VERTICAL DISTRIBUTION OF CHLOROPHYLL-A BASED ON NEURAL NETWORK

Abstract

An algorithm of estimating Vertical distribution of Chlorophyll-a (Chl-a) was evaluated based on Artificial Neural Networks (ANN) method in Hokkaido field in the northwest of Pacific Ocean. The algorithm applied to the data of SeaWiFS on OrbView-2 and AVHRR on NOAA off Hokkaido, has been applied on September 24, 1998 and September 28, 2001. Ocean color sensor provides the information of the photosynthetic pigment concentration for the upper 22% of the euphotic zone. In order to model a primary production in the water column derived from satellite, it is important to obtain the vertical profile of Chl-a distribution, because the maximum value of Chl-a concentration used to lie in the subsurface region. A shifted Gaussian model has been proposed to describe the variation of the chlorophyll-a (Chl-a) profile which consists of four parameters, i.e. background biomass (B0), maximum depth of Chl-a (zm), total biomass in the peak (h), and a measurement of the thickness or vertical scale of the peak (cr). However, these parameters are not easy to be determined directly from satellite data. Therefore, in the present study, an ANN methodology is used. Using in-situ data from 1974 to 1994 around Japan Islands, the above four parameters are calculated to derive the Chl-a concentration, sea surface temperature, mixed layer depth, latitude, longitude, and Julian days. The total of 6983 profiles of Chl-a and temperature are used for ANN. The correlation coefficients of these parameters are 0.79 (B0), 0.73 (h), 0.76 (cr) and 0.79 (zm) respectively. A site called A-linc off Hokkaido is used to evaluate Chl-a concentration in each depth. After comparing with in-situ data and ANN model, the results show good agreement relatively. Therefore, the ANN method is applicable and available tool to estimate primary production and fish resources from the space.
Keywords : Ocean color, Chlorophyll-a (Chl-a), Vertical structure, Artificial Neural Networks (ANN).