Application and analysis on
forest types and land-used classification using digital image
processing Manasanan
Kantontong Geographician (Cartographer) Office of Remote Sensing Survey and Mapping Royal Forest Department Bangkok, Thailand Abstract A study on the classification of forest types, natural vegetation distribution, land-used patterns and other artificial activities made be man was conducted at Lamphun province in northern Thailand. The purpose of the study is to make the forest land cover map, by using computer compatible tape containing Landsat-5TM data. The study areas were classified by digital image processing using Compaq Despro 386/20 and later classified by program EASI/PACE veriosn 4.2. Out of 7 bands, only 3, i.e., : band 2, band 3 and band 4 were obtained form Landsat – 5TM and interpreted data. The method applied in this study is based on the supervised and unsupervised system of Maximum Likelihood Agorithm. The result indicated that the supervised system offers a favorable result when combining with the unsupervised system. The final results clearly separates the Landsat – 5TM data into 10 categories, as follows: water, paddy field, croplands, orchard, dry dipterocarp forest, mixed deciduous forest land, evergreen forest, rock pan tree, and other types of areas. The study showed that using supervised together with unsupervised system provided better result and a possible application to other relevant uses of forest land classification. 1. Introduction Landsat Imagery In Forestry In 1973, after the Royal Forest Department (RFD) has already been a member of the National Research Council, Thailand, the data from satellite imageries, indirect visual interpretation for forest type map, existing forest map, and watershed map, etc., were brought in for forestry resources survey. Since 1983, scientific and survey technique concerning natural resources was developed. RFD has brought in computerized technique “Compaq Despro 386/20” with software EASI/PACE version 4.2 for the use of interpretation and analysis, for example, Landsat MSS, Landsat TM, spot and MOS-1. The data from satellite imageries indicate not only the changes in the forest recourses but also provide the data for solving problems and planning forest resources in the future. 2. Objectives
3.1 Area Selected Lamphun, a northern province of Thailand, is situated between the latitude 17’ 25’ – 48’45’ and the longitude 98’ 40]. It includes approximately 4,505 square kilometers. Lamphun’s boundary is far from the sea; the characteristics of the climate is the ever-changing monsoon climate and is very hot in the hot season. her climate is mainly influenced by the south-west and north-east monsoons. There are, in addition, typhoon, cyclones and depressions. Due to the regular monsoon through the year, the province has three alternating seasons, namely, the rainy season, the cold season and the dry season. 3.2 Statistical Data Rainfall and average temperature from the past 10 years ( 1980-1990 ) of Lamphun province were given from the Meteorological Department, Ministry of Prime Minister and introduced to match with the Gaussen’s theory. The formula of the dry period p<2T were brought to analyze the combination between rainfall and temperature and he distribution of rainfall per year in Lamphun province. p<2T p= the amount of the annual rainfall in average T= the mean temperature of the coldest month Owing to this formula, it is found that Lamphun’s average volume of rainfalls in 1980-1990 is 980.7 mm. and the average temperature is 26.1 c. The most rainfall, 190.2 mm, is in September. December is the coldest month of which temperature is averagely 20.8 c.; whereas the hottest temperature is 30’c. in April. The volume of rainfall and the temperature affect greatly on natural vegetation in Lamphun province. Accordingly, water performs a process of weathering, erosion and development of vegetation, and temperature effects generally seed budding, plant growing and spreading of plant roots. 4. Sources of Data
6.1 Hardware and Software Hardware
The package software “EAS/PACE Version 4.2”, produced by the PCI company, Canada of 25 main programs. 6.2 Methodology 6.2.1 Ground Truth Data Collection Stage Before analysis and classification, the field data of Lamphun province were gathered by using the method of Random Sampling, limitation of the areas of spot check on, and the FCC image taken from satellite Landat – 5TM. The ground truth information in each rest site area in the field was corrected and observed, so as to be able to define the training area of several categories for supervise classification system of MAXIMUM. LIKEHIHOOD AGORITHM. The forest land cover, other agricultural crops and artificial activities made by man collected were: water, paddy field, cropland, mixed orchard, dry dipterocarp forest, deteriorated forest land and fallow, evergreen forest, and rock pan treediperocarp forest, deteriorated forest land and fallow, evergreen forest, and rock pan tree. 6.2.2 Computer Analysis and Classification Analyzing and interpretating of Landsat – 5TM digital data (CCT, tape) were conducted by COMPAQ DESPRO 386/20 and software EASI/PACE version 4.2. Step 1: To load CCT. Tape: Data of Lamphun province were taken from path 131 row 47, and path 131 row 48, obtained by Landsat-5TM on 3 April 1990, and were loaded into hard disk. It is because he data in the CCT. Tape of Landsat-5TM in one scene do no cover the whole area of Lamphun province, therefore, the area is split into 2 scenes ( path 131 row 47 and path 131 row 48). Both scenes are recorded in separate CCT. tape, which will be read out using 3 bands (B2 : 0.52-0.60 Un, B3 : 0.63-0.69 Un, B4 : 0.76-0.90 Un ) of the spectral reflectance value by putting in the tape drive. The Lamphun parts are read by which is selected by the EASI/PACE software