This essay has been submitted by a student. This is not an example of the work written by our professional essay writers.
Worldview-2 image with composite 532 RGB bands Red, Green and Blue true /natural colors and Landsat-7 Enhanced Thematic Mapper Plus with six multispectral bands (1-5,7) were utilized in this study. The Worldview-2 image is the latest satellite in DigitalGlobe's collection and the latest improvement among sensors for the acquisition of remotely sensed imagery. It was launched on 8th October 2009 from Vandenberg Air Force Base, CA (USA) and is fully operational since 5th January 2010. The Worldview-2 image is the first very high resolution eight-bands multispectral commercial satellite providing a panchromatic sensor with spatial resolution of less than half meters (up to 0.46 ground sample distance-GSD-at nadir) and multispectral resolution in up to 1.84 meters (Hobi and Ginzler, 2012). The Worldview-2 image was obtained from the Ministry of Agriculture Republic of Indonesia in this study.
The Landsat-7 ETM+ is operated by a whiskbroom scanning multichannel radiometer was launched on the Landsat 7 satellite on 15 April 1999. It has one panchromatic band with spatial resolution of 15m, six multispectral VNIR and SWIR bands with a spatial resolution of 30m and spatial resolution of 60m for one thermal band. The ETM+ also comprises new features that make it a more useful and efficient instrument for global chance studies, land cover monitoring and assessment and large area mapping than its Landsat image predecessors (Pu et al., 2005). The Landsat-7 ETM+ image dated on June 9, 2011 was obtained from the Ministry of Forestry Republic of Indonesia and dated on May 21, 2010 was downloaded from the website of the USGS (U.S. Geological Survey) Earth Explorer ( http://earthexplorer.usgs.gov/)
Compare to other Landsat images, The Landsat-7 ETM+ has been designed with improved detector by detector calibration and improved electronics can reduce levels of banding and stripping. However, it still has spatial noise such as stripping can affect to sensor performing. Therefore, the Landsat image needs to perform gap-filling in order to clear up the strips. This was performed by "frame_and_fill_win32" software using The Landsat-7 ETM+ image dated on June 9, 2011 as the anchor (base) image and Landsat image dated on May 21, 2010 as a fill scene1.
4.1.2. Real-time kinematic differential GPS
Global Positioning System (GPS) is a satellite based positioning system developed by the Department of Defense of the United States constituted of a constellation of 24 satellites with an orbit around the world of 12 hours and their ground station. Real-time kinematic (RTK) is one measurement method of DGPS that utilizes two receiver GPS collecting data simultaneously and verifies positional information. In RTK, one receiver is usually placed at a known point as a base and the other receiver is moving along a trajectory site as a rover (Morales and Tsubouchi, 2007). Real-time kinematic (RTK) DGPS systems that create accuracies of centimeter-level of structure and vessel positioning are available (Kaplan and Hegarty, 2006). This method offers two types of solutions, float and fix. The RTK float solution needs at least 4 common satellites offering an accuracy range of approximately 20-100cm, while fixed solution requires at least five common satellites and offers an accuracy within 2cm (Morales and Tsubouchi, 2007).
In order to obtain the GPS data in this study, an RTK-DGPS survey was conducted on 15 September 2012, and data were collected from one station. Based on measurement procedure (Figure 2), one GPS receiver as a base was placed in one geodetic control point and the other receiver served as a rover collecting data along a trajectory (Morales and Tsubouchi, 2007).
The accuracy of the DGPS measurement depends on many factors, such as satellite-receiver geometry, weather condition, equipment error and biases. There are two common parameters used to indicate the accuracy of DGPS data, Dilution of Precision (DOP) and number of satellites in view (SVs). There are several kinds of DOP, including Geometry DOP, Horizontal DOP, Vertical DOP, Positional DOP and Time DOP, respectively. A lower PDOP value indicates more precise GPS data and a higher number of SVs implies more precise GPS data (Rizos, 1997) (Kaplan and Hegarty, 2006).
A set of DGPS Leica GPS1200 series using two ATX1230GG receiver type and two GX1230GG-type antennas was operated during the study (Figure...). This specification indicates a vertical accuracy of 20mm +1ppm and a horizontal accuracy of 10mm +1ppm. LEICA Geo Office version 7.0 software was used in the post-processing analysis.
4.2. Image Delineation/Interpretation and Classification
Remote sensing and GIS can have a role in the study of patterns and processes on the surface of the Earth and and to generate decision support systems. (Reddy and ebrary Inc., 2008) also noted several application areas of remote sensing and GIS such as for forestry and the environment. There are several approaches to collecting data using remote sensing technology and GIS such as a visual image interpretation, digital image classification technique and vegetation indices.
