Object Based Technique for Delineating and Mapping 15 Tree Species using VHR WorldView-2 Imagery Yaseen T. Mustafa*a, Hindav N. Habeebb a Faculty of Science, University of Zakho, Kurdistan Region–Iraq; b Directorate of Forestry, Duhok, Kurdistan Region–Iraq ABSTRACT Monitoring and analyzing forests and trees are required task to manage and establish a good plan for the forest sustainability. To achieve such a task, information and data collection of the trees are requested. The fastest way and relatively low cost technique is by using satellite remote sensing. In this study, we proposed an approach to identify and map 15 tree species in the Mangish sub-district, Kurdistan Region-Iraq. Image-objects (IOs) were used as the tree species mapping unit. This is achieved using the shadow index, normalized difference vegetation index and texture measurements. Four classification methods (Maximum Likelihood, Mahalanobis Distance, Neural Network, and Spectral Angel Mapper) were used to classify IOs using selected IO features derived from WorldView-2 imagery. Results showed that overall accuracy was increased 5-8% using the Neural Network method compared with other methods with a Kappa coefficient of 69%. This technique gives reasonable results of various tree species classifications by means of applying the Neural Network method with IOs techniques on WorldView-2 imagery. Keywords: Kurdistan Region-Iraq, Remote sensing, Supervised classification, Satellite imagery, Tree species 1. INTRODUCTION Managing and monitoring of forest has been on the agenda of forestry research in recent years. The reason is that trees play a major role in CO2 sequestration and that they contribute to the reduction of carbon emissions to the atmosphere1. They count as a source of providing aesthetics, fuel wood, natural areas, recreation, timber, and wildlife in infinite combination2, 3. Hence, it is worthwhile to monitor, manage, analyze and map trees which can be achieved by collecting a required data. Two ways are distinguish to collect trees data: traditional techniques (field survey), and modern techniques (remote sensing observation). Traditional techniques require a lot of efforts, cost, time, and the difficulty in accessing some area. However, to overcome the limitation of the traditional techniques, the modern techniques as a remote sensing need to be used4. Remote sensing technology has become increasingly important tools for mapping, inventorying, and monitoring forest resources around the world especially after the availability of very high resolution (VHR) satellite imagery5, 6. In recent years, for example, QuickBird imagery has been utilized for trees mapping using pixel-based image classification methods7, 8. The object-based image analysis (OBIA), however, is preferable to be used for the classification with VHR images instead of using the conventional pixel by pixel classification9. This is due to the fact that the high spectral variability within class decreases classification accuracy using pixel-based approaches with the VHR images. Moreover, pixel-based approaches ignore the context and the spectral values of adjacent pixels10. OBIA techniques first use image segmentation to produce image objects (IOs) that are more homogeneous regions (e.g., a tree crown), and then these IOs rather than pixels are used as the classification unit11. Several efforts have been achieved to map species composition using VHR images as IKONOS and QuickBird. However, few studies using WorldView-2 (WV2) data have been reported12, 13. Even so, researchers utilized WV2 to map not more than 7 tree species11, 13. In this context, the objective of this work is to delineate and map 15 tree species by WV2 imagery by means of IOs using four classifiers, Maximum Likelihood, Mahalanobis Distance, Neural Network, and Spectral Angel Mapper. The study were applied to a forest area in Duhok, Kurdistan Region-Iraq (Mangish sub-district), where no such a research has been achieved there yet. *yaseen.mustafa@uoz.ac; phone 0096475022553; www.uoz.ac Remote Sensing for Agriculture, Ecosystems, and Hydrology XVI, edited by Christopher M. U. Neale, Antonino Maltese, Proc. of SPIE Vol. 9239, 92390G © 2014 SPIE · CCC code: 0277-786X/14/$18 · doi: 10.1117/12.2067280 Proc. of SPIE