Research on the Retrieval Model of Non-optically Active Water Quality Parameters of Rivers based on Multi-source Remote Sensing and Meteorological Data

Zixuan DUI; Qing WANG; Min WANG; Jing ZHANG; Qianrong GU

doi:10.11873/j.issn.1004-0323.2024.1.0120

Remote Sensing Technology and Application, Volume. 39, Issue 1, 120(2024)

Research on the Retrieval Model of Non-optically Active Water Quality Parameters of Rivers based on Multi-source Remote Sensing and Meteorological Data

Zixuan DUI^1,2、*, Qing WANG³, Min WANG³, Jing ZHANG⁴, and Qianrong GU¹

Author Affiliations

¹Shanghai Carbon Data Research Center，Key Laboratory of Low - Carbon Conversion Science & Engineering，Shanghai Advanced Research Institute，Chinese Academy of Sciences，Shanghai 201210，China

²University of Chinese Academy of Sciences，Beijing 100049，China

³Shanghai Academy of Environmental Sciences，Shanghai 200233，China

⁴Jiangsu Provincial Judicial Police Officer Higher Vocational College，Nanjing 212008，China

show less

Abstract Get PDF(in Chinese)

Figures & Tables(13)

Fig. 1. Schematic diagram of support vector regression

Download full size

View in Article

Fig. 2. Schematic Diagram of Random Forest

Download full size

View in Article

Fig. 3. Schematic diagram of multi-layer perceptron

Download full size

View in Article

Fig. 4. Annual boxplots of all surface water quality monitoring data， model training data in the sample set， water quality parameters retrieved by the models with test data in the sample set

Download full size

View in Article

Fig. 5. Annual boxplots of all ground-based water quality monitoring data， time-mismatched water quality monitoring data， water quality parameters inverted by the models with time-mismatched multi-source remote sensing data

Download full size

View in Article

Fig. 6. Annual boxplots of all ground-based water quality monitoring data， space-mismatched water quality monitoring data， water quality parameters inverted by the models with space-mismatched multi-source remote sensing data

Download full size

View in Article

Fig. 7. Spatial distribution of DO， COD and NH3-H water quality parameters in a part of Yangtze and Pearl River retrieved from multi-source remote sensing data

Download full size

View in Article

Table 1. Time window for matching water quality monitoring data and multi-source remote sensing and meteorological data
View table
View in Article
Table 1. Time window for matching water quality monitoring data and multi-source remote sensing and meteorological data
数据名称哨兵2 MODIS ERA5
地表温度植被指数气溶胶
时间窗口 4 h 同一天同一天同一天 1 h

Table 2. The ground water quality monitoring sample data set filtered by spatial and temporal matching with the multi-source remote sensing and meteorological data

View table

View in Article

Table 2. The ground water quality monitoring sample data set filtered by spatial and temporal matching with the multi-source remote sensing and meteorological data

状态	水质参数	异常值处理比例	样本数	平均值/(mg $∙$ L^-1)	标准偏差	最小值/(mg $∙$ L^-1)	最大值/(mg $∙$ L^-1)
异常值剔除前	DO	-	270	8.78	2.43	3.76	18.07
	COD	-	270	2.29	2.34	0.10	15.60
	NH₃-N	-	270	0.93	1.42	0.00	10.00
异常值剔除后	DO	0.08	249	8.29	1.54	4.70	12.10
	COD	0.10	243	1.68	1.33	0.10	5.25
	NH₃-N	0.20	216	0.55	0.65	0.00	2.80

Table 3. Data set used in model robustness assessment experiments

View table

View in Article

Table 3. Data set used in model robustness assessment experiments

实验名称	数据集	水质参数/(个/组）
实验名称	数据集	DO	COD	NH₃-N
	所有地面水质监测数据	812 441	811 147	804 843
时空匹配实验	训练数据	249	243	216
时空匹配实验	测试数据	75	73	65
时间未匹配实验	水质监测数据	234 470	235 918	235 598
时间未匹配实验	多源遥感及气象数据	142	142	142
空间未匹配实验	水质监测数据	575 436	580 629	574 486
空间未匹配实验	多源遥感及气象数据	219	219	219

Table 4. Optimal hyperparameter table of three machine learning water quality parameter retrieval models
View table
View in Article
Table 4. Optimal hyperparameter table of three machine learning water quality parameter retrieval models
机器学习模型参数名称非光学活性水质参数
DO COD NH₃-N
SVR C 100 10 10
gamma 7 3 3
RF 决策树个数 15 26 15
树的深度 46 15 10
MLP 隐藏层数 6 5 5
每个隐藏层神经元数 100 100 100

