Avaliação da discrepância de modelos digitais de terreno obtidos por diferentes sensores LiDAR, algoritmos classificadores de ponto em solo e resolução espacial em um ambiente florestal

Abstract

This study aimed to evaluate the discrepancy among Digital Terrain Models (DTMs) generated from different airborne LiDAR sensors (Zenmuse L1 and L2), ground point classification algorithms, and spatial resolutions in a forested environment located in Angatuba, São Paulo State, Brazil. Flights were carried out using Remotely Piloted Aircraft Systems (RPAS) equipped with LiDAR L1 and L2 sensors under identical operational conditions, resulting in point clouds processed by the Cloth Simulation Filter (CSF), Progressive Morphological Filter (PMF), and Curvature Smoothing Surface (CSS) algorithms. Data processing was performed in RStudio and LiDAR360, generating Digital Terrain Models corresponding to each combination of sensor, algorithm, and spatial resolution, totaling eighteen derived products at 0.1 m, 0.5 m, and 1.0 m resolutions. Statistical analyses were based on 300 plots of 20 × 20 m, representing 14.93% of the study area. The Wilcoxon tests indicated statistically significant differences between L1 and L2 sensors (p < 0.01) for all algorithms, with effect sizes classified as small (d ≈ 0.10–0.18), suggesting subtle variations between platforms. Among the algorithms, CSS produced smoother surfaces, showing moderate to large differences compared to CSF (p < 10⁻³⁸; d = 0.44–0.53) and PMF (p < 10⁻⁴⁵; d = 0.67–1.57), while CSF and PMF exhibited high similarity (p < 10⁻³³; d = 0.20–0.25). Comparisons among spatial resolutions revealed no significant differences (p > 0.05; d < 0.15), demonstrating the stability of the methods across scales.