed a concept for worst-case scenario modelling for a particular investment. This would involve dropping random ignitions across a project location and modelling the spread of wildfire under different weather conditions, wind directions and so forth to determine expected average and worst-case losses. He is looking for funding and appropriate case studies to perform this modelling. Contact: Dr Thomas Smith.
Dr Mitchard was involved in a project that mapped carbon stocks in the pan-tropic regions (see Anthropogenic risk page). He has proposed determining the amount of biomass lost post fire by comparing burn scars from the MODIS remote sensor with these carbon stock maps to determine how much carbon was lost. This would be particularly useful for mixed areas of grassland and forest. Contact: Dr Edward Mitchard
Some remote sensors work at a higher resolution than others and therefore only capture the larger fires. Whilst it might be thought that these are the most damaging, in fact Prof. Balzter has found through his work that this does not always hold true. In some areas there are many small fires that collectively damage more area than the fewer larger fires. He found that the size distribution of fires varies in different ecosystems (Lehsten et al., 2014). Contact: Prof. Heiko Balzter
Dr Smith has adapted fire behaviour models (i.e. models that predict the extent and rate of spread of wildfires post ignition) for use in the UK and Europe. Such models are widely used in the US, Canada, Spain and Australia. He has also helped train fire fighters to use these fire behaviour models along with remotely sensed fire information to enhance fire-fighting. Dr Smith has experience in adapting/customising freely available off-the-shelf fire spread models (e.g. FARSITE, developed in the US; and Prometheus, developed in Canada) and developing the necessary inputs for these models (e.g. fuel maps, digital terrain maps, meteorological data) for use in novel environments (e.g. heather moorlands in the UK, and lowland forest in Denmark). Examples of training models and protocols can be found on the FIREfficient website (link). Contact: Dr Thomas Smith.
Prof. Balzter is working with a visiting scientist from Spain on self-propagating mega-fires which are increasing in the Mediterranean area. These are fires which create their own fire weather as they change the way that local winds behave, suck in local oxygen, and create uplift into fire whorls. These fire tornadoes then start moving across the landscape and are a huge risk for fire fighters. They can transport embers and sparks high into the air and create new fire outbreaks half a kilometre from their source. They are more common in California and Australia but are a new phenomenon in the Mediterranean. Urban development close to abandoned agricultural land which is starting to develop woody vegetation are also at risk. This is a new research topic and a paper is in progress but it is not a funded project. Contact: Prof. Heiko Balzter
Julia McMorrow’s expertise is in comparing the usefulness of different datasets for assessing the fire regime in Great Britain i.e. England, Wales and Scotland. She is aiming to determine the likelihood of fires of different sizes at the national and local scale. More recently she has begun work on assessing impact which will include burned area analysis and impacts on infrastructure. She has a particular interest in the new Incident Reporting System (IRS) dataset, which is a standardised record of the incidence of wildfires attended by the Fire and Rescue services in the UK since 2009. She has compared this data to that derived from the satellite-borne MODIS sensor on burned areas and active fires (available from 2002, via EFFIS and German Space Agency (DLR) Fire Service Statistics (McMorrow and Ogbechie, 2011. Link) (McMorrow, 2013). Julia used 2006 landcover data from the CORINE sensor, which covers Europe at 250 m resolution to screen out non-vegetation fires (e.g. hot chimney plumes). Updated CORINE data for 2012 will be available soon. UK landcover data at 25m resolution is available for 2007 (link) (and will be available soon for 2013) but CORINE was used as the resolution is more compatible with MODIS data. Her analyses show that whilst MODIS information is available for a longer period than IRS, it misses many smaller or less intense fires because it only has a resolution of 1 km and is affected by cloud cover and aerosols. IRS records a single ignition point, whereas MODIS records the fire front at the time of overpass which moves during successive satellite passes, and so it can be difficult to link MODIS data to the IRS ignition point. Comparison of four financial years of IRS wildfires and MODIS data for Scotland showed that only about 1 in 10 fires are detected by MODIS (Critchley and McMorrow, 2015. Link). Contact: Julia McMorrow
Julia McMorrow and her colleague, Gail Millin-Chalabi use synthetic aperture radar (SAR) to detect burn scars in peat moorland (Millin-Chalabi et al., 2014). Radar data sources include ESA’s Sentinel and Japan’s ALOS satellite’s PALSAR sensor. Radar has the advantage of penetrating cloud, particularly the C-band wavelength. The perimeter of a peat moorland fire can be detected up to 3 months after the fire. The dataset from Sentinel goes back to 2003 and is improving. The work focuses on moorland but could be adapted for forests. Airborne sensors can provide finer resolution and have been used to analyse fires on peatlands e.g. in the Peak District, but not forests. Gail uses SARscape – a tool for radar. Radar data from Sentinel is available via the Sentinel Toolbox. Contact: Julia McMorrow
Forest fire danger ratings/indices and associated risk management systems – especially application in the UK
