- Email: [email protected]
Remotely sensed characterization of forest fuel types by using satellite ASTER data
International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234 www.elsevier.com/locate/jag
Remotely sensed characterization of forest fuel types by using satellite ASTER data Rosa Lasaponara *, Antonio Lanorte National Research Council, Institute of Methodologies of Environmental Analysis, C. da S. Loja, Tito Scalo, Potenza 85050, Italy Received 9 January 2006; accepted 11 August 2006
Abstract The characterization of fuel types is very important for computing spatial fire hazard and risk and simulating fire growth and intensity across a landscape. However, due to the complex nature of fuel characteristic a fuel map is considered one of the most difficult thematic layers to build up. The advent of sensors with increased spatial resolution may improve the accuracy and reduce the cost of fuels mapping. The objective of this research is to evaluate the accuracy and utility of imagery from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) satellite imagery. In order to ascertain how well ASTER data can provide an exhaustive classification of fuel properties a sample area characterized by mixed vegetation covers was analysed. The selected sample areas has an extension at around 60 km2 and is located inside the Sila plateau in the Calabria Region (South of Italy). Fieldwork fuel type recognitions, performed before, after and during the acquisition of remote sensing ASTER data, were used as ground-truth dataset to assess the results obtained for the considered test area. The method comprised the following three steps: (I) adaptation of Prometheus fuel types for obtaining a standardization system useful for remotely sensed classification of fuel types and properties in the considered Mediterranean ecosystems; (II) model construction for the spectral characterization and mapping of fuel types based on a maximum likelihood (ML) classification algorithm; (III) accuracy assessment for the performance evaluation based on the comparison of ASTER-based results with ground-truth. Results from our analysis showed that the use ASTER data provided a valuable characterization and mapping of fuel types being that the achieved classification accuracy was higher than 90%. # 2006 Elsevier B.V. All rights reserved. Keywords: Fire; Fuel type; ASTER data
1. Introduction Wildland fires are considered one of the most important disturbance factors in natural ecosystems of the Mediterranean regions, where every year, around 45,000 forest fires break out causing the destruction of about 2.6 million ha (FAO, 2001). Several studies (see, for example, Vila et al., 2001) dealing with the effects of
* Corresponding author. E-mail address: [email protected] (R. Lasaponara). 0303-2434/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.jag.2006.08.001
fires on the vegetation within the Mediterranean basin found that fires induce significant alterations in short as well as long-term vegetation dynamics. Prevention measures, together with early warning and fast extinction, are the only methods available that can support fire fighting and limit damages caused by fires especially in regions with high ecological value, dense populations, etc. In order to limit fire damage, fire agencies need to have effective decision support tools that are able to provide timely information for quantifying fire risk. In particular, fire managers need information concerning the distribution, amount and
226
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
condition of fuels in order to improve fire prevention and modelling fire spreading and intensity. Geographic information systems (GIS) and remote sensing (RS) are considered useful tools to support different phase of fire monitoring, prevention, detection and post-fire damage evaluation. In particular, satellite remote sensing can successfully cope with different aspects of fire management problems, such as danger estimation (see, for example, review in Lasaponara, in press), fire detection (see, for example reviews performed by Cuomo et al., 2001; Lasaponara et al., 2003) burned area mapping (Lasaponara, 2006, papers therein quoted) and post-fire vegetation recovery (Telesca and Lasaponara, 2006-a,b,c; Telesca and Lasaponara, 2005; Telesca et al., 2006-d). In particular, remote sensing can provide valuable data on type (namely distribution and amount of fuels) and status of vegetation in a consistent way at different spatial and temporal scales, see, for example, (Lasaponara and Lanorte, 2006a,b,c). Since the description of fuel proprieties is usually very complex, fire managers have tried to summarize the physical parameters and spatial distribution of fuel in different classes also known as ‘‘fuel models’’ (Anderson, 1982; Burgan and Rothermal, 1984). More specifically, a fuel model has been defined as ‘‘an identifiable association of fuel elements of distinctive species, form, size, arrangement, and continuity that will exhibit characteristic fire behaviour under defined burning conditions’’ (Merrill and Alexander, 1987). The spatial distribution of the fuel characteristics can be displayed as fuel type maps. Northern Forest Fire Laboratory (NFFL) system (Albini, 1976) is the most common and well-know fuel model that was developed taking the vegetation structure and characteristic of the North-American floras in to account. Recently, in the framework of the Prometheus project (1999), a new fuel type classification system was specifically developed to better represent the fuel characteristic of the Mediterranean ecosystems (http://kentauros.rtd.algo.com.gr/promet/ index.htm, Algosystems SA, Greece). This classification is principally based on the height and density of fuel, which directly influence the intensity and propagation of wildfire. Due to the complex nature of fuel characteristic a fuel map is considered one of the most difficult thematic layers to build up (Keane et al., 2000) especially for large areas. Aerial photos have been the most common remote sensing data source traditionally used (Morris, 1970; Muraro, 1970; Oswald et al., 1999) for mapping fuel types distribution. Nevertheless, remote sensing data can be an effective data source available at different
temporal and spatial scales. For example, Lidar and have been successfully used for estimating forest canopy fuel parameters (Andersen et al., 2005). Airborne hyperspectral sensors, such as multispectral infrared and visible imaging spectrometer (MIVIS) and airborne visible/infrared imaging spectrometer (AVIRIS), demonstrated high reliability for fuel typing. Several studies have been performed for assessing spatial pattern of forest fuel using AVIRIS imagery by Jia et al. (2006), Boardman (1998) and Boardman et al. (1995). High accuracy levels (higher than 90%) were obtained from MIVIS even for extremely heterogeneous areas (see, Lasaponara et al., 2006-a,b; Lasaponara et al., in press). Satellite multispectral data can be fruitfully adopted for building up fuel type maps from global, region down to local scale. For this purpose, up to now, several satellite sensors have been used in last decades. For example, NOAA-AVHRR (Advanced Very High Resolution Radiometer) data were used by Mckinley et al. (1985) for mapping fuel types in Western United States. Landsat Thematic Mapper data were used for mapping fuels models in Yosemite National Park, USA (Van Wagtendonk and Root, 2003); and in Spain (Cohen, 1989 in California; Rian˜o et al., 2002; Salas and Chuvieco, 1994). A multisensor approach based on Spot and Landsat imager was adopted by Castro and Chuvieco (1998) to perform a classification of fuel types for Chile by using an adapted version of Anderson’s system. The accuracies obtained from these researches generally ranged from 65% to 90% (Chuvieco, 1999; Lasaponara and Lanorte, 2006a,b,c). The higher accuracy values were obtained using airborne hyperspectral MIVIS data (Lasaponara, 2006). The accuracy level is strongly related with fuel presence and spatial distribution (how many and where) and with specific environmental conditions (topography, land cover heterogeneity, etc.). The importance of using multisensor data source to map fuel model was emphasised by many authors (see, for example, Keane et al., 2001). Although the recognized feasibility of satellite sensors traditionally used for the remote characterization of fuel types, the advent of new sensors with improved spatial and spectral resolutions may improve the accuracy (Chuvieco and Congalton, 1989) and reduce the cost (Zhu and Blumberg, 2001) of forest fire fuel mapping. Up to now, fire researchers did not paid enough attention to the potentiality of using remote sensing ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) data to map fuel types and properties.
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
This research aims to investigate the usefulness of ASTER data to characterize and map fuel types in fragmented ecosystems, such as those found in the Mediterranean basin. This objective is achieved by using Prometheus model coupled with ASTER data that were analysed by using maximum likelihood (ML) classifier for a test case (located in the south of Italy) that is highly representative of Mediterranean like ecosystems. 2. Study area The study area has an extension of about 6.400 ha and it is located inside the Sila plateau in the central part of the Calabria Region. Fig. 1 shows the ASTER picture for the study area. The altitude of the study area is ranging from 1000 to 1600 m above sea level (a.s.l.). The site faces the watershed between the basins of the Arco river and its affluent. From a geological point of view the study area
227
is within the Monte Garigliano unit made up of high grade metamorphic rocks, such as graniodiorites and magmatites. The soils are mainly composed of siliceous sands, products of the chemical and mechanical alteration of acid igneous rocks, with a minor part of clays made up of kaolins and vermiculites. The more diffuse and important especially is the Calabrian Laricio pine (Pinus Nigra Var. Calabrica) that is to form pure woods and seldom mixture with other especially, caducous leaves oaks (Quercus pubescens in lower areas and Quercus cerris and Quercus frainetto in higher areas up to 1000–1200), black and Neapolitan Ontano (Alnus glutinosa, Alnus cordata) to altitudes he understood between 1000 and the 1400 m of share. In the highest altitudinal strip, above the 1400 m, they are present mixed woods of oaks and beech (Fagus sylvatica). Evident degradation forms are present, in the form of xerophytic prairies and substitution bushes. The deforested areas in this strip are generally engaged by mesophytic prairies used for pasture.
