Improved Forest Signal Detection for Space-Borne Photon-Counting LiDAR Using Automatic Machine Learning

Q: What challenge does the study address in photon-counting LiDAR over forests?

It addresses the large background photon noise from the atmosphere, ground, sun, and other radiation, which reduces signal detection quality in forest areas.

Q: How much training data does the proposed automatic machine learning approach use?

It uses a limited number of sample points, specifically 10% of the sample points for training.

Q: Which model achieves the best performance and what metric is reported?

Stacked Ensembles achieves the best mean F1 score, with a reported mean F1 of 0.957 across tests.

Aug 5, 2026

Upload by Patrick

NASA’s ICESat-2 with the Photon Counting LiDAR sensor detects signal photons at high speed using ATLAS, but it also captures substantial background photon noise from the atmosphere, ground, sun, and other radiation, especially over forested regions. The study introduces an automatic machine learning method that leverages limited training data (10% of samples) to enhance forestry data availability beyond NASA’s official products. Integrated stacked ensemble learning achieves the strongest mean F1 performance (~0.9 overall, 0.957 for the best model). Top features include kNNdist5, kNNdist10, and h, and improvements in correctly identified signal proportions are reported across five datasets, with better results in low SNR cases (<7.5) and robustness to varying topography, noise distribution, and SNR.

LanguageEnglish

CategoryResearch & Report

FormatPDF

Pages13

Reading Time33 min

AccessFree to download

Abstract

Problem: background photon noise in forest regions

Proposed method: automatic machine learning with limited training samples

Results: stacked ensemble performance and key features

Robustness: behavior under low SNR and varying conditions

Improved Forest Signal Detection for Space-Borne Photon-Counting LiDAR Using Automatic Machine Learning Bo Zhang , Li Zhang , Yong Pang , Member, IEEE, Peter North, Min Yan , Hongge Ren, Linlin Ruan , Zhenyu Yang, and Bowei Chen Abstract—NASA’s (National Aeronautics and Space Administration) ICESat-2 with a Photon Counting LiDAR (Light Detection And Ranging) Sensor sensitively detects signal photons at high speedwith an advanced detection system calledtheAdvanced Topographic Laser Altimeter System(ATLAS). However, the sensor also extracts a large amount of background photon noise coming from the atmosphere, ground, sun, or other radiation. This condition is particularly evident in forest areas. This study proposes an automatic machine learning approach to utilize data for forestry applications to improve data availability compared to NASA’sofﬁcial product. Our method uses only a very limited number (10%) of sample points for training, ensuring operational efﬁciency and training accuracy. We conclude that the integrated learning performance generally outperforms single models, and the mean F1 score of all tests is approximately 0.9. The mean F1 score of the Stacked Ensembles model is 0.957 ahead of the other models. The top three variables used in training models are kNNdist5, kNNdist10, and h. These three variables could explain 51.6% of the components of the models. Over the regions tested, the proposed method could improve the proportion of signals correctly identiﬁed by6.4%,12.2%, 2.7%, 9.3%, and 1.4% in ﬁve datasets. The model performs better in low signal-to-noise (SNR) datasets less than 7.5. The method would be largely unaffected by differences in topography, noise distribution, and SNR. The classiﬁers could correct misclassiﬁed Manuscript received 23 March 2023; revised 19 May 2023; accepted 12 June 2023. Date of publication 29 June 2023; date of current version 23 November 2023. This work was supported in part by the National Key Research and Development Program of China under Grant 2020YFE0200800, and in part by the National Natural Science Foundation of China under Grant 42001361 .(Corresponding author: Bowei Chen.) Bo Zhang, Hongge Ren, and Linlin Ruan are with the Key Laboratory of Digital Earth Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China, also with the International Research Center of Big Data for Sustainable Development Goals, Beijing 100094, China, and also with the University of Chinese Academy of Sciences, Beijing 100049, China (e-mail: [zhangbo203@mails.ucas.ac.cn](zhangbo203@mails.ucas.ac.cn); [renhg@aircas.ac.cn](renhg@aircas.ac.cn); ru [anlinlin20@mails.ucas.ac.cn](anlinlin20@mails.ucas.ac.cn)) . Li Zhang, Min Yan, and Bowei Chen are with the Key Laboratory of Digital Earth Science, Aerospace Information Research Institute, Chinese Academy of Sciences, Beijing 100094, China, and also with the International Research Center of Big Data for Sustainable Development Goals, Beijing 100094, China (e-mail: [zhangli@aircas.ac.cn](zhangli@aircas.ac.cn); [yanmin@aircas.ac.cn](yanmin@aircas.ac.cn); [chenbw@aircas.ac.cn](chenbw@aircas.ac.cn)). Yong Pang is with the Institute of Forest Resource Information Techniques, Chinese Academy of Forestry, Beijing 100091, China (e-mail: [pangy@ifrit.ac.cn](pangy@ifrit.ac.cn)). Peter North is with the Global Environmental Modelling and Earth Observation (GEMEO), Department of Geography, Swansea University, SA20LS Swansea, U.K. ([e-mail: p.r.j.north@swansea.ac.uk](e-mail: p.r.j.north@swansea.ac.uk)). Zhenyu Yangis withthe School ofMarineTechnology and Geomatics,Jiangsu Ocean University, Lianyungang 222005, China ([e-mail: yzyloloo@gmail.com](e-mail: yzyloloo@gmail.com)). Digital Object Identiﬁer 10.1109/JSTARS.2023.3290680 labels in ATL08 products and show good stability in different conditions. Index Terms—Automated machine learning, ICESat-2/ATLAS, photon point cloud ﬁltering, space-borne l

FAQs

What challenge does the study address in photon-counting LiDAR over forests?

It addresses the large background photon noise from the atmosphere, ground, sun, and other radiation, which reduces signal detection quality in forest areas.

How much training data does the proposed automatic machine learning approach use?

It uses a limited number of sample points, specifically 10% of the sample points for training.

Which model achieves the best performance and what metric is reported?

Stacked Ensembles achieves the best mean F1 score, with a reported mean F1 of 0.957 across tests.

Aug 5, 2026

Upload by Patrick