Wind power all over the world are being popularizing unlike decrease in conventional sources due to environmental issues. However, power acquired from wind is not stable during day and night, which means that intermittent due to nature of the source. Forecasting in wind power plant is very challenging compared to forecasting of production of conventional power plant. Although there are many robust and site-specific models in order to forecast wind power accurately, decrease of deviation in wind power forecasting by using statistical, physical and hybrid models is still open to new approaches. In this study, four different forecast models based numerical weather prediction (NWP) models for three different wind farms which have different atmospheric conditions are examined to improve wind farm-based power forecasting. For this purpose, wind power forecasting of the providers was categorized based on atmospheric effects, which are site temperature and turbulence. Results have been compared with real time power production from wind turbine supervisory control and data acquisition (SCADA) system. Afterwards, new method based on selecting best provider for specific condition was developed by considering atmospheric effects on power forecasting. It should be noted that the method is an engineering approach, not a new forecast model. In many cases, newly developed method has succeeded to outperform in comparison to results belonging to forecast providers. Hourly and daily wind power forecasting that have significant role in electricity market has been improved for selected wind farms by the help of an engineering approach used in this study. Same method is also implementable to another wind farm if required inputs exist.