Vol. 9239 92390G-1 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms 2. STUDY AREA AND DATA 2.1 Study Area Description Mangish is approximately located between latitudes 37o07'05″ - 36o55'27″ N and longitudes 42o48'09″- 43o13'59″ E (Figure 1) and about 489.63 km2. The maximum elevation reaches more than 1500 m above sea level in the east and the lowest elevation reaches not less than 500 m above sea level at the west part of the study area (Figure 2). It contains natural and planted trees. Land of Mangish is used for field crops (wheat and barley) and vineyards and orchards cover the foothills. The forest cover consists of different tree species including: Azarole hawthorn (Crataegus azarolus), Tera binth (Pistacia), Almand (Prunus duclis), Calabrain pine (Pinus brutia), Canary Islands Junipirus (Junipirus oxycedrus), Common fig (Ficus carica), Common walnut (Juglans regia), Gall Oak (Quercus Infectoria), Jerusalem thorn (Paliurus spina Christi), Oriental plane (Platanus orientalis), silver (Populus euphratica), Valonia Oak (Quercus aegilops), White Mulberry (Morus), White willow (Salix), and Oleaster (Elaeagnus angustifolia)14. (b) (c) I 0 I 150 Ikm 300 Figure 1. (a) Map of Iraq, (b) Map of Duhok province, (c) Map of the Mangish Legend Contour Lines DEM 5 High 512 km 10 Low : 509 Figure 2. Digital Elevation Model (DEM) image of the study area with 30 m resolution retrieved from15. 2.2 Data description The study is primarily based on field data and the WV2 data. 2. 2. 1 Field data: This data include the tree species name and their location (longitude, latitude, and altitude). A fieldwork carried out between June 19- July 20, 2013. Two Differential Global positioning system (DGPS) Proc. of SPIE Vol. 9239 92390G-2 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms devices (Leica Viva GS15 Smart Rover) were used. Prior to the fieldwork, false color composite images WV2 were brought to the field to directly locate and delineate tree species on the images for later use of determining training and validation (Table 1). A simple random sampling is used due to its satisfactory results as reported by Congalton 16. Based on the trees popularity and abundant in the study area, they have been categorized into two groups; main trees (5 species) and secondary trees (10 species). Table 1. Training and validation samples of the fifteen (main and secondary) tree species that is used in this study. Tree species Main Trees: Pinus brutia Quercus aegilops Quercus Infectoria Pistacia Juglans regia Secondary Trees: Prunus duclis Crataegus azarolus Junipirus oxycedrus Platanus orientalis Populus euphratica Salix Paliurus spina Christi Morus Ficus carica Elaeagnu angustifolia Total Training Validation 77 227 114 67 44 45 110 55 34 22 58 27 45 27 35 92 17 61 27 13 931 29 14 27 13 18 45 9 31 15 7 474 2. 2. 2 Satellite data: The unique satellite that has a high spatial resolution with 8 multispectral (MS) bands and one panchromatic (Pan) band is WV2 owed by Digital Globe. The characteristics of WV2 imagery that is used in this study is shown in Table 2. Fourteen cloud free WV2 scenes were acquired to cover the study area from 11 June to 10 July 2011. Table 2. Characteristics of the WV2 imagery that used in this study Spectral Wavelength Band width Spatial resolution bands (μm) (μm) (m) 1 Coastal 0.40–0.45 0.50 1.84 2 Blue 0.45–0.51 0.60 1.84 3 Green 0.51–0.58 0.70 1.84 4 Yellow 0.59–0.63 0.40 1.84 5 Red 0.63–0.69 0.60 1.84 6 Red Edge 0.71–0.75 0.40 1.84 7 NIR1 0.77–0.90 0.125 1.84 8 NIR2 0.86–1.04 0.128 1.84 Pan 0.46–0.80 0.350 0.5 3. METHODS The step by step methodology is shown as a flowchart in Figure 3, which consists of four main stages. Proc. of SPIE Vol. 9239 92390G-3 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms IO Process Pre-Process !Ui ú62 OuN° w0291 2Nuxbsu ybbiT w; 2Nsxbsu ybbiT w Grcv Classification Process IL9RI]I7ñ Results wubbiuS Figure 3. Summary of the research workflow 3.1 Image preprocessing The processing and the manipulation of the satellite data were done using ENVI software (V. 5.0, Exelis Visual Information Solutions Group, Boulder, CO, USA). The radiometric calibration of WV2 is already provided by DigitalGlobe 17. The following steps were implemented in the image preprocessing stage: 3.1.1. Ortho-rectification: