Proceedings of the 2nd Workshop of the EARSeL SIG on Land Use and Land Cover

MAPPING OF FUEL COVER USING REMOTE SENSING DATA Rosa Lasaponara, Antonio Lanorte Consiglio nazionale delle Ricerche – Istituto di Metodolologie di Analisi Ambientali (CNR-IMAA), Potenza, Italy; [email protected], [email protected] ABSTRACT In the context of fire management, fuel maps are essential information requested at many spatial and temporal scales for managing wildland fire hazard and risk and for understanding ecological relationships between wildland fire and landscape structure. This study aims to ascertain how well remote sensing data can characterize fuel type at different spatial scales in fragmented ecosystems. For this purpose, multisensor and multiscale remote sensing data have been processed. 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; (III) accuracy assessment for the performance evaluation based on the comparison of satellite-based results with ground-truth. Two different approaches have been adopted for fuel type mapping: the well-established classification techniques and spectral mixture analysis. Results from our investigations showed that remote sensing data can provide valuable information for the characterization and mapping of fuel types and vegetation properties at different temporal and spatial scales from global, regional to landscape level. INTRODUCTION In the context of fire management, fuel maps are essential information requested at many spatial and temporal scales for managing wildland fire hazard and risk and for understanding ecological relationships between wildland fire and landscape structure. Due to the complex nature of fuel characteristic a fuel map is considered one of the most difficult thematic layers to build up especially for large areas. Aerial photos have been the most common remote sensing data source traditionally used 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. 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 different sources of satellite imagery (i,ii,iii,iv) High accuracy levels (higher than 90%) were obtained from MIVIS and Aster data even for extremely heterogeneous areas (5,6,7). Satellite multispectral data can be fruitfully adopted for building up fuel type maps from global, region down to local scale. This study aims to ascertain how well remote sensing data can characterize fuel type at different spatial scales in fragmented ecosystems. For this purpose, multisensor and multiscale remote sensing data such as, hyperspectral MIVIS as well as Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), TM and MODIS data have been analysed for some test areas of southern Italy that are characterized by mixed vegetation covers and complex topography. Fieldwork fuel types recognitions, performed at the same time as re-

164

Center for Remote Sensing of Land Surfaces, Bonn, 28-30 September 2006

mote sensing data acquisitions, were used as ground-truth dataset to assess the results obtained for the considered test areas. Table 1: Fuel type classification developed for Mediterranean ecosystems in the framework of Prometheus project (Prometheus Project 1999). Fuel Type Fuel Type description in terms Fuel Type description in terms of vegetation class of percentage of cover typology 1

Ground fuels (cover >50%)

grass

2

Surface fuels (shrub cover grassland, shrub land (smaller than 0.3–0.6 >60%, tree cover 60%, tree cover 60%, tree cover 4 m) with a the ground fuel was removed either by preclean ground surface (shrub scribed burning or by mechanical means. This situation may also occur in closed canocover 4 m) with me- the base of the canopies is well above the dium surface fuels (shrub surface fuel layer (>0.5 m). The fuel consists essentially of small shrubs, grass, litter, and cover >30%) duff (the layer of decomposing organic materials lying immediately above the mineral soil but below the litter layer of freshly fallen twigs, needles, and leaves; the fermentation layer).

7

Tree stands (>4 m) with heavy stands with a very dense surface fuel layer surface fuels (shrub cover and with a very small vertical gap to the canopy base (30%)

METHODS The method comprised the following three steps: (I) adaptation of Prometheus fuel types vfor 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; (III) accuracy assessment for the performance evaluation based on the comparison of satellite-based results with ground-truth. Two different approaches have been adopted for fuel type mapping: the well-established classification techniques and spectral mixture analysis. Results from our investigations showed that remote sensing data can provide valuable information for the characterization and mapping of fuel types and vegetation properties at different temporal and spatial scales from global, regional to landscape level. The maximum likelihood classification (MLC) see, for example as adopted for fuel type mapping. This classification, as with other conventional hard classification techniques, assumes that all image pixels are pure. Nevertheless, this assumption is often untenable. In mixed land cover compositions, as pixels increase in size, the proportion of mixed cover type distributed

