Abstract: Every year during summer, natural and human-induced forest fires threaten the environment in the largely forested areas of the Himalayan region and the local population living near these forests. Nepal, with its multitude of forests, is one of the most forest fire-prone areas in the region. This study examines the possibility of averting forest fires, minimizing their frequency and the damage they cause, through advanced mapping of forest fire prone areas using a VHSR (very-high spatial resolution) satellite image of GeoEye-1, DEM (digital elevation data) created from topographic maps and additional data layers (e.g., precipitation, settlements). The study was conducted in Kayer Khola, Chitwan district, Nepal. The classification of the satellite image has been performed using OBIA (object-based image analysis) techniques taking into account spectral, spatial and context information as well as hierarchical properties. The land cover classification result was thereafter combined with additional data in ArcGIS, where the input layers were reclassified and all classes of the input layers ranked according to their proneness to forest fires. Fire prone areas were delineated in five classes ranging from very high to very low. The study revealed that 82% of fires occur in forest areas. This case study in Kayer Khola shows that OBIA and GIS modeling techniques can be used to successfully identify forest fire-prone areas. The mapping of forest fire-prone areas will enable forest departments in countries of the Himalayan region to delineate forest fire prone areas, which can guide the forest departments set up appropriate fire-fighting infrastructure in these areas and thus help, minimize or avert forest fires.
Key words: Remote sensing, GIS (geographic information system), segmentation, forest fire.
1. Introduction
Forest fire is an uncontrolled and quickly spreading fire in inflammable vegetation that occurs in the wilderness area and destroys forests and many other types of vegetation, as well as animal species. Not only forest fire damage, but the failure of past-fire forest restoration is also one of the major threats for the conservation of forest ecosystems. A single fire can become a wildfire and widespread rapidly, destroying an entire forest and its rich biodiversity. The development of forests is a natural ecological process that sometimes takes hundreds of years. Decades of fire suppression along with changes in land use and climate have increased the risk of wildfires in many forest areas. Wildfire risk is the potential for a forest fire to spread wildly and widely, affect lives and property and disrupt the forest’s ecological functions and attributes. A forest fire, regardless of whether it is caused by natural forces or by human activity, can be a real threat and a serious ecological disaster. It is impossible to stop nature, but it is possible to map forest fire risk zones and thereby minimize the frequency of fire and avert the damage they cause [1].
Forest fire prone areas are areas where a fire frequently inflames and can easily spread to other areas (Fig. 1). Chitwan, a large biodiversity hotspot, is located in the mid-southern part of Nepal (see Fig. 1).
At present, the entire area of Chitwan is under severe ecological threat as forest fires are increasing due to human activities namely collection of fuel wood, overexploitation of natural resources, and the development of new settlements. Increasing biotic pressure has led to rising frequency in forest fires, which has fragmented and degraded many forests. Many of these fragmented forest landscapes are highly endangered and show alarming signals of accelerated biodiversity loss. Recent published news accounts state that several forest fires happen every year, damaging many recourses [2].
A detailed understanding of the spatial patterns of natural and human-induced processes is important to be able to identify forest fire prone areas. The well-known factors that lead to fires relate to land cover/land use factors—forest, shrub land, grass land agriculture; climatic factors—temperature and rainfall, physiographic factors—elevation and aspect and human factors. Climatic regime determines the region’s vegetation and hence, plays a dominant role in creating fire prone areas. An increase in temperature increases the chances of fire, whereas rainfall and humidity have the opposite effect [3]. The drier the climate in a particular area, the more fire prone the area is likely to be. Topography is an important physiographic factor which is related to wind behavior and hence, affects proneness of an area to fire. Fire travels most rapidly up slopes and least rapidly down slopes [4]. Aspect plays a vital role in the spreading of the fire. Southern and south-western slopes exposed to the direct rays of the sun are more prone to catch fire than northern and north-eastern slope aspects [5].
In terms of topography, fire spreads out most rapidly up slopes and least rapidly down slopes [4]. Areas in the vicinity of settlements are also more prone to fires because the cultural practices of inhabitants can lead to incidental or accidental fires. Satellite remote sensing has opened up opportunities for quantitative analyses of forests and other ecosystems at all geographic and spatial scales. It has also been used effectively in the study, monitoring, and detection of forest fires. Understanding the behavior of forest fires, the factors that contribute to making an environment prone to fires and the factors that influence fire behavior, are essential for forest fire prone area mapping [6].
