In-water Algorithm (Neural Network) Motoaki Kishino ( RIKEN ) 8 November 2000 Kanazawa
PI and Co-i � Roland Doerffer � Motoaki Kishino � Helmut Schiller � Tomohiko Oishi � Heinz van de Piepen � Hiroh Satoh � Carsten Brockmann � Kouji Suzuki � J. H. M. Hakvoort � Tooru Hirawake � Herbert Siegel
NN-Training
Radiative Transfer Model = = = = × × × × L E R n w s rs R = = = = × × × × R 0 . 533 rs Q − − − − k a = = = = t R Joseph (1950) + + + + k a t = = = = + + + + k a ( a 2 b ) t t bt
Model of Optical Properties Absorption Coefficient = = = = + + + + + + + + a a a a t w c y a : Pope and Fry (1997) w = = × × = = × × * a a C c c * a : average values of the sea truth c (off Sanriku, Japan, Kishino) − − − − λ λ λ λ − − − − = = = = × × × × 0 . 014 ( 440 ) a CDOM e Roesler et al. (1989) y
Model of Optical Properties Backscattering Coefficient = = = = + + + + + + + + b b b b b b b b t w c s − − − − λ λ λ λ 4 . 32 = = = = b 0 . 00144 * Morel (1974) b w 500 550 = = = = + + + + − − − − 0 . 62 b 0 . 30 * C * ( 0 . 002 0 . 02 * ( 0 . 5 0 . 25 * log C )) * λ λ 10 λ λ bc Gordon and Morel (1983) Morel (1988) − − − − λ λ λ λ 0 . 812 = = = = b bs S 0 . 001848 * * 550 Kronfeld (1988)
Parameters and their range for the Neural Network training Parameter Unit Range mg m -3 Chlorophyll a 0.1 - 50 Inorganic Suspension g m -3 0.1 - 50 CDOM m -1 0.01 - 10 Data Set for neural network training 100,000 Data set for validation 50,000
Neural Network example for OCTS data : unit : link nLw(412) nLw(443) chlorophyll a nLw(490) suspended matter nLw(520) yellow substance nLw(565) input layer hidden layer output layer
R M S All Bands 11 Bands Chl a IOS CDOM 0%: 0.00178 0.00189 0.000168 10%: 0.0302 0.0569 0.00312 20%: 0.0520 0.141 0.00601 Except Saturation Bands 9 Bands Chl a IOS CDOM 0%: 0.00473 0.00221 0.000212 10%: 0.0218 0.135 0.00828 20%: 0.0577 0.201 0.0207
Recommend
More recommend
