Discriminating Fog and Clouds in the Namib using Reanalysis Data

Fog is a frequent phenomenon in the Namib Desert and enables the life of a highly specialised flora and fauna. The fog day frequency varies from the coast to inland areas because the offshore stratus is advected at different elevations, becoming fog when touching the ground. Two situations can be identified from the perspective of a coastal station, fog occurrence and low cloud occurrence, but the drivers leading to the one or the other are unknown.
The presented study is a first attempt to fill this knowledge gap and develop a conceptual understanding of the different processes. To identify meteorological drivers, median composites of reanalysis (ERA5) variables were prepared for fog days and for low cloud days, using satellite-derived low cloud cover and station measurements as ground truth. The composites revealed that the main drivers of fog are the synoptic pressure, the temperature inversion at the Namibian coast and the local wind system. A conceptual framework was developed for how these drivers lead to fog or low cloud occurrence. In a second step, the potential of three different machine learning algorithms was assessed, given the task to differentiate fog and low cloud events based on meteorological data.
... mehr
The best classification skill was shown by the logistic regression (67 % accuracy, 69 % F1-score), followed by random forest (64.21 % accuracy, 63 % F1-score). A convolutional neural network overfitted instantly which was attributed to the small sample size.
The presented conceptual framework of the different processes leading to fog and low cloud occurrence is the first explicit disentangling of mechanistic drivers and seasonal variation of fog and low clouds in the Namib. The results show that fog and low clouds can be differentiated based on the drivers by machine learning, which opens the perspective of modeling the fog regime in the past and future Namib Desert.