The broad expanse of the earth’s total land area, the varying nature of its terrain and wildlife in different places, makes it difficult to collect ground data in order to take up appropriate measures. Satellite remote sensing is one of the most important technologies developed to capture the earth’s surface features and be able to monitor various processes simultaneously taking place there. Satellite images therefore represent a vast resource that can be harnessed to enhance environmental mapping and fire prone areas modeling. Reliability of results depends on the type of image classification. Developments in OBIA(object-based image analysis) facilitate enhancing the quality and accuracy of feature extraction processes.
During the study, an effort was made to prepare a forest fire prone map using GIS and remote sensing techniques, and to integrate various data sets like satellite images, topographical data and climatic data. The study also attempted to exploit the capabilities of remote sensing and GIS techniques and to suggest an appropriate methodology for forest fire prone area mapping. Object-based image segmentation and classification methods were applied to identify various land uses and land cover zones using high resolution satellite images of the Kayer Khola watershed in Chitwan. The classification result has been used to improve the quality of various spatial analysis methods and detect areas prone to frequent forest fires.
2. Methods and Data
The study area, Kayer Khola Watershed, is situated in the mid-southern part of Chitwan district, Nepal and is very close to the Chitwan National Park (Fig. 2). The total area of study area is 80 km2. The boundary coordinates of the study area are 84.56°W, 84.69°E, 27.78°N and 27.67°S. Minimum elevation of the area is 239 meters and maximum elevation 1,935 meters. The study area is extremely vulnerable to forest fires. The main factors for the emergence of forest fires in the study area are (1) the dry and subtropical climate;(2) the steep topography where a major portion of the forest is located, increasing the spread of fire and (3) settlements. Although there are very few settlements within or nearby forested areas in the study area, they may still be the cause of some forest fires.
VHSR (very high spatial resolution) data from GeoEye-1 image of 2010 was used to generate a land use and land cover map. GeoEye-1 image has Panchromatic at 0.41 m Spatial Resolution and Multispectral Spectral Range 450-520 nm (blue), 520-600 nm (green), 625-695 nm (red), 760-900 nm(near infrared) 1.65 m spatial resolution. Data from other sources were also used for comparison to analyse forest fire prone areas. Digital topographic maps with a scale of 1:50,000 were provided by SoN(survey of Nepal). SoN maps became the source for a number of basic thematic layers such as sanctuary boundary, contour, drainage, settlements and roads and trek paths used for the study. Incidence of forest fires in the area from 2002 to 2010 were acquired from the FIRMS (fire information for resource management system), which integrates remote sensing and GIS technologies to deliver global MODIS hotspot fire locations and burned area information to natural resource managers and other stakeholders around the world. Global climate layers (climate grids) with a spatial resolution of one square kilometer were obtained from WorldClim.
The proneness of any area to fire depends on a host of factors such as land cover, precipitation, temperature, topography, proximity to settlements and distance to roads. Information on land cover was derived following a few steps. The acquired GeoEye-1 image was orthorectified into UTM, Zone 45 based on generated DEM from a topographic map and RPC file of the GeoEye-1 image.
DEM was generated from digital contour line using ArcInfo workstation. The process of orthorectification geometrically corrects the image and guarantees a uniform scale within the image to enable measuring. After processing the GeoEye-1 image, eCognition developer software was used for OBIA (object-based image analysis). Fig. 2 presents the workflow of image classification.
As already shown in numerous studies, an object-based approach yields better classification results with higher degree of accuracy compared to pixel-based methods, as it uses both spectral and spatial information [7-9]. A good overview on the state of the art of OBIA provides [10]. The basic step in OBIA is to derive homogeneous image objects through segmentation. Multi-resolution segmentation, a region-based, local mutual best fitting segmentation approach implemented in the eCognition software which groups areas of similar pixel values into objects, was used in this study [11]. Consequently, homogeneous areas result in larger objects, heterogeneous areas in smaller ones.
