Machine learning to predict final fire size at the time of ignition

Fires in boreal forests of Alaska are changing, threatening human health and ecosystems. Given expected increases in fire activity with climate warming, insight into the controls on fire size from the time of ignition is necessary. Such insight may be increasingly useful for fire management, especially in cases where many ignitions occur in a short time period. Here we investigated the controls and predictability of final fire size at the time of ignition. Using decision trees, we show that ignitions can be classified as leading to small, medium or large fires with 50.4±5.2% accuracy. This was accomplished using two variables: vapour pressure deficit and the fraction of spruce cover near the ignition point. The model predicted that 40% of ignitions would lead to large fires, and those ultimately accounted for 75% of the total burned area. Other machine learning classification algorithms, including random forests and multi-layer perceptrons, were tested but did not outperform the simpler decision tree model. Applying the model to areas with intensive human management resulted in overprediction of large fires, as expected. This type of simple classification system could offer insight into optimal resource allocation, helping to maintain a historical fire regime and protect Alaskan ecosystems.

Download Full-text

Factors influencing large wildland fire suppression expenditures

International Journal of Wildland Fire ◽

10.1071/wf07010 ◽

2008 ◽

Vol 17 (5) ◽

pp. 650 ◽

Cited By ~ 41

Author(s):

Jingjing Liang ◽

Dave E. Calkin ◽

Krista M. Gebert ◽

Tyron J. Venn ◽

Robin P. Silverstein

Keyword(s):

Wildland Fire ◽

Fire Suppression ◽

Burned Area ◽

Private Land ◽

Fire Size ◽

Department Of Agriculture ◽

Factors Influencing ◽

The Us ◽

Large Fires ◽

Parsimonious Model

There is an urgent and immediate need to address the excessive cost of large fires. Here, we studied large wildland fire suppression expenditures by the US Department of Agriculture Forest Service. Among 16 potential non-managerial factors, which represented fire size and shape, private properties, public land attributes, forest and fuel conditions, and geographic settings, we found only fire size and private land had a strong effect on suppression expenditures. When both were accounted for, all the other variables had no significant effect. A parsimonious model to predict suppression expenditures was suggested, in which fire size and private land explained 58% of variation in expenditures. Other things being equal, suppression expenditures monotonically increased with fire size. For the average fire size, expenditures first increased with the percentage of private land within burned area, but as the percentage exceeded 20%, expenditures slowly declined until they stabilised when private land reached 50% of burned area. The results suggested that efforts to contain federal suppression expenditures need to focus on the highly complex, politically sensitive topic of wildfires on private land.

Download Full-text

Locating Forest Management Units Using Remote Sensing and Geostatistical Tools in North-Central Washington, USA

Sensors ◽

10.3390/s20092454 ◽

2020 ◽

Vol 20 (9) ◽

pp. 2454

Author(s):

Palaiologos Palaiologou ◽

Maureen Essen ◽

John Hogland ◽

Kostas Kalabokidis

Keyword(s):

Machine Learning ◽

Forest Management ◽

Random Forests ◽

Burned Area ◽

Landsat 8 ◽

Management Units ◽

Entire Study ◽

North Central ◽

Machine Learning Classification ◽

Sentinel 2

In this study, we share an approach to locate and map forest management units with high accuracy and with relatively rapid turnaround. Our study area consists of private, state, and federal land holdings that cover four counties in North-Central Washington, USA (Kittitas, Okanogan, Chelan and Douglas). This area has a rich history of landscape change caused by frequent wildfires, insect attacks, disease outbreaks, and forest management practices, which is only partially documented across ownerships in an inconsistent fashion. To consistently quantify forest management activities for the entire study area, we leveraged Sentinel-2 satellite imagery, LANDFIRE existing vegetation types and disturbances, monitoring trends in burn severity fire perimeters, and Landsat 8 Burned Area products. Within our methodology, Sentinel-2 images were collected and transformed to orthogonal land cover change difference and ratio metrics using principal component analyses. In addition, the Normalized Difference Vegetation Index and the Relativized Burn Ratio index were estimated. These variables were used as predictors in Random Forests machine learning classification models. Known locations of forest treatment units were used to create samples to train the Random Forests models to estimate where changes in forest structure occurred between the years of 2016 and 2019. We visually inspected each derived polygon to manually assign one treatment class, either clearcut or thinning. Landsat 8 Burned Area products were used to derive prescribed fire units for the same period. The bulk of analyses were performed using the RMRS Raster Utility toolbar that facilitated spatial, statistical, and machine learning tools, while significantly reducing the required processing time and storage space associated with analyzing these large datasets. The results were combined with existing LANDFIRE vegetation disturbance and forest treatment data to create a 21-year dataset (1999–2019) for the study area.

