XGBoost-Based Inversion of Phytoplankton Pigment Concentrations from Field Measured Fluorescence Excitation Spectra

Linqi Wang; Shengqiang Wang; Deyong Sun; Junsheng Li; Yuanli Zhu; Yongjiu Xu; Hailong Zhang

doi:10.3788/AOS202242.1830002

Acta Optica Sinica, Volume. 42, Issue 18, 1830002(2022)

XGBoost-Based Inversion of Phytoplankton Pigment Concentrations from Field Measured Fluorescence Excitation Spectra

Linqi Wang¹, Shengqiang Wang^1,2、*, Deyong Sun¹, Junsheng Li², Yuanli Zhu³, Yongjiu Xu⁴, and Hailong Zhang¹

Author Affiliations

¹School of Marine Sciences, Nanjing University of Information Science and Technology, Nanjing 210044, Jiangsu, China

²State Key Laboratory of Remote Sensing Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100101, China

³Second Institute of Oceanography, MNR, Hangzhou 310012, Zhejiang, China

⁴School of Fishery, Zhejiang Ocean University, Zhoushan 316022, Zhejiang, China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Distribution histograms of eight pigment concentrations. (a) Perid; (b) 19Butfu; (c) Fucox; (d) 19Hexfu; (e) Allox; (f) Zeax;(g) Chlb; (h) Tchla

Download full size

Fig. 2. Excitation fluorescence spectrum curves measured in field

Download full size

Fig. 3. Training performances of pigment concentration inversion models based on XGBoost machine learning algorithm. (a) Perid;(b) 19Butfu; (c) Fucox; (d) 19Hexfu; (e) Allox; (f) Zeax; (g) Chlb; (h) Tchla

Download full size

Fig. 4. Validation performances of pigment concentration inversion models based on XGBoost machine learning algorithm. (a) Perid; (b) 19Butfu; (c) Fucox; (d) 19Hexfu; (e) Allox; (f) Zeax; (g) Chlb; (h)Tchla

Download full size

Fig. 5. Profile distributions of eight pigment concentrations in 32.8°N section estimated from fluorescence excitation spectra. (a) Perid;(b) 19Butfu; (c) Fucox; (d) 19Hexfu; (e) Allox; (f) Zeax; (g) Chlb; (h) Tchla

Download full size

Fig. 6. Training performances of pigment concentration inversion models based on least square regression method. (a) Perid;(b) 19Butfu; (c) Fucox; (d) 19Hexfu; (e) Allox; (f) Zeax; (g) Chlb; (h) Tchla

Download full size

Fig. 7. Validation performances of concentration inversion models based on least square regression method. (a) Perid; (b) 19Butfu;(c) Fucox; (d) 19Hexfu; (e) Allox; (f) Zeax; (g) Chlb; (h) Tchla

Download full size

Fig. 8. Comparison of accuracies of pigment concentration inversion models based on XGBoost machine learning algorithm and least square regression method. (a) Model training; (b) model validation

Download full size

Table 1. English names (symbols and full names) of phytoplankton pigments involved in this paper
View table
Table 1. English names (symbols and full names) of phytoplankton pigments involved in this paper
Symbol Pigment
Tchla Total chlorophyll a
Chlb Chlorophyll b
Fucox Fucoxanthin
Perid Peridinin
19Hexfu 19′-Hexanoyloxyfucoxanthin
19Butfu 19′-Butanoyloxyfucoxanthin
Allox Alloxanthin
Zeax Zeaxanthin

Table 2. Exponential forms of seven excitation fluorescence spectral indicators

View table

Table 2. Exponential forms of seven excitation fluorescence spectral indicators

Spectral indicator	Exponential form
X₁	$\frac{l g [F (λ_{1})] + l g [F (λ_{2})]}{l g [F (λ_{1})] / l g [F (λ_{2})]}$
X₂	$\frac{l g [F (λ_{1})] - l g [F (λ_{2})]}{l g [F (λ_{1})] / l g [F (λ_{2})]}$
X₃	$\frac{l g [F (λ_{1})] - l g [F (λ_{2})]}{l g [F (λ_{1})] + l g [F (λ_{2})]}$
X₄	$l g [F (λ)]$
X₅	$\frac{l g [F (λ_{1})]}{l g [F (λ_{2})]}$
X₆	$l g \frac{F (λ_{1})}{F (λ_{2})}$
X₇	$l g [F (λ_{1}) + F (λ_{2})]$

Table 3. Statistics of pigment concentration measured by HPLC
View table
Table 3. Statistics of pigment concentration measured by HPLC
Pigment Minimum value Maximum value Average value
Tchla 0.078 6.730 1.354
Fucox 0 2.000 0.221
Perid 0 0.616 0.048
19Hexfu 0 1.038 0.125
19Butfu 0 0.363 0.040
Allox 0 0.525 0.032
Chlb 0 1.325 0.163
Zeax 0.003 0.813 0.120

