Machine Learning: Crown Diameter Predictive Modeling for Open-Grown Trees in the Cerrado Biome, Brazil

Abstract

The Brazilian Cerrado biome is a hotspot due to its ecological importance and high diversity of fauna and flora. We aimed to develop statistical models to predict the crown diameter of open-growing trees using several forest attributes. Potential crown diameter trends in the measured trees were determined by quantile regression. Crown diameter models were developed by regression analyses, artificial neural networks, support vector machine, and random forest techniques. We evaluated 200 trees characterized into 60 species belonging to 30 botanical families. Our equation for potential crown diameter predicts the derived basal area, number of trees, and the necessary growth space of crown diameter at breast height. Artificial neural networks (with the following validation statistics: R2 = 0.90, RMSE = 1.21, MAE = 0.93, and MAPE = 16.25) predicted crown diameter more accurately than the other evaluated techniques. Modeling crown diameter via machine learning represents an important step toward the assessment of crown dynamics by species and can support the decision making of silvicultural practices and other related activities in several rural properties within the Cerrado biome.

Keywords

• computational intelligence
• crown radius
• biometrics attributes
• prediction models