Several segmentations were tested using different parameters until the results were satisfying. During class modeling information on spectral values, vegetation indices like the NDVI (normalized difference vegetation index), a Land Water Mask created through band ratioingslope and texture information were used. In a pre-processing stage, the NDVI image was created using customized features applying the formula: NDVI = (RED – IR)/(RED + IR ), where NDVI values range from -1.0 to 1.0; non vegetated areas have negative values, and vegetated areas have positive values. These values typically range from 0.2 to 0.8 for healthy vegetation.
The land and water mask was created using the formula IR/Green*100. Land and water mask index values can range from 0 to 255, but water values typically range between 0 and 50.
A class hierarchy with in classes was defined: forest, shrub land, grassland, agricultural areas, bare areas, and water areas. With each polygon assigned a specific class, a continuous land cover map can be derived. Image objects were classified using user-defined rules. Objects with an area smaller than the defined minimum mapping unit were merged with other objects. The classified land cover map of Kayer Khola was exported to a raster file format for further processing and for forest fire prone analysis and modeling. Fig. 3 shows a land cover map derived from GeoEye images.
In 2009, field mission was carried out to validate the land cover classification data. 34 sample points collected from the field and an additional 47 points collected from high resolution imagery were used to test the accuracy of the classification result. The overall accuracy was 86.54% and overall kappa statistics assigned a value of 0.6698. An error matrix is the most commonly used form of reporting site-specific accuracy as it effectively summarizes key information obtained from a sampling and response designs [12].
Several types of factors and parameters are required for forest fire prone area modeling. In delineating forest fire prone areas, all thematic layers and topographic layers like settlement, DEM, aspect and meteorological data (e.g., rainfall, temperature, etc.) were analyzed. These parameters were found to have direct correlation to forest fires. Figs. 4-6 compare fire occurrence reported from 2002 to 2010 (FIRMS) on the basis of elevation, aspect, and land cover. Aspect of the DEM (digital elevation model) and euclidean distance of settlements to forest fire occurrence in the study area were calculated. Based on the MODIS active fire occurrence data from 2002 to 2010, the classes of the different cause factors (distance of settlement, aspect, and land cover) were reclassified. To rank the classes of the input layers according to their importance to their importance as being vulnerable to fire, the forest fire occurrence point file and the reclassified data were overlaid to identify classes of particular layer with more frequent fire incidents. Classes with high fire occurrence were assigned a higher rank and classes with less positive relation to fire occurrence were ranked lower [13] (see Table 1 for rankings). The classes of each dataset were ranked in a scale of 1-5, (5 being the highest ranking, 4—high; 3—medium; 2—low and 1—very low). Low precipitation area was ranked high and high precipitation area ranked lower. Fig. 7 shows the steps to doing a forest fire prone area analysis.
Once all layers were reclassified and each assigned a rank, a model was developed to overlay this data according to defined weights in order to produce a fire prone map of Khayer Khola. Using the Weighted Overlay tool in ArcGIS model builder the input layers have been given weights that all add up to 100%. Table 1 shows the assigned weights and ratings for all layers.
Land cover layer was given the highest weight with 60% in the analysis, as even though an environment may be conducive to fire, a forest fire cannot occur unless inflammable material, such as vegetation, is present. Proximity to human activity is a key variable in predicting the probability of an ignition taking place, although this does not influence the behavior of a fire. This factor was assigned the fourth highest weight, as anthropogenic actions are the main cause of initiation of forest fires.
3. Results
Land cover plays an important role in the recognition of fire prone areas. Fig. 6 shows the percentage of forest fires occurring in different land cover classes using the ArcGIS modeller method. 82% of fires occur within the class forest, whereas 9% of fires inflame each within shrubland and agricultural land.
Forest fire prone areas were divided into five classes that range from very high to very low fire proneness. Water areas were classified as restricted areas for forest fires. Fire prone areas are mainly situated along a certain distance from settlements. The analyzed results show that 1,398 hectares of forest areas in the study area are highly fire prone (Fig. 8).
It is explicable that forest areas are classified as highly fire prone zone, as 82% of the fires occur in this land cover classification. Southwest and southeast facing slopes are more affected by fires; therefore, precautionary measures are essential in these areas to avoid human-induced fires. It is advantageous for the Forest Department to have a fire prone map in order to avert possible disaster in fire prone areas, especially forest fires that can be caused by human activity.