Download Full-text

Distribution of residual vegetation associated with large fires in Alberta

Canadian Journal of Forest Research ◽

10.1139/x87-186 ◽

1987 ◽

Vol 17 (10) ◽

pp. 1207-1212 ◽

Cited By ~ 106

Author(s):

Kevin E. Eberhart ◽

Paul M. Woodard

Keyword(s):

Shape Index ◽

Size Class ◽

Burned Area ◽

Fire Size ◽

Size Classes ◽

Size And Shape ◽

The Third ◽

Island Area ◽

Large Fires

Fire size and shape, number and size of islands of residual vegetation, amount of edge, and distances to residual vegetation were analyzed for 69 fires that burned in Alberta between 1970 and 1983. These fires ranged in size from 21 to 17 770 ha. Distribution of residual vegetation was compared among five fire size classes. Fires in the smallest size class (20–40 ha) did not contain any islands of unburned vegetation. Percent of area within the fire perimeter that was actually disturbed decreased with increasing fire size. The number of unburned islands per 100 ha was highest for the third and fourth largest fire size classes (201–400 and 401–2000 ha). Median island area per fire, fire shape index, and edge index increased with fire size. Percentages of burned area within 100, 200, 300, 400, and 500 m of residual vegetation decreased with increasing fire size. These results indicate decreased potential for natural reforestation and increased benefits to some wildlife habitats as fire size increases.

Download Full-text

Fire regime from 1973 to 2011 in north-western Patagonian grasslands

International Journal of Wildland Fire ◽

10.1071/wf15211 ◽

2016 ◽

Vol 25 (9) ◽

pp. 922 ◽

Cited By ~ 10

Author(s):

Facundo José Oddi ◽

Luciana Ghermandi

Keyword(s):

Fire Regime ◽

Terrestrial Ecosystems ◽

Fire Regimes ◽

Burned Area ◽

General Knowledge ◽

Fire Size ◽

Socioeconomic Impacts ◽

Mean Size ◽

Fire Size Distribution ◽

North Western

Fire is one of the most important disturbances in terrestrial ecosystems and has major ecological and socioeconomic impacts. Fire regime describes the variation of individual fire events in time and space. Few studies have characterised the fire regime in grasslands in spite of the importance of these ecosystems. The aim of this study was to describe the recent fire regime (from 1973 to 2011) of north-western Patagonian grasslands in terms of seasonality, frequency and burned area. Our study area covered 560 000 ha and we used a remote sensing approach combined with statistics obtained from operational databases. Fires occur during the summer in 2 of every 3 years with a frequency of 2.7 fires per year and a mean size of 823 ha. Fire size distribution is characterised by many small fires and few large ones which would respond to a distribution from the power law family. Eighty per cent of the total area affected by fire was burned in the span of a few years, which were also widespread fire years in forests and woodlands of north-western Patagonia. This work contributes to general knowledge about fire regimes in grasslands and we expect that our results will serve as a reference to further fire regime research.

Download Full-text

Quantile regression: an alternative approach to modelling forest area burned by individual fires

International Journal of Wildland Fire ◽

10.1071/wf17120 ◽

2018 ◽

Vol 27 (8) ◽

pp. 538 ◽

Cited By ~ 3

Author(s):

Baburam Rijal

Keyword(s):

Quantile Regression ◽

Forest Ecosystems ◽

Fire Regime ◽

Model Fitting ◽

Ordinary Least Squares ◽

Mean Value ◽

Mixed Effects ◽

Least Squares Regression ◽

Fire Size ◽

Large Fires

Components of a fire regime have long been estimated using mean-value-based ordinary least-squares regression. But, forest and fire managers require predictions beyond the mean because impacts of small and large fires on forest ecosystems and wildland–urban interfaces are different. Therefore, different action plans are required to manage potential fires of varying sizes that demand size-based modelling tools. The objective of this study was to compare two model-fitting techniques, namely quantile mixed-effects (QME) model and ordinary linear mixed-effects (LME) model for constructing distributions of model-predicted small and large fires. I examined these techniques by modelling the fire size of individual escaped wildfires. Results showed that the LME-predicted fire size approximately coincided to the 0.75 quantile. The LME model produced more biased predictions at the two extremes, both of which manifest great importance in forest ecosystems and fire management. Modelling the distributions for small and large fires using quantile regression can reduce such biases along with giving unbiased mean estimates. This study concludes that quantile modelling is an effective approach to complement ordinary regression that helps predict the size-based risks of individual fires more precisely, and that could allow managers to better plan resources when managing fires.

