This study aimed to map forest fire risk zones in Quang Ninh province, Vietnam using remote sensing and GIS. Forest fire data from MODIS and field surveys were compared to validate the analysis. Factors like forest type, proximity to roads and settlements, slope, and aspect were used as inputs to a weighted overlay analysis. This generated a risk map classifying the area into very low to very high risk zones. Most fire locations fell within high or very high risk areas, validating the model. Improving input data resolution and incorporating additional social and weather factors could enhance future analyses. The study effectively mapped forest fire risk to aid decision-making for forest management in Quang Ninh province.

1. JAXA Sponsored Mini Project on Utilization of Space Technology for
Disaster Mitigation 2007/2008
Forest Fire Risk Zone Mapping by using
Remote sensing and GIS
Case study in Quang Ninh province, Vietnam
by
Mr. Tran Anh Tuan
Institute of Geography, VAST
Mr. Dinh Ngoc Dat
Space Technology Institute, VAST
under the supervision of
Dr. Vivarad Phonekeo
Geo-Informatics Application Center - Asian Institute of Technology (GIC/AIT)

2. 1. Introduction
Vietnam, the tropical country with very high density of population, there are about 9
millions ethnic minorities living closed in mountainous area. They base on the forest to
earn living with very backward tradition of cultivation. The forest is contained of very
valuable natural resources, which role is important for human living. People harvest many
different kinds of forest products from nature. The forest is watershed of many big rivers,
which contain high potential of hydro-electronic. Beside, it also contains of water
resources for agriculture, industry and using of local people.
The forest in Vietnam cover about 11,785,000ha of land, in which the nature forest is
9,865,000 ha and plantation forest is 1,930,000ha. The area of forest and plant vegetation
with high risk of burning are about 6 millions ha, belong to 46 provinces. It is included
pine, casuarinas, Pomu, eucalyptus, bamboo forest and savan. However, when the severe
weather conditions in the dry season and drought happen, all kind of forests are easy to
catch fire.
1. Objectives
The objectives of the project are the following:
• To investigate the forest fire occurrence in the study area using remote sensing,
GIS data and ground truth information.
• To compare the hotspot from ground truth survey and MODIS Fire Product for
validation
• To generate forest fire risk zone map
• To find the relationship of forest fire occurrence with the social aspect in term of
human activities for better understanding and decision making.
3. Study area
Case study is selected in Quang Ninh province with coordinates approximately left top
corner: 21.66N, 106.43E degree and right bottom corner 20.72N, 108.09E degree. Quang
Ninh province has an area of more than 5,900 km2 and a population approximate
1,000,000. As it is located in the North-East of Vietnam and close to the Eastern Sea,
Quang Ninh is affected by hot and humid tropical monsoon: the rainy season from May to
September and the dry season from October to April; Annual average precipitation: 1995
mm; Average temperature: 22.90C; Maximum temperature: 40.50C; Annual average
humidity: 82 %.
Quang Ninh has 392,000 ha of forests and forest land occupies for 66 % of the total
province's area. At present the forest acreage is 153,000ha, natural forests occupies about
82 %. The area of forest and plant vegetation with high risk of burning is belong to almost
of district in Quang Ninh. It is included pine, bamboo forest, shrub.... so when the severe
weather conditions in the dry season and drought happen, all kind of forests are easy to
catch fire. As the statistic from Vietnam Forest Protection Department each year a
hundreds of ha forest has destroyed. In 2005, there are 45 of cases forest fire destroyed
about 152 ha, in 2006, 35 case and destroyed more than 360 ha including natural forest
and plantation.
Currently, Quang Ninh becomes a centre, one of the important locations of the master
plan in developing the economy of Vietnam it has great tourism potentials with Ha Long
Bay, so the natural enviromental need to protect. Therefore, forest fire risk zone mapping
have become urgent tasks.
1