Geometric correction, image registration and atmospheric correction were performed as image pre-processing in this study. The Worldview-2 image was firstly geometrically corrected by rectifying to topographic map 2010 (scale 1:25,000) to the UTM coordinate system zone 48 South projection and to the WGS 84 ellipsoid. Identified features such as lake, road intersections, buildings, plantation and permanent landmarks on the image were selected as ground control points (GCPs). A total of 10 well distributed GCPs were located for the georeferencing through the first-order polynomial method. Finally, the Worldview-2 image resample to a 1.84 meter pixel size using the nearest neighborhood algorithm with the RMSE 2,02 meters (1.09 pixels).
Afterwards, the Landsat 7 ETM+ image 2010 was rectified to the geometrically corrected Worldview-2 image through image to image rectification to the same geo-reference, UTM coordinate system zone 48 South projection and to the WGS 84 ellipsoid. A total of 10 GCPs were distributed for the georeferencing through the first-order polynomial method. Then the rectified image was then resample to a 30 meter pixel size using the nearest neighborhood algorithm with the RMSE 11.5 meters (0.38 pixels).
In order to calculate vegetation indices, the Landsat 7 ETM+ images must be converted to reflectance, physical measurement, known as an atmospheric correction. This was performed through several steps. Firstly, converting digital number (DN) data to radiance data; secondly, converting radiance data to reflectance data; finally, enforcing positive reflectances.
4.2.1 Visual Image Interpretation/ Delineation
Visual Image interpretation is a process of identifying what we see on the images and communicate the information gained from these images to others for evaluating this consequence. The analysis of a data utilizing visual image interpretation involves the use of the basic picture elements, namely tone, texture, pattern, size and shape in order to detect and identify various objects. The visual interpretation has become more important to make a spatial database for GIS (Reddy and ebrary Inc., 2008).
The visual delineation was performed by digitizing on-screen using Arc GIS on both Worldview-2 image and Landsat 7 ETM+ in 1:5.000 scale map, meaning that 1 cm on the map corresponds to 25,000 cm on the field surface. Tone, color, texture and site/location etc. are used as basic elements parameters in visual interpretation. The methodology was used the monoscopic visual interpretation of the satellite imageries for identification of forests using standard visual interpretation techniques based on photo elements. In addition, the local knowledge about the terrain has contributed considerably in interpretation such as a lake, settlement, building, road and plantation. Other auxiliary data like topographical maps, forest maps and other available details are collected and made use of for identification and mapping of forest boundary (following Menon, 1999).
4.2.2. Automated Delineation/Classification
22.214.171.124. Maximum Likelihood Classification
MLC is the most common parametric classifier method of supervised classification which assumes normal or near normal spectral distribution for each feature of interest. This method based on the probability that a pixel belongs to a particular class and takes the variability of classes into account by using the covariance matrix (Jensen, 2005).
A training dataset was created by digitizing polygons for each land cover class. Training area sample over 1,800 pixels covering 6 land cover types (Forest, plantation, shrub, cloud, lake and settlement) with same spectral characteristic of each class were used for Worldview-2 image classification and an aggregate generalization was used to smooth the classified image for reducing a "salt and pepper" effect. The training area over 3,800 pixels covering 5 land cover types (Forest, plantation, shrub, lake and settlement) with same spectral characteristic of each class were also used for Landsat 7 ETM+ image classification in this study. The plantation, shrub, lake, cloud and settlement were categorized to non forest classification .
126.96.36.199. Object-Based Classification
Object-based image classification is a classification method which encompasses not only spectral information of surrounding pixels, but also incorporates information such as texture, shape, size, directionality, and spatial distribution of features (Blascke and Strobl, 2001; Kim et al., 2009). In this study, the object-based classification was divided into two main steps using SPRING 4.2 freeware: 1) image segmentation that create image objects (or segments); and 2) image classification through creation training sites and signature classes based on image segments.
The 532 composite bands of WorldView-2 and 453 spectral bands of Landsat 7 ETM+ images were used in the segmentation. A region growing algorithm with similarity set 40 and areas (pixel) 15 method was used to create image segmentation in this study both of WorldView-2 and Landsat 7 ETM+ images. The segment images were compared with visual interpretation in order to identify the optimal parameters for segment image corresponding to different datasets. After segment the image was created for each dataset, training sites (polygons) were selected which coincide to the sample site used in Maximum Likelihood Classification. Finally, 6 land cover types (Forest, plantation, shrub, cloud, lake and settlement) of Worldview-2 image and 5 land cover types (Forest, plantation, shrub, lake and settlement) of Landsat 7 ETM+ image were resulted in this study. The plantation, shrub, lake, cloud and settlement were categorized to non forest classification.
