Bias Correction of Short-Term Minimum and Maximum Temperature Forecasts of the WRF Model by Using the Pursuit Machine

The importance of accurate forecasting in agricultural hydrometeorology is clear. This research is an approach towards the use of a tracking machine with a hidden layer for error prediction at stationary points. The predicted error will be used to modify the model output. One of the strengths of this method is the use of a meteorological variable such as maximum and minimum temperature in applications. A tracking machine with a hidden layer tracks the time series of the short-term prediction error of the maximum and minimum temperature of the model with the kernel of trigonometric functions, which is formulated as follows:
 
It provides an error prediction that will effectively modify the model prediction. This machine is compact in terms of computing. The value of the standard deviation of the statistical population of the maximum temperature during the period was 10 celsius, which shows a significant improvement from the value of 9.5 to 10.01 by the tracking machine. Also, the standard deviation of the minimum temperature was about 8.5 degrees Celsius, which was improved by the machine from 7.7 to 8.4 degrees Celsius. In this research, we use the skill score criterion, whose value will show that the skill score of the model for short-term maximum temperature has grown from a negative value with a leap to more than 0.8, which shows the significant impact of the machine in improving forecasting. The minimum temperature prediction skill score of the model will show an increase in the way of improving the prediction. The comparison of the obtained results shows that the skill score and RMSE of predicting the maximum and minimum temperature of the modification of the output of the model have increased significantly compared to the model. Also, the monthly change in the skill score indicates the effect of the chasing car on the ability to correct the forecast, especially for the short-term maximum temperature. Investigations will show that the modification of the model has a uniform overfitting in the studied period. In addition, a powerful index independent of the concept of accuracy size will be introduced and used as a method to check the reliability of the model and tracking machine outputs, which indicates the level of confidence that can be had in the model and machine outputs. In this case, the reliability of the maximum and minimum temperature predictions and the significant growth of the index have shown stability in providing the output. After bias correction, the variability of the skill score has been significantly reduced, and by reducing the amount of forecasting error, the reliability of the model forecasts has increased from 60% to more than 85%. Depending on the location and time, the WRF model's forecasting performance is different, but after bias correction, this dependence is removed, and forecasting in all regions and times has almost the same performance.