Fig. 1. ASTER picture for the study area.
228
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
Table 1 Characteristic of the ASTER spectral bands Bands
Lower edge (mm)
Upper edge (mm)
1 2 3N 3B 4 5 6 7 8 9 10 11 12 13 14
0.52 0.63 0.76 0.76 1.60 2.145 2.185 2.235 2.295 2.36 8.125 8.475 8.925 10.25 10.95
0.60 0.69 0.86 0.86 1.70 2.185 2.225 2.285 2.365 2.43 8.475 8.825 9.275 10.95 11.65
3. Data analysis 3.1. Dataset ASTER is a high resolution imaging instrument that is flying on the earth observing system (EOS) Terra satellite. It has the highest spatial resolution of all five sensors on Terra and collects data in the visible/near infrared (VNIR), short wave infrared (SWIR), and thermal infrared bands (TIR). Each subsystem is pointable in the crosstrack direction. The VNIR subsystem of ASTER is quite unique. One telescope of the VNIR system is nadir looking and two are backward looking, allowing for the construction of three-dimensional digital elevation models (DEM) due to the stereo capability of the different look angles. ASTER has a revisit period of 16 days, to any one
location on the globe, with a revisit time at the equator of every four days. ASTER collects approximately 8 min of data per orbit (rather than continuously). Among the 14 ASTER bands (see Table 1) we only considered the three channels in the VNIR region and six channels in the SWIR region, while the TIR channels were excluded. The ASTER data used for this study were acquired on 4 June 2001. Additionally, photos and air photos were obtained for the investigated area immediately before and after the acquisition of satellite ASTER data. Fieldwork fuel typing were performed using a global position system (GPS) for collecting geo-position data (latitude and longitude). Air photos and fieldwork fuel types were used as a ground-truth dataset firstly to identify the fuel types defined in the context of Prometheus system, and secondly, to evaluate performance and results obtained for the considered test area from the ASTER data processing. A total of 35 plots (corresponding to almost 1000 pixels) were selected and sampled in the field using a GPS. The selected plots were located in homogeneous fuel areas, characterized by the same vegetation species with similar density, height, and coverage. Moreover, the plots were selected also considering similar topographical features, such as slope and elevation. The acquisition of the field data before and after the acquisition of the image, maids to verify possible changes in the land cover. 3.2. Prometheus adaptation The seven fuel type classes standardized in the context of Prometheus system (see Table 2) were detailed identified and carefully verified for the study area on the basis of field works performed before,
Table 2 Fuel types classification developed (Prometheus S.V. Project 1999) for Mediterranean ecosystems in framework of the Prometheus project 1 2
Ground fuels (cover >50%) Surface fuels (shrub cover >60%, tree cover <50%)
3 4
Medium-height shrubs (shrub cover >60%, tree cover <50%) Tall shrubs (shrub cover >60%, tree cover <50%)
5
Tree stands (>4 m) with a clean ground surface (shrub cover <30%)
6
Tree stands (>4 m) with medium surface fuels (shrub cover >30%)
7
Tree stands (>4 m) with heavy surface fuels (shrub cover >30%)
Grass Grassland, shrubland (smaller than 0.3–0.6 m and with a high percentage of grassland), and clearcuts, where slash was not removed Shrubs between 0.6 and 2.0 m High shrubs (between 2.0 and 4.0 m) and young trees resulting from natural regeneration or forestation The ground fuel was removed Either by prescribed burning or by mechanical means. This situation may also occur in closed canopies in which the lack of sunlight inhibits the growth of surface vegetation The base of the canopies is well above the surface fuel layer (>0.5 m). The fuel consists essentially of small shrubs, grass, litter, and duff Stands with a very dense surface fuel layer and with a very small vertical gap to the canopy base (<0.5 m)
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
during and after the acquisition of ASTER remote sensing data. In particular, photos and air photos, taken for the investigated region before, after and during the acquisition of ASTER data, along with the fuel types recognized in the field were used for this purpose. Table 3 shows the results obtained from the adaptation of Prometheus system to the characteristics and properties of fuel types present in the investigated test
229
area. Significant patches corresponding to areas representative for each fuel class were carefully identified over the ASTER images by using geoposition data (latitude and longitude) collected during the ground surveys by means of a GPS positioning system. Pixels relating to these areas were exploited for performing the selection of adequate region of interest point (ground-truth dataset) for the seven classes (fuel
Table 3 Fuel type and vegetation typologies Fuel types
Fuel type
No fuel
Roads and houses Plowed and bare soils Woody cultivations Calcareous cliffs and detritus Course and water bodies
Fuel type 1
Xerophytic Prairies Mesophytic Prairies
Fuel type and vegetation typologies Areas at present or permanently presenting soils with no vegetation cover because of plowing or erosion phenomena Arboreal cultivations (mainly orchards, olive-groves and vineyards) Emergine rocks, landslides and generally melted sediments Natural water and artificial courses besides natural water stretches and artificial basins Grassy vegetation of primary and secondary origin with a prevalence of Bromus sp. Grassy vegetation developing when water availability is good. Middle altitude prairies and peak prairies a prevalence of Cynosurus cristatus
Fuel type 2
Substitution bushes Uncultivated soils, ferns, field boundary vegetation, woods, meadows, roads
Small shrubs with predominant presence of Cistus sp. Vegetation growing in areas no longer agriculturally used or in any case at borders of lands used for a definite and consolidate purpose
Fuel type 3
Substitution bushes, broom vegetation
Mediterranean formations of secondary origin with predominant presence of Prunus sp., Spartium junceum transition states toward forestry vegetation referable to mixed oak woods Vegetation types referable to evergreen Sclerophylls mainly degradated, Mediterranean formations with prevalence of medium-height shrubs and secondary formations scarcely covering areas no longer used for pasture
Mediterranean scrubs, garigues and shrubby prairies
Fuel type 4
Mediterranean scrubs Beech woods and mixed oak woods Ilex woods
Fuel type 5
Beech woods
Reforestation conifer woods Fuel type 6
Oak woods
Chestnut woods Fuel type 7
Mixed oak woods and reforesting trees
Vegetation types referable to evergreen Sclerophylls Mediterranean formations with prevalence of tall shrubs Young trees or small height formations of beech woods and mixed oak woods Wood formations with a prevalent presence of Quercus ilex, of which there are in the study area scarce pure stands at altitude lower than 600 m Forestry formations with predominant presence of Fagus sylvatica of great relevance in the study area since these trees cover most part of the belt between 1200 and 1800 m Areas which have been reforested mainly with Pinus nigra and Pinus halepensis Forestry formations in which, according to the altitude, with prevalence of Quercus pubescens in lower areas and Quercus cerris and Quercus frainetto in higher areas up to 1000–1200 Chestnut tree formations (Castanea sativa) Forestry formations in which Quercus sp. is associated with numerous other forestry species belonging to Acer, Fraxinus, Ostrya, Alnus, Sorbus, Malus, Crataegus, Pinus, Picea, Abies
230
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
type) with the addition of one class concerning areas having no fuel and fuel type 1 class concerning areas having unclassified pixels. The sample points of ground-truth dataset were selected in the same areas subject to direct check on field in order to be used firstly to identify the fuel types defined in the context of Prometheus system, and secondly, to evaluate performance and results obtained for the considered test area from the ASTER data processing. For this reason, pixels corresponding to the given ground-truth areas were subdivided in to testing data and training data through randomization of the pixels to 50% for every class. 3.3. Model construction and comparison The mapping of fuel types was obtained by using a supervised classification based on maximum likelihood (ML) algorithm. The ML classifier is considered one of the most important and well-known image classification methods due to its robustness and simplicity. It is wide used in vegetation and land cover mapping. Moreover, it was also tested for fuel model distribution (Rian˜o and Chuvieco, 2002). The ML method quantitatively evaluates the variance and covariance of the spectral signatures when classifying an unknown pixel assuming at the same time a Gaussian distribution of points
forming a cluster of a vegetation class. Under this assumption the distribution of a class is described by the mean vector and covariance matrix which is used to compute the statistical probability of a given pixel value being a member of a particular class. The probability for each class is calculated and the class with the highest probability is assigned the pixel (Lillesand and Kiefer, 2000). As above reported, the ML classifier is based on the assumption that different variables used in the computation are normally distributed. This assumption is generally considered acceptable for common spectral response distribution. In our case, on the basis of ground surveys and air photos, we selected the region of interest (ROI) corresponding to the considered seven fuel types, plus two additional classes related to no fuel and unclassified regions (see Fig. 2). Pixels belonging to each of the considered ROI were randomly separated into training data and testing data, used for the ML and accuracy evaluation, respectively. 4. Result ML classifier was performed twice, firstly considering solely the ASTER spectral bands, secondly also adding the normalized difference vegetation index
Fig. 2. Selection of the ROI within ASTER imagery.
