The effect of surface reflectivity on water vapor retrieval using MODIS NIR data

Different absorption techniques are suitable to retrieve the precipitable water vapor (PWV) from remote-sensing data. Some novel methods such as N/W, CIBR, LIRR etc, were proposed directly from the simplified near infrared (NIR) radiative transfer equations [Kaufman and Gao, 1992]. By using NIR "water vapor" channel (940 nm) in addition to "window" channels (865 nm and 1240 nm), two-band method and three-band method were applied to derive PWV using MODIS NIR data [Gao and Kaufman, 1998]. PWV retrieval method, with MODIS NIR band is analyzed using radiative transfer code, MODTRAN. The effect of surface nonlinear reflectivity on PWV retrieval is highlighted. Both two-band ratio method and three-band ratio method are taken into account. In a sensitivity analysis it is shown that the error due to the surface nonlinear reflectivity is very large (from 0.36 g/cm2 to 2.22 g/cm2 when PWV is from 0.2 g/cm2 to 6.2 g/cm2 ) if the reflectance ratio between "water vapor" band (940 nm band) and "windows" band (865 nm band) is not equal to unit. The error is proportional (about 14.7 times) to the transmittance (tau90 and tau865) ratio between these two bands. With Johns Hopkins University spectral library [2000], the ratio of band reflectance is calculated for hundreds kinds of surface material such as rocks, minerals, soils, vegetation and water. Most of the reflectance ratios have a considerable difference from unit. The absolute value of mean difference ranges from 0.0005 to 0.4351 for two bands method. It is from 0.0002 to 0.3089 for three bands method and smaller than two bands obviously. That indicates three-band method is superior to two-band method. To minimize the bands ratio for three bands method, the original band coefficients (0.2 and 0.8) estimated from the band position (940 nm, 865 nm and 1240 nm) are recalculated from the spectral library using least square method. The result shows that C1 = 0.8477, C2 = 1, C1 = 0.1523 and these coefficients must be obtained from the regression of whole spectral library and the sum of these two coefficients must be equal to unit