3. 4. Data collection
There are many causative factors of forest fire in Vietnam. The list below shows the data
that were collected for this study:
4.1 Digital Elevation Model
In this study, we used the SRTM of 90m resolution, which was collected from the website
of Global Landcover Faculty at the University of Maryland.
4.2 Slope and Aspect map
The slope and aspect maps were generated from the SRTM of 90m resolution.
4.3 Road map
Road map was extracted from topographic map of year 2002 with the scale of 1:100,000.
This map was obtained from the Ministry of Natural Resource and Environment
(MONRE).
4.4 Settlement points
The settlement points were extracted from the topographic map of year 2000 for the scale
of 1:100,000. This topographic map was obtained from MONRE.
4.5 Forest map
The forest map that used to this study was done for the year of 2005, with the scale of
1:1000,000 and it was collected from Forest Protection Department, Ministry of
Agricalture and Rural Development (MARD)
4.6 Forest fire inventory
This is very important data, which will be used for validation the result of the study and
evaluation the weight of each input component. Three sources of forest fire points were
collected:
• MODIS Fire Product (MOD14) of 1 km resolution was collected from MODIS
Fire Information System, Geoinformatics Center, AIT.
• 20 forest fire points for the years 2006-2007 were obtained from the field survey
records done by the Provincial Forest Protection Department of Quang Ninh.
• 10 forest fire points were collected through a survey carried out using hand held
GPS instruments during the field visit in December 2007. Accordingly, a burning
point theme was created. However, it should be noted that most of the area are
high mountainous, which is difficult to access, therefore, surveying for all of the
burning points in the study area is limited.
Based on the information obtained from the field survey, we can understand the major
causes of forest fire, which mostly are related to the human activities around the burning
points. The activities can be listed as below:
• Burning of cultivation land of ethnic minorities; straw and grass burning in rice
field near the forest boundary and sometimes the fire could be controlled and then
spread to forest.
• The illegal activities of timber, wood and other forest products exploitation also
are the cause of fire. In some forest plantation area, people deliberated to burn the
forest first and then cut and sell.
2

4. • In some cases, people deliberated to burn the forest to harm other for during the
dispute for land occupation.
In addition, we also have two scenes of ALOS/AVNIR-2 which were acquired on Jan 29,
2007 and Landsat ETM, which was acquired in year 2000. The ALOS/AVNIR-2 has
good quality, unfortunately, the images do not cover the whole study area, due to this
reason, the available ALOS data was not used in this study, only the Landsat ETM was
applied as reference for landcover and for field trip.
5. Methodology
The Forest fire risk of the study area was chosen to be assessed by a probabilistic
approach Weighted Overlay Analysis (WOA). This method requires identification of the
causal factors of forest fire and the input of the same as thematic maps. Figure 1 in the
Appendix figure shows a schematic diagram of the methodology used for this study and
the main ideas as the following:
5.1. Comparison of the ground truth with the MODIS detected hotspots
The Table 1 shows the comparision of ground truth with the MODIS detected hotspot.
From this table, there are only 10 points that matched with MODIS14 product. The reason
can be explained as below:
• The provincial department collected the burning area using simple way without
GPS
• For some burning locations could be taken by mistake due to lack of GPS.
• For some burning area which has small size, MODIS may not be able to detect.
Finally,we decided to use 10 points collected from field survey , which are realiable since
it was recorded by GPS and investigated carefully. For the other 3 points, it was obtained
from the Provincial department and it matched with MODIS Fire Product. Therefore,
these points were finalized to use for validating and allocating the input components of
model.
5.2. Allocation of the ground truth and MODIS detected hotspots to physical and
social parameters in the study area
The 13 forest fire points were overlayed to forest map, settlements, roads, slope and
aspect to find the relationship between forest fire points and the input components (Table
2). These informations will be used to determinate the risk factors in the next step.
5.3. Determination of risk factors and weight
In order to determine the risk factors and weight, we have consulted the two information
sources:
- The correlation between hotspot and input components (Table 2)
- The knowledge and experiences of the forestry experts at the forest management
in Quang Ninh who has good understanding for all of the factors that cause the
forest fire together with the background related to forestry.
Then, the factors were analysed in the following order based on the priority of
importance: Forest, Settlements, Roads, Slope and Aspect. The first classes represent
higher risk places and the last one represents low risk places, each class has different
weights as shown in the Table 4.
3