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
231
Fig. 3. Fuel type characterization obtained from ASTER data using maximum likelihood classification.
(NDVI). The accuracy levels obtained from the ASTER data processing performed with and without NDVI were substantially close each other. Fig. 3 shows the mapping of fuel type obtained for the investigated test area from the ASTER images. Fig. 4 plots the spectral patterns of the seven fuel types as obtained from ASTER spectral bands. Fuel type map presents very high user’s accuracy. For the accuracy assessment we consider the producer accuracy, user accuracy, and overall accuracy that are defined as follows. The producer’s accuracy is a measure indicating the probability that the classifier has correctly labelled an image pixel, for example, into fuel type 1 class given
Fig. 4. ASTER-based spectral patterns of the seven fuel types.
that, on the basis of ground recognition such a pixel belongs to fuel type 1 class. The user’s accuracy is a measure indicating the probability that a pixel belongs to a given class and the classifier has labelled the pixel correctly into the same given class. The overall accuracy is calculated by summing the number of pixels classified correctly and dividing by the total number of pixels. Finally, the kappa statistics (K) was also considered. It measures the increase in classification accuracy over that of pure chance by accounting for omission and commission error (Congalton and Green, 1998). Overall accuracy is computed as the sum of the number of observations correctly classified (class 1, as class 1, class 2 as class 2, etc.) divided by the total number of observations. This is equivalent to the ‘‘diagonal’’ of a square contingency table matrix divided by the total number of observations described in that contingency table (Congalton and Green, 1998). Table 4 shows the accuracy coefficients as well as the omission and commission errors. The achieved overall accuracy was 90.39%. Thus, showing that the use of remotely sensed ASTER data at high spatial and spectral resolution provided a valuable characterization and mapping of fuel types. In particular, fuel type 7 (62.12%) shows the worst producer accuracy; whereas fuel type 4 (72.73%) and
232
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
Table 4 Confusion matrix Class Fuel type Fuel type Fuel type Fuel type Fuel type Fuel type Fuel type No fuel Unclass
1 2 3 4 5 6 7
Table 5 Numerical description of the fuel User Acc. ASTER (%)
Prod. Acc. Aster (%)
74.64 84.52 82.14 72.73 100.00 99.29 87.23 95.09 100.00
85.12 85.54 84.15 86.15 100.00 99.29 62.12 88.57 98.91
ASTER overall Acc. = 90.73; ASTER kappa coefficient = 0.89.
fuel type 1 (74.64%) show the worst user accuracy. That is mainly due to the high mixture that is present between fuel types 4 and 7. Additionally, a low mixing is also present between fuel types 1 and 2, 3. As regards fuel types 7 and 4 the mixing between the two classes consists in a very strong pixel ‘‘transfer’’ from fuel type 7 toward fuel type 4 and much weaker from fuel type 4 toward fuel type 7. The ML classifier ‘‘moves’’ 27% of the pixels of fuel type 7 toward fuel type 4 and less than 8% of fuel type 4 toward fuel type 7. This fact is mainly linked to the complexity of fuel types distribution that makes impossible an accurate selection of ROI at the spatial resolution of satellite ASTER data. The level of classification accuracy is substantially determined by the light mixing between fuel type 1 with fuel types 2, 3 and the class corresponding to the No fuel. Between fuel types 1 and 2 there is a nearly analogous ‘‘exchange’’ of a pixels number and the same it happens between fuel types 1 and 3. The complexity of fuel types distribution makes an accurate selection of ROI impossible. As regards the mixing between fuel type 1 and no fuel class, ML classifier ‘‘moves’’ 10% of the pixels from no fuel toward fuel type 1 class. This is due to the presence of typical grass species in cultivated areas. The main finding of our analysis is that the characterization of fuel type obtained from ASTER data can be performed with very high accuracy levels even for heterogeneous areas characterized by complex topography features and a mixture of vegetation covers. The high classification accuracy is due to the ability to distinguish several classes that should be easily confused. In particular, fuel type 3 versus fuel type 2; fuel type 4 versus fuel type 6; fuel type 7 versus fuel type 6. The high accuracy levels achievable using ASTER data showed that such imagery can be capable to give an exhaustive classification of fuel properties,
FT1 FT2 FT3 FT4 FT5 FT6 FT7
tonne/ha
kg/m2
kcal/kg
5 9 11 30 11 8 32.5
0.5 0.9 1.1 3 1.1 0.8 3.25
4500 4700 5200 5200 4500 4500 4500
with accuracy levels very close to those obtained from airborne hypestecral MIVIS data (Lasaponara and Lanorte, 2006a,b). In order to directly adopt fuel map types in operational contexts, it is important to develop accurate ways to translate forest types and density into fuel source parameters, namely fuel load could be estimated in joule or other energy measure. Table 5 shows energy measures expressed as metric tonne per ha (as well as kg/m2 and kcal/kg) of the fuel load computed for the seven fuel types of the study area, performed following Velez Munoz (1990). 5. Conclusions ASTER data were analysed for a test area of southern Italy to ascertain how well remote sensing data can characterize fuel type and map fuel properties. Fieldwork fuel type recognitions, performed at the same time as remote sensing data acquisitions, were used as ground-truth dataset to assess the results obtained for the considered test area. Results from our analysis showed that the spatial and spectral resolution of ASTER imagery provided a valuable characterization of fuel types being that the classification accuracy was higher than 90%. The high accuracy obtained from ASTER data showed that such imagery are able to provide an exhaustive classification of fuel properties even for heterogeneous areas characterized by a complex topography and mixed vegetation covers. Results obtained from these investigations can be directly extended to other Mediterranean like ecosystems. The main finding of the performed investigation is that satellite ASTER dataset assures an high capability for fuel characterization, with accuracy levels very close to those obtained from airborne hyperspectral data (see Lasaponara et al., 2006-a, b; Lasaponara et al., in press), but strongly reduces the cost of data acquisition currently lower than $ 90 for each ASTER image.
