WQEye (v1) : A Python-based software for machine learning-based retrieval of water quality parameters from Sentinel-2 and Landsat-8/9 remote sensing data aided by Google Earth Engine
(2026) In Ecological Informatics 95.- Abstract
Despite substantial progress in Remote Sensing (RS)-based estimation of Water Quality Parameters (WQPs) using Machine Learning (ML) models, there remains a lack of end-to-end software that automates the entire workflow. To address this gap, this paper introduces WQEye (v1), an open-source software offering a user-friendly interface comprising six modular components: (1) Data loader; (2) RS sampling module supporting Sentinel-2 (S2) and Landsat-8/9 (L89) through Google Earth Engine; (3) Matching module for finding coincident in-situ and satellite observations; (4) Preprocessing module; (5) ML module supporting Random Forest (RF), Artificial Neural Networks (ANNs), and recently proposed Kolmogorov-Arnold Networks (KANs), along with... (More)
Despite substantial progress in Remote Sensing (RS)-based estimation of Water Quality Parameters (WQPs) using Machine Learning (ML) models, there remains a lack of end-to-end software that automates the entire workflow. To address this gap, this paper introduces WQEye (v1), an open-source software offering a user-friendly interface comprising six modular components: (1) Data loader; (2) RS sampling module supporting Sentinel-2 (S2) and Landsat-8/9 (L89) through Google Earth Engine; (3) Matching module for finding coincident in-situ and satellite observations; (4) Preprocessing module; (5) ML module supporting Random Forest (RF), Artificial Neural Networks (ANNs), and recently proposed Kolmogorov-Arnold Networks (KANs), along with hyperparameter tuning, model training, validation, and interpretability analysis using Shapely Additive Explanations (SHAP); and (6) Export module for generating WQP maps for spatiotemporal analysis. The reliability of WQEye's workflow was validated through the estimation of different WQPs (turbidity, chlorophyll-a, dissolved oxygen, and fluorescent dissolved organic matter) using both S2 and L89 data across five different locations in the United States. KAN consistently demonstrated statistically superior performance compared to RF and ANN across all WQPs, achieving approximately 5% lower MAPE and 6% higher R2 than ANN, the second-best model. Cross-regional assessment of the ML models further demonstrated the superior generalization capability of the KAN model compared to the other models across both the S2 and L89 datasets. Moreover, S2 imagery outperformed L89 data across all WQPs. The findings prove the reliability of WQEye's workflow for the estimation of WQPs using RS data, but the final accuracy of outputs depends on the detectability of the target WQP in RS imagery, as non-optically active parameters like dissolved oxygen yielded lower estimation accuracies compared to optically active ones. WQEye is freely available at https://github.com/ATDehkordi/WQEye to promote applied RS applications in environmental monitoring and water resource management.
(Less)
- author
- Taheri Dehkordi, Alireza
LU
; Dazi, Mostafa
; Naghibi, Amir
LU
and Hashemi, Hossein
LU
- organization
- publishing date
- 2026-05
- type
- Contribution to journal
- publication status
- published
- subject
- keywords
- Artificial neural networks, Kolmogorov–Arnold networks, Machine learning, Random Forest, Satellite data, Water quality
- in
- Ecological Informatics
- volume
- 95
- article number
- 103692
- publisher
- Elsevier
- external identifiers
-
- scopus:105032230456
- ISSN
- 1574-9541
- DOI
- 10.1016/j.ecoinf.2026.103692
- project
- The United Nations University Hub: Water in a Changing Environment (WICE)
- language
- English
- LU publication?
- yes
- additional info
- Publisher Copyright: © 2026 The Authors.
- id
- 9d188f6e-41fc-4b5a-9a33-75f7231c8ec8
- date added to LUP
- 2026-05-05 13:38:36
- date last changed
- 2026-05-11 12:33:22
@article{9d188f6e-41fc-4b5a-9a33-75f7231c8ec8,
abstract = {{<p>Despite substantial progress in Remote Sensing (RS)-based estimation of Water Quality Parameters (WQPs) using Machine Learning (ML) models, there remains a lack of end-to-end software that automates the entire workflow. To address this gap, this paper introduces WQEye (v1), an open-source software offering a user-friendly interface comprising six modular components: (1) Data loader; (2) RS sampling module supporting Sentinel-2 (S2) and Landsat-8/9 (L89) through Google Earth Engine; (3) Matching module for finding coincident in-situ and satellite observations; (4) Preprocessing module; (5) ML module supporting Random Forest (RF), Artificial Neural Networks (ANNs), and recently proposed Kolmogorov-Arnold Networks (KANs), along with hyperparameter tuning, model training, validation, and interpretability analysis using Shapely Additive Explanations (SHAP); and (6) Export module for generating WQP maps for spatiotemporal analysis. The reliability of WQEye's workflow was validated through the estimation of different WQPs (turbidity, chlorophyll-a, dissolved oxygen, and fluorescent dissolved organic matter) using both S2 and L89 data across five different locations in the United States. KAN consistently demonstrated statistically superior performance compared to RF and ANN across all WQPs, achieving approximately 5% lower MAPE and 6% higher R<sup>2</sup> than ANN, the second-best model. Cross-regional assessment of the ML models further demonstrated the superior generalization capability of the KAN model compared to the other models across both the S2 and L89 datasets. Moreover, S2 imagery outperformed L89 data across all WQPs. The findings prove the reliability of WQEye's workflow for the estimation of WQPs using RS data, but the final accuracy of outputs depends on the detectability of the target WQP in RS imagery, as non-optically active parameters like dissolved oxygen yielded lower estimation accuracies compared to optically active ones. WQEye is freely available at https://github.com/ATDehkordi/WQEye to promote applied RS applications in environmental monitoring and water resource management.</p>}},
author = {{Taheri Dehkordi, Alireza and Dazi, Mostafa and Naghibi, Amir and Hashemi, Hossein}},
issn = {{1574-9541}},
keywords = {{Artificial neural networks; Kolmogorov–Arnold networks; Machine learning; Random Forest; Satellite data; Water quality}},
language = {{eng}},
publisher = {{Elsevier}},
series = {{Ecological Informatics}},
title = {{WQEye (v1) : A Python-based software for machine learning-based retrieval of water quality parameters from Sentinel-2 and Landsat-8/9 remote sensing data aided by Google Earth Engine}},
url = {{http://dx.doi.org/10.1016/j.ecoinf.2026.103692}},
doi = {{10.1016/j.ecoinf.2026.103692}},
volume = {{95}},
year = {{2026}},
}