5. 1. Forest types were reclassified according to the burning possibility (easy or difficult to
catch fire). For example: Grass, shrub or pine are very dry, which are considered to be the
most flamable (see in table 3)
2. Distance from Settlements and Roads were evaluated to have the second highest
weight. The risk factor decreases farther from these places. It means that a zone near to
these places were evaluated to have a higher rate.
3. Slope and Aspect are also important to determine risk area. As shown in Table 2, many
points are located on slope over 35 degrees. Within this slope, fire can move fastly to
close area and burn everything on the moving way because soil on this slope is very thin
and moisture is not high. In addition, Vietnam has two main monsoon season: SouthWest
in summer and NorthEast in winter. Hence, SouthWest aspect and NorthEast aspect has
high possibility for fire ocurrence and easy to increase burn area.
4. Water body or rocky mountain area do not effect to the forest fire risk, therefore, there
is no weight to determine fire rating class.
5.4. Data Processing and Weighted Overlay Analysis
Thematics map were processed in ENVI and ArcGIS software. In this step, all data and
maps were converted and registered to the WGS84-UTM projection.
To integrate the statistics and to enable the analysing process basing on different targets,
it is necessary to standardize the data: Grouping all the collected statistics into groups of
the same ones so as they can overlay and analyse. In this study, the vector format is used
to analise. Therefore, it is necessary to convert the data from vector to raster format and to
classify raster based on unique standard.
Demand: We aim to keep the same pixel size of the raster after conversion (90m, same
resolution with SRTM-DEM). For the data like forest map, buffering settlement points,
buffering roads network, the conversion process from vector to raster is done by the
software ArcMap with the support of Spatial Analyst. (Figure 1 to Figure 6)
To integrate the information using the above formular with raster, we can use the tools in
the softwares GIS such as: “Map Calculation” or “Weighted Overlay”. To facilitate the
calculating process, the authors choose the tool WO in ArcGIS (Figure 7)
The cell values in all the file raster need to be in the integer and have the same value
“scale value”: 1-5 before Weighted Overlay. The tool used for classifying is: Reclassify
tools in Spatial Analyst.
These raster statistics are classified into the calculatable values: 1-5 and each data raster
file is evaluated by one of the weight. The cell values of the file outputs are calculated on
the following formula:
Susceptibility Index Map = [(Rated Forest x 0.4) + (Rated Settlement x 0.2)
+ (Rated Road x 0.2) + (Rated Slope x 0.1) + (Rated Aspect x 0.1)]
6. Results and discussion
Each factor was fixed with a scale value and after that we used Weight Overlay function
to build up it. On the final map (Figure 8), we classify into 5 level of forest fire risk from
Very low risk to Very high risk. 13 in 30 forest fire points from the ground truth and FPD
which has exactly posittion were putted on the final map and all thematics map to validate
and compare. The corresponding statistical summary of this study area is given in Table
5. It could be seen that there are 8 hotspots (61,53%), 3 points (23.07%) fall into Very
4

6. high risk, High risk and 2 points (15.40%) fall in to Medium risk zone, the error here is
due to the updating the statistic on the fire causing factors: the forest status, the new road
or the new residental area near the forest, in particularly one factor which is unable to be
controlled is the public awareness, uncontrolled forest burn and destroy for caltivation or
dispute, internal conflict between civil and forest management.
Some considerations from the statistical summary and comparison can be listed as below:
- Very high risk: Area in which the land cover type is plantation, shrub, grassland;
near the settlements and road; slope is more than 35 degrees on the southwest.
- High risk: Area in which the land cover is plantation; that is 1,000-2,000m far
from the settlements and less than 200m from roads; slope is between 25-35
degrees on the SouthWest or NorthEast.
- Midum risk: Area in which the forest type is secondary forest, mixture of wood
tree and bamboo; that is 2,000-3,000m far from the settlements and 200-400m
from roads; slope is between 10-25 degrees on the SouthWest or NorthEast
- Low and very low risk: An area in that forest type is rich forest, poor forest or
mangrove; that is > 3000m far from the settlements and >400m from roads; slope
is between 0-10 degrees on the SouthEast or NorthWest.
Limitations and difficulties during the Study:
Ground truth information obtained from the provincial department are not precise and
different format of baseline data that were recorded by the provincial department are time
consuming to re-arrange prior to input to the analysis.
ALOS/AVNIR-2 and PRISM data are not available for the whole province and it is the
reason that these data were not applied to the study. In this study, we used SRTM-DEM
90m resolution and Forest map 2005 with 1/1000,000 scale as input data, in this case, the
output result has a limitation of scale and quality.
The classification for the weight evaluation for each factor is highly affected to the
accuracy level of the output result. Thus, to evaluate the fire causing factors, it is
necessary to collect the knowledge of the experts on forestry as well as the relevant field
such as hydrometeorology, social statistics, etc.
7. Conclusions and recommendations
Through this study, the following conclusions were drawn:
• The forest fire occurrence in the study area were investigated using RS&GIS with
ground truth information (Obj.1)
• During the data analysis, the data from both sources (MOD14 vs. ground truth)
were compared (Obj. 2)
• Based on the information from the above two sections, the forest fire risk zone
mapping has been generated by probabilistic method using Weighted Overlay
Analysis as a tool (Obj. 3)
• In addition to the obtained forest fire risk map, burning points locations are
logically distributed, which can be seen in the result that most of hotspots are
located in the high and very high risk zones.
• During the field survey, the relationship of the forest fire occurrence with the
human activities in the study area were identified (Obj. 4)
5

