Cascaded Machine Learning of Soil Moisture and Salinity Prediction in Estuarine Wetlands Based on In Situ Internet of Things Monitoring
Abstract Estuarine wetlands, formed by the interaction of fluvial and tidal processes, exhibit complex spatiotemporal variations in soil moisture and salinity. Predicting soil moisture and salinity in estuarine wetlands is key for ecosystem management and assessing environmental impacts, while tradi...
Saved in:
| Main Authors: | , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2025-04-01
|
| Series: | Water Resources Research |
| Subjects: | |
| Online Access: | https://doi.org/10.1029/2024WR038271 |
| Tags: |
Add Tag
No Tags, Be the first to tag this record!
|
| _version_ | 1850212055084171264 |
|---|---|
| author | Jie Song Yujun Yi |
| author_facet | Jie Song Yujun Yi |
| author_sort | Jie Song |
| collection | DOAJ |
| description | Abstract Estuarine wetlands, formed by the interaction of fluvial and tidal processes, exhibit complex spatiotemporal variations in soil moisture and salinity. Predicting soil moisture and salinity in estuarine wetlands is key for ecosystem management and assessing environmental impacts, while traditional methods have limitations in resolution and complexity. The elucidation of transport pattern and prediction of water and salt in estuarine wetland soils remain significant challenges. To address these challenges and improve our ability to predict and manage wetland soil properties, this study employs an in situ Internet of Things (IoT)‐based monitoring network and a interpretable, cascaded machine learning model to predict these critical soil parameters. The IoT platform facilitates real‐time and longitudinal tracking of soil volumetric moisture content, salinity, and groundwater depth in the Yellow River Delta salt marsh wetlands, and the high‐fidelity monitoring data are used to build a two‐stage machine learning model. Artificial Neural Networks, Support Vector Machines, Random Forests (RF), and Gradient Boosting Decision Trees (GBDT) were used to develop the soil moisture and salinity prediction models. The cascaded framework, in combination with a moisture and a salinity sub‐model, which inspired by soil water and salt transport processes, was found to be an effective approach for capturing moisture‐salinity dynamics. The Gradient Boosting Decision Tree (GBDT) algorithm predicted moisture best (R2 = 0.846), while the GBDT‐RF model predicted salinity best (R2 = 0.875). To enhance model interpretability, SHAP (Shapley Additive exPlanations) analysis was applied, revealing that groundwater depth is the most significant positive driver of soil moisture, while water content is the dominant negative driver of soil salinity. These findings align with established eco‐hydrological processes, validating the models' ability to capture physically meaningful relationships. Sensitivity analysis revealed critical groundwater depth thresholds that strongly influence soil moisture and salinity. Specifically, as the water table rises, soil moisture increases to saturation at −0.5 m. Salt accumulates rapidly at −0.8 m (27% soil moisture) and becomes stable and close to seawater salinity. With real‐time in situ monitoring and the cascaded soil property prediction model, the method framework can accurately simulate and predict wetland soil moisture and salinity patterns, providing a valuable tool for monitoring and managing these vulnerable ecosystems and better understanding of wetland responses to environmental changes and supports evidence‐based conservation. |
| format | Article |
| id | doaj-art-53a6ce615a5e4af288fcd11f6998f3f0 |
| institution | OA Journals |
| issn | 0043-1397 1944-7973 |
| language | English |
| publishDate | 2025-04-01 |
| publisher | Wiley |
| record_format | Article |
| series | Water Resources Research |
