Evaluación del uso de LiDAR discreto, full-waveform y TLS en la clasificación por composición de especies en bosques mediterráneos

Abstract

<p>LiDAR technology –airborne and terrestrial- is becoming more relevant in the development of forest inventories, which are crucial to better understand and manage forest ecosystems. In this study, we assessed a classification of species composition in a Mediterranean forest following the C4.5 decision tree. Different data sets from airborne laser scanner full-waveform (ALS<sub>FW</sub>), discrete (ALS<sub>D</sub>) and terrestrial laser scanner (TLS) were combined as input data for the classification. Species composition were divided into five classes: pure Quercus ilex plots (QUI); pure Pinus halepensis dense regenerated (HALr); pure P. halepensis (HAL); pure P. pinaster (PIN); and mixed P. pinaster and Q. suber (mPIN). Furthermore, the class HAL was subdivided in low and dense understory vegetation cover. As a result, combination of ALS<sub>FW</sub> and TLS reached 85.2% of overall accuracy classifying classes HAL, PIN and mPIN. Combining ALS<sub>FW</sub> and ALS<sub>D</sub>, the overall accuracy was 77.0% to discriminate among the five classes. Finally, classification of understory vegetation cover using ALS<sub>FW</sub> reached an overall accuracy of 90.9%. In general, combination of ALS<sub>FW</sub> and TLS improved the overall accuracy of classifying among HAL, PIN and mPIN by 7.4% compared to the use of the data sets separately, and by 33.3% with respect to the use of ALS<sub>D</sub> only. ALS<sub>FW</sub> metrics, in particular those specifically designed for detection of understory vegetation, increased the overall accuracy 9.1% with respect to ALS<sub>D</sub> metrics. These analyses show that classification in forest ecosystems with presence of understory vegetation and intermediate canopy strata is improved when ALS<sub>FW</sub> and/or TLS are used instead of ALS<sub>D</sub>.</p>

Classification of Uav-Based Photogrammetric Point Clouds of Riverine Species Using Machine Learning Algorithms: A Case Study in the Palancia River, Spain

“…Some successful studies have been done using aerial and terrestrial laser scanning point clouds to classify forest species composition (Torralba et al, 2018). However, the frequency of application of these techniques is limited by their cost.…”

Processing and analysis of airborne fullwaveform laser scanning data for the characterization of forest structure and fuel properties

“…These studies generally show that CH estimations are more accurate using ALS than TLS (Hilker et al, 2010), while characterization of the foliage profile is estimated with more accuracy by TLS, especially in the lower strata (Chasmer et al, 2006;Hilker et al, 2010), where understory vegetation is found. On the other hand, other studies concur on a more accurate estimation of forest structural attributes (Anderson et al, 2016), AGB (Nie et al, 2017), stand volume (Lindberg et al, 2012), and the classification of species composition (Torralba et al, 2018).…”

“…For applications in forested environments, the useful portion of the TLS point cloud extent is often limited (10 -30 m) with a hemispherical view around the sensor. Withstanding that, many studies have demonstrated the capabilities of TLS to estimate and extract forest stand attributes (Watt and Donoghue, 2005;Moskal and Zheng, 2011;Kankare et al, 2013;Srinivasan et al, 2015;Liang et al, 2016;Ravaglia et al, 2019), and fewer on the classification of tree species (Othmani et al, 2013;Lin and Herold, 2016;Torralba et al, 2018).…”

“…Within the vertical distribution of the forest, detection of the lower strata can be challenging from ALS sensors, especially from ALSD as a results of signal occlusion from the overstory (Anderson et al, 2016). Nonetheless, ALSD has proven to discriminate presence and absence of understory vegetation with promising accuracy, e.g., with R 2 's of 0.83 (Martinuzzi et al, 2009), 0.77 (Hill and Broughton, 2009), 0.74 (Wing et al, 2012), and 0.48 (Morsdorf et al, 2010), and has been demonstrated to be more accurately estimated by ALSFW than ALSD (Hancock et al, 2017;Torralba et al, 2018). In contrast, point clouds from TLS provide a large amount of detail on understory vegetation due to the position of the sensor (Liu et al, 2017).…”

“…Furthermore, few studies have estimated understory cover with ALSFW using TLS as reference data and obtained a nRMSE of 24% (Hancock et al, 2017). Nevertheless, a limited amount of studies have compared the ability to estimate understory vegetation presence or distribution from three different laser scanning datasets (i.e., ALSFW, ALSD and TLS) (e.g., Hancock et al (2017) and Torralba et al (2018)). These studies confirmed that ALSFW and TLS are both capable of estimating the spatial distribution of understory vegetation in more detail than using ALSD data, e.g., with overall accuracies of 86.4% and 77.3%, respectively (Torralba et al, 2018).…”