Integration of remote sensing and artificial intelligence for carbon quantification in macauba palm trees

Abstract

Accurate measurement of carbon is stored in plants is essential for evaluating strategies to reduce greenhouse gas emissions. The macauba palm (Acrocomia aculeata) is a native species of South America with high potential for oil production, ecological restoration, and carbon sequestration. Traditional methods to estimate carbon stocks, such as cutting down trees and performing laboratory analysis, are destructive, expensive, and impractical on a large scale. This study tested an alternative approach that combines UAV imagery, computer vision, and artificial intelligence to estimate carbon in macauba palms without damaging the plants. High-resolution aerial images were collected at four sites in Brazil, and a deep learning model (YOLOv8s-seg) was trained to automatically detect and measure palm crowns, achieving a mean Average Precision (mAP@50) of 0.956, precision of 0.946, and recall of 0.944. Plant height was estimated from canopy height models derived from photogrammetric processing of the UAV imagery. Crown diameter and height estimates showed strong correlations with field measurements (r = 0.81 and r = 0.82, respectively), with mean absolute errors of 0.26 m for diameter and 0.61 m for height. Both parameters were used in allometric equations to calculate carbon stock. Carbon estimates ranged from less than 1 ton per hectare in young plantations to more than 190 tonnes per hectare in older stands. These results demonstrate that artificial intelligence and UAV imagery provide a fast and scalable approach for estimating structural variables of macauba palms. However, carbon stock estimates remain sensitive to the allometric equations used and should be interpreted with caution, particularly at the individual level. Therefore, the proposed approach should be understood as a non-destructive monitoring framework rather than a fully generalized carbon prediction model, while still offering a valuable tool to support sustainable agriculture, carbon markets, and land-use policies.

Plant phenotyping relevance

UAV画像、コンピュータビジョン、深層学習を用いてヤシの樹冠径・樹高を自動推定し、現地測定と比較検証している。植物の構造形質取得と炭素量推定の技術が研究の中心である。

Code and data availability

公開本文の所在を確認できませんでした。非公開または購読が必要な可能性があります。

No evidence-backed public reproduction asset is currently recorded.