MARSpline model for lead seven-day maximum and minimum air temperature prediction in Chennai, India

Abstract

In this study, a Multivariate Adaptive Regression Spline (MARS) based lead seven days minimum and maximum surface air temperature prediction system is modelled for station Chennai, India. To emphasize the effectiveness of the proposed system, comparison is made with the models created using statistical learning technique Support Vector Machine Regression (SVMr). The analysis highlights that prediction accuracy of MARS models for minimum temperature forecast are promising for short term forecast (lead days 1 to 3) with mean absolute error (MAE) less than 1 °C and the prediction efficiency and skill degrades in medium term forecast (lead days 4 to 7) with slightly above 1 °C. The MAE of maximum temperature is little higher than minimum temperature forecast varying from 0.87 °C for day-one to 1.27 °C for lag day-seven with MARS approach. The statistical error analysis emphasizes that MARS models perform well with an average 0.2 °C of reduction in MAE over SVMr models for all ahead seven days and provide significant guidance for the prediction of temperature event. The study also suggests that the correlation between the atmospheric parameters used as predictors and the temperature event decreases as the lag increases with both approaches.