Ortho-rectification is the process of removing the distortion within an image caused by terrain relief and the sensor18. This can be done using auxiliaries satellite data. For that purpose, DEM that covered the study area was used (Figure 2). It was obtained by Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) data15 from Global Data Explorer19. The ortho-rectification process was achieved twice, once for panchromatic band alone and once for multispectral bands using a ready tool in ENVI. Moreover, the ground control points were used to geometrically correct the ortho-rectified image, where 120 ground control points were used by using DGPS and projected to the Universal Transverse Mercator (UTM Zone 38N) using WGS-84 Geodetic datum. 3.1.2. Image fusion (Pan-sharpening): Pan-sharpening (PS) is the process of combining the lower-resolution imagery (spectral information) with the higher resolution image (spatial information) to produce a high resolution spectral image. It has been used to improve the classification of forests20. Several algorithms have been used to achieve the PS process, and based on the recommendation of Pu and Landry 11, the Gramm-Schmidt Spectral Sharpening algorithm adopted and used in this study. As a result, pan-sharpened 0.5 m resolution WV2 images were created by fusing the 1.84 m MS WV2 imagery with the 0.5 m Pan WV2 imagery using Gramm-Schmidt pan-sharpening method. 3.1.3. Mosaic: As the study area was covered by 14 WV2 senses, therefore, the mosaic was achieved for all these senses and a single image that represents the required study area was created. Next, the area of interest from the mosaicked image was extracted and the undesirable parts were deleted. Proc. of SPIE Vol. 9239 92390G-4 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms 3.2 Image object (IO) processing 3.2.1. Shadow index (SI): VHR satellite imagery offers a great details and information of the land with a precise map. Shadows, however, may obscure information in the image leading to the corrupted classification results. This is noticeable in the forest, for example, the shade of trees may be counted as another pattern which in turns affects to the classification results. Therefore, it is necessary to remove the shadow which is achieved by the following proposed equation. − = − where NIR and Red are the near-infrared and the Red-reflectance bands, respectively, that represent band no.7 and band no.5, respectively, in WV2. Generally, the range of SI value is between 0 and 256. Further, the shadow areas were masked out from the image. 3.2.2. Normalized difference vegetation index (NDVI): The object of interest in this study is the tree (tree crown). Therefore, it was better to remove the ground−including water, soil, and any non-vegetation objects. In this context, the following equation21 was applied: = where NIR and Red are the near infrared and Red-reflectance bands, respectively. Next, the non-vegetation areas were masked out from the images, such that an image with vegetation areas only was created. 3.2.3. Texture: One of the methods that help to interpret and identify forest stands from very high resolution imagery is texture information 22. The Grey-Level Co-occurrence Matrix (GLCM) 23 is a common algorithms for computing texture measures 5, 24. Among the fourteen GLCM texture measures, that were originally proposed by Haralick 25, angular second moment, contrast , homogeneity and variance are the most frequently used texture features for the classification of VHR imagery 26. In this study, these four features were calculated for the NIR band using 9×9 processing windows. The reason of selecting the NIR band was that it contained the greatest range in spectral brightness values. Next, we separated tree canopy from non-tree canopy (shrub and grass/lawn land, etc.) with applying a threshold on the textural feature values. The threshold value was chosen based on a trial and error approach. For example, with homogeneity and variance cases a threshold was determined based on observing a low value of homogeneity and high value of variance for the tree canopy while with grass, a high value of homogeneity and low value of variance were selected. 