Table 4. Optimal indictor forms of fluorescence excitation spectra and performances inverted by eight pigment concentrations

View table

Table 4. Optimal indictor forms of fluorescence excitation spectra and performances inverted by eight pigment concentrations

Pigment	Optimal indictor of fluorescence excitation spectrum	Training dataset			Validation dataset
Pigment	Optimal indictor of fluorescence excitation spectrum	R²	RMSE / （mg·m^-3）	MAPE /%	R²	RMSE /（mg·m^-3）	MAPE /%
Perid	X₃	0.84	0.036	39.2	0.77	0.110	49.9
19Butfu	X₆	0.94	0.025	24.1	0.67	0.024	50.6
Fucox	X₆	0.96	0.083	25.7	0.87	0.382	46.9
19Hexfu	X₆	0.78	0.103	39.2	0.68	0.125	35.8
Allox	X₅	0.96	0.030	26.5	0.86	0.037	38.2
Zeax	X₆	0.85	0.064	34.5	0.86	0.135	47.2
Chlb	X₅	0.80	0.171	41.1	0.59	0.241	64.2
Tchla	X₆	0.98	0.210	7.5	0.87	1.168	28.1

Table 5. Optimal indictor forms of fluorescence excitation spectra, best band combinations and performances inverted by eight pigment concentration based on least square regression method

View table

Table 5. Optimal indictor forms of fluorescence excitation spectra, best band combinations and performances inverted by eight pigment concentration based on least square regression method

Pigment	Optimal indictor of fluorescence excitation spectrum	Best band combination /nm	Training dataset			Validation dataset
Pigment	Optimal indictor of fluorescence excitation spectrum	Best band combination /nm	R²	RMSE /（mg·m^-3）	MAPE /%	R²	RMSE /（mg·m^-3）	MAPE /%
Perid	X₁	$λ_{1} = 570, λ_{2} = 505$	0.54	0.063	72.1	0.54	0.151	58.7
19Butfu	X₆	$λ_{1} = 505, λ_{2} = 590$	0.55	0.053	95.4	0.56	0.021	69.7
Fucox	X₁	$λ_{1} = 375, λ_{2} = 400$	0.74	0.162	94.5	0.87	0.770	68.6
19Hexfu	X₆	$λ_{1} = 375, λ_{2} = 435$	0.43	0.142	62.6	0.07	0.172	57.3
Allox	X₆	$λ_{1} = 435, λ_{2} = 505$	0.51	0.091	126.9	0.36	0.057	120.2
Zeax	X₆	$λ_{1} = 435, λ_{2} = 505$	0.42	0.119	80.0	0.46	0.186	124.5
Chlb	X₆	$λ_{1} = 505, λ_{2} = 590$	0.58	0.211	71.0	0.38	0.268	92.4
Tchla	X₇	$λ_{1} = 420, λ_{2} = 505$	0.76	0.786	43.4	0.75	0.893	45.1

Tools

Get Citation

Copy Citation Text

Linqi Wang, Shengqiang Wang, Deyong Sun, Junsheng Li, Yuanli Zhu, Yongjiu Xu, Hailong Zhang. XGBoost-Based Inversion of Phytoplankton Pigment Concentrations from Field Measured Fluorescence Excitation Spectra[J]. Acta Optica Sinica, 2022, 42(18): 1830002

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Spectroscopy

Received: Jan. 18, 2022

Accepted: Feb. 28, 2022

Published Online: Sep. 15, 2022

The Author Email: Wang Shengqiang (shengqiang.wang@nuist.edu.cn)

DOI:10.3788/AOS202242.1830002

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. English names (symbols and full names) of phytoplankton pigments involved in this paper

Table 1. English names (symbols and full names) of phytoplankton pigments involved in this paper

Table 2. Exponential forms of seven excitation fluorescence spectral indicators

Table 2. Exponential forms of seven excitation fluorescence spectral indicators

Table 3. Statistics of pigment concentration measured by HPLC

Table 3. Statistics of pigment concentration measured by HPLC

Table 4. Optimal indictor forms of fluorescence excitation spectra and performances inverted by eight pigment concentrations

Table 4. Optimal indictor forms of fluorescence excitation spectra and performances inverted by eight pigment concentrations

Table 5. Optimal indictor forms of fluorescence excitation spectra, best band combinations and performances inverted by eight pigment concentration based on least square regression method

Table 5. Optimal indictor forms of fluorescence excitation spectra, best band combinations and performances inverted by eight pigment concentration based on least square regression method