Skip to main content

Hybrid model of the near-ground temperature profile


The topic of the paper is modelling and prediction of atmospheric variables that are further used for prediction of the consequences of radioactive-material release to the atmosphere. Physics-based models of atmospheric dynamics provide an approximate description of the true nature of a dynamic system. However, the accuracy of the model’s short-term predictions and long-term forecasts, especially over complex terrain, decreases when the information at a micro-location is sought. Integration of a physics-based model with a statistical model for enhancing the prediction power is proposed in the paper. Gaussian Processes models can be used to identify the mapping between the system input and output measured values. With the given mapping function, we can provide one-step ahead prediction of the system output values together with its uncertainty, which can be used advantageously. In this paper, we combine a physics-based model with a Gaussian-process model to identify air temperature from measurements at different atmospheric surface layers as a dynamic system and to make short-term predictions as well as long-term forecasts.

This is a preview of subscription content, access via your institution.

Fig. 1
Fig. 2
Fig. 3
Fig. 4
Fig. 5
Fig. 6
Fig. 7
Fig. 8
Fig. 9
Fig. 10


  • Božnar MZ, Mlakar P, Grašič B (2012) Short-term fine resolution WRF forecast data validation in complex terrain in Slovenia. Int J Environ Pollut 50(1–4):12–21

    Article  Google Scholar 

  • Breznik B, Božnar MZ, Mlakar P, Tinarelli G (2003) Dose projection using dispersion models. Int J Environ Pollut 20(1–6):278–285

    Article  CAS  Google Scholar 

  • Emeis S, Schäfer K, Münkel C, Friedl R, Suppan P (2012) Evaluation of the interpretation of ceilometer data with RASS and radiosonde data. Bound Layer Meteorol 143(1):25–35

    Article  Google Scholar 

  • Fochesatto GJ (2015) Methodology for determining multilayered temperature inversions. Atmos Meas Techniq 8(5):2051–2060

    Article  Google Scholar 

  • Gangwar RK, Mathur AK, Gohil B, Basu S (2014) Neural network based retrieval of atmospheric temperature profile using AMSU—a observations. Int J Atmos Sci.

    Article  Google Scholar 

  • Gradišar D, Grašič B, Božnar MZ, Mlakar P, Kocijan J (2016) Improving of local ozone forecasting by integrated models. Environ Sci Pollut Res 23(18):18439–18450

    Article  CAS  Google Scholar 

  • Grašič B, Božnar MZ, Mlakar P, Kocijan J (2018) Validation of numerically forecasted vertical temperature profile with measurements for dispersion modelling. Int J Environ Pollut 64(1–3):22–34

    Article  Google Scholar 

  • Kocijan J (2016) Modelling and control of dynamic systems using Gaussian process models. Springer International Publishing, Cham

    Book  Google Scholar 

  • Kocijan J, Hančič M, Petelin D, Božnar MZ, Mlakar P (2015) Regressor selection for ozone prediction. Simul Model Pract Theory 54:101–115

    Article  Google Scholar 

  • Kocijan J, Gradišar D, Božnar MZ, Grašič B, Mlakar P (2016) On-line algorithm for ground-level ozone prediction with a mobile station. Atmos Environ 131:326–333

    Article  CAS  Google Scholar 

  • Kocijan J, Gradišar D, Stepančič M, Božnar MZ, Grašič B, Mlakar P (2018) Selection of the data time interval for the prediction of maximum ozone concentrations. Stoch Environ Res Risk Assess 32(6):1759–1770

    Article  Google Scholar 

  • Kononenko I, Šimec E, Robnik-Šikonja M (1997) Overcoming the myopia of inductive learning algorithms with RELIEFF. Appl Intell 7(1):39–55

    Article  Google Scholar 

  • Korn GA (2007) Advanced dynamic-system simulation: model-replication techniques and Monte Carlo simulation. Wiley, London

    Book  Google Scholar 

  • MacKay DJ (1998) Introduction to Gaussian processes. NATO ASI Series F Comput Syst Sci 168:133–166

    Google Scholar 

  • Matthews DG, Alexander G, Van Der Wilk M, Nickson T, Fujii K, Boukouvalas A, León-Villagrá P, Ghahramani Z, Hensman J (2017) GPflow: a Gaussian process library using TensorFlow. J Mach Learn Res 18(1):1299–1304

    Google Scholar 

  • May R, Dandy G, Maier H (2011) Review of input variable selection methods for artificial neural networks. In: Suzuki K (ed) Artificial neural networks—methodological advances and biomedical applications. InTech, Rijeka, pp 19–44

    Google Scholar 

  • Ramaswamy V, Hurrell J, Meehl G, Phillips A, Santer B, Schwarzkopf M, Seidel D, Sherwood S, Thorne P, Karl T et al (2006) Why do temperatures vary vertically (from the surface to the stratosphere) and what do we understand about why they might vary and change over time? In: Temperature trends in the lower atmosphere: steps for understanding and reconciling differences, vol synthesis and assessment product 1.1, US Climate Change Science Program/Subcommittee on Global Change Research, Washington, D.C., chap 1, pp 15–28

  • Rasmussen CE, Nickisch H (2015) The GPML toolbox version 3.6

  • Rasmussen CE, Williams CK (2006) Gaussian processes for machine learning. MIT Press, Cambridge

    Google Scholar 

  • Rekhy A, Shneider MN, Miles RB (2018) Temperature profiling of the atmosphere by filtered Rayleigh scattering. In: 2018 Aerodynamic measurement technology and ground testing conference, p 3630

