A 3D point cloud dataset of pecan tree woody crown architecture acquired using consumer-grade LiDAR

Overview

This repository contains a comprehensive 3D point cloud dataset detailing the woody crown architecture of 54 mature pecan trees (Carya illinoinensis, cv. 'Western'). Acquired in situ within commercial orchards in Ciudad Delicias, Chihuahua, Mexico, this dataset establishes a foundational 3D benchmark to advance precision agriculture methodologies and high-throughput phenotyping.

To evaluate the viability of democratized, consumer-grade sensors and to promote algorithmic experimentation, all spatial data was generated using a Mobile Terrestrial Laser Scanner (MTLS)—an iPad Pro 11-inch (M1 processor) equipped with a solid-state LiDAR sensor.

Dataset Structure

The repository maps a complete computational processing pipeline, offering the point clouds at various stages of refinement. This allows users to access both the raw, noisy geometries and the definitively curated architectures. The data is provided in standard .ply and .las formats and is organized into the following stages:

Raw / Preprocessed Instances: Initial spatial geometries exhibiting inherent sensor noise and positional artifacts (the "noise dome") captured via a pedestrian-based kinematic scanning protocol.

02_Filtered_Density_Based: Point clouds filtered using local volume density descriptors.

03_Filtered_Curvature_Optimized: Point clouds filtered using Gaussian curvature descriptors (optimized parameters).

04_Filtered_Curvature_Aggressive: Point clouds processed with an aggressive curvature threshold.

05_QSM_Ready_Final: Definitively curated, pristine woody architectures prepared for Quantitative Structure Models (QSM).

Photographic Reference: In-situ optical ground-truth images (.jpg, .png) serving as a qualitative structural reference for each tree.

Methodology

Data acquisition followed a pedestrian-based kinematic scanning protocol. The operator navigated a continuous 360-degree circular trajectory (approx. 3 m radius) around each tree at a sub-walking speed (~3 km/h). The MTLS was mounted on a 1-meter tripod extension to capture the upper canopy. Computational filtering was systematically executed utilizing Python-based preprocessing alongside scalar geometric descriptors in CloudCompare to systematically isolate the woody structure from sensor noise artifacts.

Scientific Value and Potential Use

This dataset serves a dual scientific purpose:

For Computer Scientists & Data Engineers: It provides a complex, highly entangled woody structure benchmark to train, test, and optimize novel denoising and outlier-removal algorithms tailored for low-cost, noisy sensor data.

For Agronomic Researchers: It empowers the development of Data-Driven decision-support tools for orchard management—such as simulating precise pruning constraints (e.g., 30% woody volume removal)—thereby entirely circumventing the need for destructive structural assessments.