Derivation of consistent, continuous daily river temperature data series by combining remote sensing and water temperature models

Tavares, Matheus Henrique; Cunha, Augusto Hugo Farias da; Marques, David da Motta; Ruhoff, Anderson Luís; Fragoso, Carlos Ruberto; Munar, Andrés Maurício; Bonnet, Marie‐Paule

doi:10.1016/j.rse.2020.111721

Cited by 23 publications

(13 citation statements)

References 97 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…Most publications reporting findings regarding FNN performance for stream water temperature tested only a small set of FNN hyperparameter combinations, mostly chosen by trial and error (e.g. Piotrowski et al, 2015;Rabi et al, 2015;Abba et al, 2017;Zhu et al, 2018;Temizyurek and Dadaser-Celik, 2018;Zhu et al, 2019d). Our results show the extremely an large influence of hyperparameters, therefore rendering any trialand-error approach insufficient and certainly non-optimal.…”

Section: Discussionmentioning

confidence: 79%

“…It was already applied in multiple studies over a range of catchments and generally had an improved performance compared to linear regression models (e.g. Piccolroaz et al, 2016;Yang and Peterson, 2017;Piotrowski and Napiorkowski, 2018;Zhu et al, 2019d;Piotrowski and Napiorkowski, 2019;Tavares et al, 2020). air2stream uses the inputs T a and Q and was derived from simplified physical relationships expressed as ordinary differential equations for heat-budged processes.…”

Section: Air2streammentioning

confidence: 99%

“…FNNs have already been applied in numerous stream water temperature prediction studies, which range from simple one hidden layer models (e.g. Risley et al, 2003;Bélanger et al, 2005;Chenard and Caissie, 2008;McKenna et al, 2010;Hadzima-Nyarko et al, 2014;Rabi et al, 2015;Zhu et al, 2018;Temizyurek and Dadaser-Celik, 2018) to multiple hidden-layer models with a hyperparameter optimization (Sahoo et al, 2009) and to more complex FNN (hybrid) architectures and ensembles of FNNs (e.g. DeWeber and Wagner, 2014;Piotrowski et al, 2015;Abba et al, 2017;Graf et al, 2019;Zhu et al, 2019a, b).…”

Section: Introductionmentioning

confidence: 99%

See 2 more Smart Citations

Machine-learning methods for stream water temperature prediction

Feigl

Lebiedzinski

Herrnegger

et al. 2021

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

Abstract. Water temperature in rivers is a crucial environmental factor with the ability to alter hydro-ecological as well as socio-economic conditions within a catchment. The development of modelling concepts for predicting river water temperature is and will be essential for effective integrated water management and the development of adaptation strategies to future global changes (e.g. climate change). This study tests the performance of six different machine-learning models: step-wise linear regression, random forest, eXtreme Gradient Boosting (XGBoost), feed-forward neural networks (FNNs), and two types of recurrent neural networks (RNNs). All models are applied using different data inputs for daily water temperature prediction in 10 Austrian catchments ranging from 200 to 96 000 km2 and exhibiting a wide range of physiographic characteristics. The evaluated input data sets include combinations of daily means of air temperature, runoff, precipitation and global radiation. Bayesian optimization is applied to optimize the hyperparameters of all applied machine-learning models. To make the results comparable to previous studies, two widely used benchmark models are applied additionally: linear regression and air2stream. With a mean root mean squared error (RMSE) of 0.55 ∘C, the tested models could significantly improve water temperature prediction compared to linear regression (1.55 ∘C) and air2stream (0.98 ∘C). In general, the results show a very similar performance of the tested machine-learning models, with a median RMSE difference of 0.08 ∘C between the models. From the six tested machine-learning models both FNNs and XGBoost performed best in 4 of the 10 catchments. RNNs are the best-performing models in the largest catchment, indicating that RNNs mainly perform well when processes with long-term dependencies are important. Furthermore, a wide range of performance was observed for different hyperparameter sets for the tested models, showing the importance of hyperparameter optimization. Especially the FNN model results showed an extremely large RMSE standard deviation of 1.60 ∘C due to the chosen hyperparameters. This study evaluates different sets of input variables, machine-learning models and training characteristics for daily stream water temperature prediction, acting as a basis for future development of regional multi-catchment water temperature prediction models. All preprocessing steps and models are implemented in the open-source R package wateRtemp to provide easy access to these modelling approaches and facilitate further research.

show abstract

Section: Discussionmentioning

confidence: 79%

Section: Air2streammentioning

confidence: 99%

Section: Introductionmentioning

confidence: 99%

See 1 more Smart Citation

Machine-learning methods for stream water temperature prediction

Feigl

Lebiedzinski

Herrnegger

et al. 2021

Hydrol. Earth Syst. Sci.