and kept in the hard disk. The selected 3 bands are in spectral range that are appropriate for analyzing the forest data and other agricultural crops, etc. Step 2: To create the FCC image on the screen : The 3 bands of spectra are overlaid to create the row data image ( or false color composite image ) of Lamphun province that can be seen in many colors on the color monitor. Step 3: Enhancement for details: In order to extend the pixel values for designed band 2, 3 and 4, the enhancement procedure must be performed. As the original histogram with narrow contrast will depend on the number of original pixel that may be adjusted by the linear contrast streatch values from 0 to 255. The main idea of the enhancement is to investigate details of cleariness of data appeared on the screen. However the enhancement, row data corrected from the training area will be used, instead. Step 5: To merge data into Mosaic Pattern : The 2 scenes data ( path 131 row 47 and path 131 row 48 ) are retrieved from the data base and merged together into the mosaic patter. Step 5: Geometric correction: Changeable satellite orbits among various different time and unsteady sensor records produce the unreliable positions of the specified object on earth’s surface. The reliability can be made via GCP (Ground Control Point) which is relatred to the geographical referring points and compatible with those employed in UTM (Universal Traverse Mercrator). Step 6: Calculation of statiscal values: After geometric correction, the image has been corrected for the right position. The areas are done by dividing for training areas according to all spot checks in the field : water, paddy field, cropland, mixed orchard, dry dipteocarp forest, deteriorated forest land and fallow, evergreen forest and rock pan tree. Then the microcomputer will calculate only the statistical value in the training area, such as, Mean Vector, Standard Deviation, Covariance Matrix, of each category. Step 7 : Maximum likelihood agorithm classification: All pixels in the image were classified by the supervised classification system of the Maximum Likelihood Agorithm that the Mean Vector and Covariance were determined by the hypothesis that the data distribution is normal. By comparing each pixel value with the training area, the probability of the whole province was calculated. The Maximum Likelihood classification method gives an accurate value but consumes timing. SUPERVISED CLASSIFICATION SYSTEM: The data analysis was done from the ground survey in Lamphun province which was called “Training area”. According to the training area, It is a representative of characteristics as shown in satellite imagery. The specification of the training area can be flexible to data analysis technique depending on the calculated statistical values. The training area specification is a model or representative of the characteristic data, such as : Mean, Standard Deviation, and Covarience Matrix values which were specified in order to identify the total area by using that statistical values. The result after classification by supervised system yields 9 classes and unclassified ( unknown ) pixels. Step 8 : To assign colors for the 9 classes. Step 9: To smooth the data by low pass filter method. Step 10 : To cut boundary Map of Lamphun Province. Step 11: To select only unclassified pixels : After classification by the supervised system, there are still unclassified pixels. It would be due to the insufficient ground truth data sets of the given 9 categories which had been set up. The information given in the training area categories does not represent enough statistical calculation ( Mean Vector, Standard Deviation, and Coverience Matrix ) by the computer. The statistical calculation applied is based on the basis of Maximum Likelihood method by the which the probability density function and the degree of accuracy of the pixels in each training set are used. Step 12: To classify unclassified pixels by unsupervised classification: The computer system group all these pixels into 7 group. Each group has a new statistical value. Step 13: To have field check for 7 groups: The data in 7 groups were to field check and compared together with statistical value of the supervised system. Step 15: To append partly the 7 classes to the same position of the 9 previous classes and unclassified pixels. Step 15: To reglassify using the supervised system: After that process, the supervised classification was applied again and those new groups that can no be grouped into the 9 classes will them put into their own new class named “others”. Step 16: To reassign colors for the 10 classes. Step 17: To produce a thematic ( digital ) map of lamphun province. Step 18: To check map accuracy in the field in order to correct the details. Step 19: To calculate each area of 10 classes. Step 20: To reproduce a new varified thematic map. Result
Conclusion Using CCT. tape containing LANDSAT-5TM data, “COMPAQ DESPRO” 386/20, a computerized technique with software EASI/PACE version 4.2 and by the Supervised and Unsupervised Classification System in order to make a forest land cover map of Lamphun province, it is found that the study represents and effective and accurate result. The data from satellite imageries indicate both changes in local forestry and data of planning and solving problems in forest resources in the future. Recommendation To study forest resources and changes using satellite imageries combining with automation technique of other scientific technology is recommended to be further performed in other provinces because it renders effective consequences. Informations obtained from the studies conducted in several provinces will represent both the actual and future forest resources for the government awareness on forestry management and planning in the long run. References
|