Analyzing Fire Behavior and Calibrating a Fire Growth Model in a Seasonally Dry Tropical Forest Area

Abstract

Dry broad-leaved seasonal forests are widespread in Southeast Asia. They are characterized by drought deciduous tree species, which are adapted to a severe dry season that lasts several months each year. Forest fires are frequent in this vegetation type. To further understanding of fire behavior and fire impact, a series of fire field experiments implemented in the Huay Kha Khaeng (HKK) Wildlife Sanctuary (Uthai Thani Province, Thailand) between 2008 and 2016 was analyzed. A fire behavior model based on the Canadian Fire Behavior Prediction System (Prometheus) was calibrated using the experimental data for the deciduous dipterocarp forest fuel type. The model was then tested on a remotely observed large wildfire in Thailand. Our results confirm the slow fire spread and low to moderate fire intensities observed for this forest type in earlier studies. The fire spread model performs well compared to satellite observations but tends to overestimate area burned and fuel consumption and, consequently, fire emissions when used in air pollution models. Our results indicate that widely used global databases may substantially overestimate fuel consumption and hence fire emissions for this forest type.

Full text

Analyzing fire behavior and calibrating a fire growth model in a seasonal dry tropical forest area
a. Gernot Ruecker, ZEBRIS Geo-IT GmbH, Germany, gruecker@zebris.com
b. Veerachai Tanpipat, Upper ASEAN Wildland Fire Special Research Unit, Forestry Research Center,
Faculty of Forestry, Kasetsart University, Thailand, fforvrc@ku.ac.th and iamtanpipat@hotmail.com
c. Kobsak Wanthongchai, Upper ASEAN Wildland Fire Special Research Unit, Forestry Research Center,
Faculty of Forestry, Kasetsart University, Thailand, fforksw@ku.ac.th
Final submitted version
Published in:
Ruecker, G., Tanpipat, V., Wanthongchai, K. (2023). Analyzing Fire Behavior and Calibrating a Fire Growth
Model in a Seasonally Dry Tropical Forest Area. In: Vadrevu, K.P., Ohara, T., Justice, C. (eds) Vegetation
Fires and Pollution in Asia. Springer, Cham. https://doi.org/10.1007/978-3-031-29916-2_14
Abstract
Dry broad leaved seasonal forests are widespread in South-East Asia. They are characterized by drought-
deciduous tree species which are adapted to a severe dry season that lasts several months each year.
Forest fires are frequent in this vegetation type. To further understanding of fire behavior and fire impact,
a series of fire field experiments implemented in the Huay Kha Khaeng (HKK) Wildlife Sanctuary (Uthai
Thani Province, Thailand) between 2008 and 2016 were analyzed and a fire behavior model based on the
Canadian Fire Behavior Prediction System (Prometheus) was calibrated using the experimental data for
the Deciduous Dipterocarp Forest fuel type. The model was then tested on a remotely observed large
wildfire in Thailand. Our results confirm the slow fire spread and low to moderate fire intensities observed
for this forest type in earlier studies. The fire spread model performs well when compared to satellite
observations but tends to overestimate area burned and fuel consumption and consequently fire
emissions when used in air pollution models. Our results furthermore indicate that widely used global
databases may substantially overestimate fuel consumption and hence fire emissions for this forest type.
1. INTRODUCTION
1.1 Fire in Dry and Deciduous Dipterocarp forests
Seasonally dry tropical and subtropical forests are one of the global biomes identified by Olsen et al.
2001 (Olson et al., 2001). These forests are characterized by a severe dry season that lasts several months
each year. Dry broad leaved seasonal forests are often characterized by drought-deciduous tree species
(Bullock, Mooney and Medina, 1995). The remaining cover of this forest type has been estimated to be
roughly 1 million km² globally (Miles et al., 2006). In South-East Asia, dry deciduous dipterocarp (DDF)
and mixed (deciduous and non-deciduous) forests (MDF) are key forest types. They are restricted to areas
with a total annual rainfall of 1,000 -1,500 mm and a pronounced dry season and are characterized
by great seasonal changes in tree phenology (Rundel and Boonpragob, 1995). DDF often have an open
canopy allowing abundant grass cover.
Although leaves of the dominant species are quite thick and large, virtually all dominant species in DDF
shed their leaves during the dry season. Fires are frequent (with return intervals from one to three years)
and feed on the leaf litter and grasses. The dominant dipterocarp tree species show morphological
adaptations to fire such as thick barks, and germination and seed dispersal are also adapted to fire (Baker

