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 4000 m 1.00 Very 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