Download Full-text

Modelling the spatial patterns of ignition causes and fire regime features in southern France: implications for fire prevention policy

International Journal of Wildland Fire ◽

10.1071/wf15205 ◽

2016 ◽

Vol 25 (7) ◽

pp. 785 ◽

Cited By ~ 28

Author(s):

Thomas Curt ◽

Thibaut Fréjaville ◽

Sébastien Lahaye

Keyword(s):

Urban Areas ◽

Fire Regime ◽

Seasonal Distribution ◽

Environmental Drivers ◽

Burned Area ◽

Global Changes ◽

Prevention Policy ◽

Fire Policy ◽

Large Fires ◽

Spatial Hotspots

A good knowledge of the spatiotemporal patterns of the causes of wildfire ignition is crucial to an effective fire policy. However, little is known about the situation in south-eastern France because the fire database contains unreliable data. We used data for cases with well-established causes from 1973–2013 to determine the location of spatial hotspots, the seasonal distribution, the underlying anthropogenic and environmental drivers and the tendency of five main causes to generate large fires. Anthropogenic ignitions were predominant (88%) near human settlements and infrastructures in the lowlands, whilst lightning-induced fires were more common in the coastal mountains. In densely populated urban areas, small summer fires were predominating, due to the negligence of private individuals around their homes or accidental ignitions near infrastructures. In rural hinterlands, ignitions due to negligence by professionals generate many medium-sized fires from autumn to spring. Intentional and accidental ignitions contribute the most to the total burned area and to large fires. We conclude that socioeconomic factors partially control the fire regime, influencing the timing, spatial distribution and potential size of fires. This improved understanding of why, where and when ignitions occur provides the opportunity for controlling certain causes of ignitions and adapting French policy to global changes.

Download Full-text

Modelling the role of fires in the terrestrial carbon balance by incorporating SPITFIRE into the global vegetation model ORCHIDEE – Part 1: simulating historical global burned area and fire regimes

Geoscientific Model Development ◽

10.5194/gmd-7-2747-2014 ◽

2014 ◽

Vol 7 (6) ◽

pp. 2747-2767 ◽

Cited By ~ 58

Author(s):

C. Yue ◽

P. Ciais ◽

P. Cadule ◽

K. Thonicke ◽

S. Archibald ◽

...

Keyword(s):

20Th Century ◽

Burned Area ◽

Data Sets ◽

Observation Data ◽

Fire Size ◽

Vegetation Model ◽

Data Set ◽

Large Fires ◽

Global Vegetation ◽

Area Data

Abstract. Fire is an important global ecological process that influences the distribution of biomes, with consequences for carbon, water, and energy budgets. Therefore it is impossible to appropriately model the history and future of the terrestrial ecosystems and the climate system without including fire. This study incorporates the process-based prognostic fire module SPITFIRE into the global vegetation model ORCHIDEE, which was then used to simulate burned area over the 20th century. Special attention was paid to the evaluation of other fire regime indicators such as seasonality, fire size and fire length, next to burned area. For 2001–2006, the simulated global spatial extent of fire agrees well with that given by satellite-derived burned area data sets (L3JRC, GLOBCARBON, GFED3.1), and 76–92% of the global burned area is simulated as collocated between the model and observation, depending on which data set is used for comparison. The simulated global mean annual burned area is 346 Mha yr−1, which falls within the range of 287–384 Mha yr−1 as given by the three observation data sets; and is close to the 344 Mha yr−1 by the GFED3.1 data when crop fires are excluded. The simulated long-term trend and variation of burned area agree best with the observation data in regions where fire is mainly driven by climate variation, such as boreal Russia (1930–2009), along with Canada and US Alaska (1950–2009). At the global scale, the simulated decadal fire variation over the 20th century is only in moderate agreement with the historical reconstruction, possibly because of the uncertainties of past estimates, and because land-use change fires and fire suppression are not explicitly included in the model. Over the globe, the size of large fires (the 95th quantile fire size) is underestimated by the model for the regions of high fire frequency, compared with fire patch data as reconstructed from MODIS 500 m burned area data. Two case studies of fire size distribution in Canada and US Alaska, and southern Africa indicate that both number and size of large fires are underestimated, which could be related with short fire patch length and low daily fire size. Future efforts should be directed towards building consistent spatial observation data sets for key parameters of the model in order to constrain the model error at each key step of the fire modelling.