and Bunyavejchewin, 2006). Consequently, a study in the aftermath of the El Nino – Southern Oscillation
(ENSO) event of 1997/1998 that had devastating consequences of logged lowland dipterocarp forests in
Indonesia (Siegert et al., 2001), did not find substantial impacts in mixed and dry deciduous dipterocarp
forests in our study area (Baker, Bunyavejchewin and Robinson, 2009; Baker and Bunyavejchewin, 2009).
However, high fire frequencies or intensities may be critical as slow growing tree seedlings are vulnerable
to fire (Wanthongchai, Bauhus and Goldammer, 2014), and a fire mediated tree-recruitment bottleneck
has been postulated for DDF making them very similar to savannas (Nguyen, Murphy and Baker, 2019).
Fire experiments have been carried out to characterize fires in MDF and DDF in Northern and central
Thailand. Junpen et al (Junpen et al., 2013) report on fire experiments in Chiang Mai province in Northern
Thailand. Fuel loads were on average 3.9 t/ha; fires had low rates of spread between 0.51 and 2.55 m/min
and fire intensities between 40 and close to 400 kW/m. In fire experiments conducted in DDF in our study
area, the Huai Kha Khaeng Wildlife Sanctuary, (Wiriya and Kaitpraneet, 2009) found fuel loads between
0.7 and 4.6 t/ha. Head fire rate of spread was on average 1.9 m/min (0.57 to 3.94 m/min) and fire intensity
190 kW/m (36 to 372 kW/m) with flame lengths of 0.9 m (041 to 1.44 m). In summary, fires lit in typical
conditions (high relative humidity, high temperatures, low wind speed) in mixed and dry deciduous
dipterocarp forests are spreading slowly and burn at very low to low intensity (Wiriya and Kaitpraneet,
2009; Wanthongchai, Goldammer and Bauhus, 2011).
1.2 Fire Behavior and the Canadian Forest Fire Behavior Prediction System
Fire behavior modelling provides a means to predict different parameters of fire behavior (e.g. rate of
spread, intensity, direction of spread). The Canadian Fire Weather Index (CFWI) is a set of weather indices
used to characterize fire danger and to predict fire behavior for a given fuel type (Stocks et al., 1989),
which is possible in conjunction with the Canadian Fire Behavior Prediction system (CFBP) (Forestry
Canada Fire Danger Group, 1992). CFBP can be used to estimate fire behavior for fuel types which have
been calibrated against the CFWI and its sub-indices.
Although originally designed and calibrated for Canadian fuel types, the CFWI is widely used outside
Canada to predict fire danger, since the indices are found useful to predict the probability of fire
occurrence (Bianchi and Defosse, 2013; Jong et al., 2016; Steenkamp et al.; Dimitrakopoulos, Bemmerzouk
and Mitsopoulos, 2011). It is also calculated on a global scale (Field et al., 2014; Field, 2020; Vitolo et al.,
2019). CFWI is used across South-East Asia in an adapted version to assess fire weather in the region
(Groot et al., 2006; Manomaiphiboon et al., 2017)). However, to use the CFBP to predict fire behavior for
fuel types outside Canada, these fuel types need to be identified and calibrated (Dymond, Roswintiarti
and Brady, 2004; Fogarty et al.; Pearce, Anderson and Clifford, 2008).
To provide spatially explicit estimates, the Canadian Fire Behavior Prediction system has been coupled
with a fire propagator which enables the modelling of fire spread for single fire events. This fire spread
model is publicly available under the name of Prometheus (Tymstra et al., 2010). In Prometheus, fires are
propagated from points making up the vertices of a perimeter polygon (or an ignition point). Using the
Prometheus fire growth model, it is possible to predict the position of a fire front at a given time under
given weather, fuel and terrain conditions, as well as fire intensity and rate of spread. Here, we describe
an approach using remotely sensed and field data to first calibrate the CFBP model for a dry and mixed
dipterocarp forest type and then test it using the Prometheus fire growth simulator.
2. Study Area
The Huay Kha Khaeng (HKK) Wildlife Sanctuary is located at Uthai Thani Province, Thailand. It is a UNESCO
World Heritage site since December 1991. There are four types of forests in the HKK Wildlife Sanctuary,
namely dry evergreen forest, mixed deciduous forest, dry deciduous dipterocarp forest, and savanna
forest. Forest fires in this area are typically surface fires. The HKK Wildlife Sanctuary regularly witnesses a
long period of continuous burning, sometimes lasting up to two months (Baker, Bunyavejchewin and
Robinson, 2009). Fires occur from mid-December to late April with the peak fire season in March. Air

temperatures averaged from 1979-2020 (Hersbach et al., 2020) show an annual mean maximum
temperature of 32.3 °C with highest temperature (35.4 °C) in April, annual mean temperature is 24.0 °C.
Average precipitation is 1,663 mm. Mean relative air humidity varies from 79% in April to 93% in
September.
3. Data and Methods
2.3. Fire experiments
Field data were collected in a total of 71 annual fire experiments (Figure 1) implemented between 2008
and 2016 in DDF in the HKK area. Experiments were carried out during the height of the fire season from
February through April of each year and on different sites within the study area. Experiments comprised
four different slope classes (0-10, 10-20,0- 30 and 30- 40 degree) to assess the effect of slope on the fire
propagation.
Fuel consumption was estimated by sampling fuels before and after the burn. Sampling was done by
collecting all material (litter and live plants) contained in 50 by 50 cm sampling frames. Fuel was weighed
in the field and then oven dried at HKK research station to determine fine fuel moisture content.
The methodology of the experimental burns was similar to other work in the area (Junpen et al., 2013;
Wiriya and Kaitpraneet, 2009). The experimental plots were circular, having a 30 m radius. Fires were
ignited using a point ignition in the center of the circle. Fire rate of spread was assessed by measuring
fire arrival times at evenly spaced measuring rods placed in a star shaped pattern oriented in the eight
main compass directions.
(Insert fig. 1 about here)
From these data, head, flanking and tail fire rate of spread were determined. Flame height was estimated
from the height of the flames at the poles when the fire front was passing. Fire intensity was calculated
according to Byram by multiplying head fire rate of spread with the fuel consumption and the low heat
of combustion (Alexander, 1982), and low heat of combustion was assumed to be 18,330 kJ/kg (Junpen
et al., 2013).
2.1 Active fire, fuel map and fire weather data
We used Landsat 8 derived fire fronts as ignitions for the fire spread model, and Suomi National Polar-
orbiting Partnership (Suomi-NPP) Visible Infrared Imaging Radiometer Suite (VIIRS) data to assess its
spread. Landsat 8 is flying in a sun-synchronous orbit and has a local equatorial overpass time at about
10 am. The Landsat Operational Land Imager (OLI) sensor features short-wave infrared bands at 30 m
resolution which are suited to detect active fires with a flaming component (Schroeder et al., 2015; Kumar
and Roy, 2018; Sofan et al., 2020). Landsat flies about three and a half hours ahead of S-NPP VIIRS which
has a local equatorial overpass time of about 1:30 am and 1:30 pm. Schroeder et al. (Schroeder et al.,
2014; Schroeder, 2017) developed an algorithm that blends the features of the two bands VIIRS Mid-
wave infrared (MWIR) bands into a single fire detection product that also provides fire radiative power
(FRP), which is related to fuel consumption rates (Wooster et al., 2005). This relationship can be used to
estimate fuel consumption and fire smoke emissions from space; it has been used in a large number of
studies (Ichoku and Kaufman, 2005; Vermote et al., 2009; Ellicott et al., 2009; Roberts et al., 2011; Mota
and Wooster, 2018) and is the theoretical basis of the European Copernicus Atmospheric Monitoring
Service (Kaiser et al., 2012). Fire fronts were extracted from Landsat using the Shortwave Infrared bands
employing the algorithm by (Schroeder et al., 2015). We used the publicly available 30 m DEM produced
by the shuttle radar topography mission (SRTM) (Farr et al., 2007) to derive slope, aspect and height
above sea level data for our study area. We used the official Thai land cover data (Land Development

