A Local Index of Cloud Immersion in Tropical Forests Using Time-Lapse Photography

Abstract:

Data on the frequency, duration and elevation of cloud immersion is essential to improve estimates of cloud water deposition in water budgets in cloud forests. Here, we present a methodology to detect local cloud immersion in remote tropical forests using time-lapse photography. A simple approach is developed to detect cloudy conditions in photographs within the canopy where image depth during clear conditions may be less than 10 meters and moving leaves and branches and changes in lighting are unpredictable. A primary innovation of this study is that cloudiness is determined from images without using a reference clear image and without minimal threshold value determination or human judgment for calibration. Five sites ranging from 600 to 1000 meters elevation along a ridge in the Luquillo Critical Zone Observatory, Puerto Rico were each equipped with a trail camera programmed to take an image every 30 minutes since March 2014. Images were classified using four selected cloud-sensitive image characteristics (SCICs) computed for small image regions: contrast, the coefficient of variation and the entropy of the luminance of each image pixel, and image colorfulness. K-means clustering provided reasonable results to discriminate cloudy from clear conditions. Preliminary results indicate that 79-94% (daytime) and 85-93% (nighttime) of validation images were classified accurately at one open and two closed canopy sites. The euclidian distances between SCICs vectors of images during cloudy conditions and the SCICs vector of the centroid of the cluster of clear images show potential to quantify cloud density in addition to immersion. The classification method will be applied to determine spatial and temporal patterns of cloud immersion in the study area. The presented approach offers promising applications to increase observations of low-lying clouds at remote mountain sites where standard instruments to measure visibility and cloud base may not be practical.