Machine learning algorithms predict bark coverages on saguaro cacti (Carnegiea gigantea)

Abstract

Bark coverages lead to the premature death of saguaro cacti (Carnegiea gigantea). Bark coverages on surfaces of more than 20 species of tall, long-lived cacti occur throughout the Americas. This study analyzed data of bark coverages on twelve surfaces on 1149 saguaro cacti from four sampling periods over a period of 23 years. Bark coverage starts on equator-facing surfaces and progresses to polar-facing surfaces. For this population, 45 % of all cacti were dead by the end of the 23-year period of study. In order to understand how bark coverages are related to cactus death, four machine learning programs were used to analyze the data obtained. Decision trees predicted rates of bark coverages and death with time as a variable with probabilities of 90 % and 99 %, respectively. Trough surfaces were better predictors of bark coverage and cactus death than crest surfaces. Left troughs of both polar- and west-facing surfaces were the best predictors of bark coverage and cactus death with probabilities above 99 %. The four machine learning programs gave complimentary results to present a unified view to predict future bark coverages over time and predict eventual cactus death with more than 99 % accuracy.