Abstract - Two methods of cloud masking tuned to Malaysian conditions have been developed, based on spectral analysis and Principal Components Analysis (PCA) of MODIS data. In the spectral approach, thresholds were applied to four reflective bands (1, 2, 3, and 4), three thermal bands (29, 31 and 32), the band 2/band_1 ratio, and the difference between band 29 and 31 in order to detect cloud. The PCA test applied a threshold to the first principal component derived from the seven quantities used for spectral analysis. Cloud detections were compared with the standard MODIS cloud mask, and their accuracy was assessed using reference images and geographical information on the study area.
Keywords: Cloud masking, spectral analysis, Principal Components Analysis , reflectance, brightness temperature
Typically, 50% of the Earth’s surface is covered by cloud at any time, where cloud is defined as a visible mass of condensed water droplets or ice crystals suspended in the atmosphere above the Earth's surface. In remote sensing, clouds are generally characterized by higher reflectance and lower temperature than the background. Thick opaque cloud blocks almost all information from the surface or near surface, while thin cloud has some physical characteristics similar to other atmospheric constituents. Misinterpretation of cloud may result in inaccuracy of various remote sensing applications, ranging from land cover classification to retrieval of atmospheric constituents (e.g. in air pollution studies).
Several cloud detection and masking studies have been reported in the literature. However, most of these algorithms were designed for the global scale (Rossow & Gardner 1993; Ackerman et al 2002), and little effort has been devoted to optimising regional methods. Some regional cloud masking algorithms have been designed for high, low and mid latitude regions, and these customised cloud masking algorithms tend to work best for these regions (Saunders 1986; Logar et al 1998). Little serious effort has been applied to the equatorial regions, especially South-east Asia (Franca & Cracknell 1995; Bendix et al 2004). This study considers this issue for the particular case of Malaysia. It uses MODIS Terra data to examine the spectral behaviour of cloud, identify effective MODIS bands for cloud detection and determine suitable cloud detection and masking methods in this region.
2. MATERIALS AND METHODS
This study is based on the MOD021KM product from MODIS Terra. A major advantage of MODIS is its wide range of spectral bands, with 36 spectral bands covering the visible, near infrared and thermal infrared wavelengths. In addition, MODIS, with its swath width of 2330 km, is capable of recording every point on the Earth at least once every two days and has an equatorial crossing time of 10:30 a.m. local time. Thus it can cover the whole study area (Peninsular Malaysia) in a single day pass with a high frequency of revisit....