3.3 Image classification Image classification is defined as the process of extracting differentiated classes or themes (e.g., trees species) from raw remotely sensed satellite data 27. Generally, this technique is categorized into two groups: unsupervised and supervised classification. In this study, through using OBIA method, the supervised classification was adopted and investigated with four algorithms (classifiers): Maximum Likelihood (ML), Mahalanobis Distance (MahaD), Spectral Angle Mapper (SAM), Neural Networks (NN). 4. RESULTS AND DISCUSSION 4.1 Trees without shadow Figure 4(c) shows the created mask area from PS image which has two values only 0 (black area) and 1 (white area), whereas, Figure 4(d) shows the final image after applying the SI mask (i.e., masking out the shaded area from the image). The threshold value for SI was determined experimentally as in this study and was chosen to be 223. The very white color attached to the trees in Figure 4(b) represents the shade of the tree crown. Meanwhile, the same area is shown in Figure 4(c) as a black color area which was used as a masked area with 0 values. The final output of this process is shown in Figure 4(d) where it is very evident that the shaded area was removed and turns into an image without shadow. It should be mentioned that the result of this process did not remove the shade of the tree crown only; it further removed the shade of other objects; for example, the shade of rocks. However, this has had no a major influence on the main target of this study. Proc. of SPIE Vol. 9239 92390G-5 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms (a) (b) 20 m 10 m (c) (d) t Figure 4: A small portion of the study area showing the results of the SI process. (a) image after PS process; (b) resulted image from applying Eq. 1 (c) built mask image created after selecting a threshold; (d) image without shadow, created after applying the mask. 4.2 Identify and mapping tree species Figure 5 shows the NDVI map of the whole study area. It appears in a grey color because it consists of one band, which is created based on two WV2 bands (Eq. 2). In this map, the very white color represents the vegetation area, while the dark grey and black represent the non-vegetation area. Figure 5: (a) Color composite satellite imagery of Mangish area; (b) NDVI Image of Mangish was made using Eq. 2. Figure 6 shows the results of this process for a small portion of the study area. The image after removing the shadow is shown in Figure 6(a), while the calculated NDVI image from SI image is shown in Figure 6(b). Meanwhile, Figure 6(c) Proc. of SPIE Vol. 9239 92390G-6 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms shows the created mask image from NDVI image. This image was created after selecting a threshold value for NDVI which was 0.32 in this study. The final result of this process is shown in Figure 6(d) which includes the vegetation canopy only in the image without urban, water, roads, and soil. L k 10 m (a) (b) (c) (d) li e A ee 1 i 20 m Figure 6: A portion of a study area showing the results of the NDVI process. (a) image without shadow; (b) NDVI image; (c) built mask image resulted after selecting a threshold; (d) image with vegetation canopy only. Although, NDVI helps to distinguish between vegetation and non-vegetation, grass and trees cannot be distinguished. This is because NDVI of a dense (or healthy) grass may have the same (or close) value of NDVI of the tree crown as shown in Figure 6(d). Therefore, considering the texture feature as an additional process overcomes this limitation. Figure 7 shows examples of four textural features calculated using GLCM algorithm for NIR band of the WV2 image. These features contributed to the process of separation tree canopy and non-tree canopy. Therefore, a threshold was used for that purpose. However, the criteria used in selecting a proper threshold were depending on a combination of at least two textures feature measurements. For instance, with the tree canopy the value of angular second moment and homogeneity decreased to a very low value while simultaneously the value, of contrast and variance increased to become a very high value. This is noticeable in Figure 8(a*) and (a**) with their attached chart. Next, the grass was masked out from the whole image and kept tree canopy. Figure 8(c) shows the final results of this process for such a particular portion of the study area. It contains tree canopy only with no other information around the trees which turns into what is so called IO that represent tree crown. Next, the image becomes ready for the classification process. Figure 9 shows the IO of the whole study area. Proc. of SPIE Vol. 9239 92390G-7 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms (b) (a) (d) (c) I i il 20 m Figure 7: GLCM textures (a) angular second moment; (b) contrast; (c) homogeneity; (d) variance. 