188.8.131.52. Vegetation Indices
Vegetation indices are gained by combinations of several spectral values of images that are mathematically recombined in such a way as to yield a single value indicating the quantity or energy of vegetation within a pixel (Campbell, 1996).
Normalized Difference Vegetation Index (NDVI), Soil-Adjusted Vegetation Index (SAVI), Optimized Soil-Adjusted Vegetation Index (OSAVI), Enhanced Vegetation Index (EVI), and Green Red Vegetation Index (GRVI) was proposed for forest delineation in this study through Map algebra of GIS and ERDAS ERmapper 2011 software.
NDVI was calculated as the following equation by (Rouse et al., 1974):
Where, NDVI = Normalized Difference Vegetation Index
ðœŒNIR = Near InfraRed band
ðœŒRed = Visible Red band
EVI was calculated as the following equation by(Huete et al., 1997):
Where, EVI = Enhanced Vegetation Index
ðœŒNIR = Near InfraRed band
ðœŒRed = Visible Red band
ðœŒBlue = Visible Blue band
SAVI was calculated as the following by (Huete, 1988):
4.3. Accuracy Assessment
This study consists of two main parts: delineation accuracy assessment (line) and classification accuracy assessment (area). Root Mean Square error was used as tool for accuracy assessment in forest boundary delineation due to its capacity to encompass both random and systematic errors introduced during data production (Nikolakopoulos et al., 2006). RMSE was used as a standard statistical tool for analyzing DEM accuracy by the the USGS and was also utilized in other studies (Hirt et al., 2010; Mouratidis et al., 2010; Miliaresis and Paraschou, 2011).
For RMSE calculation purposes, a Cartesian coordinate system was used to compare the distance of x (longitude) at the same y (latitude) position, as well as the distance of y at the same x locations, at 100 m intervals of two different forest delineations extracted from the various delineation methods and various satellite images (following pryde et. al., 2007). This procedure was performed in ArcGIS by overlaying and intersecting 100 m interval grid cells to each generated forest delineation.
RMSE was calculated as:
Where, RMSE = Root Mean Square Error
n = The number of verifying points
yi = The forecast values of the parameter
Å·i = The corresponding verifying values
The accuracy was also assessed using standard criteria of the classification accuracy including of producer's accuracy, user's accuracy, overall accuracy and the kappa coefficient which computed from an error matrix. Using an error matrix to represent the accuracy assessment has been recommended by many researchers (Congalton, 1991; Congalton and Green, 1999; Jensen, 2005; Ismail and Jusoff, 2008). The error matrix is a square composed of numbers set out in rows and columns which express the number of sample units (i.e., pixel, cluster of pixels, or polygon) defined in a particular category relative to the actual category as verified on the field (Congalton, 1991). In the error matrix, the numbers in the rows are the data from the classified remotely sensed data, while the numbers in the columns are the reference data in this study.
In this study, producer's accuracy was calculated by dividing the number of correctly classified pixel/area in a category by the total number of pixels/area of that category as obtained from the reference data (column total in the matrix). The statistic value of producer's accuracy refers to the probability of a reference pixel/area being correctly classified in the image . Producer's accuracy offers how well a certain area can be classified (Jensen, 2005).
The user's accuracy was calculated by dividing the number of correctly classified pixel/area in a category by the total number of pixels/area of that category on the classified image (row total in the matrix error). The result of user's accuracy represents the probability of a pixel/area classified on the image actually represent that category on the field and is a measure of commission error. The overall accuracy was defined as the total number of correctly classified pixels/areas (some of the major diagonal) divided by the total number of pixels/areas in the error matrix (Jensen, 2005).
Kappa coefficient was estimated from a Khat statistic which is calculated as:
Where, Khat = Kappa coefficient
N = The total area of sites in the matrix
r = The number of rows in the matrix
xii = The number of observations in row i and column i
x+i = The total area for row i
xi+ = The total area for column i
The values of Kappa coefficient can range from -1 to 1.Based on (Jensen, 2005), the possible ranges of the Kappa coefficient values can categorized into three groups: a value is more than 0.80 represents to strong agreement or good classification performance; a value between 0.40 and 0.80 represents to moderate agreement or moderate classification performance; and a value less than 0.40 represents to poor agreement or poor classification performance.