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
References Albini, F.A., 1976. Estimating wildfire behaviour and effects. General Technical Report. USDA Forest Service, Intermountain Forest and Range Experiment Station INT-30. Ogden, UT, pp. 1–92. Andersen, H.E., McGaughey, R.J., Robert, J., Reutebuch, S.E., 2005. Estimating forest canopy fuel parameters using Lidar data. Remote Sens. Environ. 4, 441–449. Anderson, H.E., 1982. Aids to determining fuels models for estimating fire behaviour. USDA Forest Service, Intermountain Forest and Range Experiment Station General Technical Report INT-122. Ogden, UT, 22 pp. Boardman, J.W., 1998. Leveraging the high dimensionality of AVIRIS data for improved sub-pixel target unmixing and rejection of false positives: mixture tuned matched filtering. Summaries of the Seventh JPL Airborne Geoscience Workshop, JPL Publication 97-1. NASA Jet Propulsion Lab., Pasadena, CA, pp. 55–56. Boardman, J.W., Kruse, F.A., Green, R.O., 1995. Mapping target signatures via partial unmixing of AVIRIS data. Summaries of the Fifth JPL Airborne Geoscience Workshop, JPL Publication 95-1. NASA Jet Propulsion Lab., Pasadena, CA, pp. 23–26. Burgan, R., Rothermal, R.C., 1984. BEHAVE: fire behaviour prediction and fuel modelling system-FUEL subsystem. USDA Forest Service General Technical Report INT-167, 126 pp. Castro, R., Chuvieco, E., 1998. Modelling forest fire danger from geographic information system. Geocarto Int. 13, 15–23. Chuvieco, E., Congalton, R.G., 1989. Application of remote sensing and geographic information systems to forest fire hazard mapping. Remote Sens. Environ. 29, 147–159. Chuvieco, E., 1999. Remote Sensing of Large Wildfires in European Mediterranean Basin. Springer-Verlag, Berlin, p. 122. Cohen, W.B., 1989. Potential utility of the TM tasseled cap multispectral data transformation for crown fire hazard assessment. In: ASPRS/ACSM Annual Convention Proceedings: Agenda for the 90s, vol. 3, Baltimore, MA, pp. 118–127. Congalton, R.G., Green, K., 1998. Assessing the accuracy of remotely sensed data. CRC Press, Lewis Publishers, Boca Raton, Florida, USA. Cuomo, V., Lasaponara, R., Tramutoli, V., 2001. Evaluation of a new satellite-based method for forest fire detection. Int. J. Remote Sens. 22 (9), 1799–1826. FAO, 2001. Global forest fire assessment 1990–2000. Forest Resources Assessment Programme, Working Paper No. 55, http://www.fao.org/forestry/fo/fra/docs/Wp55_eng.pdf. Jia, G.J., Ingrid, I.C., Goetz, A.F.H., Kaufmann, M.R., Kindel, B.C., 2006. Assessing spatial patterns of forest fuel using AVIRIS data. Remote Sens. Environ. 3/4, 318–327. Keane, R.E., Mincemoyer, S.A., Schmidt, K.A., Long, D.G., Garner, J.L., 2000. Mapping vegetation and fuel for fire management on the Gila National Forest Complex. USDA Forest Service General Technical Report RMRS-GTR-46-CD, New Mexico. Keane, R.E., Burgan, R., Van Wagtendonk, J., 2001. Mapping wildland fuels for fire management across multiple scales: integrating remote sensing, GIS, and biophysical modelling. Int. J. Wildl. Fire 10 (3/4), 301–319. Lasaponara, R., in press. Inter-comparison of AVHRR-based fire danger estimation methods. International J. Remote Sens. 26 (5), 853–870. Lasaponara, R., 2006. Estimating Spectral separability of satellite derived parameters for burned areas mapping in the Calabria Region by using SPOT-vegetation data. Ecol. Model. 196, 265– 270.
233
Lasaponara, R., Lanorte, A. 2006a Multiscale fuel type characterization by using multisensor remote sensing data for the Mediterranean ecosystems of Southern Italy. IEMSS conference July, 9–13, Burlington, Vermont USA, in press. Lasaponara R., Lanorte, A. 2006b Remote characterization of fuel type using satellite data, SPIE conference, Stoccolma-12–14 September, in press. Lasaponara R, A Lanorte 2006b Multispectral fuel type characterization based on remote sensing data and Prometheous model ICFFR Conference, Coimbra, November 2006. Lasaponara, R., Lanorte, A., Pignatti, S., 2006-a. Characterization and mapping of fuel types for the Mediterranean ecosystems of Pollino National Park in the Southern Italy by using Hyperspectral MIVIS data. Earth Interactions 10, 1–11. Lasaponara, R., Lanorte, A., Pignatti, S., 2006-b. Multiscale fuel type mapping in fragmented ecosystems: preliminary results from Hyperspectral MIVIS and Multispectral Landsat TM data. Int. J. Remote Sens. 27 (3), 587–593. Lasaponara R., Lanorte, A., Pignatti, S., in press. Hyperspectral analysis of fuel types for Mediterranean ecosystems, Bull. Am. Meteorol. Soc. Lasaponara, R., Cuomo, V., Macchiato, M.F., Simoniello, T., 2003. A self-adaptive algorithm based on AVHRR multitemporal data analysis for small active fire detection. Int. J. Remote Sens. 24 (8), 1723–1749. Lillesand, T.M., Kiefer, R.W., 2000. Remote Sensing and Image Interpretation. Wiley, New York. Mckinley, R.A., Chine, E.P., Werth, L.F., 1985. Operational Fire Fuels Mapping with NOAA-AVHRR data. American Society for Photogrammetry and Remote Sensing, Falls Church, VA, pp. 295–304. Merrill, D.F., Alexander, M.E., 1987. Glossary of Forest Fire Management Terms. National Research Council of Canada, Committee for Forest Fire Management, Ottawa. Morris, W.G., 1970. Photo inventory of fine logging slash. Photogrammet. Eng. 36, 1252–1256. Muraro, S.J., 1970. Slash Fuel Inventories from 70 mm Low-level Photography. Canadian Forest Service, Ottawa, Ontario, p. 63. Oswald, B.P., Fancher, J.T., Kulhavy, D.L., Reeves, H.C., 1999. Classifying fuels with aerial photography in East Texas. Int. J. Wildl. Fire 9 (2), 301–319. Prometheus, S.V., Project, 1999. Management techniques for optimisation of suppression and minimization of wildfire effects. System Validation. European Commission, contract number ENV4-CT980716. Rian˜o, D., Chuvieco, E., 2002. Generation of fuel type maps from Landsat-TM images and auxiliary data in Mediterranean ecosystem. Department of Geography, Alcala´ de Henares University, Alcala´ de Henares, Spain. Rian˜o, D., Chuvieco, E., Salas, J., Palacios-Orueta, A., Bastarrika, A., 2002. Generation of fuel type maps from Landsat TM images and ancillary data in Mediterranean ecosystems. Canadian Journal of Forest Resources 32, 1301–1315. Salas, J., Chuvieco, E., 1994. Geographic information system for wildland fire risk mapping. Wildfire 3 (2), 7–13. Telesca, L., Lasaponara, R., 2005. Discriminating dynamical patterns in burned and unburned vegetational covers by using SPOT-VGT NDVI data. Geophys. Res. Lett. 32, L21401 2005, doi:10.1029/ 2005GL024391. Telesca, L., Lasaponara, R., 2006-a. Pre-and post-fire behavioural trends revealed in satellite NDVI time series. Geophys. Res. Lett., 33, L14401 2006, doi:10.1029/2006GL026630
234
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
Telesca, L., Lasaponara, R., 2006-b. Vegetational patterns in burned and unburned areas investigated by using the detrended fluctuation analysis. Physic. A. 368, 531–535. Telesca, L., Lasaponara, R., 2006-c. Fire-induced variability in satellite SPOT-VGT NDVI vegetational data. Int. J. Remote Sens. 27, 3087–3095. Telesca, L., Lasaponara, R., Lanorte, A., 2006-d. Discrimination of vegetational patterns in burned and unburned areas. Int. J. Nonlinear Sci. Numer. Simul. 7 (3), 279–284. Van Wagtendonk, J.W., Root, R.R., 2003. The USE of multitemporal Landsat normalized difference vegetation index (NDVI) data for
mapping fuels models in Yosemite National Par k, USA. Int. J. Remote Sens. 24, 1639–1651. Velez Munoz, R., 1990. Algunas observaciones para una selvicoltura preventive de incendios forestales. Ecologia, Fuera de Serie 1, 561–571. Vila, M., Lloret, F., Ogheri, E., Terradas, J., 2001. Positive fire–grass feedback in Mediterranean basin shrubland. For. Ecol. Manage. 147, 3–14. Zhu, G., Blumberg, D.G., 2001. Classification using ASTER data and SVM algorithms; the case study of Beer Sheva, Israel. Remote Sens. Environ. 80, 233–240.