200 Grass (a*) 140.8 100 Texture Values 150 192 50 0 ASM CON HO OM am 86 70 VAR (b) (a) Ia (c) . r 4 700 Tree 600 500 (a**) 200 300 400 443.2 r 71.3 4 28.8 0 100 Texture Values i 639.5 20 m ASM CON HO VAR Figure 8: (a) image with tree canopy and non-tree canopy (a*) example of the non-tree canopy (grass) with its chart; (a**) example of the tree canopy with its chart; (b) original image; (c) final image after texture process (IO raster). Proc. of SPIE Vol. 9239 92390G-8 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms Jkm 10 Figure 9: WV2 image of the study area containing IO (tree crown) only. 4.3 IO classification Four different classification algorithms (ML, SAM, MahaD, NN) were used to perform digital image classification of tree species. Each classification followed by a sequence of training, and evaluation. The results of each classification were evaluated using confusion matrix. This procedure was implemented and evaluated for both groups (main and secondary trees) independently. Next, a better classifier performance was selected based on the accuracy criteria that were resulted from confusion matrix. The resulted classification accuracy (overall accuracy-OA, and Kappa coefficientKC) gained from each classifier is reported in Table 3. Table 3. The resulted classification accuracy of implementing four classifiers for main and secondary trees. Secondary Main Trees Criteria ML MahaD SAM NN OA (%) 73.31 73.63 71.41 77.50 KC 0.65 0.65 0.62 0.75 OA (%) 64.52 41.72 30.46 76.00 KC 0.52 0.30 0.17 0.63 From Table 3 we noticed that the best accuracy of both groups (main and secondary) is NN where the OA accuracy of main and secondary trees is 77.5% and 76%, with KC of 0.75 and 0.63, respectively. Therefore based on this result the NN was adapted for further process of creating trees map. Figure 10 is a graphical comparison between all classifiers in terms of their accuracy. It shows that NN classifier gives a better accuracy for both main and secondary trees. The resulted producer accuracy (Prod. Acc.) and the user accuracy (User Acc.) from the confusion matrix by NN classifier for the main and secondary trees are reported in the Table 4. The values of Prod. Acc. And User Acc. were varying from species to species. For example, the Prod. Acc. of Pinus brutia was 41.43, while it was 74.00 for Quercus aegilops. This might be due to the shape and size of the tree leaves (as broad vs needle) that has an influence of the reflected value that was detected by the satellite sensor. In addition, the low Prod. Acc. of Pinus brutia is due to the interference between the reflected values from Pinus brutia with the reflected values of some other species which has a dark color (or close to black). However, this value (Prod. Acc.) was much lower before removing the tree crown shadow, because the reflected values from the shade of the tree crown gave the similar reflected values of Pinus brutia. Proc. of SPIE Vol. 9239 92390G-9 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms Overall Accuracy (%) 1.0 100 (a ) 73.63 73.31 77.5 76 Kappa Coefficient Main Trees Secondary Trees 0.8 80 Main Trees Secondary Trees (b ) 0.75 71.41 0.65 60 0.6 64.52 0.65 0.63 0.62 0.52 40 0.4 41.72 0.3 0.17 0 0.0 20 0.2 30.46 ML MahaD SAM NN ML MahaD SAM NN Figure 10. Resulted (a) Overall Accuracy (OA %) and (b) Kappa Coefficient (KC) from four classifiers of the main and secondary trees. Table 4. Producer's Accuracy (Prod. Acc.) and User's Accuracy (User Acc.) resulted from implementing NN classifier for the main and secondary trees. NN Secondary Main Species Pinus brutia Quercus aegilops Quercus Infectoria Pistacia Juglans regia Prunus duclis Crataegus azarolus Junipirus oxycedrus Platanus orientalis Populus euphratica Salix Paliurus spina Christi Morus Ficus carica Elaeagnu angustifolia Prod. Acc.