Department, 2015) as the basis of our fuel map. The deciduous forest class of the land cover map was
assigned the new fuel model “Deciduous Dipterorcarp Forest”.
Fire weather data were retrieved from the Global Fire Weather database (Field et al., 2014; Field, 2020)
(GFWD) for the dates and grid cell of the fire experiments for the years 2008-2016. The data represent
conditions at local noon. For the field data, recordings of the 2m air temperature and relative humidity
were available for the time the burnings were implemented, but no wind measurements. The exact hour
of burning was not recorded. Therefore, we used the GFWD data and assigned the (noon) CFWI data of
the corresponding grid cell to each experiment.
Meteorological data for the 2016 fire season (January to April) were retrieved from the ERA Global
reanalysis database (Hersbach et al., 2020). The data were fed into the R-Version of the Canadian Fire
Weather Index calculation algorithm (Wang et al., 2017) to calculate daily and hourly values of the CFWI
weather indices which were then used to drive the Prometheus fire simulator.
2.3 Experimental calibration for a deciduous dipterocarp forest fuel type
The relationship between rate of spread (ROS) and the initial spread index (ISI) of the CFWI exerts the
main control for fire behavior prediction in CFBP. ISI is derived from the fine fuel moisture code (FFMC),
reflecting the curing of fine fuels, and the wind speed. In CFBP, the relationship between ISI and ROS is
derived from experimental data through fitting a double exponential function of the form:
      󰇛󰇜 (1)
to the data (Forestry Canada Fire Danger Group, 1992). In equation 1, RSI is the equilibrium rate of spread,
and a, b and c are fuel type specific constants which have been determined for Canadian fuels using
regression models from experiments. The double exponential function results for most fuels in an S-
shaped curve of variable slope steepness. The steeper the slope before levelling off for very high ISI
values, the stronger is the increase in rate of spread with increasing ISI. To derive the a, b and c
parameters, a nonlinear least squares (NLS) regression is used to fit the double exponential model to the
data.
2.4 Burned area
We initialize the Prometheus fire spread model using Landsat 8 derived fire fronts as ignitions. Since the
Landsat satellite overpass happens at any time after the real ignition(s) of the fire, the area already burned
has to be taken into consideration to prevent the model to simulate fire growth in areas already burned,
especially to prevent the fire from burning backwards. Already burned areas where therefore derived
from the Landsat data. The Landsat Collection 1 data were downloaded from the Landsat Data archive
and converted to top of the atmosphere reflectance. An important requirement for the burned area map
was that the methods applied should be as robust as possible with regards to pixels contaminated by
smoke, which is especially the case if the Short-Wave Infrared bands of Landsat are used. Two spectral
indices were calculated, the widely used the Normalized burn ratio (NBR) (García and Caselles, 1991), and
the Mid-Infrared Burn Index (MIRBI) (Smith et al., 2007; Trigg and Flasse, 2010; Trigg and Flasse, 2001).
To avoid misclassifications of cloud shadows, and mask clouds, a cloud mask was applied using the FMask
algorithm (Zhu, Wang and Woodcock, 2015).
Mapping of burned areas was done by a) detecting changes in the two indices between two acquisitions
and b) using a single-scene threshold for MIRBI.
The two scenes used for change detection were co-registered and detection was based on threshold
values 0.3 for the difference MIRBI (t2 – t1) and of 0 for difference NBR (t2 – t1). Pixels above the threshold
for MIRBI difference and below the threshold for NBR difference were classified as burned. The single
scene threshold for MIRBI was 1.4. This threshold was necessary since clouds and smoke may obscure
scenes and it is then not possible to obtain two observations of the same spot for change detection.

