Retrieving Lake Chla concentration from remote Sensing: Sampling time matters
Remote sensing is a promising technology for global water eutrophication monitoring, and the quantity and representativeness of the observed samples influence its accuracy. Due to limitations in high costs in situ monitoring, current remote sensing of lakes often relies on models trained with water...
| Main Authors: | , , , , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Elsevier
2024-01-01
|
| Series: | Ecological Indicators |
| Subjects: | |
| Online Access: | http://www.sciencedirect.com/science/article/pii/S1470160X23014322 |
| _version_ | 1797454352076505088 |
|---|---|
| author | Yufeng Yang Xikang Hou Wei Gao Feilong Li Fen Guo Yuan Zhang |
| author_facet | Yufeng Yang Xikang Hou Wei Gao Feilong Li Fen Guo Yuan Zhang |
| author_sort | Yufeng Yang |
| collection | DOAJ |
| description | Remote sensing is a promising technology for global water eutrophication monitoring, and the quantity and representativeness of the observed samples influence its accuracy. Due to limitations in high costs in situ monitoring, current remote sensing of lakes often relies on models trained with water quality samples from a single season. However, the generalization ability of these single-season trained models to other seasons has not been extensively studied. In this study, focusing on 38 major lakes in the Yangtze River Basin, we utilized monthly chlorophyll-a (Chla) monitoring data and Sentinel-2 image datasets from 2016 to 2021 to assess the impact of different seasonal data-trained models on the performance of Chla concentration retrieval in lakes. The results indicate that: (a) The sampling time for lake Chla significantly affects the performance of the retrieval model. Based on the seasonal data, the trained models show an R2 range of 0.523 to 0.699, with a Bias ranging from −7.06 % to 7.74 %, while the models trained with full-season samples perform between the various seasonal models. (b) When models built with seasonal data are applied to full-season data retrieval, there is a noticeable decrease in performance. The R2 of the spring-winter data trained model drops by 18.2 %, with RMSL and MAPE increasing by 11.4 % and 4.2 %, respectively. Similarly, the model trained from summer-autumn data shows a 4.02 % decrease in R2, with RMSLE and MAPE rising by 2.76 % and 13.56 %, respectively. This suggests that using models established from seasonal samples to infer full-season Chla concentrations can amplify errors. (c) Compared to seasonal models, full-season models based on different seasonal sampling exhibit better performance (R2 = 0.585, RMSLE = 0.337, Bias = -3.12 %, MAPE = 34.71 %). R2, RMSLE, and MAPE are all superior to seasonal models. Therefore, when performing full-time and long-term retrieval of Chla concentrations in water bodies, it is necessary to use data sampled in different seasons to reduce the errors introduced by seasonal sampling. (d) Chla retrieval models were established for 38 typical lakes and reservoirs in the Yangtze River Basin using the full-time dataset, revealing that 25 % (7/28) of the lakes and 20 % (2/10) of the reservoirs showed significant alteration trends. The number of lakes with increasing Chla concentrations exceeded those with decreasing concentrations, indicating that there is still a substantial risk of eutrophication in the lakes and reservoirs of the Yangtze River Basin. |
| first_indexed | 2024-03-09T15:36:02Z |
| format | Article |
| id | doaj.art-6d6fa616d6c9476791a02bdd81bcece3 |
| institution | Directory Open Access Journal |
| issn | 1470-160X |
| language | English |
| last_indexed | 2024-03-09T15:36:02Z |
| publishDate | 2024-01-01 |
| publisher | Elsevier |
| record_format | Article |
| series | Ecological Indicators |
| spelling | doaj.art-6d6fa616d6c9476791a02bdd81bcece32023-11-26T05:12:11ZengElsevierEcological Indicators1470-160X2024-01-01158111290Retrieving Lake Chla concentration from remote Sensing: Sampling time mattersYufeng Yang0Xikang Hou1Wei Gao2Feilong Li3Fen Guo4Yuan Zhang5Guangdong Provincial Key Laboratory of Water Quality Improvement and Ecological Restoration for Watersheds, School of Ecology, Environment and Resources, Guangdong University of Technology, Guangzhou 510006, ChinaState Environmental Protection Key Laboratory of Environmental Criteria and Risk Assessment, Chinese Research Academy of Environmental Sciences, Beijing 100012, ChinaGuangdong Provincial Key Laboratory of Water Quality Improvement and Ecological Restoration for Watersheds, School of Ecology, Environment and Resources, Guangdong University of Technology, Guangzhou 510006, China; Corresponding author.Guangdong Provincial Key Laboratory of Water Quality Improvement and Ecological Restoration for Watersheds, School of