165

Proceedings of the 2nd Workshop of the EARSeL SIG on Land Use and Land Cover

at pixel level will likewise increase and information at the sub-pixel level will be of increasing interest. Consequently, in fragmented landscapes conventional “hard” image classification techniques provide only a poor basis for the characterization and mapping of fuel types giving, in the best case, a compromised accuracy, or, in the worst case, a totally incorrect classification. In these conditions, the use of spectral mixture analysis (SMA) can reduce the uncertainty in hard classification techniques since it is able to capture, rather than ignore, subpixel heterogeneity. The SMA allows for classifying the proportions of the ground cover types (endmember classes) covered by each individual pixel. End-member classes can be taken from “pure” pixels within an image or from spectral libraries. Over the years, different models of spectral mixtures have been proposed (vi) that is based on the assumption that the spectrum measured by a sensor is a linear combination of the spectra of all components within the pixels. On the basis of ground surveys and air photos, we selected the region of interest (ROI) corresponding to the considered 7 fuel types, plus 2 additional classes related to no fuel and unclassified regions. Pixels belonging to each of the considered ROI were separated randomly into data training and data testing, used for the MLC and accuracy evaluation respectively. Results obtained from different remote data sources were compared on the basis of the achieved accuracy levels. 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 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. Overall accuracy is computed as the sum of the number of observations correctly classified (class1, 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. RESULTS & DISCUSSION Study area description The selected study area (figure 1) extends inside the National Park of Pollino in the Basilicata Region (Southern Italy). It is characterized by complex topography with altitude varying from 400 m to 1900 m above sea level (asl) and mixed vegetation covers. Between 400 and 600 m natural vegetation is constituted by the Mediterranean scrubs, xeric prairies and Mediterranean shrubby formations. In the strip included between 600 and 1000-1200 m the characteristic vegetation is represented by poor populations of Quercus pubescens and from extensive woods of Turkey oaks (Quercus cerris); evident degradation Figure 1. Arrow and forms are present, in the form of xerophytic prairies and substi- black box indicate the tution bushes. The higher horizons are constituted by beech study area. woods (Fagus sylvatica) which arrive up to 1900 m: the deforested areas in this strip are generally engaged by mesophytic prairies used for pasture.

166

Center for Remote Sensing of Land Surfaces, Bonn, 28-30 September 2006

The remotely sensed characterization of fuel types was performed by adopting as reference the fuel types classification (Table 1) developed for Mediterranean ecosystems in the framework of the Prometheus project (Prometheus Project 1999). RESULTS Two different approaches were adopted for mapping fuel types, (I) the well-established Maximum Likelihood Classification (MLC) and the spectral mixture analysis. On the basis of ground surveys and air photos, we selected the Region Of Interest (ROI) corresponding to the considered 7 fuel types, plus 2 additional classes related to no Fuel and unclassified areas. Pixels belonging to each of the considered ROI were randomly separated into data training and data testing, used for the MLC and accuracy evaluation respectively. Figure 2 shows the ROI selected for the considered classes for MIVIS, ASTER, and TM data. Due to the heterogeneity of the investigated area, the ROI for MODIS data were selected by using a region larger than those shown in Figure 2.

Table 3a: Accuracy coefficients obtained from MIVIS data. Class

Prod. User Acc.(%) Acc.(%) Fuel type 1 98.29 88.41 Fuel type 2 89.20 71.71 Fuel type 3 49.09 84.13 Fuel type 4 96.24 73.04 Fuel type 5 100.00 99.92 Fuel type 6 95.44 99.52 Fuel type 7 94.64 98.35 No fuel 99.14 98.68 Unclassified 99.38 100.00 Overall Accuracy = 90.3964% Kappa Coefficient = 0.8905

The evaluation of results obtained for the different data set by applying the MLC and MTMF was performed by using the following traditional accuracy measurements, producer accuracy, user accuracy and overall accuracy. The producer accuracy is a measure indicating the probability that the classifier has labelled an image pixel for example into Fuel type 1 Class given that the ground truth is Fuel type 1 Class. Table 3b: Accuracy coefficients obtained from ASTER data. The user accuracy is a measCommi Omission User Producer ure indicating the probability Class (%) ssion AccuAccuthat a pixel is for example (%) racy.(%) racy(%) Fuel type 1 Class given that 93.95 88.91 11.09 6.05 the classifier has labelled the Fuel type 1 Fuel type 2 46.00 44.09 55.91 54.00 pixel into Fuel type 1 Class. Fuel type 3 59.13 77.71 22.29 40.87 Overall accuracy is calculated Fuel type 4 70.33 54.07 45.93 29.67 by summing the number of Fuel type 5 95.70 88.58 11.42 4.30 pixels classified correctly and Fuel type 6 86.24 93.44 6.56 13.76 dividing by the total number of Fuel type 7 82.39 82.95 17.05 17.61 pixels. No fuel 99.05 99.05 0.95 0.95 Unclassi88.00 100.00 0.00 12.00 In order to compare the results obtained from the differ- fied ent data set, “a single date” Overall Accuracy = 81.9384% processing was performed. Kappa Coefficient= 0.7849 Firstly, results obtained from the different acquisition dates were very close each other. Secondly, as expected, data (MIVIS and ASTER) at higher spatial resolution substantially provide very close levels of accuracy from both MLC and MTMF. For this reason, Table 2 solely shows the accuracy levels obtained from the MLC algorithm. In the case of TM and MODIS data processing, results from our preliminary analysis have showed that the use of unmixing technique allows us for improving at around 7% and 5% the overall accuracy level obtained respectively for TM and MODIS compared to the results ob-