Since this algorithm had to be more conservative to avoid false alarms, the error of omission is higher in
areas where it was applied. The change detection results were retained only when no clouds or cloud
shadows were detected in both scenes, whereas the single step results were always retained.
3.4 Model setup
To perform a test of the usability of the calibrated fuel model, we assessed a fire front system in HKK
using the Prometheus fire growth model.
Fig 2 shows a flow chart describing the modelling process. Burned areas were inserted into the fuel maps
and classified as non-fuel. Evergreen forests – found on the higher areas – was also classified as non-fuel
as it was observed that fires rarely spread into this forest type. To successfully run the model in
Prometheus and avoid memory limitations, the ignition points derived from Landsat were clustered
based on their spatial proximity, and the model was run on each of these clusters. Landsat 8-derived fire
fronts observed during two Landsat 8 overpasses on 02.03.2016 at 03:43 UTC (09:43 local time) and on
18th March 2016 at 03:43 UTC (09:43 local time) were use as ignitions for the model. Each cluster was
assigned the weather data of the nearest weather grid point. Hourly CFWI data were directly fed into
Prometheus. The model was then run for 15 hourly time steps starting from the time of the Landsat
overpass until shortly after the VIIRS nighttime overpass at approximately 1:30 local time.
(Insert fig. 2 about here)
4. RESULTS AND DISCUSSION
4.1 Field experiments
Figure 3 shows the main results of the fire experiments in boxplot charts. A total of 71 experiments were
implemented. Weather conditions were dry during all burns and FFMC above 90 in all cases. Wind speed
was slow with speeds between 6.6 and a maximum of 13.3 km/h, resulting in a low ISI between 8.6 and
14.9. ROS was 1.06 +/- 0.61 m/min across all slope classes. Faster spread on steeper slopes was only
observed in the experiments implemented in the steepest slope class (above 40 °), where mean ROS was
1.6 m/min (+/- 08.8 m/min). Mean ROS for all other slope classes was below 1 m/min. These low rates of
spread confirm observations in HKK and Northern Thailand by (Wiriya and Kaitpraneet, 2009) and
(Junpen et al., 2013) described above.
FC was very similar across all slope classes with a mean of 5.4 t/ha (+/- 1.26 t/ha), maximum of 8.9 t/ha
and a minimum of 2.66 t/ha. This is higher than the FC of 3.1 t/ha reported by (Junpen et al., 2013)
1
in a
drier area of Northern Thailand, but similar to measurements in HKK by (Wanthongchai, Goldammer and
Bauhus, 2011) who reported FC range from 4.3 to 8.1 t/ha, depending on the frequency of fire at the
investigated sites. Also in HKK, (Wiriya and Kaitpraneet, 2009) reported mean fuel loads of 4.64 t/ha (no
FC is given). These values are an order of magnitude lower than those given for dry tropical forests in the
Global Fuel Consumption Database (van Leeuwen et al., 2014), which is lacking examples from South East
Asia. In this compilation, only examples from dry tropical forest in Brazil and Mexico are cited with an FC
of 61 and 91 t/ha, respectively. Both of these examples, however, were from studies investigating
conversion from forests to pasture (Kauffman, Cummings and Ward, 1998; Kauffman et al., 2003), while
in South East Asia, frequent fires with low FC occur without conversion. To compare the experimental FC
values to another global database, the 0.25x0.25 degree resolution Global Fire Emissions Database
(GFED4s) (van der Werf et al., 2017), we downloaded the 2016 data and calculated FC from burned area
1
These authors report Fuel load (FL) and combustion completeness (CC). For comparison, we calculated average
FC as FL * CC

fraction, grid cell size and the dry matter (DM) emissions provided in the dataset. FC values in GFED over
the HKK in February and March 2016 were about 150 t/ha, i.e. about 30 times higher than the FC values
obtained from the experiments. A possible reason for this overestimation is that GFED classifies fires over
tropical forest as “deforestation fires” if they show repeated (persistent) fire detections over a longer time
period (van der Werf et al., 2010), and indeed the five 0.25° grid cells covering HKK showed a
deforestation fire fraction between 73% and 99% whereas no deforestation occurred in this protected
area. Persistent fire detections of HKK – and most probably over other mixed and dry deciduous
dipterocarp forests - may be a result of the specific fire behavior with slow moving fires that in fact
consume relatively little fuel while causing persistent fire detections. Hence, this classification method
and the subsequent modelling of deforestation fires over this forest type may lead to a systematic bias
in GFED.
Fire intensities were low to moderate with a mean of 170 kW/m (+/- 123 kW/m). Like fire rate of spread,
FI was only markedly influenced by slope when slope steepness was above 40°. In these slopes mean FI
was 260 kW/m, whereas FI ranged between an average 130 and 160kW/m in the other slope classes.
Flame heights were 0.65 m on average (+/- 0.59 m), again with higher flame heights on steep slopes
(0.77 m +/- 0.35 m). We tested a widely used, simple formula after (Alexander, 1982) to estimate flame
length from fire intensity against field measured flame heights, and found a reasonable correlation (slope
0.89, intercept -0.09, r² 0.49), see figure 4. Consequently, modelling flame length from modelled fire
intensity based on this formula may be a usable output of the fire spread model, e.g. to spatially estimate
probabilities of recruitment bottlenecks as described by (Nguyen, Murphy and Baker, 2019).
(Insert fig. 3 about here)
(Insert fig. 4 about here)
The nonlinear least squares regression model for derivation of a, b and c parameters was fitted using the
Levenberg-Marquardt algorithm as implemented in the R minpack.lm package (Moré, 1978) for the CFBP
model.
As can be seen in Figure 5, there is little correlation between ROS and ISI, and correspondingly the fit
was poor, and the resulting curve has a steep initial slope and levels off quickly. Due to the generally low
spread rates, the weak relationship between ISI and ROS did not have a very significant effect on the
model results, as the rate of spread across all ISI conditions was close to 1 m/min. We used the derived
parameters to drive the fire spread model in the next step (a = 3.2, b = -7E-05, c = 0.16). Using these
parameters, both conditions with high wind speeds and steep slopes may result in underprediction of
rate of spread due to the early flattening of the ISI-ROS relationship. Inaccurate modelling of steep slopes
may arise from the way CFBPS handles fire spread on slopes which is done through increasing the ISI
value used for ROS prediction with increasing slope steepness.
(Insert fig. 5 about here)
4.2 Model results
Modelling outputs for the fire clusters for 2nd and 18th March is presented in figure 6, showing the initial
Landsat fire detections at about 10:45 am, the final fire perimeters at about 1:30 am local time on the
next day, and the VIIRS and MODIS fire detections. To assess model performance, we assess spatial
proximity of the VIIRS night-time detections to the modelled fire perimeters at the time of the VIIRS
overpass. We only used the 375 m resolution detections associated with the last timestep since, due to

