عنوان مقاله [English]
Weather forecasting and monitoring systems based on numerical weather forecasting models have been increasingly used to manage issues related to meteorology and agriculture. Using more accurate daily average wind speed (10m) and relative humidity forecasts can be helpful in this regard. But systematic and random errors in the model affect the accuracy of forecasts. In this study, the model errors during the 5 and 14 days training period in the same climate areas on the points of the network where the observations are available were calculated. Then the errors were generalized on all points of the network using the cokriging interpolation method. This preserves the model forecasts for other points of the network and only error values are applied to them. To better evaluate the model, the spatial and temporal distribution of daily average wind speed (10m) and relative humidity forecast errors were also investigated over Iran. Observed daily wind speed and relative humidity data from 560 meteorological stations for the period 1/11/2019 to 1/2/2021 were used to evaluate the WRF model performance. The WRF model was run daily at 12UTC, with a forecast time of 120 hours, and first 12 hours of each run was consider as the model spin-up time and was not used in errors calculation. In order to correct wind speed and relative humidity forecast errors for next three days (forecasts of 36, 60 and 84 hours), the forecasts for each day in the period of 11/1/ 2019 to 1/2/2021, was extracted from the model outputs. In order to evaluate the error correction method, the skill score index was used. The validation results of the error correction method showed that the absolute mean error value, correlation coefficient and RMSE improved after the error correction compared to results that were before the error correction, which showed that the error correction method can be used for other network points that did not contain observational data. In general after correction, the RMSE for wind speed and relative humidity forecasts could decrease by 13% and 18%, and the skill score could increase to a maximum of 160% and 308%, respectively. Value of correlation coefficient, after correcting the model error, was significantly increased, compared to the raw model output. In general skill score for the raw wind speed and relative humidity forecast for more than 50% of the days was more than -0.5 and -0.3, but after corrections were increased to 0.2, 0.4 respectively. Without exception, all climatic regions after error correction have higher skill scores than before error correction, so that the model skill score for most climatic regions after error correction was reached above zero for more than 75% of the days. The results showed that errors of the model in different months, places and climatic zones did not have a uniform distribution. In general, the model underestimated the wind speed and overestimated the relative humidity in most areas. In general, the lowest skill scores for relative humidity forecasts occurred in the colder months of November to February in most climatic zones. The 14-day error correction method did not improve the modeling skill score much compared to the 5-day error correction method, and they acted almost similarly. Knowing the spatial and temporal distribution of model forecast error can be helpful for researchers to have an overview of the areas (and months) where the model forecast error can be high or low.