Prediction of river water temperature: a comparison between a new family of hybrid models and statistical approaches

Abstract

Abstract:River water temperature is a key physical variable controlling several chemical, biological and ecological processes. Its reliable prediction is a main issue in many environmental applications, which however is hampered by data scarcity, when using datademanding deterministic models, and modelling limitations, when using simpler statistical models. In this work we test a suite of models belonging to air2stream family, which are characterized by a hybrid formulation that combines a physical derivation of the key equation with a stochastic calibration of parameters. The air2stream models rely solely on air temperature and streamflow, and are of similar complexity as standard statistical models. The performances of the different versions of air2stream in predicting river water temperature are compared with those of the most common statistical models typically used in the literature. To this aim, a dataset of 38 Swiss rivers is used, which includes rivers classified into four different categories according to their hydrological characteristics: low-land natural rivers, lake outlets, snow-fed rivers and regulated rivers. The results of the analysis provide practical indications regarding the type of model that is most suitable to simulate river water temperature across different time scales (from daily to seasonal) and for different hydrological regimes. A model intercomparison exercise suggests that the family of air2stream hybrid models generally outperforms statistical models, while cross-validation conducted over a 30-year period indicates that they can be suitably adopted for long-term analyses.

StreamFlow 1.0: an extension to the spatially distributed snow model Alpine3D for hydrological modelling and deterministic stream temperature prediction

“…Statistical models require less input data and are usually easier to use, but their lack of physical basis is often seen as a limit to the validity of their predictions in the context of climate change studies (e.g. Piccolroaz et al, 2016). On the contrary, more credit is generally given to the long-term forecasts of the deterministic (2000) hourly, daily forested catchments in Canada UBC Morrison et al (2002) hourly large river basins GISS GCM Ferrari et al (2007) monthly large river basins SWAT Ficklin et al (2012) daily, monthly medium-to large-scale catchments MIKE-SHE MIKE11 Loinaz et al (2013) hourly (Ficklin et al, 2014) -than that of the statistical models.…”

“…On the contrary, more credit is generally given to the long-term forecasts of the deterministic (2000) hourly, daily forested catchments in Canada UBC Morrison et al (2002) hourly large river basins GISS GCM Ferrari et al (2007) monthly large river basins SWAT Ficklin et al (2012) daily, monthly medium-to large-scale catchments MIKE-SHE MIKE11 Loinaz et al (2013) hourly (Ficklin et al, 2014) -than that of the statistical models. It should be mentioned that an intermediate sort of model, referred to as hybrid, has recently been developed (Gallice et al, 2015;Toffolon and Piccolroaz, 2015) and shown by Piccolroaz et al (2016) to be suitable for climate change studies. As opposed to the separate simulation of discharge and stream temperature, the coupled modelling of the two offers new perspectives to investigate the effects of climate change on mountain hydrology (e.g.…”

Waning habitats due to climate change: the effects of changes in streamflow and temperature at the rear edge of the distribution of a cold-water fish

“…Their main objections to regressive methods arose when modelling reaches of regulated rivers, but this is not our case. In addition, our model improves the models that were tested in both studies (Arismendi et al, 2014;Piccolroaz et al, 2016). Performance indicators of our models produce good results, showing that the models are sufficiently competent.…”

“…Arismendi et al (2014) hold that regression models based on air temperature can be inadequate for projecting future stream temperatures because they are only surrogates for air temperature, whereas Piccolroaz et al (2016) argued that the adequacy depends on the hydrological regime, type of model and the timescale analysis. Their main objections to regressive methods arose when modelling reaches of regulated rivers, but this is not our case.…”

Peer Review #3 of "Modelling daily water temperature from air temperature for the Missouri River (v0.1)"

“…Rabi et al, 2015) have been developed successfully for data relating to different time 99 scales in the past years. Although these statistical models, which link water to air temperatures 100 provide quite simple approaches for water temperature prediction, other statistical models, such 101 as Box Jenkins and non-parametric models (Benyahya et al, 2007), and hybrid statistical-102 physical based models as air2water (Toffolon and Piccolroaz, 2015;Piccolroaz et al, 2016) Piotrowski et al, 2015). DeWeber 107 and Wagner (2014) developed an ANN ensemble model to predict the mean daily water 108 temperature using air temperature, landform attributes and forested land cover as predictors.…”