Daily NDVI was obtained from the AVHRR Land Pathfinder data set (8 km resolution; available at URL: http://daac.gsfc.nasa.gov/; (Goward et al., 1991)) between 1995 and 2000 for the Yellow River Basin in China (Figure 1). The region consists of three typical landforms: the Qinghai-Tibet Plateau, the Loess Plateau and the alluvial plain (Yang et al., 2004). The reader is referred to Yang et al. (2004) for a detailed description of the area. The Pathfinder AVHRR daily NDVI, which has been corrected for Rayleigh scattering and ozone absorption but not corrected for atmospheric water vapor (James and Kalluri, 1994), is considered as the NDVI data set with the influence of clouds in this study although it is not the real-time AVHRR data on board the satellite. It is worth mentioning that even the AVHRR Land Pathfinder data set are not sufficiently stable enough to present clouds variations in a day, it is expected to present daily variations of cloud status and it is the only NDVI data source before 2000. The land use data was obtained from U.S. Geological Survey Land Use data set which was based on 1-km AVHRR data spanning April 1992 through March 1993 (Loveland et al., 1999; Anderson et al., 1976). The elevation data was drawn from HYDRO1k, a geographic database developed at the U.S. Geological Survey's (USGS) EROS Data Center and providing comprehensive and consistent global coverage of topographically derived data sets (Team, 2003).
There are many ways to try to get the cloud free NDVI (Roerink et al., 2000; Holben, 1986; Verhoef et al., 1996). A maximum composite way is used in this study in order to simplify the calculation. The daily NDVI values are composed at each grid of the study area. The composite is generated by comparing the NDVI values from consecutive daily values in each month. The pixel with the highest NDVI value for the month is chosen as the date for the inclusion in the composite. The corrected NDVI for each day of the study period were then interpolated using cubic spline from the composite monthly data. The corrected NDVI is considered as the "true" NDVI without the influence of clouds.
Therefore, the ratio of the satellitic real-time NDVI and the derived "true" NDVI is recognized as the consequence of clouds influence. The NDVI Clouds Index (NCI) is defined as the ratio of the terms (NDVIr - NDVImin) and (NDVIc - NDVImin) (Figure 2):
| NCI = (NDVIr - NDVImin)/(NDVIc - NDVImin) | (1) |
where NDVIr is daily real-time NDVI; NDVIc is daily corrected NDVI and NDVImin is the minimum value of NDVI during the whole study period. Note that the NCI must be limited so that NCI is less than 1.0.
Daily observed cloud amount and actual duration of sunshine at 120 meteorological stations inside and close to the study area was obtained from the China Meteorological Administration (CMA) (Figure 1). In FAO Penman-Monteith equation, the solar radiation, Rs, is calculated with the Angstrom formula which relates solar radiation to extraterrestrial radiation and relative sunshine duration (Allen et al., 1998):
| Rs = (as + bs×n/N)Ra | (2) |
where n is actual duration of sunshine (hour); N is maximum possible duration of sunshine or daylight hours (hour); as is regression constant expressing the fraction of extraterrestrial radiation reaching the earth on overcast days; (as + bs) is fraction of extraterrestrial radiation reaching the earth on clear days; Ra is extraterrestrial radiation. The item SCI = n/N roughly reflects the clouds influence on solar radiation.
The daily NCI values are related to the two cloud indices based on ground observations, cloud amount and Sunshine Clouds Index (SCI), at the 120 meteorological stations. The NDVI data can provide distributed NCI values over the whole study area, but the ground stations can only give point observations. The surface climate data from station observations are usually interpolated to get the gridded dataset in meteorology (New et al., 1999). As an application, the observed cloud amount and SCI values at the stations were interpolated to all over the study area using a thin plate spline algorithm. The daily NCI values over the study area were then related to gridded cloud amount and SCI during the same period, from 1995 to 2000. The relationship variation over land cover was analyzed according to the U.S. Geological Survey Land Use. The Hydro1k DEM data was used to observe relationship variation with elevation.
A running for the comparison in each month was also performed to observe seasonal variation of the relationship. The daily NCI values of one special month were selected and related to cloud amount and SCI during 1995 - 2000. The monthly variation of the relationship was demonstrated.