% User Acc.% 41.43 74.00 51.82 88.39 96.50 78.26 63.43 49.56 51.79 80.00 77.59 61.00 48.06 81.17 58.90 81.03 82.93 52.17 62.60 98.41 54.68 53.46 100.0 58.52 90.68 79.22 46.9 61.36 65.10 72.30 Furthermore, some other issues require further work. For instance, using a spectroradiometer device in the field survey to measure the reflectance value of the tree species may help to identify which WV2 band (bands combination) is best matching with each tree species. This may aid to avoid the problem of having two classes for one tree species, which in turns improves the result accuracy. Further, other types of classifier such as Support Vector Machine might be useful to use and to be investigated for such a study. The classification map of the main and secondary tree species that are resulted from NN classifier is shown in Figure 11 with some sample locations. Most of the tree species (Pinus brutia, Quercus aegilops, and Morus) in Mangish shown in Figure 11(b), (c), and (d) are classified correctly. We can notice that all other land types (as buildings and streets) were not taken in the classification process. This indicates that our procedure reported in Figure 3 of identifying the tree crown Proc. of SPIE Vol. 9239 92390G-10 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms was successful. Figure 11(b) shows only one species which is Quercus aegilops. Meanwhile, a mixed species (Quercus aegilops, Pinus brutia, Morus) appears in Figure 11(c) and (d). The occupied area of each tree species is determined and calculated. Table 5 shows the area in hectares of each tree species. (b(a) ) (c ) (d ) (a) Mangish area Pistada Junipiers oxycednrs K Pa /iurus spina K Pious bruta K Juglans regia Platanus orainfa /is K Quemas Aegi/ops K Prunus duc/is Popu /us euphratca 06 Quercus /nfectorla K Crataegus azaro/us K Sa/ix Morus Ficus cana Elaeagnu angusfih /ia Figure 11: (a) Classification map of all tree species within Mangish; (b) Quercus aegilops species within forest land; (c) Quercus aegilops, and Pinus brutia species within mixed land type; (d) Quercus aegilops, Pinus brutia, and Morus species within urban land type. Table 5: Area in hectares of the (main and secondary) tree species in the study area Tree species Main Trees: Pinus brutia Quercus aegilops Quercus Infectoria Pistacia Juglans regia Secondary Trees: Prunus duclis Crataegus azarolus Junipirus oxycedrus Platanus orientalis Populus euphratica Salix Paliurus spina Christi Morus Ficus carica Elaeagnu angustifolia Total Area (ha) 100.39 1378.83 454.17 731.85 27.13 13.40 5.90 0.19 4.49 188.45 89.24 0.03 99.17 40.30 0.01 Proc. of SPIE Vol. 9239 92390G-11 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms 5. CONCLUSION In this work 15 tree species were identified and mapped in Mangish, Duhok, Kurdistan Region-Iraq. This is achieved using VHR WV2 imagery and an advanced methodology that served the objective of this study. The study leads to the following conclusions: 1- Fifteen tree species were identified and mapped with a satisfactory accuracy within Mangish, Duhok, Kurdistan Region-Iraq. 2- The WV2 sensor has a great impact in improving the accuracy of vegetation classification and tree species recognition. 3- The NN classifier gave more realistic result among four other supervised classifiers. 4- Given the presented method and results in this study, this approach can be applied in classifying and mapping several and different kind of trees. REFERENCES [1] [2] [3] [4] [5] [6] [7] [8] [9] [10] [11] [12] [13] [14] [15] [16] [17] Wamelink, G. W. W., Wieggers, H. J. J., Reinds, G. J. et al., “Modelling impacts of changes in carbon dioxide concentration, climate and nitrogen deposition on carbon sequestration by European forests and forest soils,” Forest Ecology and Management, 258(8), 1794-1805 (2009). Grebner, D. L., Bettinger, P., and Siry, J. P., [Chapter 1 - A brief history of forestry and natural resource management] Academic Press, San Diego(2013). Yang, J., “Urban forestry in challenging environments,” Urban Forestry & Urban Greening, 11(2), 103-104 (2012). Tasoulas, E., Varras, G., Tsirogiannis, I. et al., “Development of a GIS application for urban forestry management