Status, advancements and prospects of deep learning methods applied in forest studies

International Journal of Applied Earth Observation and Geoinformation, Год журнала: 2024, Номер 131, С. 103938 - 103938

Опубликована: Июнь 4, 2024

Язык: Английский

---

Sub-meter tree height mapping of California using aerial images and LiDAR-informed U-Net model

Remote Sensing of Environment, Год журнала: 2024, Номер 305, С. 114099 - 114099

Опубликована: Март 18, 2024

Tree canopy height is one of the most important indicators forest biomass, productivity, and ecosystem structure, but it challenging to measure accurately from ground space. Here, we used a U-Net model adapted for regression map all trees in state California with very high-resolution aerial imagery 0.6 m USDA-NAIP program. The was trained using models computed LiDAR data as reference, along corresponding RGB-NIR NAIP images collected 2020. We evaluated performance deep-learning 42 independent 1 km2 areas across various types landscape variations California. Our predictions tree heights exhibited mean error 2.9 showed relatively low systematic bias entire range present In 2020, taller than 5 covered ∼ 19.3% successfully estimated up 50 without saturation, outperforming existing products global models. approach allowed reconstruction three-dimensional structure individual observed nadir-looking optical airborne imagery, suggesting robust estimation mapping capability, even presence image distortion. These findings demonstrate potential large-scale monitoring height, well biomass estimation, imagery.

Язык: Английский

---

Canopy height model and NAIP imagery pairs across CONUS

Scientific Data, Год журнала: 2025, Номер 12(1)

Опубликована: Фев. 22, 2025

Язык: Английский

---

Estimating canopy height in tropical forests: Integrating airborne LiDAR and multi-spectral optical data with machine learning

Sustainable Environment, Год журнала: 2025, Номер 11(1)

Опубликована: Фев. 25, 2025

Язык: Английский

---

FORMS: Forest Multiple Source height, wood volume, and biomass maps in France at 10 to 30 m resolution based on Sentinel-1, Sentinel-2, and Global Ecosystem Dynamics Investigation (GEDI) data with a deep learning approach

Earth system science data, Год журнала: 2023, Номер 15(11), С. 4927 - 4945

Опубликована: Ноя. 2, 2023

Abstract. The contribution of forests to carbon storage and biodiversity conservation highlights the need for accurate forest height biomass mapping monitoring. In France, are managed mainly by private owners divided into small stands, requiring 10 50 m spatial resolution data be correctly separated. Further, 35 % French territory is covered mountains Mediterranean which very extensively. this work, we used a deep-learning model based on multi-stream remote-sensing measurements (NASA's Global Ecosystem Dynamics Investigation (GEDI) lidar mission ESA's Copernicus Sentinel-1 Sentinel-2 satellites) create canopy map France 2020 (FORMS-H). second step, with allometric equations fitted National Forest Inventory (NFI) plot data, created 30 above-ground density (AGBD) (Mg ha−1) (FORMS-B). Extensive validation was conducted. First, independent datasets from airborne laser scanning (ALS) NFI thousands plots reveal mean absolute error (MAE) 2.94 FORMS-H, outperforms existing models. Second, FORMS-B validated using two inventory Renecofor permanent network GLORIE MAE 59.6 19.6 Mg ha−1, respectively, providing greater performance than other AGBD products sampled over France. Finally, compared FORMS-V (for volume) wood volume estimations at ecological region scale obtained an R2 0.63 m3 ha−1. These results highlight importance coupling technologies recent advances in computer science bring material insights climate-efficient management policies. Additionally, our approach open-access having global coverage high temporal resolution, making maps reproducible easily scalable. FORMS can accessed https://doi.org/10.5281/zenodo.7840108 (Schwartz et al., 2023).

Язык: Английский

---

Forestry Applications of Space-borne LiDAR Sensors: A Worldwide Bibliometric Analysis

Опубликована: Янв. 8, 2024

The 21st century has seen the launch of new space-borne sensors based on LiDAR (light detection and ranging) technology developed in second half 20th century. was initially to integrate laser-focused imaging with capability determine distances through measurement signal return times, utilizing suitable data acquisition electronics. Nowadays, these have transformed into robust instruments, offering novel opportunities for mapping terrain, canopy heights, estimating above-ground biomass (AGB) across local regional scales. This work aims analyze scientific impact large-scale for-est retrieve 3D information, monitor forest degradation, estimate AGB, model key ecosystem variables such as primary productivity biodiversity. In this way, a worldwide bibliometric analysis topic carried out up 412 publications in-dexed Scopus database during period 2004-2022. results showed that number published documents increased exponentially last five years, coinciding commis-sioning two space missions: Ice, Cloud Land Elevation Satellite (ICESat-2) Global Ecosystem Dynamics Investigation (GEDI). These missions are providing since 2018 2019, respectively. journal demonstrated highest field "Remote Sensing," among leading contributors, top countries terms publica-tions were USA, China, UK, France, Germany. realm prominent research in-stitutions, France boasted six, USA had four, China three, while UK Canada each one. upward trajectory recorded from 2004 2022 catego-rizes subject under investigation highly trending topic, particularly within context enhancing administration resources engaging global climate treaty frameworks mandating surveillance reporting carbon stocks forests. recent August Terrestrial Carbon Monitoring (TECMS; State Administration Forestry Grassland), along planned coming years three sensors, Multi-footprint Observation Im-ager (Japan Aerospace Exploration Agency), BIOMASS P-band Synthetic Aperture Radar (SAR) (European Space Surface Topography (LIST; NASA), will greatly contribute expanding ability map systems at very large context, integration data, including imagery, SAR, LiDAR, is anticipated steer upcoming years.

