Field Data Collection Land Use Classification

25 Figure 5. The flow chart of field data collection

3.4.2 Land Use Classification

Time-series data of LANDSAT images 2002, 2005, and 2008 were used in this research in order to generate the information about time-series data of land use in Siak District. The land use categories were generated by image classification process that was done in each LANDSAT image. In order to have a good classification result, there were some processes done which can be divided into two processes: image pre-processing and image processing. Image pre-processing consists of processes to prepare image data for subsequent analysis that attempts to correct or compensate for systematic errors. LANDSAT images delivered to the user may consist of some errors which may be caused by the atmospheric or sensor condition when capturing data from the earth surface. The Landsat 7 scan-line corrector SLC, a mechanism designed to correct the undersampling of the primary scan mirror, failed on May 31, 2003. With the SLC now permanently turned off SLC-OFF, the ETM+ is losing approximately 22 of the data due to the increased scan gap Scaramuzza, et al. 2004. Thus, in this research LANDSAT images might need to be subjected to several corrections, such as SLC-OFF Gap Filling and geometric correction. The SLC-OFF Gap Filling process was done to each LANDSAT 7 ETM+ SLC-OFF 26 image by using Frame and Fill tool which has been developed by Richard Irish in affiliation with NASA Goddard Space Flight Center. The corrections made were different for each image considering the condition of the images. LANDSAT 7 ETM+ Image 2005 SLC-OFF Gap Filling Supervised Classification Accuracy Assesment Yes No Field Check Data Base Map of Riau Province from Ministry of Forestry - Image Registration - Rectification K Nearest Neigbors Image Pre-processing Image Processing Segmentation- based Image Processing LANDSAT 7 ETM+ Image 2002 LANDSAT 7 ETM+ Image 2008 Land Use Map 2005 Land Use Map 2002 Land Use Map 2008 3 4 1 2 Land Use Classification Geometric Correction Define Training Data 1 2 4 3 Field Data Collection Land Use Classification Land Use Change Detection Land Use Change Modeling Research Activities: Figure 6. The flow chart of land use classification process Image processing done in this activity was image classification of LANDSAT images which has been conducted by using supervised classification with K-Nearest Neighbors method. K-nearest neighbor is a supervised learning algorithm which works based on minimum distance from the query instance to the training samples to determine the K-nearest neighbors, where the result of new instance query is classified based on majority of K-nearest neighbor category Teknomo 2006. The minimum distance for continuous variables can be calculated by using Euclidean distance which examines the root of square 27 differences between coordinates of a pair of objects. The training data for supervised classification were derived by using image segmentation technique that segments an image into regions of pixels. Image segmentation is the process of partitioning an image into segments by grouping neighboring pixels with similar feature values. Land use data from previous years and GPS data that were taken from field data collection activity functioned as guidance in order to define the training data for supervised classification. The types of land use in this research refer to the criteria of Top-level Land Use Categories defined by IPCC 2003 which have been described in Table 1. The classified images produced from the supervised classification may contain classification errors which the land use identified in particular area may not represent the real condition land use on the images. This situation may be happen due to the quality of LANDSAT images used. The post-classification should be applied in order to correct its classification errors. The post- classification procedures which may be done are either by conducting the Majority technique, or the manual correction to each error. The manual correction would be done by delineating the errors and re-classify it into the correct land use. These procedures would also be applied to re-classify the cloud and shadow. At the end of this activity, accuracy assessment has been done in order to evaluate the quality of land use maps which have been produced. The accuracy assessment was done by creating the error matrix, accuracy report and the Kappa Analysis. The error matrix simply compares the reference points to the classified points in a c × c matrix, where c is the number of classes including class 0. The accuracy report calculates statistics of the percentages of accuracy, based upon the results of the error matrix Leica Geosystem 2005. There are three types of accuracy which were produced in the accuracy report: Producer’s Accuracy, User’s Accuracy and Overall Accuracy. The Kappa analysis is a discrete multivariate technique used in accuracy assessment to statistically determine if one error matrix is significantly different from another Bishop et al. 1975 in Congalton and Green 2009. This measure of agreement or accuracy is based on the difference between the actual agreement in the error matrix and the chance agreement that is indicated by the row and column totals. The Kappa analysis has 28 become a standard component of most every accuracy assessment Congalton and Green 2009. By considering the actual condition of LANDSAT images used to generate the information of land use categories of Siak District, the facts are as following: 1. LANDSAT images that were used are SLC-OFF Gap Filled images which were patched by some LANDSAT images which have different characteristics according to the time acquisition and atmospheric conditions, 2. Siak District covered by three different scenes might also have different characteristics, and 3. Siak District lays on equator zone and it is known that this area is often covered by cloud and shadow. The accuracy assessment for this research was set to meet the following requirements: Table 4. Accuracy Assessment Requirement for this research Producer’s Accuracy PA PA 70, for each land use category User’s Accuracy UA UA 70, for each land use category Overall Accuracy OA OA 80 Kappa Coefficient K K 0.8 The producer’s accuracy was produced by dividing the number of correct pixels for a class by the actual number of reference pixels for that class, while the user’s accuracy was produced by dividing the number of correct pixels for a class by the total pixels assigned to that class. The overall accuracy was produced by dividing the number of correct pixels for all classes by the total number of sample pixels for all classes. The Kappa coefficient expresses the proportionate reduction in error generated by a classification process compared with the error of a completely random classification Leica Geosystem 2005, and it was calculated based upon the result of error matrix and the accuracy report. 29 Illustration of Error Matrix Mathematical Expression for Accuracy Assessment Figure 7. Illustration of Error Matrix and Mathematical Expression for Accuracy Assessment

3.4.3 Land Use Change Detection

Land use map 2002 of Siak District was compared with land use map 2005 and also land use map 2005 with land use map 2008 in order to detect land use transitions during 2002 – 2005 and 2005 - 2008. Two products have been resulted from this land use change detection: land use change maps, and transition and probability matrices of land use change 2002 – 2005 and 2005 - 2008. Land use change map shows the transitions of land uses spatially, whereas the matrix shows the aggregation of land use transitions and the probability of land uses transforming to other land uses. In the further activities, the information of land use transitions 2002 – 2005 would be used as dependent variable in MLR model analysis, and the land use change map 2005 – 2008 would be used as reference map in model validation. The flow chart of land use change detection can be seen in Figure 8. 30 F C G W S O F FF FC FG FW FS FO C CF CC CG CW CS CO G GF GC GG GW GS GO W WF WC WG WW WS WO S SF SC SG SW SS SO O OF OC OG OW OS OO 2005 Land Use 2 2 FF FC FG FW FS FO CF CC CG CW CS CO GF GC GG GW GS GO WF WC WG WW WS WO SF SC SG SW SS SO OF OC OG OW OS OO Figure 8. The flow chart of land use change detection

3.4.4 Land Use Change Modeling

Land use change modeling in this research used statistical Multinomial Logistic Regression MLR model in order to determine the significant variables that might drive the land use change and also develop the final model of land use change in Siak District. The research assumed that the change of land use during 2002 – 2005 was driven by some relevant factors which existed in 2002, and the land use change during 2005 – 2008 was driven by the same factors which existed in 2005. The MLR model of land use change 2002 – 2005, that has been produced, was used to project land use transitions 2005 – 2008 by inputting significant independent variables that existed in 2005. After that the projected land use transitions 2005 – 2008 would be validated to the actual land use change map 2005 – 2008 that has been produced in the land use change detection activity.

a. Data Preparation

The procedure of data layer preparation is the most fundamental process which is time-consuming trial and error process, and it was done in order to fit the