Prof. Moffat has significant expertise in approaches to managing forest fire risk, the use of Forest Fire Danger Rating Systems, and, in particular, their usefulness and applicability to the UK context. He compared approaches in New Zealand with the UK, as both countries have similar temperate maritime climates and levels of wildfire occurrence. Whilst the UK’s approach is based around contingency planning and response, NZ’s encompasses a broader risk management framework to reduce and prepare for fire, as well as respond and recover from it (Moffat and Pearce, 2013a). His work has included a detailed evaluation of Forest Fire Danger Rating Systems, such as the Canadian and New Zealand approaches (adapted from the former). These approaches link fire indices (such as the Fire Weather Index module of the Canadian system), with fire behaviour and prediction modules (such as New Zealand’s fire behaviour predication module) to support operational activities. He has looked at how such approaches might be developed in the UK using the current Met Office Fire Severity Index (and/or outputs from the Natural Hazards Partnership) – (Moffat and Pearce, 2013b) (Full report: link). An effective assessment of medium to high fire risk is a precursor to obtaining co-funding from the EU Rural Development Regulation to protect forests from fire and support recovery. The absence of an effective system in the UK prevents such funding being obtained. Contact: Prof. Andy Moffat
Mark de Jong, working for Prof. Wooster, was seconded to the Met Office as part of the NERC PURE funding to develop better risk estimates for UK wildfires. Under the Crow Act (2000), the Canadian Fire Weather Index (CFWI) is used in the UK to determine the level of fire risk in the UK on a daily basis and landowners must close land that is at extreme risk of fire due to extremely dry conditions. However the thresholds for triggering this closure were uniform across the country, whereas a dry period in northern Britain that might be anomalous might be perfectly normal for southern Britain. The project worked to define local thresholds by determining the 80th, 90th, 95th, 97th and 99th percentiles at 2km resolution for the entire UK for each season. It then compared this to actual fires so that the daily forecasts could be set in local and historical context. 7,000 wildfires were analysed and the new calibrated index was valued at extreme for the majority of these fires whereas the old approach did not capture them. Contact: Prof. Martin Wooster
A team from Manchester University consisting of Julia McMorrow, Aleksandra Kazmierczak and Jonathan Aylen, worked on a NERC-funded scoping project to evaluate the potential use of Wildfire Threat Analysis (WTA) for the UK, in conjunction with the Forestry Commission using a forest-urban interface case study of the Swinley Forest Fire 2011. WTA is used in Canada and New Zealand and sees the threat of fire as a combination of: ‘risk of ignition’ (RoI), ‘hazard’ of fire spread, and ‘values at risk’ (VaR) which includes valuations for forest, health, infrastructure etc.. RoI and VaR modules were successfully developed in a GIS system via a structured evaluation process involving 2 workshops with a wide range of stakeholders. A hazard module could not be developed but Dr Smith (KCL) provided multiple runs of a fire behaviour model as an alternative to model where the fire might spread to and how fast. The WTA work is seeking further funding to complete but has outlined the potential for this approach. Further details see the post project report: ‘Wildfire Threat Analysis (WTA): NERC-funded scoping project with Forestry Commission.’ Julia’s primary contribution focused on using IRS data to determine the risk of ignition (RoI) factor. Primary contact: Julia McMorrow
Julia McMorrow would like to develop a national risk of ignition and wildfire hazard map along with colleagues at KCL as a follow-up to the Wildfire Threat Analysis project (see previous section). Julia would define risk of ignition using IRS data and geographic factors derived from the Swinley project. Hazard information would cover how intense a fire might be and how dangerous, using factors such as fire climate (provided at 2km resolution by Mark de Jong, KCL); slope (taken from a digital elevation model. fires move uphill more quickly than downhill) and fuel (from a landcover map). The final dimension for a full WTA would be the risk of a fire starting and the hazard of it spreading in relation to the value of assets at risk. The value of ecosystem services such as biodiversity is difficult to quantify and values at risk are best understood and evaluated at the local scale so, at present this, would not be done at the national scale. The project would focus instead on using the risk of ignition and hazard to identify those gridcells at most risk, which will pinpoint where more detailed analysis of VaR is needed. Local values at risk could be determined by e.g. mapping houses around proposed forest stands. Contact: Julia McMorrow
Prediction of future UK fires under climate change scenarios (moorland approach that could be adapted to forests)
Julia McMorrow and colleagues, including Jonathan Aylen from Manchester Business School, have experience of developing future forecasts of the incidence of wildfire in moorlands. Daily weather, day of the week (as a proxy for density of human ignition sources) and fire data, if available for sufficiently long periods (~30 years), can be statistically modelled to give the probability of fire occurring (Albertson et al., 2009). This approach has not yet been done for forests, however, they would be interested in developing similar approaches for forests in the UK if there was a location with a sufficiently long fire record. They have also combined this statistical model for the Peak District National Park with the UKCP09-based weather generator to predict future fire trends based on the relationship between weather, people and historic fires as modelled by Probit regression (Albertson et al., 2010. Link). It is not yet possible to determine if this might work in other areas as there is insufficient fire data for most other locations. All fires need to be recorded, not just the large ones, and daily weather data is needed. Contact: Julia McMorrow
Prof Moffat was the lead author of the first Defra Climate Change Risk Assessment for the Forestry sector in the UK and has a broad level of expertise on climate change risks to UK forests. His personal focus is on the risks to forestry and land regeneration from wildfire in the UK. Fire has recently been deprioritised as a risk due to the decline in large scale planting and the focus on threats to more mature trees such as wind and pests and diseases amongst other reasons. However, the last three decades have seen increasing wildfires in the UK and climate change is likely to significantly increase this further (Moffat et al., 2012). Contact: Prof. Andy Moffat
ALBERTSON, K., AYLEN, J., CAVAN, G. & MCMORROW, J. 2009. Forecasting the outbreak of moorland wildfires in the English Peak District. Journal of Environmental Management, 90, 2642-2651.
