| Author | Paudyal, Dipak Ram |
| Call Number | AIT Diss. no. NR-94-01 |
| Subject(s) | Agriculture--Tropics--Remote sensing
|
| Note | A disse11ation submitted in pa11ial fulfillment of the requirements for the degree of Doctor of Engineering. |
| Publisher | Asian Institute of Technology |
| Series Statement | Dissertation ; no. NR-94-01 |
| Abstract | The vast amount of spacebome Synthetic Ape1ture Radar (SAR) data now available demand appropriate methods of image interpretation and analysis. This study attempts to understand the backscattering characteristics of agricultural crops, mainly rice and sugarcane, in the province ofKanchanabmi in West Thailand situated in the tropics of SE Asia. Other prominent cover types include water, urban areas, bushes and shmbs. Use is made of both C- and L- band spacebome SAR data. This study fiuther attempts to evaluate and refine existing methods to digitally analyze SAR images, with a view towards application in agiiculture for crop acreage inventory and crop monito1ing pm-poses. The C-band ERS-1 data used in this study are processed at the National Remote Sensing Agency (NRSA) at Hyderabad, India and Deutsche Forschungsanstalt fiir Luft-Und Raumhalut (DLR), in Ge1many. The L-band JERS-1 SAR data is processed at National Space Development and Administration (NASDA), Japan. A full understanding of the nature of SAR data is required before attempting to use the data for agiicultural pm-poses. The general the01y of SAR imaging and image statistics, the effect of speckle and presence of texture pa1ticularly related to the SAR data used for the study is first described. At the beginning the feasibility of using single date SAR images for land-cover disc1imination is investigated. It is found that single date SAR images both in the C-band and L-band are inadequate to completely disc1iminate lice and sugarcane from each other and from other land-cover categ01ies. Water is discriminated from the rest of the categ01ies in late November and the mban areas in the Febma1y imaging date. The use of multi.temporal SAR images showed the greatest promise in cover type disc1imination. The potential of multitemporal SAR data was investigated with study of the temporal backscatter profile for each land-cover type. It was found that of all the landcover types only water, urban areas and paddy could be readily disc1iminated based on the temporal backscatter profile. In addition paddy provided a unique temporal signature cmve. Crop classification possibilities of the multitemporal data in both JERS-1/ERS-1 and ERS-1/ERS-1 configurations showed that two images acquired in early October and late November are sufficient to disc1iminate paddy from other categoiies. Sugarcane can be disc1iminated from other land-cover types, with the exception of shmbs, because of similar backscatter responses. It is found that an additional diy season date is required for the best discdmination of sugarcane and shmb. A systematic evaluation of SAR speckle filters was canied out based on several qualitative and quantitative analyses. Land-cover separability before and after filte1ing was used as a major criteiion for the filter evaluations. It was found that the speckle specific filters were supe1ior to the non-speckle specific filters. Some refinements in the existing algodthms are proposed. The refined MAP filter was found to be best for visual analysis. The modified Lee Filter was found to be the best for fiuther digital analysis such as for classification. Both first and second order images statistics were used to investigate the presence of texture in SAR data. Discriminat01y potential in the ERS-1 PRI data based on first order statistics was negligible. The NRSA processed ERS-1 data and JERS-1 SAR showed evidence of texture but this is attributed to the quoted number of looks being higher than estimated from measurements on the actual Image data. Second order texture statistics based on GLC matrices provided better disc1iminato1y potential. It was found that Contrast, Entropy and Ine1tia features are positively conelated whereas Angular Second Moment, Inverse Difference Moment and Homogeneity exhibited an inverse relationship with the 01iginal image tone. Land-cover separability pa1ticularly for water and urban areas in some GLC derived texture features was found to be superior than in the 01iginal tone images. It was found that use of speckle filters and texture features considerably increases the accuracy in the classification of agiicultural crops. Maximum likelihood classification based on tone alone was outpe1fo1med by combined tone and texture based classification. Classification using filtered images shows improvement over unfiltered ones. Best results were obtained using combined filtered and texture images. Finally, a knowledge based classification of multitemporal SAR images for land-cover application in general and agricultural application in pa1ticular, by adapting methodologies more commonly applied to digital segmentation is presented. The knowledge requirements are obtained from the temporal signature va1iations of the various land-cover types present in the area, with signature cwves being based on either radiometry or the texture of the objects concerned. Use of textural information is made only in those cases when the use of object radiometiy proves to be inadequate. Results showed the classification accuracy attained with this method is similar to that based on combined tone and texture. This must be regarded as satisfacto1y as this method requires minimum input from the user and modifications to the algorithm could be made to suit different environmental and cultural settings. It was found that acreage invent01y of rice crops is possible as early as late November, which is about one month before the actual dee ha1vest. However, acreage estimation of sugarcane needs at least an additional dty season image besides the wet season acquisition. In summa1y, it is found that existing images analysis techniques are generally sufficient for digital analysis and interpretation of SAR data for large scale applications such as in agdculture, provided a good knowledge and understanding of the backscatte1ing behavior of different crop/land-cover types for different pe1iods of the year is available. |
| Year | 1994 |
| Corresponding Series Added Entry | Asian Institute of Technology. Dissertation ; no. NR-94-1 |
| Type | Dissertation |
| School | School of Environment, Resources, and Development (SERD) |
| Department | Department of Development and Sustainability (DDS) |
| Academic Program/FoS | Natural Resources Management (NRM) |
| Chairperson(s) | Apisit Eiumnoh, |
| Examination Committee(s) | Aschbacher, J.;Nualchawee K.;Upasena, H. S.;Loof, R.;Schumann, R.;Quegan, Shaun; |
| Scholarship Donor(s) | Government of Australia; |
| Degree | Thesis (Ph.D.) - Asian Institute of Technology, 1994 |