Rice Crop Monitoring in Zhaoqing, China using RADARSAT SAR - Initial Results
Gordon C. Staples and Jeff Hurley
RADARSAT International
Richmond, British Columbia
Canada
Email : gstaples@rsi.ca
Abstract RADARSAT Standard mode imagery of Zhaoqing, China was acquired on march 5 and June 10, 1996. The acquisitions were during the Early-rice regional cropping calendar, with the first acquisition during the sowing stage, and the second acquisition during the heading stage. Rice and banana crops, aquaculture ponds, and urban area land-use classes were identified. Aquaculture ponds used for the production of fish and ducks had no significant backscatter change between acquisitions, but ponds used for growing gorgon euryale showed a significant backscatter increase. Rice paddies were also identified. The backscatter change for the rice paddies was variable, and was attributed to inhomogeneous growing conditions and variability of the vegetative phase. The growth stage between tillering and heading was identified as a key period for image acquisition if the rice-yield estimates are required.
1.0 Introduction
RADARSAT, which was lunched in November 1995, is equipped with a C-band HH-polarized synthetic aperture radar (SAR). One of the potential applications for RADARSAT imagery is for operational rice monitoring. With RADARSAT, a number of key requirements to bring this application to an operational level are readily attainable. These requirements include. (1) SAR backscatter, which is sensitive to plant type and growth (Brown et al., 1993), and has been shown to responsive to rice growth and development (Le Toan et al., 1989 & 1995; Kurosu et al., 1995; Aschbacher and Paudyal, 1993); (2) SAR is relatively insensitive to atmospheric effects, so image acquisition is weather independent; and (3) RADARSATs variable incidence angel and beam modes dramatically increase the re-visit frequency, and provide variable resolutions (RSI, 1995; Raney et al., 1993).
The objectives of this study were to assess the ability of RADARSAT to discriminate land-use classes, and to better understand the backscatter response as a function of the rice crop growth. The backscatter as a function of time is required so future, multitemporal acquisitions, can be planned to coincide with rice-growth stages based on knowledge of the local cropping calendar. The Zhaoqing test site was selected because it was also a test site during the GlobeSAR program (Ryerson et al., 1994). During the GlobeSAR program, airborne radar imagery was acquired and ground truth information was collected. The ground truth information was used to interpret the RADARSAT imagery.
2. The Study Area
Sihui County, in the province of Guangdong, is located in the southeast corner of China, in the Southern Uplands geographic region. This region is fairly mountainous with the only level terrain located in the delta of the River Xijiang. Rainfall in the southeast corner is the heaviest in China, averaging 100 to 200 cm per year, which is suitable for rain-fed, double-cropping of rice. In Guangdong province, attempts have been made to grow as many as there crops per year, but in terms of the added resources required, the addition of a third crop has not proved economical (Barker et al., 1985).
The double cropping system consists of Early rice and Late rice. Sowing for Early rice begins in March with harvest in early July. In the case of Late rice, sowing begins in early July with harvest in early November. The cropping calendar (Typical dates) for Early rice is shown in Table 1. Inter annual variability for both Early rice and Late rice does occur, with deviations that are largely related to the availability of water.
Table 1. Early rice cropping calendar for the Zhaoqing region. RADARSAT acquisition dates are indicated in brackets.
| Sowing |
Transplanting |
First Tillering |
Heading |
Harvest |
| March 1 (March 5) |
April 11 |
April 21 |
June 11 (June 10) |
July 11 |
3.0 Data
3.1 RADARSAT imagery
RADARSAT Standard 5 (S5) beam mode imagery was acquired in descending node on March 5 and June 10, 1996. The data were recorded on one of RADARSAT's on-board recorders, and downlinked for processing at the Canadian Data processing Facility in gatineau, Quebec. The data were processed to Path Image level (RSI, 1995). The RADARSAT Standard beam modes have nominal image coverage of 100 km x 100 km, 25 m resolution, and incidence angles that range from 20o to 49o.For the S5 mode, the range resolution varies between 23.6 m in near range to 20.7 m in far range, with 27 m nominal azimuth resolution (4 looks). The incidence angel varies from 36o in near range to 42o in far range.
3.2 Image Analysis Methodology
Manipulation of the imagery was performed using PCI's EASI/PACE software on a SunSPARC Station 20. All images were read from CEOS format using the EASI/PACE RADARSAT autoloading functionality.
The original format of all files was 16-bit unsigned, with gave a possible range of Digital Numbers (DN) of 0 to 655335. To reduce the working size of these files, and to prepare the images for printing and file export, each image was scaled to 8-bit unsigned date with a 0-255 DN range.
The merging of the images was accomplished using PCI's GCPWorks. The images were not geocoded individually, but each image was geolocated relative to the first Standard Beam image. The process is more commonly known as image-to-image correction. The Standarh mode image acquired March 5 was used as a baseline, and the second image (June 10) was geolocated to the first image. A total of thirteen ground control points were used resulting in a root-mean-square error of less than half a pixel in the x and y direction. A first-order transformation with cubic resampling was used.