ALBERTSON, K., AYLEN, J., CAVAN, G. & MCMORROW, J. 2010. Climate change and the future occurrence of moorland wildfires in the Peak District of the UK. Climate Research, 45, 105-118.
BALZTER, H., TANSEY, K., KADUK, J., GEORGE, C., GERARD, F., GONZALEZ, M. C., SUKHININ, A. & PONOMAREV, E. 2010. Fire/Climate Interactions in Siberia. In: BALZTER, H. (ed.) Environmental Change in Siberia: Earth Observation, Field Studies and Modelling. Dordrecht: Springer.
FAO 2010. Global Forest Resources Assessment 2010. FAO Forestry Paper 163. Rome: Food and Agriculture Organization of the United Nations.
CRITCHLEY, T., MCMORROW, J. (2015) A Comparison of Fire Service IRS and MODIS-Detected Vegetation Fires in Scotland. Poster presented at Wildfires 2015, Cambuslang, Glasgow, 10-11 Nov. (Link)
KHALEFA, E., SMIT, I. P. J., NICKLESS, A., ARCHIBALD, S., COMBER, A. & BALZTER, H. 2013. Retrieval of Savanna Vegetation Canopy Height from ICESat-GLAS Spaceborne LiDAR With Terrain Correction. Ieee Geoscience and Remote Sensing Letters, 10, 1439-1443.
LEHSTEN, V., TANSEY, K., BALZTER, H., THONICKE, K., SPESSA, A., WEBER, U., SMITH, B. & ARNETH, A. 2009. Estimating carbon emissions from African wildfires. Biogeosciences, 6, 349-360.
LEHSTEN, V., DE GROOT, W. J., FLANNIGAN, M., GEORGE, C., HARMAND, P. & BALZTER, H. 2014. Wildfires in boreal ecoregions: Evaluating the power law assumption and intra-annual and interannual variations. Journal of Geophysical Research-Biogeosciences, 119, 14-23.
MCMORROW, J. 2013. MODIS-detected fire regime in Great Britain: potential and challenges of validating against national fire incident data. In: Tansey, Kevin. Quantifying the Environmental Impact of Forest Fires: EARSeL Forest Fire Special Interest Group workshop; 15 Oct 2013-17 Oct 2013; Coombe Abbey, Warwickshire. Leicester: University of Leicester; 2013. p. 136-140. Link
MCMORROW, J. & OGBECHIE, O. 2011. MODIS-detected fire regime for Great Britain; 2007-2011. In Wildfire 2011; 14 Oct 2013-15 Oct 2013; Buxton, Derbyshire. Ripon: Rural Development Initiatives (RDI).
MILLIN-CHALABI, G., MCMORROW, J. & AGNEW, C. 2014. Detecting a moorland wildfire scar in the Peak District, UK, using synthetic aperture radar from ERS-2 and Envisat ASAR. International Journal of Remote Sensing, 35, 54-69.
MOFFAT, A. J., MORISON, J. I. L., NICOLL, B. & BAIN, V. 2012. Climate Change Risk Assessment for the Forestry Sector. UK 2012 Climate Change Risk Assessment. Crown Copyright.
MOFFAT, A. J. & PEARCE, H. G. 2013a. Contrasting approaches to forest fire risk in New Zealand and Great Britain. Scottish Forestry, 67 (4), 10-17.
MOFFAT, A. J. & PEARCE, H. G. 2013b. Harmonising approaches to evaluation of forest fire risk. A report by Forest Resarch & Scion, supported by Tranzfor.
SCHELHAAS, M. J., NABUURS, G. J. & SCHUCK, A. 2003. Natural disturbances in the European forests in the 19th and 20th centuries. Global Change Biology, 9, 1620-1633.