Table 5. DO， COD and NH₃-N retrieval results of SVR， RF and MLP models for 14 combinations of input features

View table

View in Article

Table 5. DO， COD and NH₃-N retrieval results of SVR， RF and MLP models for 14 combinations of input features

序号	输入数据	指标	DO			COD			NH₃-N
序号	输入数据	指标	SVR	RF	MLP	SVR	RF	MLP	SVR	RF	MLP
1	MSI	R²	0.837	0.845	0.832	0.649	0.699	0.656	0.242	0.447	0.485
		MAE/(mg $∙$ L^-1)	0.308	0.393	0.426	0.477	0.506	0.474	0.183	0.144	0.139
		RMSE/(mg $∙$ L^-1)	0.402	0.384	0.414	0.594	0.513	0.612	0.100	0.071	0.066
2	MSI+VI	R²	0.837	0.852	0.357	0.650	0.685	0.744	0.191	0.523	0.353
		MAE/(mg $∙$ L^-1)	0.302	0.375	0.729	0.485	0.502	0.415	0.188	0.139	0.151
		RMSE/(mg $∙$ L^-1)	0.407	0.366	1.711	0.596	0.537	0.444	0.106	0.063	0.080
3	MSI+LST	R²	0.866	0.860	0.764	0.611	0.654	0.605	0.232	0.292	0.386
		MAE/(mg $∙$ L^-1)	0.306	0.384	0.465	0.506	0.546	0.543	0.182	0.173	0.154
		RMSE/(mg $∙$ L^-1)	0.336	0.347	0.594	0.654	0.592	0.703	0.100	0.092	0.078
4	MSI+VI+LST	R²	0.890	0.879	0.841	0.671	0.673	0.717	0.251	0.232	0.325
		MAE/(mg $∙$ L^-1)	0.282	0.359	0.377	0.472	0.527	0.430	0.184	0.168	0.154
		RMSE/(mg $∙$ L^-1)	0.274	0.300	0.397	0.560	0.558	0.491	0.098	0.096	0.084
5	MSI+AOD	R²	0.872	0.863	0.856	0.676	0.671	0.722	0.307	0.507	0.432
		MAE/(mg $∙$ L^-1)	0.300	0.363	0.368	0.470	0.514	0.435	0.165	0.139	0.143
		RMSE/(mg $∙$ L^-1)	0.320	0.341	0.360	0.553	0.564	0.476	0.091	0.064	0.072
6	MSI+WS	R²	0.802	0.851	0.835	0.648	0.699	0.738	0.289	0.488	0.536
		MAE/(mg $∙$ L^-1)	0.344	0.385	0.377	0.464	0.495	0.431	0.172	0.143	0.129
		RMSE/(mg $∙$ L^-1)	0.497	0.364	0.417	0.599	0.514	0.449	0.091	0.066	0.061
7	MSI+AOD+WS	R²	0.868	0.829	0.733	0.704	0.669	0.781	0.334	0.529	0.502
		MAE/(mg $∙$ L^-1)	0.293	0.402	0.454	0.439	0.515	0.398	0.171	0.137	0.137
		RMSE/(mg $∙$ L^-1)	0.330	0.431	0.647	0.507	0.565	0.383	0.087	0.061	0.065
8	MSI+VI+LST+AOD+WS	R²	0.889	0.868	0.853	0.670	0.660	0.742	0.221	0.278	0.383
		MAE/(mg $∙$ L^-1)	0.289	0.378	0.337	0.487	0.528	0.414	0.184	0.168	0.153
		RMSE/(mg $∙$ L^-1)	0.278	0.332	0.367	0.564	0.578	0.445	0.099	0.092	0.079
9	MSI+LST+WS	R²	0.853	0.839	0.695	0.674	0.642	0.721	0.249	0.320	0.436
		MAE/(mg $∙$ L^-1)	0.320	0.411	0.506	0.479	0.542	0.446	0.178	0.161	0.149
		RMSE/(mg $∙$ L^-1)	0.371	0.404	0.780	0.557	0.618	0.482	0.099	0.087	0.073
10	MSI+LST+AOD	R²	0.896	0.866	0.837	0.644	0.656	0.748	0.314	0.314	0.458
		MAE/(mg $∙$ L^-1)	0.276	0.369	0.368	0.506	0.543	0.438	0.175	0.165	0.143
		RMSE/(mg $∙$ L^-1)	0.263	0.333	0.409	0.611	0.589	0.428	0.088	0.089	0.070
11	MSI+LST+AOD+WS	R²	0.891	0.872	0.857	0.707	0.646	0.736	0.273	0.226	0.381
		MAE/(mg $∙$ L^-1)	0.283	0.370	0.333	0.441	0.542	0.424	0.180	0.170	0.152
		RMSE/(mg $∙$ L^-1)	0.275	0.320	0.350	0.506	0.613	0.470	0.094	0.098	0.078
12	MSI+VI+WS	R²	0.835	0.831	0.817	0.677	0.690	0.764	0.217	0.482	0.293
		MAE/(mg $∙$ L^-1)	0.305	0.382	0.358	0.464	0.502	0.404	0.185	0.145	0.162
		RMSE/(mg $∙$ L^-1)	0.412	0.416	0.466	0.553	0.531	0.412	0.103	0.066	0.092
13	MSI+VI+AOD	R²	0.869	0.856	0.867	0.692	0.667	0.714	0.313	0.459	0.382
		MAE/(mg $∙$ L^-1)	0.291	0.371	0.337	0.439	0.511	0.419	0.175	0.146	0.146
		RMSE/(mg $∙$ L^-1)	0.324	0.360	0.337	0.529	0.571	0.498	0.088	0.069	0.077
14	MSI+VI+AOD+WS	R²	0.879	0.836	0.844	0.690	0.676	0.724	0.307	0.509	0.412
		MAE/(mg $∙$ L^-1)	0.291	0.396	0.344	0.452	0.502	0.409	0.177	0.139	0.145
		RMSE/(mg $∙$ L^-1)	0.303	0.410	0.391	0.526	0.561	0.483	0.089	0.064	0.077