167

Proceedings of the 2nd Workshop of the EARSeL SIG on Land Use and Land Cover

tained by applying the MLC algorithm. Some improvements may be achieved by merging class 2 with class 3 due to the fact that such classes refer to similar vegetation typologies that are substantially differentiated by the highly Table 3c: Accuracy coefficients obtained from TM data. CLASS Fuel type 1 Fuel type 2 Fuel type 3 Fuel type 4 Fuel type 5 Fuel type 6 Fuel type 7 No fuel

MCL Producer User Accuracy (%) Accuracy (%) 89.74 71.43 82.26 68.00 62.77 36.65 92.06 85.29 76.87 62.09 67.37 77.56 48.39 63.83 77.26 97.23 Overall Accuracy = 73.83% Kappa Coefficient = 0.6763

MTMF Producer User Accuracy (%) Accuracy (%) 76.92 100.00 80.65 73.53 75.53 52.99 92.06 89.23 80.95 80.41 81.36 82.76 56.45 81.40 91.69 95.66 Overall Accuracy = 83.63% Kappa Coefficient = 0.7924

Table 3d: Accuracy coefficients obtained from MODIS data. Classification Sensor Class Fuel type 1 Fuel type 2 Fuel type 3 Fuel type 4 Fuel type 5 Fuel type 6 Fuel type 7 No fuel Unclassified Overall Accuracy Kappa Coefficient

ML classification TM Prod. Acc.(%) User Acc.(%) 70.83 83.61 46.60 41.03 56.47 67.88 62.96 46.48 78.77 42.59 44.33 76.22 54.00 63.68 95.68 94.51 89.13 73.21 62.48%

MTMF TM Prod. Acc.(%) User Acc.(%) 80.56 70.30 48.54 52.08 54.31 68.11 74.60 59.75 77.40 53.05 52.84 75.25 71.60 69.92 95.68 94.51 89.13 100.00 68.85%

0.5735

0.6443

Final remarks Hyperspectral MIVIS, Landsat-TM, Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER), and MODIS have been processed by using MLC and MTMF for fuel type mapping performed in the fragmented ecosystems of Pollino national park. Fieldwork fuel types 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 areas. As expected, results from the higher spatial resolution data namely MIVIS and ASTER imagery substantially provide very close levels of accuracy from both MLC and MTMF higher than 89% and higher than 78% respectively for MIVIS and ASTER data. Whereas, in the case of TM and MODIS data processing, results from our analysis showed that the use of unmixing technique allows us for improving at around 7% and 12% the accuracy level obtained respectively for TM ( k coefficient from 57% to 64%) and MODIS (k coefficient from 67% to 79%) applying the widely used classification algorithms. Results obtained from our analyses confirmed the effectiveness of MLA in handling spectral mixture problems, and, in particular, showed that the selection of appropriate end-members 168

Center for Remote Sensing of Land Surfaces, Bonn, 28-30 September 2006

and image bands to be processed is crucial for developing high quality fraction images using LSMA. The use of spectral signature can provide a fundamental aid to obtain satisfactory results from the classification process. REFERENCES

i Lasaponara R., and A. Lanorte, 2006. “Remotely sensed characterization of forest fuel

types by using satellite ASTER data” International Journal of Applied Earth Observations and Geoinformation (In press JAG 139 Editorial reference: JAG-D-06-00010). ii Lanorte A., and Lasaponara R., 2006. “Fuel type characterization based on coarse reso-

lution Modis satelite data. Forest@ (in press). iii Lasaponara R., A. Lanorte, S. Pignatti,2006 Multiscale fuel type mapping in fragmented ecosystems: preliminary results from Hyperspectral MIVIS and Multispectral Landsat TM data, Int. J. Remote Sens., vol. 27 (3) pp. 587-593. iv Lasaponara R., A. Lanorte, S. Pignatti, 2006. 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. Vol. 10, pp. 1-11. v PROMETHEUS, S.V. PROJECT, 1999, Management techniques for optimisation of suppression and minimization of wildfire effects. System Validation. European Commission. Contract number ENV4-CT98-0716. vi Ichku, C., and Karnieli, A., (1996), A review of mixture modelling techniques for sub-pixel land cover estimation. Remote Sensing Reviews, 13:161–186.

169