Remotely sensed characterization of forest fuel types by using satellite ASTER data Rosa Lasaponara *, Antonio Lanorte National Research Council, Institute of Methodologies of Environmental Analysis, C. da S. Loja, Tito Scalo, Potenza 85050, Italy Received 9 January 2006; accepted 11 August 2006
Abstract The characterization of fuel types is very important for computing spatial fire hazard and risk and simulating fire growth and intensity across a landscape. However, due to the complex nature of fuel characteristic a fuel map is considered one of the most difficult thematic layers to build up. The advent of sensors with increased spatial resolution may improve the accuracy and reduce the cost of fuels mapping. The objective of this research is to evaluate the accuracy and utility of imagery from the Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) satellite imagery. In order to ascertain how well ASTER data can provide an exhaustive classification of fuel properties a sample area characterized by mixed vegetation covers was analysed. The selected sample areas has an extension at around 60 km2 and is located inside the Sila plateau in the Calabria Region (South of Italy). Fieldwork fuel type recognitions, performed before, after and during the acquisition of remote sensing ASTER data, were used as ground-truth dataset to assess the results obtained for the considered test area. The method comprised the following three steps: (I) adaptation of Prometheus fuel types for obtaining a standardization system useful for remotely sensed classification of fuel types and properties in the considered Mediterranean ecosystems; (II) model construction for the spectral characterization and mapping of fuel types based on a maximum likelihood (ML) classification algorithm; (III) accuracy assessment for the performance evaluation based on the comparison of ASTER-based results with ground-truth. Results from our analysis showed that the use ASTER data provided a valuable characterization and mapping of fuel types being that the achieved classification accuracy was higher than 90%. # 2006 Elsevier B.V. All rights reserved. Keywords: Fire; Fuel type; ASTER data
1. Introduction Wildland fires are considered one of the most important disturbance factors in natural ecosystems of the Mediterranean regions, where every year, around 45,000 forest fires break out causing the destruction of about 2.6 million ha (FAO, 2001). Several studies (see, for example, Vila et al., 2001) dealing with the effects of
* Corresponding author. E-mail address: [email protected] (R. Lasaponara). 0303-2434/$ – see front matter # 2006 Elsevier B.V. All rights reserved. doi:10.1016/j.jag.2006.08.001
fires on the vegetation within the Mediterranean basin found that fires induce significant alterations in short as well as long-term vegetation dynamics. Prevention measures, together with early warning and fast extinction, are the only methods available that can support fire fighting and limit damages caused by fires especially in regions with high ecological value, dense populations, etc. In order to limit fire damage, fire agencies need to have effective decision support tools that are able to provide timely information for quantifying fire risk. In particular, fire managers need information concerning the distribution, amount and
226
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
condition of fuels in order to improve fire prevention and modelling fire spreading and intensity. Geographic information systems (GIS) and remote sensing (RS) are considered useful tools to support different phase of fire monitoring, prevention, detection and post-fire damage evaluation. In particular, satellite remote sensing can successfully cope with different aspects of fire management problems, such as danger estimation (see, for example, review in Lasaponara, in press), fire detection (see, for example reviews performed by Cuomo et al., 2001; Lasaponara et al., 2003) burned area mapping (Lasaponara, 2006, papers therein quoted) and post-fire vegetation recovery (Telesca and Lasaponara, 2006-a,b,c; Telesca and Lasaponara, 2005; Telesca et al., 2006-d). In particular, remote sensing can provide valuable data on type (namely distribution and amount of fuels) and status of vegetation in a consistent way at different spatial and temporal scales, see, for example, (Lasaponara and Lanorte, 2006a,b,c). Since the description of fuel proprieties is usually very complex, fire managers have tried to summarize the physical parameters and spatial distribution of fuel in different classes also known as ‘‘fuel models’’ (Anderson, 1982; Burgan and Rothermal, 1984). More specifically, a fuel model has been defined as ‘‘an identifiable association of fuel elements of distinctive species, form, size, arrangement, and continuity that will exhibit characteristic fire behaviour under defined burning conditions’’ (Merrill and Alexander, 1987). The spatial distribution of the fuel characteristics can be displayed as fuel type maps. Northern Forest Fire Laboratory (NFFL) system (Albini, 1976) is the most common and well-know fuel model that was developed taking the vegetation structure and characteristic of the North-American floras in to account. Recently, in the framework of the Prometheus project (1999), a new fuel type classification system was specifically developed to better represent the fuel characteristic of the Mediterranean ecosystems (http://kentauros.rtd.algo.com.gr/promet/ index.htm, Algosystems SA, Greece). This classification is principally based on the height and density of fuel, which directly influence the intensity and propagation of wildfire. Due to the complex nature of fuel characteristic a fuel map is considered one of the most difficult thematic layers to build up (Keane et al., 2000) especially for large areas. Aerial photos have been the most common remote sensing data source traditionally used (Morris, 1970; Muraro, 1970; Oswald et al., 1999) for mapping fuel types distribution. Nevertheless, remote sensing data can be an effective data source available at different
temporal and spatial scales. For example, Lidar and have been successfully used for estimating forest canopy fuel parameters (Andersen et al., 2005). Airborne hyperspectral sensors, such as multispectral infrared and visible imaging spectrometer (MIVIS) and airborne visible/infrared imaging spectrometer (AVIRIS), demonstrated high reliability for fuel typing. Several studies have been performed for assessing spatial pattern of forest fuel using AVIRIS imagery by Jia et al. (2006), Boardman (1998) and Boardman et al. (1995). High accuracy levels (higher than 90%) were obtained from MIVIS even for extremely heterogeneous areas (see, Lasaponara et al., 2006-a,b; Lasaponara et al., in press). Satellite multispectral data can be fruitfully adopted for building up fuel type maps from global, region down to local scale. For this purpose, up to now, several satellite sensors have been used in last decades. For example, NOAA-AVHRR (Advanced Very High Resolution Radiometer) data were used by Mckinley et al. (1985) for mapping fuel types in Western United States. Landsat Thematic Mapper data were used for mapping fuels models in Yosemite National Park, USA (Van Wagtendonk and Root, 2003); and in Spain (Cohen, 1989 in California; Rian˜o et al., 2002; Salas and Chuvieco, 1994). A multisensor approach based on Spot and Landsat imager was adopted by Castro and Chuvieco (1998) to perform a classification of fuel types for Chile by using an adapted version of Anderson’s system. The accuracies obtained from these researches generally ranged from 65% to 90% (Chuvieco, 1999; Lasaponara and Lanorte, 2006a,b,c). The higher accuracy values were obtained using airborne hyperspectral MIVIS data (Lasaponara, 2006). The accuracy level is strongly related with fuel presence and spatial distribution (how many and where) and with specific environmental conditions (topography, land cover heterogeneity, etc.). The importance of using multisensor data source to map fuel model was emphasised by many authors (see, for example, Keane et al., 2001). Although the recognized feasibility of satellite sensors traditionally used for the remote characterization of fuel types, the advent of new sensors with improved spatial and spectral resolutions may improve the accuracy (Chuvieco and Congalton, 1989) and reduce the cost (Zhu and Blumberg, 2001) of forest fire fuel mapping. Up to now, fire researchers did not paid enough attention to the potentiality of using remote sensing ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer) data to map fuel types and properties.
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
This research aims to investigate the usefulness of ASTER data to characterize and map fuel types in fragmented ecosystems, such as those found in the Mediterranean basin. This objective is achieved by using Prometheus model coupled with ASTER data that were analysed by using maximum likelihood (ML) classifier for a test case (located in the south of Italy) that is highly representative of Mediterranean like ecosystems. 2. Study area The study area has an extension of about 6.400 ha and it is located inside the Sila plateau in the central part of the Calabria Region. Fig. 1 shows the ASTER picture for the study area. The altitude of the study area is ranging from 1000 to 1600 m above sea level (a.s.l.). The site faces the watershed between the basins of the Arco river and its affluent. From a geological point of view the study area
227
is within the Monte Garigliano unit made up of high grade metamorphic rocks, such as graniodiorites and magmatites. The soils are mainly composed of siliceous sands, products of the chemical and mechanical alteration of acid igneous rocks, with a minor part of clays made up of kaolins and vermiculites. The more diffuse and important especially is the Calabrian Laricio pine (Pinus Nigra Var. Calabrica) that is to form pure woods and seldom mixture with other especially, caducous leaves oaks (Quercus pubescens in lower areas and Quercus cerris and Quercus frainetto in higher areas up to 1000–1200), black and Neapolitan Ontano (Alnus glutinosa, Alnus cordata) to altitudes he understood between 1000 and the 1400 m of share. In the highest altitudinal strip, above the 1400 m, they are present mixed woods of oaks and beech (Fagus sylvatica). Evident degradation forms are present, in the form of xerophytic prairies and substitution bushes. The deforested areas in this strip are generally engaged by mesophytic prairies used for pasture.