Table 6. The best retrieval performance of water quality parameters that can be obtained by models of different methods

View table

View in Article

Table 6. The best retrieval performance of water quality parameters that can be obtained by models of different methods

方法	指标	非光学活性水质参数
方法	指标	DO	COD	NH₃-N
MLR	R²	0.383	0.066	0.120
	MAE/(mg $∙$ L^-1)	1.027	1.046	0.251
	RMSE/(mg $∙$ L^-1)	1.650	1.561	0.132
SVR	R²	0.896	0.707	0.334
	MAE/(mg $∙$ L^-1)	0.276	0.441	0.171
	RMSE/(mg $∙$ L^-1)	0.263	0.506	0.087
RF	R²	0.879	0.699	0.529
	MAE/(mg $∙$ L^-1)	0.359	0.495	0.137
	RMSE/(mg $∙$ L^-1)	0.300	0.514	0.061
MLP	R²	0.867	0.781	0.536
	MAE/(mg $∙$ L^-1)	0.337	0.398	0.129
	RMSE/(mg $∙$ L^-1)	0.337	0.383	0.061

Tools

Get Citation

Copy Citation Text

Zixuan DUI, Qing WANG, Min WANG, Jing ZHANG, Qianrong GU. Research on the Retrieval Model of Non-optically Active Water Quality Parameters of Rivers based on Multi-source Remote Sensing and Meteorological Data[J]. Remote Sensing Technology and Application, 2024, 39(1): 120

Download Citation

EndNote(RIS)BibTex Plain Text

Set citation alerts for article

Save article for my favorites

Paper Information

Category: Research Articles

Received: Oct. 13, 2022

Accepted: --

Published Online: Jul. 22, 2024

The Author Email: Zixuan DUI (duizx@sari.ac.cn)

DOI:10.11873/j.issn.1004-0323.2024.1.0120

Topics

laser devices and laser physics

Lasers and Laser Optics

Laser physics

laser manufacturing

Instrumentation, Measurement and Metrology

Table 1. Time window for matching water quality monitoring data and multi-source remote sensing and meteorological data

Table 1. Time window for matching water quality monitoring data and multi-source remote sensing and meteorological data

Table 2. The ground water quality monitoring sample data set filtered by spatial and temporal matching with the multi-source remote sensing and meteorological data

Table 2. The ground water quality monitoring sample data set filtered by spatial and temporal matching with the multi-source remote sensing and meteorological data

Table 3. Data set used in model robustness assessment experiments

Table 3. Data set used in model robustness assessment experiments

Table 4. Optimal hyperparameter table of three machine learning water quality parameter retrieval models

Table 4. Optimal hyperparameter table of three machine learning water quality parameter retrieval models

Table 5. DO， COD and NH3-N retrieval results of SVR， RF and MLP models for 14 combinations of input features

Table 5. DO， COD and NH3-N retrieval results of SVR， RF and MLP models for 14 combinations of input features

Table 6. The best retrieval performance of water quality parameters that can be obtained by models of different methods

Table 6. The best retrieval performance of water quality parameters that can be obtained by models of different methods

Table 5. DO， COD and NH₃-N retrieval results of SVR， RF and MLP models for 14 combinations of input features

Table 5. DO， COD and NH₃-N retrieval results of SVR， RF and MLP models for 14 combinations of input features