Download Full-text

Human-related ignitions concurrent with high winds promote large wildfires across the USA

International Journal of Wildland Fire ◽

10.1071/wf17149 ◽

2018 ◽

Vol 27 (6) ◽

pp. 377 ◽

Cited By ~ 21

Author(s):

John T. Abatzoglou ◽

Jennifer K. Balch ◽

Bethany A. Bradley ◽

Crystal A. Kolden

Keyword(s):

Vapour Pressure ◽

Vapour Pressure Deficit ◽

Fire Risk ◽

Burned Area ◽

Fuel Moisture ◽

Wind Speeds ◽

The Us ◽

The Usa ◽

Large Fires ◽

High Winds

Large wildfires (>40 ha) account for the majority of burned area across the contiguous United States (US) and appropriate substantial suppression resources. A variety of environmental and social factors influence wildfire growth and whether a fire overcomes initial attack efforts and becomes a large wildfire. However, little is known about how these factors differ between lightning-caused and human-caused wildfires. This study examines differences in temperature, vapour pressure deficit, fuel moisture and wind speed for large and small lightning- and human-caused wildfires during the initial days of fire activity at ecoregion scales across the US. Large fires of both human and lightning origin occurred coincident with above-normal temperature and vapour pressure deficit and below-normal 100-hour dead fuel moisture compared with small fires. Large human-caused wildfires occurred, on average, coincident with higher wind speeds than small human-caused wildfires and large lightning-caused wildfires. These results suggest the importance of winds in driving rapid fire growth that can allow fires to overcome many of the factors that typically inhibit large human-caused fires. Additionally, such findings highlight the interplay between human activity and meteorological conditions and the importance of incorporating winds in modelling large-fire risk in human-dominated landscapes.

Download Full-text

Detection of COVID-19 Infection from Routine Blood Exams with Machine Learning: a Feasibility Study

10.1101/2020.04.22.20075143 ◽

2020 ◽

Cited By ~ 2

Author(s):

Davide Brinati ◽

Andrea Campagner ◽

Davide Ferrari ◽

Massimo Locatelli ◽

Giuseppe Banfi ◽

...

Keyword(s):

Machine Learning ◽

Gold Standard ◽

False Negative ◽

White Blood Cells ◽

Standard Test ◽

Tree Model ◽

Web Based ◽

Machine Learning Classification ◽

Blood Tests ◽

Routine Blood

AbstractBackgroundThe COVID-19 pandemia due to the SARS-CoV-2 coronavirus, in its first 4 months since its outbreak, has to date reached more than 200 countries worldwide with more than 2 million confirmed cases (probably a much higher number of infected), and almost 200,000 deaths. Amplification of viral RNA by (real time) reverse transcription polymerase chain reaction (rRT-PCR) is the current gold standard test for confirmation of infection, although it presents known shortcomings: long turnaround times (3-4 hours to generate results), potential shortage of reagents, false-negative rates as large as 15-20%, the need for certified laboratories, expensive equipment and trained personnel. Thus there is a need for alternative, faster, less expensive and more accessible tests.Material and methodsWe developed two machine learning classification models using hematochemical values from routine blood exams (namely: white blood cells counts, and the platelets, CRP, AST, ALT, GGT, ALP, LDH plasma levels) drawn from 279 patients who, after being admitted to the San Raffaele Hospital (Milan, Italy) emergency-room with COVID-19 symptoms, were screened with the rRT-PCR test performed on respiratory tract specimens. Of these patients, 177 resulted positive, whereas 102 received a negative response.ResultsWe have developed two machine learning models, to discriminate between patients who are either positive or negative to the SARS-CoV-2: their accuracy ranges between 82% and 86%, and sensitivity between 92% e 95%, so comparably well with respect to the gold standard. We also developed an interpretable Decision Tree model as a simple decision aid for clinician interpreting blood tests (even off-line) for COVID-19 suspect cases.DiscussionThis study demonstrated the feasibility and clinical soundness of using blood tests analysis and machine learning as an alternative to rRT-PCR for identifying COVID-19 positive patients. This is especially useful in those countries, like developing ones, suffering from shortages of rRT-PCR reagents and specialized laboratories. We made available a Web-based tool for clinical reference and evaluation1.

Download Full-text