Fig. 1. ASTER picture for the study area.
228
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
Table 1 Characteristic of the ASTER spectral bands Bands
Lower edge (mm)
Upper edge (mm)
1 2 3N 3B 4 5 6 7 8 9 10 11 12 13 14
0.52 0.63 0.76 0.76 1.60 2.145 2.185 2.235 2.295 2.36 8.125 8.475 8.925 10.25 10.95
0.60 0.69 0.86 0.86 1.70 2.185 2.225 2.285 2.365 2.43 8.475 8.825 9.275 10.95 11.65
3. Data analysis 3.1. Dataset ASTER is a high resolution imaging instrument that is flying on the earth observing system (EOS) Terra satellite. It has the highest spatial resolution of all five sensors on Terra and collects data in the visible/near infrared (VNIR), short wave infrared (SWIR), and thermal infrared bands (TIR). Each subsystem is pointable in the crosstrack direction. The VNIR subsystem of ASTER is quite unique. One telescope of the VNIR system is nadir looking and two are backward looking, allowing for the construction of three-dimensional digital elevation models (DEM) due to the stereo capability of the different look angles. ASTER has a revisit period of 16 days, to any one
location on the globe, with a revisit time at the equator of every four days. ASTER collects approximately 8 min of data per orbit (rather than continuously). Among the 14 ASTER bands (see Table 1) we only considered the three channels in the VNIR region and six channels in the SWIR region, while the TIR channels were excluded. The ASTER data used for this study were acquired on 4 June 2001. Additionally, photos and air photos were obtained for the investigated area immediately before and after the acquisition of satellite ASTER data. Fieldwork fuel typing were performed using a global position system (GPS) for collecting geo-position data (latitude and longitude). Air photos and fieldwork fuel types were used as a ground-truth dataset firstly to identify the fuel types defined in the context of Prometheus system, and secondly, to evaluate performance and results obtained for the considered test area from the ASTER data processing. A total of 35 plots (corresponding to almost 1000 pixels) were selected and sampled in the field using a GPS. The selected plots were located in homogeneous fuel areas, characterized by the same vegetation species with similar density, height, and coverage. Moreover, the plots were selected also considering similar topographical features, such as slope and elevation. The acquisition of the field data before and after the acquisition of the image, maids to verify possible changes in the land cover. 3.2. Prometheus adaptation The seven fuel type classes standardized in the context of Prometheus system (see Table 2) were detailed identified and carefully verified for the study area on the basis of field works performed before,
Table 2 Fuel types classification developed (Prometheus S.V. Project 1999) for Mediterranean ecosystems in framework of the Prometheus project 1 2
Ground fuels (cover >50%) Surface fuels (shrub cover >60%, tree cover <50%)
3 4
Medium-height shrubs (shrub cover >60%, tree cover <50%) Tall shrubs (shrub cover >60%, tree cover <50%)
5
Tree stands (>4 m) with a clean ground surface (shrub cover <30%)
6
Tree stands (>4 m) with medium surface fuels (shrub cover >30%)
7
Tree stands (>4 m) with heavy surface fuels (shrub cover >30%)
Grass Grassland, shrubland (smaller than 0.3–0.6 m and with a high percentage of grassland), and clearcuts, where slash was not removed Shrubs between 0.6 and 2.0 m High shrubs (between 2.0 and 4.0 m) and young trees resulting from natural regeneration or forestation The ground fuel was removed Either by prescribed burning or by mechanical means. This situation may also occur in closed canopies in which the lack of sunlight inhibits the growth of surface vegetation The base of the canopies is well above the surface fuel layer (>0.5 m). The fuel consists essentially of small shrubs, grass, litter, and duff Stands with a very dense surface fuel layer and with a very small vertical gap to the canopy base (<0.5 m)
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
during and after the acquisition of ASTER remote sensing data. In particular, photos and air photos, taken for the investigated region before, after and during the acquisition of ASTER data, along with the fuel types recognized in the field were used for this purpose. Table 3 shows the results obtained from the adaptation of Prometheus system to the characteristics and properties of fuel types present in the investigated test
229
area. Significant patches corresponding to areas representative for each fuel class were carefully identified over the ASTER images by using geoposition data (latitude and longitude) collected during the ground surveys by means of a GPS positioning system. Pixels relating to these areas were exploited for performing the selection of adequate region of interest point (ground-truth dataset) for the seven classes (fuel
Table 3 Fuel type and vegetation typologies Fuel types
Fuel type
No fuel
Roads and houses Plowed and bare soils Woody cultivations Calcareous cliffs and detritus Course and water bodies
Fuel type 1
Xerophytic Prairies Mesophytic Prairies
Fuel type and vegetation typologies Areas at present or permanently presenting soils with no vegetation cover because of plowing or erosion phenomena Arboreal cultivations (mainly orchards, olive-groves and vineyards) Emergine rocks, landslides and generally melted sediments Natural water and artificial courses besides natural water stretches and artificial basins Grassy vegetation of primary and secondary origin with a prevalence of Bromus sp. Grassy vegetation developing when water availability is good. Middle altitude prairies and peak prairies a prevalence of Cynosurus cristatus
Fuel type 2
Substitution bushes Uncultivated soils, ferns, field boundary vegetation, woods, meadows, roads
Small shrubs with predominant presence of Cistus sp. Vegetation growing in areas no longer agriculturally used or in any case at borders of lands used for a definite and consolidate purpose
Fuel type 3
Substitution bushes, broom vegetation
Mediterranean formations of secondary origin with predominant presence of Prunus sp., Spartium junceum transition states toward forestry vegetation referable to mixed oak woods Vegetation types referable to evergreen Sclerophylls mainly degradated, Mediterranean formations with prevalence of medium-height shrubs and secondary formations scarcely covering areas no longer used for pasture
Mediterranean scrubs, garigues and shrubby prairies
Fuel type 4
Mediterranean scrubs Beech woods and mixed oak woods Ilex woods
Fuel type 5
Beech woods
Reforestation conifer woods Fuel type 6
Oak woods
Chestnut woods Fuel type 7
Mixed oak woods and reforesting trees
Vegetation types referable to evergreen Sclerophylls Mediterranean formations with prevalence of tall shrubs Young trees or small height formations of beech woods and mixed oak woods Wood formations with a prevalent presence of Quercus ilex, of which there are in the study area scarce pure stands at altitude lower than 600 m Forestry formations with predominant presence of Fagus sylvatica of great relevance in the study area since these trees cover most part of the belt between 1200 and 1800 m Areas which have been reforested mainly with Pinus nigra and Pinus halepensis Forestry formations in which, according to the altitude, with prevalence of Quercus pubescens in lower areas and Quercus cerris and Quercus frainetto in higher areas up to 1000–1200 Chestnut tree formations (Castanea sativa) Forestry formations in which Quercus sp. is associated with numerous other forestry species belonging to Acer, Fraxinus, Ostrya, Alnus, Sorbus, Malus, Crataegus, Pinus, Picea, Abies
230
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
type) with the addition of one class concerning areas having no fuel and fuel type 1 class concerning areas having unclassified pixels. The sample points of ground-truth dataset were selected in the same areas subject to direct check on field in order to be used firstly to identify the fuel types defined in the context of Prometheus system, and secondly, to evaluate performance and results obtained for the considered test area from the ASTER data processing. For this reason, pixels corresponding to the given ground-truth areas were subdivided in to testing data and training data through randomization of the pixels to 50% for every class. 3.3. Model construction and comparison The mapping of fuel types was obtained by using a supervised classification based on maximum likelihood (ML) algorithm. The ML classifier is considered one of the most important and well-known image classification methods due to its robustness and simplicity. It is wide used in vegetation and land cover mapping. Moreover, it was also tested for fuel model distribution (Rian˜o and Chuvieco, 2002). The ML method quantitatively evaluates the variance and covariance of the spectral signatures when classifying an unknown pixel assuming at the same time a Gaussian distribution of points
forming a cluster of a vegetation class. Under this assumption the distribution of a class is described by the mean vector and covariance matrix which is used to compute the statistical probability of a given pixel value being a member of a particular class. The probability for each class is calculated and the class with the highest probability is assigned the pixel (Lillesand and Kiefer, 2000). As above reported, the ML classifier is based on the assumption that different variables used in the computation are normally distributed. This assumption is generally considered acceptable for common spectral response distribution. In our case, on the basis of ground surveys and air photos, we selected the region of interest (ROI) corresponding to the considered seven fuel types, plus two additional classes related to no fuel and unclassified regions (see Fig. 2). Pixels belonging to each of the considered ROI were randomly separated into training data and testing data, used for the ML and accuracy evaluation, respectively. 4. Result ML classifier was performed twice, firstly considering solely the ASTER spectral bands, secondly also adding the normalized difference vegetation index
Fig. 2. Selection of the ROI within ASTER imagery.