the slow spread rate, detections e. g. at noon would often fall so close to the original ignition that a
meaningful evaluation would not be possible. The position of the fire fronts derived from modelling were
generally close to the coincident VIIRS fire detections and often within the positional accuracy of the
VIIRS data. Most of the VIIRS fire detections (119 out of 125 on 2nd March 2016) over the HKK area were
within 750 m (i. e. two pixel sizes) of the modelled final perimeters. Detections outside of the modelled
perimeters seemed to be associated with new ignitions of that day (i. e. these ignitions could not be
modelled). On 2nd March, 44 % or 95 of the detected pixel’s center coordinate were within 250 m of the
modelled fire perimeter, and 71% or 155 VIIRS fire pixels centers were within 500 m of the modelled final
perimeter. Similarly, on 18th March, 47 % or 209 pixels of the VIIRS detections were within 250 m and
73% or 326 detections were within 500 m of the modelled perimeters for that day.
While active fires were detected over all clusters, not all frontal systems within the individual clusters had
detections, and a substantial number of the smaller fires detected by Landsat did not evolve as modelled.
This is not surprising since, due to the coarseness of the input fuel map, minor barriers were not
considered in the model. At the low intensities observed, even small barriers, fuel discontinuities or moist
areas can stop fire spread as the fire is not able to jump these obstacles.
One potential application of automated fire behavior modelling is to improve air quality forecasts. To
achieve such a forecast, fire emissions would need to be modelled based on fuel consumption rates using
the fire propagation model.
To test the suitability of our model for such a task, the Fire Radiative Power detected by infrared remote
sensors can be used as a proxy to fuel consumption rate. The conversion between total fuel consumption
and simulated FRE (i.e. the FRP summed over the time interval, unit MJ) can be done using a scaling
factor of 0.368 MJ/kg (Wooster et al., 2005). Fuel consumption for hourly intervals can thus be estimated
from the total fuel consumption and the fire arrival time. Modelled Fire Radiative Energy (MJ) for each
hourly interval was estimated by multiplying the modelled burned area for the interval by the modelled
fuel consumption and applying the above-mentioned scaling factor to convert to FRE. Average FRP (unit
MW) for the hourly time interval was then calculated by dividing hourly FRE by 3600.
Satellite-derived fire emissions estimates based on FRP are often systematically lower than those
obtained by fuel consumption modelling based on burned area, and this effect is stronger under canopy
cover than in open landscapes (Roberts et al., 2018). To explore the magnitude of this effect, a scaling
factor derived from comparison of satellite data to FC data in the GFCD by (Andela et al., 2016) was
applied to simulate the FRP detected by the satellite sensor. To constrain FRP estimates, the uncertainty
of the field-derived fuel consumption estimates was applied to the modeled FRP (95% confidence
interval).
Both fire detections by the MODIS and VIIRS sensors were used to compare against the model output.
Fig 7 shows the results for the same clusters as depicted in Fig 6. Not all clusters had fire detections for
all possible satellite overpasses. Observed FRP was lower than modelled FRP in most cases. Especially at
night, modelled FRP was substantially higher than observed FRP. Most likely, the most important source
for this discrepancy is the overestimation of rate of spread and fuel consumption at nighttime for
significant parts of the fire fronts, and that many of the smaller fire fronts detected by Landsat in the
morning were not active any more in the evening, whereas in the model these ignitions were sustained
and kept growing. Thus, in clusters 2, 3 and 9 of Figure 7, observed FRP was within the (wide) bounds of
modelled FRP.
(Insert fig. 6 about here)
(Insert fig. 7 about here)

5. CONCLUSIONS AND OUTLOOK
We tested the possibility of calibrating the CFBP for deciduous dipterocarp forests using globally
available weather data, remote sensing data, and data from field experiments.
Our results indicate that even during drought, fire rate of spread for fires in dry dipterocarp forests is
slow, and fire intensity is in general low to moderate. Fuel consumption is similar to that found in other
studies in the area, and substantially lower than given for this (broad) forest type in an international
reference database, which is based on deforestation fires (van Leeuwen et al., 2014).
The poor fit of the ISI-ROS NLS regression may be partly caused by the inaccurate weather data, and the
small range of observed values. This relatively small variability of weather conditions may also cause a
relatively small range of fire behavior in deciduous dipterocarp forests. To establish fire behavior further
and to better calibrate the fire growth models for practical purposes, a larger study on a wider range of
samples and fire situations will be necessary.
Despite this limitation, the comparison of modelled fire growth with remote sensing observations of
actual wildfires in the same area was reasonably accurate for those fires that continued spreading during
the modelled time span. A number of the (mostly smaller) fire fronts detected in the morning did not
continue to burn in the evening at the end of the model run. This led to an overestimate of the area
burned and of total fuel consumption especially at night and would equally lead to an overestimation of
emissions if the model was applied for smoke forecasting. We relate this to the fact that, due to their low
intensity, many fires are easily stopped at natural barriers that are not in the fuel map data, and hence
improvements in the accuracy and resolution of the fuel map could possibly improve outputs
substantially. Our results also indicate that the fuel consumption and emission estimates for the particular
fire observed in our study area in the Global Fire Emissions database (van der Werf et al., 2017) is an
order of magnitude higher than the FC found in our field data, and that this overestimate in GFED may
be systematic and is possibly caused by the repeat detection of the slow-spreading, low intensity fires
which are classified as deforestation fires in GFED.
Acknowledgments: G.R. acknowledges financial support by the ZIM program of the German Ministry of
Economy, grant number 16KN052420.
Author contributions: Conceptualization, G.R., V.T., K.W.; methodology, G.R., V.T., K.W.; formal analysis,
G.R., V.T., K.W.; data curation, G.R., V.T., K.W.; writing, G.R., V.T., funding acquisition, G.R. All authors have
read and agreed to the published version of the manuscript.
References
Alexander, M. E. (1982) 'Caculating and interpreting forest fire intensities', Canadian Journal of Botany,
60(349-357).
Andela, N., van der Werf, G. R., Kaiser, J. W., van Leeuwen, T. T., Wooster, M. J. and Lehmann, C. E. R. (2016)
'Biomass burning fuel consumption dynamics in the tropics and subtropics assessed from satellite',
Biogeosciences, 13(12), pp. 3717-3734.
Baker, P. J. and Bunyavejchewin, S. (2006) 'Bark thickness and the influence of forest fire on tree
population structures in a seasonal evergreen tropical forest', Natural History Bulletin of the Siam Society,
54(2), pp. 215-225.
Baker, P. J. and Bunyavejchewin, S. (2009) 'Fire behavior and fire effects across the forest landscape of
continental Southeast Asia', Tropical fire ecology: Springer, pp. 311-334.

