Acta Optica Sinica, Volume. 43, Issue 24, 2401007(2023)
Chlorophyll Profile Retrieval Algorithm Based on Oceanographic Lidar and BP Neural Network
Figures & Tables(17)
Fig. 4. Chlorophyll concentration profile and simulated lidar echo results. (a) Chlorophyll concentration profile from Argo-BGC; (b) lidar profile from Monte Carlo simulation using Fig.3(a)
Fig. 5. Comparison of filtering algorithms before and after processing. (a) Results of RANSAC; (b) lidar echoes after noise processing
Fig. 7. Comparison of inversion results and dataset labels (chlorophyll profile) of LIMC-BPNN and PR Chla models with different concentration ranges and depths. (a) (d) (g) (j) (m) Inversion results of low mass concentration water; (b) (e)(h)(k)(n) inversion results of medium mass concentration water; (c) (f) (i) (l) (o) inversion results of high mass concentration water
Fig. 8. Errors of different conditions (depths, concentration ranges, algorithms). (a) RE; (b) RMSE; (c) R; (d) ME
Fig. 9. Comparisons between retrieved and in-situ in different sites. (a) B3 site; (b) G1 site; (c) G2 site
Table 1. MSE with different layer numbers of network structure
View tableTable 1. MSE with different layer numbers of network structure
Layer number 2 3 4 5 Best MSE /（mg/m3） 0.305 0.096 0.181 0.198
Table 2. MSE with different batch sizes
View tableTable 2. MSE with different batch sizes
Batch size 16 32 64 128 Best MSE /（mg/m3） 0.202 0.147 0.177 0.272
Table 3. Model building environment description
View tableTable 3. Model building environment description
Operating system Programming
language
Environment Framework GPU Windows10 Python 3.8 Spyder Pytorch RTX A5000
Table 4. Chlorophyll peak range statistics
View tableTable 4. Chlorophyll peak range statistics
Chlorophyll concentration range Value （0.0，5.0］ 19183 （5.0， 10.0］ 420 （10.0，15.0］ 100 （15.0， 20.0］ 43 （20.0， 25.0］ 20 （25.0， 30.0］ 27 （30.0， 35.0］ 6 >35.0 50
Table 5. Parameters for simulation of semi-analytic Monte Carlo
View tableTable 5. Parameters for simulation of semi-analytic Monte Carlo
Parameter Value Laser wavelength /nm 486 Telescope diameter /m 0.1 Platform height /m 2000 Field of view /mrad 25 Phase function Fournier-Fornad Transmission photon counts 106 Maximum scattering times 10 Profile resolution /m 0.1
Table 6. Average errors of different algorithms
View tableTable 6. Average errors of different algorithms
Algorithm RE /% RMSE /（mg/m3） ME /（mg/m3） R PR-Chla 56.73 0.616 0.331 0.724 LIMC-BPNN 22.51 0.253 0.118 0.904
Table 7. Comparison of average errors between two models at different depths
View tableTable 7. Comparison of average errors between two models at different depths
Depth /m RE /% RMSE /（mg/m3） ME /（mg/m3） R PR-Chla LIMC-BPNN PR-Chla LIMC-BPNN PR-Chla LIMC-BPNN PR-Chla LIMC-BPNN 0-10 50.04 17.00 0.371 0.089 0.157 0.037 0.782 0.991 10-20 44.82 16.03 0.281 0.174 0.152 0.063 0.808 0.959 20-30 39.26 20.35 0.421 0.259 0.201 0.108 0.767 0.923 30-40 56.53 23.56 0.795 0.334 0.432 0.160 0.673 0.874 40-50 92.98 35.60 1.209 0.410 0.715 0.222 0.587 0.775
Table 8. Average errors of LIMC-BPNN retrieved to the measurement value in situ
View tableTable 8. Average errors of LIMC-BPNN retrieved to the measurement value in situ
Site RE /% RMSE /
（mg/m3）
ME /
（mg/m3）
R B3 19.88 0.094 0.068 0.958 G1 10.24 0.037 0.026 0.988 G2 10.36 0.045 0.028 0.965
Tools
Get Citation
Copy Citation Text
Ning Tie, Bingyi Liu. Chlorophyll Profile Retrieval Algorithm Based on Oceanographic Lidar and BP Neural Network[J]. Acta Optica Sinica, 2023, 43(24): 2401007
Paper Information
Category: Atmospheric Optics and Oceanic Optics
Received: Apr. 10, 2023
Accepted: May. 31, 2023
Published Online: Dec. 8, 2023
The Author Email: Liu Bingyi (liubingyi@ouc.edu.cn)
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20