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
231
Fig. 3. Fuel type characterization obtained from ASTER data using maximum likelihood classification.
(NDVI). The accuracy levels obtained from the ASTER data processing performed with and without NDVI were substantially close each other. Fig. 3 shows the mapping of fuel type obtained for the investigated test area from the ASTER images. Fig. 4 plots the spectral patterns of the seven fuel types as obtained from ASTER spectral bands. Fuel type map presents very high user’s accuracy. For the accuracy assessment we consider the producer accuracy, user accuracy, and overall accuracy that are defined as follows. The producer’s accuracy is a measure indicating the probability that the classifier has correctly labelled an image pixel, for example, into fuel type 1 class given
Fig. 4. ASTER-based spectral patterns of the seven fuel types.
that, on the basis of ground recognition such a pixel belongs to fuel type 1 class. The user’s accuracy is a measure indicating the probability that a pixel belongs to a given class and the classifier has labelled the pixel correctly into the same given class. The overall accuracy is calculated by summing the number of pixels classified correctly and dividing by the total number of pixels. Finally, the kappa statistics (K) was also considered. It measures the increase in classification accuracy over that of pure chance by accounting for omission and commission error (Congalton and Green, 1998). Overall accuracy is computed as the sum of the number of observations correctly classified (class 1, as class 1, class 2 as class 2, etc.) divided by the total number of observations. This is equivalent to the ‘‘diagonal’’ of a square contingency table matrix divided by the total number of observations described in that contingency table (Congalton and Green, 1998). Table 4 shows the accuracy coefficients as well as the omission and commission errors. The achieved overall accuracy was 90.39%. Thus, showing that the use of remotely sensed ASTER data at high spatial and spectral resolution provided a valuable characterization and mapping of fuel types. In particular, fuel type 7 (62.12%) shows the worst producer accuracy; whereas fuel type 4 (72.73%) and
232
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
Table 4 Confusion matrix Class Fuel type Fuel type Fuel type Fuel type Fuel type Fuel type Fuel type No fuel Unclass
1 2 3 4 5 6 7
Table 5 Numerical description of the fuel User Acc. ASTER (%)
Prod. Acc. Aster (%)
74.64 84.52 82.14 72.73 100.00 99.29 87.23 95.09 100.00
85.12 85.54 84.15 86.15 100.00 99.29 62.12 88.57 98.91
ASTER overall Acc. = 90.73; ASTER kappa coefficient = 0.89.
fuel type 1 (74.64%) show the worst user accuracy. That is mainly due to the high mixture that is present between fuel types 4 and 7. Additionally, a low mixing is also present between fuel types 1 and 2, 3. As regards fuel types 7 and 4 the mixing between the two classes consists in a very strong pixel ‘‘transfer’’ from fuel type 7 toward fuel type 4 and much weaker from fuel type 4 toward fuel type 7. The ML classifier ‘‘moves’’ 27% of the pixels of fuel type 7 toward fuel type 4 and less than 8% of fuel type 4 toward fuel type 7. This fact is mainly linked to the complexity of fuel types distribution that makes impossible an accurate selection of ROI at the spatial resolution of satellite ASTER data. The level of classification accuracy is substantially determined by the light mixing between fuel type 1 with fuel types 2, 3 and the class corresponding to the No fuel. Between fuel types 1 and 2 there is a nearly analogous ‘‘exchange’’ of a pixels number and the same it happens between fuel types 1 and 3. The complexity of fuel types distribution makes an accurate selection of ROI impossible. As regards the mixing between fuel type 1 and no fuel class, ML classifier ‘‘moves’’ 10% of the pixels from no fuel toward fuel type 1 class. This is due to the presence of typical grass species in cultivated areas. The main finding of our analysis is that the characterization of fuel type obtained from ASTER data can be performed with very high accuracy levels even for heterogeneous areas characterized by complex topography features and a mixture of vegetation covers. The high classification accuracy is due to the ability to distinguish several classes that should be easily confused. In particular, fuel type 3 versus fuel type 2; fuel type 4 versus fuel type 6; fuel type 7 versus fuel type 6. The high accuracy levels achievable using ASTER data showed that such imagery can be capable to give an exhaustive classification of fuel properties,
FT1 FT2 FT3 FT4 FT5 FT6 FT7
tonne/ha
kg/m2
kcal/kg
5 9 11 30 11 8 32.5
0.5 0.9 1.1 3 1.1 0.8 3.25
4500 4700 5200 5200 4500 4500 4500
with accuracy levels very close to those obtained from airborne hypestecral MIVIS data (Lasaponara and Lanorte, 2006a,b). In order to directly adopt fuel map types in operational contexts, it is important to develop accurate ways to translate forest types and density into fuel source parameters, namely fuel load could be estimated in joule or other energy measure. Table 5 shows energy measures expressed as metric tonne per ha (as well as kg/m2 and kcal/kg) of the fuel load computed for the seven fuel types of the study area, performed following Velez Munoz (1990). 5. Conclusions ASTER data were analysed for a test area of southern Italy to ascertain how well remote sensing data can characterize fuel type and map fuel properties. Fieldwork fuel type recognitions, performed at the same time as remote sensing data acquisitions, were used as ground-truth dataset to assess the results obtained for the considered test area. Results from our analysis showed that the spatial and spectral resolution of ASTER imagery provided a valuable characterization of fuel types being that the classification accuracy was higher than 90%. The high accuracy obtained from ASTER data showed that such imagery are able to provide an exhaustive classification of fuel properties even for heterogeneous areas characterized by a complex topography and mixed vegetation covers. Results obtained from these investigations can be directly extended to other Mediterranean like ecosystems. The main finding of the performed investigation is that satellite ASTER dataset assures an high capability for fuel characterization, with accuracy levels very close to those obtained from airborne hyperspectral data (see Lasaponara et al., 2006-a, b; Lasaponara et al., in press), but strongly reduces the cost of data acquisition currently lower than $ 90 for each ASTER image.