Baker, P. J., Bunyavejchewin, S. and Robinson, A. P. (2009) 'The impacts of large-scale, low-intensity fires
on the forests of continental South-east Asia', International Journal of Wildland Fire, 17(6), pp. 782-792.
Bianchi, L. O. and Defosse, G. E. (2013) 'Ignition probability of fine dead surface fuels in native
Patagoniaforests of Argentina'.
Bullock, S. H., Mooney, H. A. and Medina, E. (1995) Seasonally dry tropical forests. Cambridge University
Press.
Department, L. D. (2015) 'Land Cover Map of Thailand' (Accessed.
Dimitrakopoulos, A. P., Bemmerzouk, A. M. and Mitsopoulos, I. D. (2011) 'Evaluation of the Canadian fire
weather index system in an eastern Mediterranean environment', Meteorological Applications, 18(1), pp.
83-93.
Dymond, C. C., Roswintiarti, O. and Brady, M. (2004) 'Characterizing and mapping fuels for Malaysia and
western Indonesia', International Journal of Wildland Fire, 13, pp. 323-334.
Ellicott, E., Vermote, E., Giglio, L. and Roberts, G. (2009) 'Estimating biomass consumed from fire using
MODIS FRE', Geophysical Research Letters, 36(13), pp. L13401.
Farr, T. G., Rosen, P. A., Caro, E., Crippen, R., Duren, R., Hensley, S., Kobrick, M., Paller, M., Rodriguez, E.
and Roth, L. (2007) 'The shuttle radar topography mission', Reviews of geophysics, 45(2).
Field, R. D. (2020) 'Evaluation of Global Fire Weather Database reanalysis and short-term forecast
products', Natural Hazards and Earth System Sciences, 20(4), pp. 1123-1147.
Field, R. D., Spessa, A. C., Aziz, N. A., Camia, A., Cantin, A., Carr, R., de Groot, W. J., Dowdy, A. J., Flannigan,
M. D., Manomaiphiboon, K., Pappenberger, F., Tanpipat, V. and Wang, X. (2014) 'Development of a global
fire weather database for 1980–2012', Natural Hazards and Earth System Sciences Discussions, 2(10), pp.
6555-6597.
Fogarty, L. G., Pearce, H. G., Catchpole, W. R. and Alexander, M. E. 'Adoption vs. adaptation: lessons from
applying the Canadian forest fire danger rating system in New Zealand'. 1998, 1011-1028.
García, M. J. L. and Caselles, V. (1991) 'Mapping burns and natural reforestation using Thematic Mapper
data', Geocarto International, 6(1), pp. 31-37.
Groot, W. J. d., Field, R. D., Brady, M. A., Roswintiarti, O. and Mohamad, M. (2006) 'Development of the
Indonesian and Malaysian Fire Danger Rating Systems', Mitigation and Adaptation Strategies for Global
Change, 12(1), pp. 165-180.
Group, F. C. F. D. (1992) Development and Structure of the Canadian Forest Fire Behaviour Prediction
System, Ottawa: Forestry Canada - Science and Sustainable Development Directorate.
Hersbach, H., Bell, B., Berrisford, P., Hirahara, S., Horányi, A., Muñoz‐Sabater, J., Nicolas, J., Peubey, C.,
Radu, R. and Schepers, D. (2020) 'The ERA5 global reanalysis', Quarterly Journal of the Royal
Meteorological Society, 146(730), pp. 1999-2049.
Ichoku, C. and Kaufman, Y. J. (2005) 'A method to derive smoke emission rates from MODIS fire radiative
energy measurements', IEEE Transactions on Geoscience and Remote Sensing, 43(11), pp. 2636-2649.
Jong, M. C. d., Wooster, M. J., Kitchen, K., Manley, C., Gazzard, R. and McCall, F. F. (2016) 'Calibration and
evaluation of the Canadian Forest Fire Weather Index (FWI) System for improved wildland fire danger
rating in the United Kingdom', Natural Hazards and Earth System Sciences, 16(5), pp. 1217-1237.
Junpen, A., Garivait, S., Bonnet, S. and Pongpullponsak, A. (2013) 'Fire spread prediction for deciduous
forest fires in Northern Thailand', Science Asia, 39, pp. 535-545.