| spelling | doaj-art-53a6ce615a5e4af288fcd11f6998f3f02025-08-20T02:09:25ZengWileyWater Resources Research0043-13971944-79732025-04-01614n/an/a10.1029/2024WR038271Cascaded Machine Learning of Soil Moisture and Salinity Prediction in Estuarine Wetlands Based on In Situ Internet of Things MonitoringJie Song0Yujun Yi1Key Laboratory for Water and Sediment Science Ministry of Education School of Environment Beijing Normal University Beijing ChinaKey Laboratory for Water and Sediment Science Ministry of Education School of Environment Beijing Normal University Beijing ChinaAbstract Estuarine wetlands, formed by the interaction of fluvial and tidal processes, exhibit complex spatiotemporal variations in soil moisture and salinity. Predicting soil moisture and salinity in estuarine wetlands is key for ecosystem management and assessing environmental impacts, while traditional methods have limitations in resolution and complexity. The elucidation of transport pattern and prediction of water and salt in estuarine wetland soils remain significant challenges. To address these challenges and improve our ability to predict and manage wetland soil properties, this study employs an in situ Internet of Things (IoT)‐based monitoring network and a interpretable, cascaded machine learning model to predict these critical soil parameters. The IoT platform facilitates real‐time and longitudinal tracking of soil volumetric moisture content, salinity, and groundwater depth in the Yellow River Delta salt marsh wetlands, and the high‐fidelity monitoring data are used to build a two‐stage machine learning model. Artificial Neural Networks, Support Vector Machines, Random Forests (RF), and Gradient Boosting Decision Trees (GBDT) were used to develop the soil moisture and salinity prediction models. The cascaded framework, in combination with a moisture and a salinity sub‐model, which inspired by soil water and salt transport processes, was found to be an effective approach for capturing moisture‐salinity dynamics. The Gradient Boosting Decision Tree (GBDT) algorithm predicted moisture best (R2 = 0.846), while the GBDT‐RF model predicted salinity best (R2 = 0.875). To enhance model interpretability, SHAP (Shapley Additive exPlanations) analysis was applied, revealing that groundwater depth is the most significant positive driver of soil moisture, while water content is the dominant negative driver of soil salinity. These findings align with established eco‐hydrological processes, validating the models' ability to capture physically meaningful relationships. Sensitivity analysis revealed critical groundwater depth thresholds that strongly influence soil moisture and salinity. Specifically, as the water table rises, soil moisture increases to saturation at −0.5 m. Salt accumulates rapidly at −0.8 m (27% soil moisture) and becomes stable and close to seawater salinity. With real‐time in situ monitoring and the cascaded soil property prediction model, the method framework can accurately simulate and predict wetland soil moisture and salinity patterns, providing a valuable tool for monitoring and managing these vulnerable ecosystems and better understanding of wetland responses to environmental changes and supports evidence‐based conservation.https://doi.org/10.1029/2024WR038271soil water and saltin situ monitoringinternet of thingsmachine learningYellow River Delta |
| spellingShingle | Jie Song Yujun Yi Cascaded Machine Learning of Soil Moisture and Salinity Prediction in Estuarine Wetlands Based on In Situ Internet of Things Monitoring Water Resources Research soil water and salt in situ monitoring internet of things machine learning Yellow River Delta |
| title | Cascaded Machine Learning of Soil Moisture and Salinity Prediction in Estuarine Wetlands Based on In Situ Internet of Things Monitoring |
| title_full | Cascaded Machine Learning of Soil Moisture and Salinity Prediction in Estuarine Wetlands Based on In Situ Internet of Things Monitoring |
| title_fullStr | Cascaded Machine Learning of Soil Moisture and Salinity Prediction in Estuarine Wetlands Based on In Situ Internet of Things Monitoring |
| title_full_unstemmed | Cascaded Machine Learning of Soil Moisture and Salinity Prediction in Estuarine Wetlands Based on In Situ Internet of Things Monitoring |
| title_short | Cascaded Machine Learning of Soil Moisture and Salinity Prediction in Estuarine Wetlands Based on In Situ Internet of Things Monitoring |
| title_sort | cascaded machine learning of soil moisture and salinity prediction in estuarine wetlands based on in situ internet of things monitoring |
| topic | soil water and salt in situ monitoring internet of things machine learning Yellow River Delta |
| url | https://doi.org/10.1029/2024WR038271 |
| work_keys_str_mv | AT jiesong cascadedmachinelearningofsoilmoistureandsalinitypredictioninestuarinewetlandsbasedoninsituinternetofthingsmonitoring AT yujunyi cascadedmachinelearningofsoilmoistureandsalinitypredictioninestuarinewetlandsbasedoninsituinternetofthingsmonitoring |