planning,” Procedia Technology, 8(0), 70-80 (2013). Kim, M., M., M., and Warner, T. A., “Forest type mapping using object-specific texture measures from multispectral Ikonos imagery: Segmentation Quality and image classification issues,” Photogrammetric Engineering & Remote Sensing, 75(7), 819-829 (2009). Czerwinski, C. J., King, D. J., and Mitchell, S. W., “Mapping forest growth and decline in a temperate mixed forest using temporal trend analysis of Landsat imagery, 1987–2010,” Remote Sensing of Environment, 141(0), 188-200 (2014). Salehi, B., Zhang, Y., and Zhong, M., “A Combined Object- and Pixel-Based Image Analysis Framework for Urban Land Cover Classification of VHR Imagery,” Photogrammetric Engineering & Remote Sensing, 79(11), 999-1014 (2013). Arenas-Castro, S., Fernández-Haeger, J., and Jordano-Barbudo, D., “Evaluation and Comparison of QuickBird and ADS40-SH52 Multispectral Imagery for Mapping Iberian Wild Pear Trees (Pyrus bourgaeana, Decne) in a Mediterranean Mixed Forest,” Forests, 5(6), 1304-1330 (2014). Chen, G., Hay, G. J., Carvalho, L. M. T. et al., “Object-based change detection,” International Journal of Remote Sensing, 33(14), 4434-4457 (2012). Dingle Robertson, L., and King, D. J., “Comparison of pixel- and object-based classification in land cover change mapping,” International Journal of Remote Sensing, 32(6), 1505-1529 (2011). Pu, R., and Landry, S., “A comparative analysis of high spatial resolution IKONOS and WorldView-2 imagery for mapping urban tree species,” Remote Sensing of Environment, 124(0), 516-533 (2012). Chen, Q., [Comparison of Worldview-2 and IKONOS-2 imagery for identifying tree species in the habitat of an endangered bird species in Hawaii] DigitalGlobe Longmont, CO, USA, (2011). Immitzer, M., Atzberger, C., and Koukal, T., “Tree Species Classification with Random Forest Using Very High Spatial Resolution 8-Band WorldView-2 Satellite Data,” Remote Sensing, 4(9), 2661-2693 (2012). Al-doski, H. S., [A study of the quantity of interception and chemistry of gross rainfall, throughfall, stemflow and soil in natural and artificial stands of Zawita Pine] Uninversity of Duhok, Duhok(2012). ASTER-GDEM, [ASTER Global Digital Elevation Model (GDEM)], (2013). Congalton, R. G., “A comparison of sampling schemes used in generating error matrices for assessing the accuracy of maps generated from remotely sensed data,” Photogrammetric Engineering & Remote Sensing, 54, 593-600 (1988). DigitalGlobe, [The benefits of the 8 spectral bands of WorldView-2,], (2009). Proc. of SPIE Vol. 9239 92390G-12 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms [18] [19] [20] [21] [22] [23] [24] [25] [26] [27] EXELIS, [ENVI Classic Orthorectifying Aerial Photographs], (2013 a). USGS, [The Shuttle Radar Topography Mission], (2005). Kosaka, N., Akiyama, T., Bien, T. et al., "Forest type classification using data fusion of multispectral and panchromatic high-resolution satellite imageries." 4, 2980-2983. Fontana, F. M. A., Coops, N. C., Khlopenkov, K. V. et al., “Generation of a novel 1 km NDVI data set over Canada, the northern United States, and Greenland based on historical AVHRR data,” Remote Sensing of Environment, 121(0), 171-185 (2012). Lillesand, T. M., Kiefer, R. W., and Chipman, J. W., [Remote sensing and image interpretation] John Wiley & Sons, Hoboken, NJ(2008). Haralick, R. M., Shanmugam, K., and Dinstein, I. H., “Textural Features for Image Classification,” IEEE Transactions on Systems, Man and Cybernetics, SMC-3(6), 610-621 (1973). Hay, G. J., Niemann, K. O., and McLean, G. F., “An object-specific image texture analysis of H-resolution forest imagery,” Remote Sensing of Environment, 55(2), 108-122 (1996). Haralick, R. M., “Statistical and structural approaches to texture,” Proceedings of the IEEE, 67(5), 786-804 (1979). Salehi, B., Zhang, Y., Zhong, M. et al., “A review of the effectiveness of spatial information used in urban land cover classification of VHR imagery,” International Journal of Geoinformatics, 8(2), 16 (2012). Jensen, J. R., [Introductory digital image processing : a remote sensing perspective] Prentice Hall, Upper Saddle River, N.J.(2005). Proc. of SPIE Vol. 9239 92390G-13 Downloaded From: http://proceedings.spiedigitallibrary.org/ on 03/28/2015 Terms of Use: http://spiedl.org/terms