Kaiser, J. W., Heil, A., Andreae, M. O., Benedetti, A., Chubarova, N., Jones, L., Morcrette, J. J., Razinger, M.,
Schultz, M. G., Suttie, M. and van der Werf, G. R. (2012) 'Biomass burning emissions estimated with a
global fire assimilation system based on observed fire radiative power', Biogeosciences, 9(1), pp. 527-554.
Kauffman, J. B., Cummings, D. L. and Ward, D. E. (1998) 'Fire in the Brazilian Amazon 2. Biomass, nutrient
pools and losses in cattle pastures', Oecologia, 113(3), pp. 415-427.
Kauffman, J. B., Steele, M. D., Cummings, D. L. and Jaramillo, V. J. (2003) 'Biomass dynamics associated
with deforestation, fire, and, conversion to cattle pasture in a Mexican tropical dry forest', Forest Ecology
and Management, 176(1-3), pp. 1-12.
Kumar, S. S. and Roy, D. P. (2018) 'Global operational land imager Landsat-8 reflectance-based active fire
detection algorithm', International Journal of Digital Earth, 11(2), pp. 154-178.
Manomaiphiboon, K., Tanpipat, V., Nhuchaiya, P., Jaroonrattanapak, N. and Buaniam, C. 'An Operational
Fire Danger Forecast System for Lower Mekong River Region: Technical Concepts and Current
Implementation'. Regional Haze and Climate Change Management, Chiang Mai: Haze Free Thailand
Project.
Miles, L., Newton, A. C., DeFries, R. S., Ravilious, C., May, I., Blyth, S., Kapos, V. and Gordon, J. E. (2006) 'A
global overview of the conservation status of tropical dry forests', Journal of biogeography, 33(3), pp. 491-
505.
Moré, J. J. (1978) 'The Levenberg-Marquardt algorithm: implementation and theory', Numerical analysis:
Springer, pp. 105-116.
Mota, B. and Wooster, M. J. (2018) 'A new top-down approach for directly estimating biomass burning
emissions and fuel consumption rates and totals from geostationary satellite fire radiative power (FRP)',
Remote sensing of environment, 206, pp. 45-62.
Nguyen, T. T., Murphy, B. P. and Baker, P. J. (2019) 'The existence of a fire‐mediated tree‐recruitment
bottleneck in an Asian savanna', Journal of Biogeography, 46(4), pp. 745-756.
Olson, D. M., Dinerstein, E., Wikramanayake, E. D., Burgess, N. D., Powell, G. V. N., Underwood, E. C.,
D'Amico, J. A., Itoua, I., Strand, H. E. and Morrison, J. C. (2001) 'Terrestrial Ecoregions of the World: A New
Map of Life on EarthA new global map of terrestrial ecoregions provides an innovative tool for conserving
biodiversity', BioScience, 51(11), pp. 933-938.
Pearce, H. G., Anderson, S. A. J. and Clifford, V. R. (2008) A manual for predicting fire behaviour in New
Zealand fuels. Scion Rural Fire Research Group.
Roberts, G., Wooster, M. J., Freeborn, P. H. and Xu, W. (2011) 'Integration of geostationary FRP and polar-
orbiter burned area datasets for an enhanced biomass burning inventory', Remote Sensing of
Environment.
Roberts, G., Wooster, M. J., Lauret, N., Gastellu-Etchegorry, J. P., Lynham, T. and McRae, D. (2018)
'Investigating the impact of overlying vegetation canopy structures on fire radiative power (FRP) retrieval
through simulation and measurement', Remote sensing of environment, 217, pp. 158-171.
Rundel, P. W. and Boonpragob, K. (1995) 'Dry forest ecosystems of Thailand', Seasonally dry tropical
forests, pp. 93-123.
Schroeder, W. 2017. Visible Infrared Imaging Radiometer Suite (VIIRS) 375 m & 750 m Active Fire
Detection Data Sets Based on NASA VIIRS Land Science Investigator Processing System (SIPS)
Reprocessed Data—Version 1. 2017.
Schroeder, W., Oliva, P., Giglio, L. and Csiszar, I. (2014) 'The New VIIRS 375 m active fire detection data
product: Algorithm description and initial assessment', Remote Sensing of Environment, 143, pp. 85-96.

