Tracking fog occurrence and drivers in a mountainous Costa Rican rainforest using phenological camera imagery

Fog patterns were determined using web camera images collected at half-hour intervals and uploaded to the PhenoCam network. These images were analyzed to determine fog presence and intensity using the K-Means iterative algorithm, as implemented in Python. Atmospheric conditions were clustered into five different categories: clear, overcast, light fog, medium fog, and heavy fog. Ecohydrological variable data was gathered from sensors placed within the forest and at a nearby weather station. The quantified fog data was then compared with the ecohydrological variables; the diurnal patterns of fog and precipitation were determined over the entire dataset and during dry and wet months. In April, rain was present 2% of the time and fog was present in 68% of the images and in September rain was present 18% of the time and fog was present in 40% of the images. Occurrence of heavy fog conditions are consistently higher in January and December but daily appeared to be highest in the early mornings. A generalized linear model was used to relate fog occurrence with temperature, relative humidity, solar radiation, and wind speed.