Using Multi Resolution Images
Overall and Kappa accuracy
Accuracy assessment on forest boundary area
Figure 2. Flowchart for the accuracy assessment of visual/manual and automated delineation/classification with multi satellite images
Figure 2. A set of DGPS Leica GPS1200 series including two ATX1230GG receiver type and two GX1230GG-type antennas operated during the studyhttp://t3.gstatic.com/images?q=tbn:ANd9GcRU_WOWIRP2hO6f7hIh49KvtqH2MU6W1fM1TyuCRqTYNozvEvzLKQ
Figure 2. A schematic concept of RTK-DGPS survey (a) where the base receiver GPS is placed in one point (b) while the another ones as a rover receiver GPS is moving to around study site in a trajectory by walking (c).
(b)IMG_0803.JPGIMG_0811.JPGstudy area 1.jpgIMG_0803.JPGIMG_0811.JPG
Campbell, J.B., 1996. Introduction to remote sensing. Guilford Press, New York.
Congalton, R.G., 1991. A Review of Assessing the Accuracy of Classifications of Remotely Sensed Data. Remote Sens Environ 37, 35-46.
Congalton, R.G., Green, K., 1999. Assessing the accuracy of remotely sensed data : principles and practices. Lewis Publications, Boca Raton, FL.
Hirt, C., Filmer, M.S., Featherstone, W.E., 2010. Comparison and validation of the recent freely available ASTER-GDEM ver1, SRTM ver4.1 and GEODATA DEM-9S ver3 digital elevation models over Australia. Aust J Earth Sci 57, 337-347.
Hobi, M.L., Ginzler, C., 2012. Accuracy assessment of digital surface models based on WorldView-2 and ADS80 stereo remote sensing data. Sensors (Basel) 12, 6347-6368.
Huete, A.R., 1988. A soil-adjusted vegetation index (SAVI). Remote Sens Environ 25, 295-309.
Huete, A.R., Liu, H.Q., Batchily, K., van Leeuwen, W., 1997. A comparison of vegetation indices over a global set of TM images for EOS-MODIS. Remote Sens Environ 59, 440-451.
Ismail, M.H., Jusoff, K., 2008. Satellite Data Classification Accuracy Assessment Based from Reference Dataset. International Journal of Computer and Information Science and Engineering 2, 386-392.
Jensen, J.R., 2005. Introductory digital image processing : a remote sensing perspective. Prentice Hall, Upper Saddle River, N.J.
Kaplan, E.D., Hegarty, C., 2006. Understanding GPS : principles and applications. Artech House, Boston.
Miliaresis, G.C., Paraschou, C.V.E., 2011. An evaluation of the accuracy of the ASTER GDEM and the role of stack number: a case study of Nisiros Island, Greece. Remote Sensing Letters 2, 127-135.
Morales, Y., Tsubouchi, T., 2007. DGPS, RTK-GPS and StarFire DGPS Performance Under Tree Shading Environments. Integration Technology, 2007. ICIT '07. IEEE International Conference on, pp. 519-524.
Mouratidis, A., Briole, P., Katsambalos, K., 2010. SRTM 3'' DEM (versions 1, 2, 3, 4) validation by means of extensive kinematic GPS measurements: a case study from North Greece. Int. J. Remote Sens. 31, 6205-6222.
Nikolakopoulos, K.G., Kamaratakis, E.K., Chrysoulakis, N., 2006. SRTM vs ASTER elevation products. Comparison for two regions in Crete, Greece. Int J Remote Sens 27, 4819-4838.
Pu, R., Yu, Q., Gong, P., Biging, G.S., 2005. EO-1 Hyperion, ALI and Landsat 7 ETM+ data comparison for estimating forest crown closure and leaf area index. Int J Remote Sens 26, 457-474.
Reddy, M.A., ebrary Inc., 2008. Textbook of remote sensing and geographical information systems. BS Publications, Hyderabad [India].
Rizos, C., 1997. Principles and practice of GPS surveying / Chris Rizos. School of Geomatic Engineering, University of New South Wales, Kensington, N.S.W. :.
Rouse, J.W., Jr., Haas, R.H., Deering, D.W., Schell, J.A., Harlan, J.C., 1974. Monitoring the Vernal Advancement and Retrogradation (Green Wave Effect) of Natural Vegetation. Final Report. A & M University, Texas. Greenbelt.