7. Future plan:
• The high-resolution data of ALOS/PRISM will be applied in the next step for
generating better DEM and the application of ALOS/PRISM and ALOS/AVNIR-
2 using sharpening technique will improve the hotspot detection during the visual
interpretation.
• The result can be improved by having more casual factors which related to Social
activities (number of population in the village, population careers, etc.) and
Weather conditions (air temperature, humidity, rainfall, wind directions/speed,
etc.)
6

8. ACKNOWLEDGEMENT
We would like to acknowledge the financial and technical support provided by the Japan
Aerospace Exploration Agency (JAXA) and the Geoinformatics Center (GIC), Asian
Institute of Technology (AIT) for this study. We appreciate the encouragement and useful
suggestions received from Dr. Lal Samarakoon, Director, GIC, Dr. Manzul Hazarika
Senior Research Specialist, GIC, Dr. Vivarad Phonekeo, Researcher, GIC and Dr. Wataru
Takeuchi, Institute of Industrial Science (IIS), University of Tokyo. We also acknowledge
the support provided in collecting data by the Forest Protection Department of Vietnam
and Institute of Geography, Space Technology Institute – Vietnamese Academy of
Science and Technology.
7

9. REFERENCES
1. From Wikipedia, the free encyclopedia. Analytic Hierarchy Process
http://en.wikipedia.org/wiki/Analytic_Hierarchy_Process
2. Malczewski, J. (1999) GIS and mutlicriteria decision analysis. New York: John
Wiley & Sons.
3. Quang N. H. 2005. Forest fire prevention and fire fighting in Vietnam. Report of
Forest Protection Department, Ministry of Agriculture and Rural Development (by
personal communication)
4. Esra Erten, Vedat Kurgun and Nebiye Musaoglu, 2004. Forest fire risk zone
mappig from satellite imagery and GIS A case study.
http://www.cartesia.org/geodoc/isprs2004/yf/papers/927.pdf
5. Biswajeet Pradhan, Mohamad Arshad Bin Awang. Application of Remote Sensing
and GIS for forest fire susceptibility mapping using likelihood ratio model.
http://www.gisdevelopment.net/application/environment/ffm/mm031_1.htm
6. T. Fung, C.Y. Jim. Application of remote sensing/GIS in analysis of hill fire
impact on vegetation resources in HongKong. Geoscience and Remote Sensing
Symposium, 1993. IGARSS apos;93. Better Understanding of Earth
Environment., International Volume , Issue , 18-21 Aug 1993 Page(s):2073 - 2075
vol.4
7. Vuong Van Quynh – Vietnam Forestry University. Use of remote sensing data for
early detection of forest fire in U Minh and Tay Nguyen. Space Generation
actively contributes to the UN workshop in Hanoi, Vietnam. (November 2007).
8. Wildland Fire Risk and Hazard Severity Assessment Form
http://www.wildfirelessons.net/documents/ArkRiskAssessment_050119.pdf
9. MODIS Fire Information System, Asian Institute of Technology.
http://www.geoinfo.ait.ac.th:80/mod14/
10. Website: Quang Ninh Province
http://www.halong.com/halongcom/index/e_index.asp
8

