Assessment of spatiotemporal dynamics of diurnal fog occurrence in subtropical montane cloud forests

Montane cloud forests (MCFs) are frequently immersed in low-altitude clouds or fog. The diurnal (defined as 07:00-16:50 local time) cycle of fog is particularly important for regulating the carbon, water and energy cycles of these ecosystems. Elevated temperatures may alter the spatiotemporal dynamics of fog and have cascading impacts on MCFs. Therefore, systematic monitoring of fog occurrence is essential for understanding the ramifications of climate change on these unique forests. This study aims to quantify three years (2018-2020) of diurnal fog occurrence with high spatiotemporal (5 km, 10 minutes) resolutions in subtropical MCF in northeast Taiwan. Four open-sky observation stations were installed along an elevation gradient (1151, 1514, 1670 and 1811 m a.s.l.) within the cloud band to record meteorological data including rainfall, air temperature and relative humidity. We also acquired spatiotemporally-corresponding photosynthetic photon flux density from the geostationary Himawari-8 satellite and derived solar zenith angle for each station. We utilized these ground and satellite meteorological attributes to model fog occurrence using seven machine learning methods. By referring to time-lapse images, the performance of random forests was determined to be superior compared to other approaches and was therefore selected to quantify spatiotemporal dynamics of fog occurrence. Fog was determined to be more abundant in terms of probability, frequency and duration in mid-elevations when compared to the lower and higher ends. Temporal analysis demonstrated that overall seasonality was pronounced with higher fog abundance in the afternoons and cold months but varied from station to station. In addition, three-year fog duration and event variability for each month were notable for all stations; the foggiest station was at 1670 m a.s. l. This study demonstrates the feasibility of using machine learning to quantify spatiotemporal dynamics of fog using cross-scale meteorological attributes, which may facilitate monitoring the impact of climate change on MCFs.

Automated summary

This study quantifies the diurnal fog occurrence in subtropical mountain cloud forests in northeast Taiwan using machine learning methods. The results show that fog is more abundant in mid-elevations, with pronounced seasonality and higher fog abundance in the afternoons and cold months. Machine learning can be an effective tool for monitoring the impact of climate change on cloud forests.