Ecology, Environment and Resources, Guangdong University of Technology, Guangzhou 510006, ChinaGuangdong Provincial Key Laboratory of Water Quality Improvement and Ecological Restoration for Watersheds, School of Ecology, Environment and Resources, Guangdong University of Technology, Guangzhou 510006, ChinaGuangdong Provincial Key Laboratory of Water Quality Improvement and Ecological Restoration for Watersheds, School of Ecology, Environment and Resources, Guangdong University of Technology, Guangzhou 510006, ChinaRemote sensing is a promising technology for global water eutrophication monitoring, and the quantity and representativeness of the observed samples influence its accuracy. Due to limitations in high costs in situ monitoring, current remote sensing of lakes often relies on models trained with water quality samples from a single season. However, the generalization ability of these single-season trained models to other seasons has not been extensively studied. In this study, focusing on 38 major lakes in the Yangtze River Basin, we utilized monthly chlorophyll-a (Chla) monitoring data and Sentinel-2 image datasets from 2016 to 2021 to assess the impact of different seasonal data-trained models on the performance of Chla concentration retrieval in lakes. The results indicate that: (a) The sampling time for lake Chla significantly affects the performance of the retrieval model. Based on the seasonal data, the trained models show an R2 range of 0.523 to 0.699, with a Bias ranging from −7.06 % to 7.74 %, while the models trained with full-season samples perform between the various seasonal models. (b) When models built with seasonal data are applied to full-season data retrieval, there is a noticeable decrease in performance. The R2 of the spring-winter data trained model drops by 18.2 %, with RMSL and MAPE increasing by 11.4 % and 4.2 %, respectively. Similarly, the model trained from summer-autumn data shows a 4.02 % decrease in R2, with RMSLE and MAPE rising by 2.76 % and 13.56 %, respectively. This suggests that using models established from seasonal samples to infer full-season Chla concentrations can amplify errors. (c) Compared to seasonal models, full-season models based on different seasonal sampling exhibit better performance (R2 = 0.585, RMSLE = 0.337, Bias = -3.12 %, MAPE = 34.71 %). R2, RMSLE, and MAPE are all superior to seasonal models. Therefore, when performing full-time and long-term retrieval of Chla concentrations in water bodies, it is necessary to use data sampled in different seasons to reduce the errors introduced by seasonal sampling. (d) Chla retrieval models were established for 38 typical lakes and reservoirs in the Yangtze River Basin using the full-time dataset, revealing that 25 % (7/28) of the lakes and 20 % (2/10) of the reservoirs showed significant alteration trends. The number of lakes with increasing Chla concentrations exceeded those with decreasing concentrations, indicating that there is still a substantial risk of eutrophication in the lakes and reservoirs of the Yangtze River Basin.http://www.sciencedirect.com/science/article/pii/S1470160X23014322Remote sensingInland waterEutrophicationTemporal representativenessYangtze River BasinMachine Learning |
| spellingShingle | Yufeng Yang Xikang Hou Wei Gao Feilong Li Fen Guo Yuan Zhang Retrieving Lake Chla concentration from remote Sensing: Sampling time matters Ecological Indicators Remote sensing Inland water Eutrophication Temporal representativeness Yangtze River Basin Machine Learning |
| title | Retrieving Lake Chla concentration from remote Sensing: Sampling time matters |
| title_full | Retrieving Lake Chla concentration from remote Sensing: Sampling time matters |
| title_fullStr | Retrieving Lake Chla concentration from remote Sensing: Sampling time matters |
| title_full_unstemmed | Retrieving Lake Chla concentration from remote Sensing: Sampling time matters |
| title_short | Retrieving Lake Chla concentration from remote Sensing: Sampling time matters |
| title_sort | retrieving lake chla concentration from remote sensing sampling time matters |
| topic | Remote sensing Inland water Eutrophication Temporal representativeness Yangtze River Basin Machine Learning |
| url | http://www.sciencedirect.com/science/article/pii/S1470160X23014322 |
| work_keys_str_mv | AT yufengyang retrievinglakechlaconcentrationfromremotesensingsamplingtimematters AT xikanghou retrievinglakechlaconcentrationfromremotesensingsamplingtimematters AT weigao retrievinglakechlaconcentrationfromremotesensingsamplingtimematters AT feilongli retrievinglakechlaconcentrationfromremotesensingsamplingtimematters AT fenguo retrievinglakechlaconcentrationfromremotesensingsamplingtimematters AT yuanzhang retrievinglakechlaconcentrationfromremotesensingsamplingtimematters |