A Simple Metod for the Cloud Detection over Land Using Daytime AVHRR Data
MyoungSeok Suh, Kwangmi Jang, KyoungYoon park
Systems Integration Dept. Systems Engineering Research Institute,
P.O. Box 1, Yoousung-ku, Taejeon 305-600, Korea
Tel : (82)-42-869-1577, Fax : (82)-42-869-1549
E-mail :smsd@seri.re.kr
Abstract
A simple 5 step threshold method using different combination on channels was developed to detect the cloud-contaminated pixels from NOAA-14/AVHRR daytime imagery. The first two thresholds were applied to the infra-red(channel 4) and visible (channel 1) imagery to detect the low temperature pixels (high or middle -level clouds) and the high reflectance pixels (low-level clouds). To detect the cloud edge and small cumulus cloud, spatial coherency test using local standard deviation (LSD) was applied to near infra-red images. And split-window threshold(the brightness temperature difference between channel 4 and 5) was applied to detect the optically thin cirrus. Finally, inverse relation between visible and infra-red imagery was tested to detect the other cloud contaminated pixels.
We applied this algorithm to the NOAA-14/AVHRR day time imagery for the April 1 to June 30, 1997. The quality of this algorithm was good in the visual about the composting periods showed that this algorithm was very stable in the composting period changes.
1. Introduction
NDVI (Normalized Difference Vegetation Index) from NOAA/AVHRR imagery data is the most widely used among many satellites imagery data. The magnitude of NDVI can indicate the status and amount of vegetation, but there are so many factors (e.g., cloud, sun-sensor-target geometry, atmospheric condition, sensor degradation) unrelevant to vegetation dynamics which affect the magnitude of NDVI. Cloud is the most critical factor among them. Reflectance and emissivity of land surface vary spatially and temporally. Accurate cloud detection over land using NOAA/AVHRR imagery dta is a complicated task (Simpson and Gobat, 1996). Numerous cloud detection methods using AVHRR data have been already reported. They can be separated into two methods using AVHRR data have been already reported. They can be separated into two categories. One is clustering or pattern recognition methods relying on histogram analysis (Desbois et atl., 1982; Ebert, 1987; Seze and Desbois, 1987; Simpson and Gobat,
1996). The other is thresholding techniques which are applied to each pixel (Reynolds and Vonder haar, 1977; Schiffer and Rossow, 1983; Saunders and Kriebel, 1988; Derrien et al., 1993; Suh and Park, 1993).
A simple 5 step threshold method using different combination of channels was developed to detect the cloud contaminated pixels from NOAA-14/AVHRR day time imagery. The work presented here was based on Saunders and kriebel(1988) and Derrien et al., (1993), but our method differed from those of Saunders et al. and Derrien et al. in several impotent respects. Although we adopted some threshold value from them, we made a reference data(clear sky radiance : brightness temperature and reflectance, local standard devation of them) quite different ways using only NOAA/AVHRR imagery data.
We applied this algorithm to the NOAA-14/AVHRR day imagery for the April 1 to June 30 of 1997. The results were analyzed through visual intercomparison with the enhanced visible and infrared imagery. Sensitivities of the composting period and the magnitude of threshold was studied. And we investigated the proper composting period for the cloud free NDVI data using this results.
