Improved Turbidity Estimation from Local Meteorological Data for Solar Resourcing and Forecasting Applications

Abstract

This work presents a new method to estimate atmospheric turbidity with improved accuracy in estimating clear-sky irradiance. The turbidity is estimated by machine learning algorithms using commonly measured meteorological data including ambient air temperature, relative humidity, wind speed and atmospheric pressure. The estimated turbidity is then served as input to the Ineichen-Perez clear-sky model to estimate clear-sky global horizontal irradiance (GHI) and direct normal irradiance (DNI). When compared with the original Ineichen-Perez model who uses interpolated turbidity from the monthly climatological means, our turbidity estimation better captures its daily, seasonal, and annual variations. When using the improved turbidity estimation in the Ineichen-Perez model, the root mean square error (RMSE) of clear-sky GHI is reduced from 24.02 W/m 2 to 9.96 W/m 2 . The RMSE of clear-sky DNI is deceased from 76.40 W/m 2 to 29.53 W/m 2 . The presented method is also capable to estimate turbidity in partially cloudy days with improved accuracy, evidenced by that the corresponding estimated clear-sky irradiance has smaller deviation from measured irradiance in the cloudless time instants. In sum, the proposed method brings new insights about turbidity estimation in both clear and partially cloudy days, providing support to solar resourcing and forecasting.

Keywords: Clear-sky irradiance, turbidity estimation, meteorological measurements, machine learning methods