Authors: Mohammad Zeynoddin, Hossein Bonakdari
Download file: https://library.csbe-scgab.ca/docs/meetings/2021/CSBE21530.pdf
Published in: CSBE-SCGAB Technical Conferences » 5th CIGR and AGM Quebec City 2021 » Special Session on Hydrological Modelling: a Tool for Resilient and Sustainable Agriculture
Download RAW file:
Description: Soil temperature plays a critical role in the fields of hydrology, environmental science, ecology, soil science, meteorology, geotechnics and agronomy. In the present study the main purpose is to evaluate the performances of a novelty structured hybrid support vector machine model, in prediction of daily soil temperatures at six different depths from 5 to 100 cm. Wavelet analysis is also applied in order to pre-process the time series of meteorological data and obtain more accurate results. The modeling is carried out using the widely available input variables of maximum air temperature (Tmax), minimum air temperature (Tmin), evaporation from pan (EP) and sunshine duration (SD). The results are compared with Multi-Layer Perceptron Artificial Neural Network (MLP-ANN). All the criteria indicated that the results obtained by the SVM-WAVE are more precise than those of the MLP-ANN. The SVM-WAVE by average indices of coefficient of determination R2 0.953, Root mean squared error RMSE 1.928, Mean absolute error MAE 1.452, Mean absolute percentage error MAPE 9.970%, Average performance error APE 6.318% and Nash-Sutcliffe efficiency NS 94.865 outperformed the MLP-ANN by R2 0.947, RMSE 2.407, MAE 1.858, MAPE 11.913%, APE 8.123% and NS 93.022. The proposed method shows promising results for obtaining precise ST models with comprehensible conventional methods.
Conference name: 5th CIGR International Conference and CSBE-SCGAB AGM 2021, Quebec City,QC, 11-14 May 2021.
Session name: SPECIAL SESSION Hydrological modelling II
Publication type: Presentation
Language 1: en
Rights: Canadian Society for Bioengineering