A fire risk zone map would enable the Forest Department to set up appropriate fire-fighting infrastructure in areas identified as highly fire-prone. Such a map would help in planning infrastructure like main roads, subsidiary roads, inspection paths, among others, and may lead to a reliable communication and transport system to efficiently fight small and large forest fires.
4. Conclusion
The study thus demonstrated an efficient way to determine forest fire prone areas through the combined use of remote sensing and GIS. OBIA and GIS could be used to identify fire-prone areas in other study area of Nepal. Forest fire is difficult to control, but it is possible to minimize their frequency and avert associated damage they can cause using forest fire prone area mapping. Satellite images represent a vast resource for significantly enhancing environmental mapping and fire modeling. The study attempted to generate an accurate land cover map using object based image analysis with high spatial resolution satellite images within a short time frame. GIS analysis has taken into consideration a wide range of suitability parameters in identifying fire-prone areas. The mapping of forest fire prone areas will be helpful for the Forest Department in Nepal and in other Himalayan countries as it would enable the department to set up appropriate fire-fighting infrastructure in forest fire prone areas.
Acknowledgments
The author expresses his gratitude to all those who gave him encouragement to complete this work, particularly Dr. Stefan Lang who facilitated his working place at Z_GIS; Dr. Shahnawaz, Director of UNIGIS International for South and Southeast Asia, for facilitating the author’s come to Salzburg, Austria and for stimulating suggestions and encouragement which helped him during the research and writing of this report. Special thanks to Dr. Andreas Schild, Director General of ICIMOD, and Mr. Basanta Shrestha, Division Head of MENRIS, ICIMOD, for financial contribution in the completion of this work and for allowing the author to join the program. Cordial thanks to Hammad Gilani for field photographs.
References
[1] E.C. Council Regulation No. 1257/1999 on Support for Rural Development from the European Agricultural Guidance and Guarantee Fund (E.A.G.G.F.), 1999, p. 23.
[2] E.C. Council Regulation No. 1698/2005 on Support for Rural Development by the European Agricultural Fund for Rural Development (E.A.F.R.D.), 2005, p. 40.
[3] M.G.R. Cannel, Growing trees to sequester carbon in the UK: Answers to some questions, Forestry 72 (1999) 237-247.
[4] I.J. Batemann, A.A. Lovett, Estimating and valuing the carbon sequestered in softwood and hardwood trees, timber production and forest soils in Wales, Journal of Environmental Management 60 (2000) 310-323.
[5] B.H.J. D. Jong, R. Tipper, G. Montoya-Gomez, An economic analysis of the potential for carbon sequestration by forests: Evidence from Southern Mexico, Ecological Economics 33 (2000) 313-327.
[6] D.W.M. Kenney, D. Yemshanov, G. Fox, E. Ramlal, Cost estimates for carbon sequestration from fast growing poplar plantations in Canada, Forest Policy and Economics 6 (2004) 345-358.
[7] C. Price, The Theory and Application of Forest Economics, Basil Blackwell, Corporate Headquarters, United States, 1989, p. 402.
[8] D.L. Civco, J.D. Hurd, E.H. Wilson, M. Song, Z. Zhang, A comparison of land use and land cover change detection methods, in: Annual Conference and FIG XXII Congress, Washington DC, 2002, pp. 22-26.
[9] J. Harken, R. Sugumaran, Classification of Iowa wetlands using an airborne hyperspectral image: A comparison of spectral angle mapper classifier and an object-oriented approach, Canadian Journal of Remote Sensing 31 (2005) 167-174.
[10] T. Blaschke, I. Tomljenovi?, Lidarscapes and OBIA, ASPRS 2012 Annual Conference Sacramento, California, 2012.
[11] L. Boskos, N. Grigoriades, Volume tables for poplar plantations of clone I-214 in plain areas of Drama’s region, Greece, Geotechnical Scientific 2 (1996) 16-26.
[12] N.B. Bajracharya, K. Uddin, Study of land cover dynamics in sagarmatha national park and buffer zone, in: 10th International Symposium on Hill Mountain Remote Sensing Cartography, Graz, 2010, pp. 125-132.
[13] M.H. Jo, M.B. Lee, S.Y. Lee, Y.W. Jo, S. Baek, The development of forest fire forecasting system using Internet GIS and satellite remote sensing [Onine], 2002, http://www.gisdevelopment.net.