  • Ren WW, Yang T, Huang CS, Xu CY, Shao QX (2018) Improving monthly streamflow prediction in alpine regions: integrating hbv model with bayesian neural network. Stoch Environ Res Risk Assess 32(12):3381–3396

    Article  Google Scholar 

  • Shi JQ, Choi T (2011) Gaussian process regression analysis for functional data. CRC Press, Boca Raton

    Book  Google Scholar 

  • Skamarock WC, Klemp JB, Dudhia J, Gill DO, Barker M, Duda KG, Huang XY, Wang W, Powers JG (2008) A description of the advanced research WRF version 3. Tech. rep., National Center for Atmospheric Research

  • Von Stosch M, Oliveira R, Peres J, de Azevedo SF (2014) Hybrid semi-parametric modeling in process systems engineering: past, present and future. Comput Chem Eng 60:86–101

    Article  CAS  Google Scholar 

  • Worsnop RP, Scheuerer M, Hamill TM, Lundquist JK (2018) Generating wind power scenarios for probabilistic ramp event prediction using multivariate statistical post-processing. Wind Energy Sci 3(1):371–393

    Article  Google Scholar 

  • Zhang Y, Bocquet M, Mallet V, Seigneur C, Baklanov A (2012a) Real-time air quality forecasting, part I: history, techniques, and current status. Atmos Environ 60:632–655

    Article  CAS  Google Scholar 

  • Zhang Y, Bocquet M, Mallet V, Seigneur C, Baklanov A (2012b) Real-time air quality forecasting, part II: state of the science, current research needs, and future prospects. Atmos Environ 60:656–676

    Article  CAS  Google Scholar 

Download references


The authors acknowledge the project “Method for the forecasting of local radiological pollution of atmosphere using Gaussian process models”, ID L2-8174, and research core funding No. P2-0001, which were financially supported by the Slovenian Research Agency. We are grateful to the Krško NPP for the measurement data from their automatic measuring system. The discussions and technical assistance with data processing by Martin Stepančič are gratefully acknowledged.

Author information

Authors and Affiliations


Corresponding author

Correspondence to Juš Kocijan.

Additional information

Publisher's Note

Springer Nature remains neutral with regard to jurisdictional claims in published maps and institutional affiliations.

Appendix: Performance measures

Appendix: Performance measures

We use the following statistical measures for the assessment:

  • The normalised root-mean-square error—NRMSE

    $$\begin{aligned} {{\mathrm {NRMSE}}}=1-\frac{\Vert {{\mathbf {y}}} -\mathbf {\mu }\Vert ^2}{\Vert {{\mathbf {y}}}-E({{\mathbf {y}}})\Vert ^2}, \end{aligned}$$

    where \({{\mathbf {y}}}\) the vector of validation values, \(\mathbf {\mu }\) the vector of mean predicted values, \(E({{\mathbf {y}}})\) the mean value of \({{\mathbf {y}}}\).

    NRMSE is 1 for a perfect match and \(-\infty\) for a very bad match of the validation and mean predicted values.

  • The standardised mean-squared error—SMSE

    $$\begin{aligned} {{\mathrm {SMSE}}}=\frac{1}{N}\frac{\sum _{i=1}^N(E({\hat{y}}_i)-y_i)^2}{\sigma _y^2}, \end{aligned}$$

    where \(\sigma _y^2\) the variance of the observations.

    SMSE is a frequently used standardised measure for the accuracy of predictions’ mean values with values between 0 and 1, where the value 0 is the result of a perfect model.

  • The Pearson’s correlation coefficient—PCC:

    $$\begin{aligned} {{\mathrm {PCC}}}=\frac{\sum _{i=1}^N(E({\hat{y}}_i) -E(\hat{{{\mathbf {y}}}}))(y_i-E({{\mathbf {y}}}))}{N\sigma _y\sigma _{{\hat{y}}}}, \end{aligned}$$

    where \(E(\hat{{{\mathbf {y}}}})\) the expectation, i.e., the mean value, of the vector of predictions, \(\sigma _y\) the standard deviation of the observations, \(\sigma _{{\hat{y}}}\) the standard deviation of the predictions.

    PCC is a measure of associativity and is not sensitive to bias. Its value is between \(-1\) and \(+1\), with ideally linearly correlated values resulting in a value of 1.

  • The mean standardised log loss—MSLL (Rasmussen and Williams 2006):

    $$\begin{aligned} {{\mathrm {MSLL}}}= \,& {} \frac{1}{2N}\sum _{i=1}^N\left[ \ln (\sigma _i^2)+ \frac{(E({\hat{y}}_i)-y_i)^2}{\sigma _i^2}\right] \\&-\frac{1}{2N}\sum _{i=1}^N\left[ \ln (\sigma _y^2)+ \frac{(y_i-E({{\mathbf {y}}}))^2}{\sigma _y^2}\right] , \end{aligned}$$

    where \(\sigma ^2_i\) the prediction variance in the i-th step, \(E({{\mathbf {y}}})\) the expectation, i.e., the mean value, of the vector of the observations.

    MSLL is a standardised measure suited to predictions in the form of random variables. It weights the prediction error more heavily when it is accompanied by a smaller prediction variance. The MSLL is approximately zero for the not very good models and negative for the better ones.

Rights and permissions

Reprints and Permissions

About this article

Verify currency and authenticity via CrossMark

Cite this article

Kocijan, J., Perne, M., Mlakar, P. et al. Hybrid model of the near-ground temperature profile. Stoch Environ Res Risk Assess 33, 2019–2032 (2019).

Download citation

  • Published:

  • Issue Date:

  • DOI:


  • Hybrid model
  • Vertical temperature profile
  • Physics-based model
  • Statistical modelling
  • Gaussian-process model