|
Landuse Change Analysis of Rutbeek Recreational Area, Netherlands Md. Rejaur Rahman GIS Specialist and Urban Planner Asian Disaster Preparedness Center (ADPC), Bangladesh E-mail: reja.83@gmail.com Published: November 2009 Abstract Landuse change is a common issue. Due to population growth and increase the productive activities made the changes. The processes of the landuse changes are slow, for that it is difficult to identify the changes. But Remote Sensing (RS) technology and spectral image made it easier. On this paper the landuse changes of the Rutbeek recreational area identified using RS data and finally ensure the accuracy also done the field check. For this study used Landsat ETM image of 25 May 2001, IKONOS image of 03 April 2000 (both pan and MS bands), ASTER VNIR Image of 12 September 2006 and Topo Map of the area. After the analysis found that the overall classification accuracy is 61.97 percent. Introduction The objective of the project is to do a supervised image classification of a Landsat ETM image of the Rutbeek recreational area, and to incorporation ground truth data collected during a fieldtrip for accuracy assessment. Collected parametric signatures using two different tools, digitized polygon and seed growing, based on topographic map and visual IKONOS interpretation. Five classes were defined, namely water, forest, grassland, bare/arable crop, and heather. Evaluate the training data using five different methods to check whether there is any mixed pixels that could confuse classification process. Perform maximum likelihood parametric classification using supervised and fuzzy classification methods. Collecting ground truth data from the fieldwork. Combine all the ground truth data and use this data for accuracy assessment. After the analysis found that the overall classification accuracy is 61.97 percent. The major causes of low accuracy rate were the different year data sources are used. (Map of 2001 versus field check in 2009), time or season of the base image and field check year was not same (Map was in May 2001 and field check in October) and landuse change due to the long time of the base map. Location of the Area The Rutbeek recreational area is located South of Enschede. The terrain is flat to almost flat, with sandy soils. In the middle of the area, an artificial lake is found. The lake is surrounded by “swamp areas”, “beaches”, parking lots and other recreational infra-structure. The agricultural land use is mainly maize, cereals and grassland. The Rutbeek is a recreational area of approximately 120. In 1975 the layout of Het Rutbeek as a recreational area began around an erratic former sand excavation lake. Now it is a transformed area of 120 hectares with trees, water parts, with beaches, laying meadows, grass land, and leisure facilities. The area is made up of one fen and five different beaches. On the South Western side of the lake there is a place for nudist recreation. The area is shown with boards. It is clearly signposted also. There is a foot path around the lake. The nudist beach of approximately 3 hectares is mainly made up of laying meadows, on the water side there is a small beach. There is a toilet block on the nudist beach, further there are no facilities. On the textile beach (after approx 1000 meters) there is a kiosk available. (Source: www.naaktstrandje.nl) Available Image 
Methodology  Figure 1: Methodology of the study Collection of Spectral Signature Samples (parametric) with the Signature Editor using two different methods: Two methods of collecting spectral signature one is user defined polygon using the polygon tool and other is using seed growing tool. In the table 1 showed the advantages of these two methods.   Figure 2: Spectral Signature Collected using two method ( Mean Plot) Tools used and tools preferred in evaluating data: A feature space plot allows you to determine the spectral location of surface features within your image. In some ways this is similar to the signature plots that you made earlier. However, it differs in that we examine the association between 2 bands in a scatter plot and locate within that scatter plot, the spectral location of various features (agric., trees, water, etc.). Examining the feature space is important because it allows us to make quantitative comparisons between cover types on the ground. This process is the exploratory phase of classification.  Figure 3: Feature space band 1 - 4, 3 - 4 with standard deviation = 2 (overlapped feature space between bare/arable crop, heather & forest) (A,B), standard deviation=1 (no overlapped) (C,D) Mapping a thematic layer into a feature space image can be useful for evaluating the validity of the parametric and nonparametric decision boundaries of a classification. Here we used band 1-4 and band 3-4 of pixel frequency positioning. Figure 3.A and 3.B shows the standard deviation =2 of pixel frequency and Figure 3.C and 3.D shows the standard deviation =2. The lower value of standard deviation will give less overlapped between frequency classes. The standard deviation will generate the range of ellipses to the mean of pixel frequency. By analyzing the ellipse graphs for all band pairs, we can determine which signatures and which bands provide accurate classification results. Evaluate Contingency Below the contingency matrix of pixel of training sample. This evaluation performs maximum likelihood which accuracy of using this method is 99.59%. The other rule has performed almost similar accuracy value.  Figure 4: Contingency matrix of each class (in pixels) Histograms A histogram frame part is an advanced frame part that is used in the contrast adjustment tools. It is a graphic that shows the histogram of a raster layer or other group of file values. It may also show a lookup table or other graph relative to the histogram. Some histogram frame parts are for viewing only, but others provide an interactive user interface that lets you manipulate the data. This document explains that user interface.  Figure 5: Histogram of the different band of the Landsat ETM Image Accuracy Assessment from Collected Field Data Accuracy assessment is a general term for comparing the classification to geographical data that are assumed to be true, in order to determine the accuracy of the classification process. Usually, the assumed-true data are derived from ground truth data. We performed 161 ground truth checking points on the area of study. Examples of Land use Classes in the field:   Results from the Accuracy Assessment: The accuracy of our classification after we validated with the ground truth checking is 61.49 %. This lower accuracy was due to not only lack of signature training which needed to repeated over and over but also the different interpretation of each class on the ground site between surveyors. Accuracy matrix of supervised classification from collected field data  Causes of low accuracy are
The Kappa coefficient expresses the proportionate reduction in error generated by a classification process compared with the error of a completely random classification. Our overall Kappa statistics of maximum likelihood supervised classification is 0.4385 implies that the classification process is avoiding 43.85 % of the errors that a completely random classification generates. How to interpret Kappa Kappa is always less than or equal to 1. A value of 1 implies perfect agreement and values less than 1 imply less than perfect agreement. In rare situations, Kappa can be negative. This is a sign that the two observers agreed less than would be expected just by chance. It is rare that we get perfect agreement. Different people have different interpretations as to what is a good level of agreement. At the bottom of this page is one interpretation, provided on page 404 of Altman DG. Practical Statistics for Medical Research. (1991) London England: Chapman and Hall. Here is one possible interpretation of Kappa.
 Discussion Ground Checking is an important part of quality assessment of the image classification. In this process image class is observed from field for different classes and later it is compared with the classified image to get the accuracy assessment. After performing the accuracy assessment is it observed that forest and water has high accuracy, Grass and heather has moderate accuracy and arable land has poor accuracy.
Reference
Page 1 of 1
|