Assessment of ERS-l SAR Data
for Rice Crop Mapping and Monitoring Supan Karnchanasutham, Dr
.Apichart Pongsihadldchai Abstract Office of Agricultural Economics, chatuchak, Bangkok 10900, Thailand Chockchai Rodprom National Research Council of Thailand, chatuchak, Bangkok 10900, Thailand The objective of the study is to evaluate the capabilities of BRS-l SAR data for monitoring of rice planting acreage and its growth. The study area was about lOO Km in Tha Muang district, Kanchanaburi province. The ERS-l PRI. acquired during 35 day repeat orbit were used in the study. These data were acquired on 20 August 1993, 29 October 1993 and 3 December 1993 respectively. The study area was divided into 10 test sites and each of which was randomly selected 6 sample areas for extensive measurements of rice plant parameters height, density, net weight, yield etc.) as wen as auxiliary infomtation (wind, rain, and growth stage etc.). A Global Positioning System (GPS) was used to locate and register test site boundaries and calculate the area of each test site. The BRS-l data were geometrically corrected to topographic map at scale 1 :50,000. The field boundaries of all of the surveyed test site were digitised and superimposed on the geometrically corrected data which is filtered by MAP filter method. The result of the study revealed that it was very difficult to identify the rice crop if only single data was used. The rice planted area classified from date composite was found to be 20,253 hectares or about 27% of the total study area. The rice mapping accuracy was 78% while the overall accuracy was 79%. The very bright of backscattering coefficient (dB) was observed in August. This was due to the fact that the muddy broadcast rice was adopted in this area. The water level in the field for this type of planting method was very limited. The closer to the harvesting date of the rice crop is, the better classification of planted area would be obtained. Introduction Rice is the most important crop in Thailand in terms of acreage, number of, farmers and export earning. The ability to monitor and forecast its production is therefore vary crucial for short term policy determination. One major problem in utilising satellite data for crop area estimation is due to the cloud covers of the areas on the imagery because the needed data is in the rainy season. The capability of ERS-1 satellite that can see through cloud or all-weather conditions will make the estimation of crop area possible. The combination of SAR and other satellite data such as TM and SPOT will be very useful for crop classification in the area where cloud is always present all year round. 2.0bjectives To evaluate the capabilities of ERS-l SAR data for monitoring of rice planting acreage and its growth in Kanchanabw-i province. 3.Equipment and data acquisition
An 10X10 square kilometers area located in Amphoe Tha Muang Kanchanaburi province, Thailand between latitude 99o 37 44.1 E -99o 43 17.22 E and :'~, longttude13o 50 58N -13o 56 5 N was as the study area. This area IS part of the Mae Klong irrigation Project of the Mae Klong river basin which is one of the largest river basins in Thailand. 5.Methodology 5.1 Site selection and ground data collection the study area was divided in to 10 sites and for each site 6 samples were randomly selected . thus , in all three are 60 samples which were used for ground data parameters collection . for each time of satellite overpass (20 Agust 1993 , 29 October 1993 and 3 December 1993 ) the same data parameters were collected as follows ;
5.2 Image rectification Rectification is the process of projecting the data onto the plane making it conform to a map projection system. Assigning map coordinates to the image data is georeferencing. Since all map projection systems are associated with map coordinates. Five Ground Control Points (GCP) are specific pixels in the image data for which the output map coordinates are known. GCP consists of two x,y pairs of coordinates. source coordinates and reference coordinates. The resampling method used in this study is the nearest neighbor method. 5.3 Speckle filtering The Map filter method was applied in all 3 different dates on SAR data. 5.4 Ctilibration of backscatter coefficient , The calibration of backscatter coefficient (so) in this study used the following equations below. 5.5 Image processing Supervise classification with Maximum Likelihood method was performed on 3 different dates eRS-l SAR data and classified into 7 classes. .namely: rice sugarcanee,bushed, water, city, mountain 1 (bright) and inountain2 (dark). 5.6 Accuracy assessment An error matrix which compares the classified data with the reference data ( aerial photography) was computed. It contains parameters such as error of omission, error of commision , map accuracies and overall (global) accuracy. 6. Results and discussion 6.1 Site acrage and general information The minimum and maximum site acreage in the study area measured by using GPS were 6.57 hectares and 11.43 hectares respectively. The rice varieties planted in this area were SP60 (Suphan 60) and RD23 (Rice Department No.23). This rice crop was planted in August and harvested in December 6.2 Image rectifcation Five ground control points (GCP) were used for geometric collection by map (1:50,000) to image (ERS-l SAR images). The resampling method used is the nearest neighbor and the pixel size is 3Ox30 M2 6.3 Speckle filtering Map filtering tehnique was performed on 3 ERS-l SAR data. 6.4 Backscattering coefficient (dB) Due to Earth View (EV) software which was originally planned to be used for this project can not operate on 16 bits and thus can not draw field boundaries and overlay on the image. To solve this problem the ERS-l SAR data were converted to 8 bits and the IDRISI software was used instead to find Backscattering coefficient using equation in item 5.4. The average of rice backscattering coefficients of 20 August, 29 October and 8 December 1993 a; were -7.4, -9.8 and -8.5 respectively. 6.5 Image classification Supervise classification technique was performed on 3 different ERS-l SAR data: 20 Aug. 1993,29 Oct. 1993, and 3 Dec.1993 and divided into 7 classes as already mentioned. Since it is very difficult to classify the rice crop if only single date data is used. Therefore, these 3 dates data were combined and used as a basis for classification. 6.6 Accuracy assessment It was found that the rice mapping accuracy was 78% while the over all accuracy was79%. 7.Conclusion
The recommendation for future work are as follows:
|