Schroeder, W., Oliva, P., Giglio, L., Quayle, B., Lorenz, E. and Morelli, F. (2015) 'Active fire detection using
Landsat-8/OLI data', Remote Sensing of Environment.
Siegert, F., Ruecker, G., Hinrichs, A. and Hoffmann, A. A. (2001) 'Increased damage from fires in logged
forests during droughts caused by El Nino', Nature, 414, pp. 437-440.
Smith, A., Drake, N., Wooster, M., Hudak, A., Holden, Z. A. and Gibbons, C. (2007) 'Production of Landsat
ETM+ reference imagery of burned areas within Southern African savannahs: comparison of methods
and application to MODIS', International Journal of Remote Sensing, 28(12), pp. 2753-2775.
Sofan, P., Bruce, D., Jones, E., Khomarudin, M. R. and Roswintiarti, O. (2020) 'Applying the Tropical
Peatland Combustion Algorithm to Landsat-8 Operational Land Imager (OLI) and Sentinel-2 Multi
Spectral Instrument (MSI) Imagery', Remote Sensing, 12(23), pp. 3958.
Steenkamp, K. C., Wessels, K. J., van den Bergh, F., McFerren, G. A., Frost, P. E., Lai, C. W. and Swanepoel,
D. 'Quantitative comparison of fire danger index performance using fire activity'. 2012: IEEE, 3375-3378.
Stocks, B. J., Lynham, T. J., Lawson, B. D., Alexander, M. E., Wagner, C. E. V., McAlpine, R. S. and Dube, D.
E. (1989) 'Canadian forest fire danger rating system: an overview', The Forestry Chronicle, 65(4), pp. 258-
265.
Trigg, S. and Flasse, S. (2001) 'An evaluation of different bi-spectral spaces for discriminating burned
shrub-savannah', International Journal of Remote Sensing, 22(13), pp. 2641-2647.
Trigg, S. and Flasse, S. (2010) 'An evaluation of different bi-spectral spaces for discriminating burned
shrub-savannah', International Journal of Remote Sensing, 22(13), pp. 2641-2647.
Tymstra, C., Bryce, R. W., Wotton, B. M., Taylor, S. W. and Armitage, O. B. (2010) 'Development and
structure of Prometheus: the Canadian wildland fire growth simulation model', Natural Resources Canada,
Canadian Forest Service, Northern Forestry Centre, Information Report NOR-X-417.(Edmonton, AB).
van der Werf, G. R., Randerson, J. T., Giglio, L., Collatz, G. J., Mu, M., Kasibhatla, P. S., Morton, D. C., DeFries,
R. S., Jin, Y. and van Leeuwen, T. T. (2010) 'Global fire emissions and the contribution of deforestation,
savanna, forest, agricultural, and peat fires (1997–2009)', Atmospheric Chemistry and Physics, 10(23), pp.
11707-11735.
van der Werf, G. R., Randerson, J. T., Giglio, L., van Leeuwen, T. T., Chen, Y., Rogers, B. M., Mu, M., van
Marle, M. J. E., Morton, D. C., Collatz, G. J., Yokelson, R. J. and Kasibhatla, P. S. (2017) 'Global fire emissions
estimates during 1997–2016', Earth System Science Data, 9(2), pp. 697-720.
van Leeuwen, T. T., van der Werf, G. R., Hoffmann, A. A., Detmers, R. G., Rücker, G., French, N. H. F.,
Archibald, S., Carvalho Jr, J. A., Cook, G. D., de Groot, W. J., Hély, C., Kasischke, E. S., Kloster, S., McCarty,
J. L., Pettinari, M. L., Savadogo, P., Alvarado, E. C., Boschetti, L., Manuri, S., Meyer, C. P., Siegert, F., Trollope,
L. A. and Trollope, W. S. W. (2014) 'Biomass burning fuel consumption rates: a field measurement
database', Biogeosciences, 11(24), pp. 7305-7329.
Vermote, E., Ellicott, E., Dubovik, O., Lapyonok, T., Chin, M., Giglio, L. and Roberts, G. J. (2009) 'An approach
to estimate global biomass burning emissions of organic and black carbon from MODIS fire radiative
power', Journal of Geophysical Research, 114(D18), pp. D18205.
Vitolo, C., Di Giuseppe, F., Krzeminski, B. and San-Miguel-Ayanz, J. (2019) 'A 1980–2018 global fire danger
re-analysis dataset for the Canadian Fire Weather Indices', Scientific data, 6(1), pp. 1-10.
Wang, X., Wotton, B. M., Cantin, A. S., Parisien, M.-A., Anderson, K., Moore, B. and Flannigan, M. D. (2017)
'cffdrs: an R package for the Canadian forest fire danger rating system', Ecological Processes, 6(1), pp. 1-
11.
Wanthongchai, K., Bauhus, J. and Goldammer, J. G. (2014) 'Effects of past burning frequency on woody
plant structure and composition in dry dipterocarp forest', Thai J. For, 33(3), pp. 109-130.

Wanthongchai, K., Goldammer, J. G. and Bauhus, J. (2011) 'Effects of fire frequency on prescribed fire
behaviour and soil temperatures in dry dipterocarp forests', International Journal of Wildland Fire, 20(1),
pp. 35-45.
Wiriya, K. and Kaitpraneet, S. (2009) 'Fire behavior in dry deciduous dipterocarp forest at Huai Kha Khaeng
Wildlife Sanctuary, Uthai Thani province', Agriculture and Natural Resources, 43(5), pp. 41-47.
Wooster, M. J., Roberts, G., Perry, G. L. W. and Kaufman, Y. J. (2005) 'Retrieval of biomass combustion
rates and totals from fire radiative power observations: FRP derivation and calibration relationships
between biomass consumption and fire radiative energy release', Journal of Geophysical Research,
110(D24).
Zhu, Z., Wang, S. and Woodcock, C. E. (2015) 'Improvement and expansion of the Fmask algorithm: cloud,
cloud shadow, and snow detection for Landsats 4–7, 8, and Sentinel 2 images', Remote Sensing of
Environment, 159, pp. 269-277.

Figure 1: Fire experiments in the Huay Kha Khaeng (HKK) Wildlife Sanctuary. Left: fuel sampling. Right:
Fire experiment with measuring rod for ROS and flame height.
Figure 2: Fire behavior model data processing workflow

Figure 3: Boxplots of main fire characteristics of the experimental fires. Top left: Head fire rate of spread,
top right: fuel consumption, bottom left: fire intensity, bottom right: flame height. Arrows to the right of
boxplot indicate mean (*) and standard deviation (arrow).

Figure 4: Flame length estimated after Byram versus field observed flame height. The line indicates the
regression line.
Figure 5 : ISI - ROS function derived from the experimental fires in the HKK. The line indicates the ROS(ISI) function
obtained through NLS fitting.

Figure 6: Model output and satellite fire detections: left column: 02. March 2016, right column: 18th March
2016. Background colors: Light gray: Deciduous Dipterocarp Forest, medium gray: already burned
area/non-fuel, dark gray: Evergreen Dipterocarp Forest, White: water/no data. Square markers: MODIS
fire detections, circle markers: VIIRS fire detections, triangle markers: VIIRS fire detections at the end of
the simulation run. Axes coordinates are UTM zone 47 N coordinates.

Figure 7: Model-derived FRP vs. observed FRP for 03rd March, 2016 (left column) and 18th March, 2016
(right column). Black line with circle dots: FRP estimated following Wooster et al., 2005). Black dotted
lines 95% confidence interval for the estimate of fuel consumption. Blue solid line: FRP estimated
applying the correction factor of Andela et al. (2016), blue dotted lines: bounds of this FRP estimate
defined by 95% confidence interval of field-derived FC.