Hour-scale cloud cover forecasting using WRF and an infrared all-sky imager

Numerical weather prediction (NWP) models such as the Weather Research and Forecasting (WRF) model allow to forecast cloud processes at high spatio-temporal resolutions of a few kilometres and minutes. Nevertheless, WRF is known to underestimate the presence of clouds, which has negative impacts on solar irradiance forecasts for example. Therefore, drawing the forecasts closer to observations of clouds can be an efficient means for improving hour-scale cloudiness forecasts. Determining precise initial conditions from both observations and model output in terms of clouds is a challenge. This study explores the use of four-dimensional data assimilation (FDDA) nudging to take into account cloud cover observations from an infrared all-sky imager. Experiments with two WRF domains (9 km and 3 km grid spacing), driven by Global Forecasting System (GFS) forecasts, are performed over Central Europe. The cloud cover observations are determined by the Sky InSight, an all-sky infrared imager, installed at the Lindenberg Meteorological Observatory – Richard-Assmann-Observatory (MOL-RAO) in eastern Germany. Compared to observations in the visible range, this instrument has the advantage to deliver cloud cover observations of the same quality at day and night time. The observations are compared to cloud cover values from the WRF background. Depending on the difference between the two, the WRF background humidity profile at the location of the imager is either increased, decreased or unchanged to obtain a new profile for the nudging. Cloud base height values are determined by a ceilometer and empirical values are used for the cloud top height to limit the vertical extent of the derived humidity profiles to be nudged. Reuniwatt’s custom Meteosat Second Generation (MSG) Satellite Application Facility NoWCasting (SAFNWC) cloud products are used to evaluate the impact of the nudging on WRF forecasts of cloud cover. The fast and cost-effective nudging method leads to an improvement of WRF cloud cover forecasts. It could be easily upscaled to a large amount of ground-based camera observations.