10. APPENDIX FIGURES
Fig. 1: Methodology flowchart
DEM (SRTM 90m) Road map Settlement points
Buffering roads Buffering
Forest map Slope map Aspect map
settlement points
network
Determination
of risk factors & weight
Data processing and Overlay analysis
MODIS Fire
Product
Forest fire risk Index
Ground truth
data
No
Reclassify / Validation
Yes
Forest fire risk zone map
Fig. 2: Forest reclassification map
9

11. Fig. 3: Buffering settlement points
Fig. 4: Road buffering network
10

12. Fig. 5: Slope map
Fig. 6: Aspect map
11

13. Fig. 7: Weighted Overlay Analysis
Fig. 8: Forest Fire Risk Zone Mapping
12

14. APPENDIX FIGURES
Table 1: Comparison of the ground truth with the MODIS detected hotspots
1. GROUND TRUTH OBTAINED FROM FIELD SURVEY
Ground Truth MODIS Distance Burn area
Point Name Date Time lat long lat long Date Time Fire conf. (km) Ha Km2
P1 21/09/2007 10:00 21 106.93 20.99 106.92 21/09/2007 03:43 96 1.5 6 0.06
P2 13/05/2007 n/a 21 106.95 n/a n/a n/a n/a n/a 30.5 0.305
P3 xx/02/2007 n/a 20.99 106.94 20.99 106.92 8/2/2007 06:04 81 2 n/a n/a
P4 5/12/2007 21 106.9 n/a n/a n/a n/a n/a 6.7 0.067
P5 12/1/2007 21.56 107.93 n/a n/a n/a n/a n/a 28.1 0.281
P6 30/01/2007 18:00 21.35 107.33 21.35 107.45 30/01/2007 06:22 93 11.4 3 0.03
P7 29/01/2007 n/a 21.35 107.33 21.36 107.34 30/01/2007 06:10 76 1.4 30 0.3
P8 27/11/2007 n/a 21.56 107.84 21.56 107.8 27/11/2007 23:39 24 1.3 40 0.4
P9 27/11/2007 n/a 21.55 107.84 21.55 107.83 27/11/2007 23:39 69 1.5 14 0.14
P10 27/11/2007 n/a 21.541 107.845 21.54 107.84 27/11/2007 23:39 92 0.5 50 0.5
2. GROUND TRUTH OBTAINED FROM FOREST PROTECTION DEPARTMENT (FPD)
Ground Truth MODIS Distance Burn area
Point Name Date Time lat long lat long Date Time Fire conf. (km) Ha Km2
N1 08/01/2006 n/a 21.48371 107.72521 n/a n/a n/a n/a n/a
N2 09/01/2006 n/a 21.48256 107.71947 n/a n/a n/a n/a n/a
N3
N4
11/01/2006
17/05/2006
n/a
n/a
21.52389
21.52311
107.72567
107.7464
n/a
n/a
NO DATA
n/a
n/a
n/a
n/a
n/a
n/a
n/a
n/a
N5 15/10/2006 n/a 21.51403 107.72483 n/a n/a n/a n/a n/a
N6 30/10/2006 n/a 21.51071 107.70759 n/a n/a n/a n/a n/a
N7 30/12/2006 n/a n/a n/a 21.7 107.25 29/12/2006 06:10 94 n/a 4 0.04
N8 17/12/2006 21.5135 107.70829 n/a n/a n/a n/a n/a
N9 30/12/2006 n/a 21.34893 107.46587 21.37 107.23 29/12/2006 06:10 76 24.6 2.9 0.029
N10 30/12/2006 n/a 21.35549 107.32044 21.37 107.23 29/12/2006 06:10 76 9.5 2.4 0.024
N11 29/01/2007 n/a 21.35299 107.32138 n/a n/a n/a n/a n/a
N12 29/01/2007 n/a 21.35167 107.3248 n/a n/a n/a n/a n/a
N13 29/01/2007 n/a 21.35105 107.32786 n/a n/a n/a n/a n/a
N14 29/01/2007 n/a 21.35028 107.33164 n/a n/a n/a n/a n/a
N15 29/01/2007 n/a 21.34823 107.33232 n/a n/a n/a n/a n/a
N16 30/01/2007 n/a 21.34823 107.33232 21.35 107.42 28/01/2007 06:22 93 9.1 2 0.02
N17 30/01/2007 n/a n/a n/a n/a n/a n/a n/a n/a
N18 18/04/2007 21.51864 107.72264 n/a n/a n/a n/a n/a
N19 09/02/2007 21.381 107.9335 n/a n/a n/a n/a n/a
N20 22/09/2007 21.00656 106.93026 n/a n/a n/a n/a n/a
D a ta S o u rc e T o ta l p o in ts M a tc h e d p o in ts
F ie ld s u rve y (D e c 1 4 -1 6 , 2 0 0 7 ) 10 7
F o re s t P ro te c tio n D e p a rtm e n t 14 3
T o ta l 24 10
Table 2: Allocate the matched hotspots of ground truth
and MOD14 to input components
No Lat Lon Forest Setllements Roads Slope Aspect
1 21.00 106.93 Pine < 1000 < 200 > 35 SouthWest
2 21.00 106.95 Pine < 1000 < 200 > 35 SouthWest
3 20.99 106.94 Shrub < 1000 < 200 > 35 NorthEast
4 21.00 106.90 Pine 1000 – 2000 < 200 > 35 SouthWest
5 21.56 107.93 Plantation 2000 – 3000 < 200 > 35 NorthEast
6 21.35 107.33 Plantation < 1000 < 200 > 35 SouthWest
7 21.35 107.33 Plantation < 1000 < 200 > 35 SouthWest
8 21.56 107.84 Plantation 2000 – 3000 < 200 10 – 25 SouthEast or NorthWest
9 21.55 107.84 Plantation 1000 – 2000 < 200 25 – 35 NorthEast
10 21.54 107.84 Plantation < 1000 < 200 10 – 25 SouthWest
11 21.34 107.46 Shrub < 1000 < 200 10 – 25 SouthWest
12 21.35 107.32 Shrub < 1000 200 – 400 10 – 25 SouthEast or NorthWest
13 21.34 107.33 Shrub 1000 – 2000 < 200 > 35 SouthWest
13