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
References Albini, F.A., 1976. Estimating wildfire behaviour and effects. General Technical Report. USDA Forest Service, Intermountain Forest and Range Experiment Station INT-30. Ogden, UT, pp. 1–92. Andersen, H.E., McGaughey, R.J., Robert, J., Reutebuch, S.E., 2005. Estimating forest canopy fuel parameters using Lidar data. Remote Sens. Environ. 4, 441–449. Anderson, H.E., 1982. Aids to determining fuels models for estimating fire behaviour. USDA Forest Service, Intermountain Forest and Range Experiment Station General Technical Report INT-122. Ogden, UT, 22 pp. Boardman, J.W., 1998. Leveraging the high dimensionality of AVIRIS data for improved sub-pixel target unmixing and rejection of false positives: mixture tuned matched filtering. Summaries of the Seventh JPL Airborne Geoscience Workshop, JPL Publication 97-1. NASA Jet Propulsion Lab., Pasadena, CA, pp. 55–56. Boardman, J.W., Kruse, F.A., Green, R.O., 1995. Mapping target signatures via partial unmixing of AVIRIS data. Summaries of the Fifth JPL Airborne Geoscience Workshop, JPL Publication 95-1. NASA Jet Propulsion Lab., Pasadena, CA, pp. 23–26. Burgan, R., Rothermal, R.C., 1984. BEHAVE: fire behaviour prediction and fuel modelling system-FUEL subsystem. USDA Forest Service General Technical Report INT-167, 126 pp. Castro, R., Chuvieco, E., 1998. Modelling forest fire danger from geographic information system. Geocarto Int. 13, 15–23. Chuvieco, E., Congalton, R.G., 1989. Application of remote sensing and geographic information systems to forest fire hazard mapping. Remote Sens. Environ. 29, 147–159. Chuvieco, E., 1999. Remote Sensing of Large Wildfires in European Mediterranean Basin. Springer-Verlag, Berlin, p. 122. Cohen, W.B., 1989. Potential utility of the TM tasseled cap multispectral data transformation for crown fire hazard assessment. In: ASPRS/ACSM Annual Convention Proceedings: Agenda for the 90s, vol. 3, Baltimore, MA, pp. 118–127. Congalton, R.G., Green, K., 1998. Assessing the accuracy of remotely sensed data. CRC Press, Lewis Publishers, Boca Raton, Florida, USA. Cuomo, V., Lasaponara, R., Tramutoli, V., 2001. Evaluation of a new satellite-based method for forest fire detection. Int. J. Remote Sens. 22 (9), 1799–1826. FAO, 2001. Global forest fire assessment 1990–2000. Forest Resources Assessment Programme, Working Paper No. 55, http://www.fao.org/forestry/fo/fra/docs/Wp55_eng.pdf. Jia, G.J., Ingrid, I.C., Goetz, A.F.H., Kaufmann, M.R., Kindel, B.C., 2006. Assessing spatial patterns of forest fuel using AVIRIS data. Remote Sens. Environ. 3/4, 318–327. Keane, R.E., Mincemoyer, S.A., Schmidt, K.A., Long, D.G., Garner, J.L., 2000. Mapping vegetation and fuel for fire management on the Gila National Forest Complex. USDA Forest Service General Technical Report RMRS-GTR-46-CD, New Mexico. Keane, R.E., Burgan, R., Van Wagtendonk, J., 2001. Mapping wildland fuels for fire management across multiple scales: integrating remote sensing, GIS, and biophysical modelling. Int. J. Wildl. Fire 10 (3/4), 301–319. Lasaponara, R., in press. Inter-comparison of AVHRR-based fire danger estimation methods. International J. Remote Sens. 26 (5), 853–870. Lasaponara, R., 2006. Estimating Spectral separability of satellite derived parameters for burned areas mapping in the Calabria Region by using SPOT-vegetation data. Ecol. Model. 196, 265– 270.
233
Lasaponara, R., Lanorte, A. 2006a Multiscale fuel type characterization by using multisensor remote sensing data for the Mediterranean ecosystems of Southern Italy. IEMSS conference July, 9–13, Burlington, Vermont USA, in press. Lasaponara R., Lanorte, A. 2006b Remote characterization of fuel type using satellite data, SPIE conference, Stoccolma-12–14 September, in press. Lasaponara R, A Lanorte 2006b Multispectral fuel type characterization based on remote sensing data and Prometheous model ICFFR Conference, Coimbra, November 2006. Lasaponara, R., Lanorte, A., Pignatti, S., 2006-a. Characterization and mapping of fuel types for the Mediterranean ecosystems of Pollino National Park in the Southern Italy by using Hyperspectral MIVIS data. Earth Interactions 10, 1–11. Lasaponara, R., Lanorte, A., Pignatti, S., 2006-b. Multiscale fuel type mapping in fragmented ecosystems: preliminary results from Hyperspectral MIVIS and Multispectral Landsat TM data. Int. J. Remote Sens. 27 (3), 587–593. Lasaponara R., Lanorte, A., Pignatti, S., in press. Hyperspectral analysis of fuel types for Mediterranean ecosystems, Bull. Am. Meteorol. Soc. Lasaponara, R., Cuomo, V., Macchiato, M.F., Simoniello, T., 2003. A self-adaptive algorithm based on AVHRR multitemporal data analysis for small active fire detection. Int. J. Remote Sens. 24 (8), 1723–1749. Lillesand, T.M., Kiefer, R.W., 2000. Remote Sensing and Image Interpretation. Wiley, New York. Mckinley, R.A., Chine, E.P., Werth, L.F., 1985. Operational Fire Fuels Mapping with NOAA-AVHRR data. American Society for Photogrammetry and Remote Sensing, Falls Church, VA, pp. 295–304. Merrill, D.F., Alexander, M.E., 1987. Glossary of Forest Fire Management Terms. National Research Council of Canada, Committee for Forest Fire Management, Ottawa. Morris, W.G., 1970. Photo inventory of fine logging slash. Photogrammet. Eng. 36, 1252–1256. Muraro, S.J., 1970. Slash Fuel Inventories from 70 mm Low-level Photography. Canadian Forest Service, Ottawa, Ontario, p. 63. Oswald, B.P., Fancher, J.T., Kulhavy, D.L., Reeves, H.C., 1999. Classifying fuels with aerial photography in East Texas. Int. J. Wildl. Fire 9 (2), 301–319. Prometheus, S.V., Project, 1999. Management techniques for optimisation of suppression and minimization of wildfire effects. System Validation. European Commission, contract number ENV4-CT980716. Rian˜o, D., Chuvieco, E., 2002. Generation of fuel type maps from Landsat-TM images and auxiliary data in Mediterranean ecosystem. Department of Geography, Alcala´ de Henares University, Alcala´ de Henares, Spain. Rian˜o, D., Chuvieco, E., Salas, J., Palacios-Orueta, A., Bastarrika, A., 2002. Generation of fuel type maps from Landsat TM images and ancillary data in Mediterranean ecosystems. Canadian Journal of Forest Resources 32, 1301–1315. Salas, J., Chuvieco, E., 1994. Geographic information system for wildland fire risk mapping. Wildfire 3 (2), 7–13. Telesca, L., Lasaponara, R., 2005. Discriminating dynamical patterns in burned and unburned vegetational covers by using SPOT-VGT NDVI data. Geophys. Res. Lett. 32, L21401 2005, doi:10.1029/ 2005GL024391. Telesca, L., Lasaponara, R., 2006-a. Pre-and post-fire behavioural trends revealed in satellite NDVI time series. Geophys. Res. Lett., 33, L14401 2006, doi:10.1029/2006GL026630
234
R. Lasaponara, A. Lanorte / International Journal of Applied Earth Observation and Geoinformation 9 (2007) 225–234
Telesca, L., Lasaponara, R., 2006-b. Vegetational patterns in burned and unburned areas investigated by using the detrended fluctuation analysis. Physic. A. 368, 531–535. Telesca, L., Lasaponara, R., 2006-c. Fire-induced variability in satellite SPOT-VGT NDVI vegetational data. Int. J. Remote Sens. 27, 3087–3095. Telesca, L., Lasaponara, R., Lanorte, A., 2006-d. Discrimination of vegetational patterns in burned and unburned areas. Int. J. Nonlinear Sci. Numer. Simul. 7 (3), 279–284. Van Wagtendonk, J.W., Root, R.R., 2003. The USE of multitemporal Landsat normalized difference vegetation index (NDVI) data for
mapping fuels models in Yosemite National Par k, USA. Int. J. Remote Sens. 24, 1639–1651. Velez Munoz, R., 1990. Algunas observaciones para una selvicoltura preventive de incendios forestales. Ecologia, Fuera de Serie 1, 561–571. Vila, M., Lloret, F., Ogheri, E., Terradas, J., 2001. Positive fire–grass feedback in Mediterranean basin shrubland. For. Ecol. Manage. 147, 3–14. Zhu, G., Blumberg, D.G., 2001. Classification using ASTER data and SVM algorithms; the case study of Beer Sheva, Israel. Remote Sens. Environ. 80, 233–240.