Extended Abstract (432K)P4.12
Dust detection and quantification from MODIS IR bands using Artificial Neural Network (ANN) model
Sang-Sam Lee, Seoul National University, Seoul, South Korea; and B. J. Sohn
Aerosol optical depth has been retrieved using artificial neural network (ANN) method from MODIS Aqua data. The applications were carried out on East Asia domain (20°N-55°N, 90°E-145°E) in spring (MAM) 2006. We used ANN model, the so-called MLP (Multi-Layer Perceptron) allowing a feed-forward backpropagation to retrieve aerosol optical depth at 550nm from MODIS measurements of brightness temperatures in the IR bands. The advantage of IR bands is that we can monitor dust distributions over bright surface and in nighttime. First, dust detection process by IR bands is carried out. Each threshold values of brightness temperature differences (BTD) between 3.7, 8.6, 11, and 12 µm were determined empirically. BTD(11-12) appears a general tool for dust detection but it shows unstable aspect spatiotemporally, e.g., land/ocean and day/night. The BTD(8.6-11) and BTD(3.7-11) bands were applied to correct the difference of land/ocean and day/night, respectively. To train and validate the ANN model, we used the brightness temperatures (MYD021KM), surface emissivities, and aerosol (MYD04_L2) product of MODIS Aqua. The input variables of the ANN model are sixteen brightness temperatures (band 20, 21, 22, 23, 24, 25, 27, 28, 29, 30, 31, 32, 33, 34, 35, and 36), sensor zenith angle from MYD021KM, surface temperature, and topography from SRTM30. The band-averaged surface emissivities were obtained from ASTER spectral library for each IGBP surface type. The target variable is aerosol optical depth of MYD04_L2 which is applied Deep Blue algorithm. Correlation coefficient between ANN model result and MODIS AOD on 0440UTC 8 April 2006 case was found to be 0.89. Other case studies and statistical analysis will be presented at the conference.
Poster Session 4, Radar and Icing Posters
Thursday, 24 January 2008, 9:45 AM-11:00 AM, Exhibit Hall B
Previous paper