15. Table 3: Forest reclassification
Forest type Level of risk Fire rating classes
Barren with Shrub 5
Barren with Sparse tree 5
Very high
Barren with Grassland 5
Milpa 5
Plantation forest 4
High
Plantation forest for special products 4
Bamboo 3
Secondary forest 3 Mideum
Mixture of wood tree and bamboo 3
Poor forest (IIIA1) 2
Mangrove forest 2 Low
Settlement 2
Medium forest 1
Very low
Rich forest 1
Bare Rocky mountain No risk
Rocky mountain No risk
Agriculture land No risk No risk
Other land types No risk
Water body No risk
Table 4: Evaluation of Susceptibility Coefficient
Causal Factors Weight Class Factors Fire Rating classes
level 5 5.00 Very high
level 4 4.00 High
Forest 0.40 level 3 3.00 Medium
level 2 2.00 Low
level 1 1.00 Very low
<1000 m 5.00 Very high
1000-2000 m 4.00 High
Settlements 0.20 2000-3000 m 3.00 Medium
3000-4000 m 2.00 Low
>4000 m 1.00 Very low
<200 m 5.00 Very high
200-400 m 4.00 High
Roads 0.20 400-800 m 3.00 Medium
800-1000 m 2.00 Low
>1000 m 1.00 Very Low
> 35 degree 5.00 Very high
25-35 degree 4.00 High
Slope 0.10 10-25 degree 3.00 Medium
5-10 degree 2.00 Low
< 5 degree 1.00 Very low
SouthWest 5.00 Very high
Aspect 0.10 NorthEast 3.00 Medium
NorthWest and SouthEast 1.00 Very low
14

16. Table 5: Statistical Summary
Risk level No. of hotspots Percent (%) Area (%)
Very high 8.00 61.53 8.69
High 3.00 23.07 39.99
Medium 2.00 15.40 37.41
Low 0.00 0.00 13.06
Very low 0.00 0.00 0.85
Total 13.00 100.00 100.00
15