View full text Add to dashboard Cite

show abstract

“…Furthermore, the ACPC that uses profiles with 1 • × 1 • (~100 km) horizontal resolution showed good results. Although these profiles have a coarser horizontal resolution, previous studies have been finding satisfactory results using the ACPC, even surpassing other methods [8,16,38,52,53]. It is important to note that in this study the WRF resulting profiles were extracted at 12 UTC to match with the available radiosonde data.…”

Section: Application To Rte-based Lst Retrievalmentioning

confidence: 98%

Atmospheric Correction of Thermal Infrared Landsat Images Using High-Resolution Vertical Profiles Simulated by WRF Model

Diaz

Santos

Käfer

et al. 2021

The 4th International Electronic Conference on Atmospheric Sciences

View full text Add to dashboard Cite

Atmospheric profiles are key inputs in correcting the atmospheric effects of thermal infrared (TIR) remote sensing data for estimating land surface temperature (LST). This study is a first insight into the feasibility of using the weather research and forecasting (WRF) model to provide high-resolution vertical profiles for LST retrieval. WRF numerical simulations were performed to downscaling NCEP climate forecast system version 2 (CFSv2) reanalysis profiles, using two nested grids with horizontal resolutions of 12 km (G12) and 3 km (G03). We investigated the use of these profiles in the atmospheric correction of TIR data applying the radiative transfer equation (RTE) inversion single-channel approach. The MODerate resolution atmospheric TRANsmission (MODTRAN) model and Landsat 8 TIRS10 (10.6–11.2 µm) band were taken for the method application. The accuracy evaluation was performed using in situ radiosondes in Southern Brazil. We included in the comparative analysis the atmospheric correction parameter calculator (ACPC; NASA) web tool and profiles directly from the NCEP CFSv2 reanalysis. The atmospheric correction parameters from ACPC, followed by CFSv2, had better agreement with the ones calculated using in situ radiosondes. When applied into the RTE to retrieve LST, the best results (RMSE) were, in descending order: CSFv2 (0.55 K), ACPC (0.56 K), WRF G12 (0.79 K), and WRF G03 (0.82 K). The findings suggest that increasing the horizontal resolution of reanalysis profiles does not particularly improve the accuracy of RTE-based LST retrieval. However, the WRF results are yet satisfactory and promising, encouraging further assessments. We endorse the use of the well-known ACPC and also recommend the NCEP CFSv2 reanalysis profiles for TIR remote sensing atmospheric correction and LST single-channel retrieval.

show abstract

“…Air2stream (Toffolon and Piccolroaz, 2015) is a hybrid model for predicting river water temperature, which combines a physical based structure with a stochastic parameter calibration. It was already applied in multiple studies over a range of catchments and generally had an improved performance compared to linear regression models (e.g., Piccolroaz et al, 2016;Yang and Peterson, 2017;Piotrowski and Napiorkowski, 2018;Zhu et al, 2019d;Piotrowski and Napiorkowski, 2019;Tavares et al, 2020). Air2stream uses the inputs T a and Q and was derived from simplified physical relationships expressed as ordinary differential equations for heat budged processes.…”

Section: Air2streammentioning

confidence: 99%

Comment on hess-2020-670

Heddam¹

2021

View full text Add to dashboard Cite

Water temperature in rivers is a crucial environmental factor with the ability to alter hydro-ecological as well as socio-economic conditions within a catchment. The development of modelling concepts for predicting river water temperature is and will be essential for an effective integrated water management and the development of adaptation strategies to future global changes (e.g. climate change). This study tests the performance of 6 different machine learning models: step-wise linear regression, Random forest, eXtreme Gradient Boosting (XGBoost), Feedforward neural networks (FNN), and two types of Recurrent neural networks (RNN). All models are applied using different data inputs for daily water temperature prediction in 10 Austrian catchments ranging from 200 km 2 to 96000 km 2 and exhibiting a wide range of physiographic characteristics. The evaluated input data sets include combinations of daily means of air temperature, runoff, precipitation and global radiation.Bayesian optimization is applied to optimize the hyperparameters of all applied machine learning models. To make the results comparable to previous studies, two widely used benchmark models are applied additionally: linear regression and air2stream.With a mean root mean squared error (RMSE) of 0.55 • C the tested models could significantly improve water temperature prediction compared to linear regression (1.55 • C) and air2stream (0.98 • C). In general, the results show a very similar performance of the tested machine learning models, with a median RMSE difference of 0.08 • C between the models. From the 6 tested machine learning models both FNNs and XGBoost performed best in 4 of the 10 catchments. RNNs are the best performing models in the largest catchment, indicating that RNNs are mainly performing well when processes with long-term dependencies are important. Furthermore, a wide range of performance was observed for different hyperparameter sets for the tested models, showing the importance of hyperprameter optimization. Especially the FNN model results showed an extremely large RMSE standard deviation of 1.60 • C due to the chosen hyperparamerters.This study evaluates different sets of input variables, machine learning models and training characteristics for daily stream water temperature prediction, acting as a basis for future development of regional multi-catchment water temperature prediction models. All preprocessing steps and models are implemented into the open source R package wateRtemp, to provide easy access to these modelling approaches and facilitate further research. IntroductionWater temperature in rivers should not be considered only as a physical property, since it is a crucial environmental factor and a substantial key element for water quality and aquatic habitats. In particular, it influences riverine species by governing e.g.

show abstract

Derivation of consistent, continuous daily river temperature data series by combining remote sensing and water temperature models

Cited by 23 publications

References 97 publications

Machine-learning methods for stream water temperature prediction

Machine-learning methods for stream water temperature prediction

Atmospheric Correction of Thermal Infrared Landsat Images Using High-Resolution Vertical Profiles Simulated by WRF Model

Comment on hess-2020-670

Contact Info

Product

Resources

About