Xun Zhao , Jianbo Qi , Jingyi Jiang , Shangbo Liu , Haifeng Xu , Simei Lin , Zhexiu Yu , Linyuan Li , Huaguo Huang
{"title":"Fine-scale retrieval of leaf chlorophyll content using a semi-empirically accelerated 3D radiative transfer model","authors":"Xun Zhao , Jianbo Qi , Jingyi Jiang , Shangbo Liu , Haifeng Xu , Simei Lin , Zhexiu Yu , Linyuan Li , Huaguo Huang","doi":"10.1016/j.jag.2024.104285","DOIUrl":null,"url":null,"abstract":"<div><div>Leaf chlorophyll content (LCC) retrieval from remote sensing imagery is essential for monitoring vegetation growth and stress in the agroforestry industry. Many remote sensing inversion methods for estimating LCC primarily rely on 1D radiative transfer models (RTMs) that abstract canopies into horizontal layers or simple geometric primitives. Yet, this methodology faces challenges when applied to heterogeneous canopies, particularly in fine-scale mapping where each pixel's reflectance is significantly influenced by its surroundings, e.g. crown shadows. While 3D RTMs hold promise for addressing these challenges by explicitly describing complex canopy structures, their computational demands and the complexity involved in parameterizing detailed 3D structures limit the generation of extensive training datasets, requiring simulations across numerous parameter combinations. In this study, we used a semi-empirically accelerated 3D RTM, termed Semi-LESS, with a 1D residual network to accurately retrieve leaf chlorophyll content (LCC) from UAV images and LiDAR data at a 3-m resolution. We first reconstructed structures of forest plots using UAV LiDAR point cloud, based on which, UAV images with varying leaf and soil optical properties are simulated using the Semi-LESS. Subsequently, a training dataset consisting LCC and its corresponding reflectance was generated from the simulated UAV images by focusing on sunlit pixels. A 1D residual network is trained using the training dataset for LCC estimation. For comparison, we also trained an estimation model using a dataset generated from PROSAIL. The results show that estimation model trained with Semi-LESS surpasses PROSAIL in retrieving LCC from both simulation datasets and field measurements of two forest plots. The RMSE of Semi-LESS was 5.40–6.92 µg/cm<sup>2</sup> for simulation datasets and 8.21–9.76 µg/cm<sup>2</sup> for field measurements, whereas PROSAIL exhibited lower accuracy with an RMSE of 7.76–9.83 µg/cm<sup>2</sup> for simulation datasets and 12.76–13.06 µg/cm<sup>2</sup> in field measurements. The results demonstrate that Semi-LESS coupled with deep learning is reliable and has great potential for LCC mapping using UAV images, which is particularly useful for fine-scale applications such as crop and orchard monitoring. This approach also highlights the impact of shadows on LCC retrieval.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"135 ","pages":"Article 104285"},"PeriodicalIF":7.6000,"publicationDate":"2024-11-27","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":"0","resultStr":null,"platform":"Semanticscholar","paperid":null,"PeriodicalName":"International journal of applied earth observation and geoinformation : ITC journal","FirstCategoryId":"1085","ListUrlMain":"https://www.sciencedirect.com/science/article/pii/S1569843224006411","RegionNum":0,"RegionCategory":null,"ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":null,"EPubDate":"","PubModel":"","JCR":"Q1","JCRName":"REMOTE SENSING","Score":null,"Total":0}
引用次数: 0
Abstract
Leaf chlorophyll content (LCC) retrieval from remote sensing imagery is essential for monitoring vegetation growth and stress in the agroforestry industry. Many remote sensing inversion methods for estimating LCC primarily rely on 1D radiative transfer models (RTMs) that abstract canopies into horizontal layers or simple geometric primitives. Yet, this methodology faces challenges when applied to heterogeneous canopies, particularly in fine-scale mapping where each pixel's reflectance is significantly influenced by its surroundings, e.g. crown shadows. While 3D RTMs hold promise for addressing these challenges by explicitly describing complex canopy structures, their computational demands and the complexity involved in parameterizing detailed 3D structures limit the generation of extensive training datasets, requiring simulations across numerous parameter combinations. In this study, we used a semi-empirically accelerated 3D RTM, termed Semi-LESS, with a 1D residual network to accurately retrieve leaf chlorophyll content (LCC) from UAV images and LiDAR data at a 3-m resolution. We first reconstructed structures of forest plots using UAV LiDAR point cloud, based on which, UAV images with varying leaf and soil optical properties are simulated using the Semi-LESS. Subsequently, a training dataset consisting LCC and its corresponding reflectance was generated from the simulated UAV images by focusing on sunlit pixels. A 1D residual network is trained using the training dataset for LCC estimation. For comparison, we also trained an estimation model using a dataset generated from PROSAIL. The results show that estimation model trained with Semi-LESS surpasses PROSAIL in retrieving LCC from both simulation datasets and field measurements of two forest plots. The RMSE of Semi-LESS was 5.40–6.92 µg/cm2 for simulation datasets and 8.21–9.76 µg/cm2 for field measurements, whereas PROSAIL exhibited lower accuracy with an RMSE of 7.76–9.83 µg/cm2 for simulation datasets and 12.76–13.06 µg/cm2 in field measurements. The results demonstrate that Semi-LESS coupled with deep learning is reliable and has great potential for LCC mapping using UAV images, which is particularly useful for fine-scale applications such as crop and orchard monitoring. This approach also highlights the impact of shadows on LCC retrieval.
期刊介绍:
The International Journal of Applied Earth Observation and Geoinformation publishes original papers that utilize earth observation data for natural resource and environmental inventory and management. These data primarily originate from remote sensing platforms, including satellites and aircraft, supplemented by surface and subsurface measurements. Addressing natural resources such as forests, agricultural land, soils, and water, as well as environmental concerns like biodiversity, land degradation, and hazards, the journal explores conceptual and data-driven approaches. It covers geoinformation themes like capturing, databasing, visualization, interpretation, data quality, and spatial uncertainty.