Язык: Английский

---

Assimilating Sentinel-2 data in a modified vegetation photosynthesis and respiration model (VPRM) to improve the simulation of croplands CO2 fluxes in Europe

International Journal of Applied Earth Observation and Geoinformation, Год журнала: 2024, Номер 127, С. 103666 - 103666

Опубликована: Янв. 29, 2024

In Europe, the heterogeneous features of crop systems with majority small to medium sized agricultural holdings, and diversity rotations, require high-resolution information estimate cropland Net Ecosystem Exchange (NEE) its two main components Gross (GEE) Respiration (RECO). this context, paper presents an assimilation Sentinel-2 indices eddy covariance measurements at selected European flux sites in a new modified version Vegetation Photosynthesis Model (VPRM). VRPM is data-driven model simulating CO2 fluxes previously applied using satellite-derived vegetation from Moderate Resolution Imaging Spectroradiometer (MODIS). This study proposes modification VPRM by including explicit soil moisture stress function GEE changing equation RECO. It also compares results driven S2 instead MODIS. The parameters are calibrated eddy-covariance data. All possible optimization scenarios include use initial vs. proposed VPRM, S2, or MODIS indices, finally choice calibrating single set against observations all types, per type, one site. Then, we focus analysis on improvement distinct for different types optimized without distinction types. Our findings are: (1) superiority over simulations, leading root mean squared error (RMSE) NEE less than 3.5 μmolm-2s-1 compared 5 (2) better performances significant RECO, (3) when crop-type lumped together, lower RMSE Akaike criterion (AIC), despite larger number parameters. Associated availability land cover maps, data parameterization presented study, provide step forward upscaling carbon scale.

Язык: Английский

---

Retrieval of grassland aboveground biomass across three ecoregions in China during the past two decades using satellite remote sensing technology and machine learning algorithms

International Journal of Applied Earth Observation and Geoinformation, Год журнала: 2024, Номер 130, С. 103925 - 103925

Опубликована: Май 21, 2024

The aboveground biomass (AGB) is closely linked to the carbon cycle in grassland ecosystems worldwide. Accurately quantifying AGB variations thus essential for assessing sequestration and its feedback on climate change. Although many studies have investigated AGB, they are limited local areas few research efforts been attempted estimate at large scales with constraint of situ quadrat harvested AGB. In this study, we used multi-source satellite remote sensing data from 2000 2021 abundant harvest quadrats explore estimation methods then analyze spatiotemporal patterns various types across China's three ecoregions. results indicate that: (1) temporal resolution improvement a higher correlation between remotely sensed NDVI Therefore, MODIS MCD43A4 dataset has better fit harvesting data. (2) Compared statistical methods, machine learning algorithms exhibit high accuracy estimating Among them, random forest (RF) model performs most robustly, highest R2 0.83 (explaining 83 % variation AGB), lowest RMSE 43.84 gm−2. (3) multi-year average annual maximum decreases southeast northwest, temperate steppe region having highest, followed by alpine vegetation desert region. (4) While approximately 61.94 pixels show an increasing trend 2021, significant (P < 0.05) changes mainly concentrated eastern each ecoregion. Our study presents valuable framework using datasets. Additionally, it provides robust product grasslands contributing our understanding long-term ecosystems.

Язык: Английский

---

Retrieval performance of mangrove tree heights using multiple machine learning regression models and UAV-LiDAR point clouds

International Journal of Digital Earth, Год журнала: 2024, Номер 17(1)

Опубликована: Авг. 19, 2024

Mangroves are vital coastal ecosystems that provide crucial links between land and sea. Tree height is a key indicator for assessing mangroves' health status. Currently, there still numerous challenges in estimating mangrove tree height. In this study, multiple deep learning shallow machine regression models were developed to accurately estimate using multi-dimensional Light Detection Ranging (LiDAR) point clouds their derivatives. We constructed novel CNN_RepMLP model mapping. also further verified the applicability of different types heights, explored influence LiDAR-derived features on inversion accuracy heights. The results indicated following: (1) displayed satisfactory performance exhibited better robustness generalization ability than convolutional neural network (CNN) model. (2) Among feature combinations, combining variables with intensity can not only mitigate negative impact models, but enhance accuracy. (3) ensemble framework ExtraTrees as meta-model make use differences complementarities single base trees compared other models. (4) Multiple based UAV-LiDAR point-cloud-derived suitable outperformed CNN stacking had more detailed differentiation terms Its prediction realistically reflect spatial characteristics

Язык: Английский

---

Enhancing high-resolution forest stand mean height mapping in China through an individual tree-based approach with close-range lidar data

Earth system science data, Год журнала: 2024, Номер 16(11), С. 5267 - 5285

Опубликована: Ноя. 14, 2024

Abstract. Forest stand mean height is a critical indicator in forestry, playing pivotal role various aspects such as forest inventory, sustainable management practices, climate change mitigation strategies, monitoring of structure changes, and wildlife habitat assessment. However, there currently lack large-scale, spatially continuous maps. This primarily due to the requirement accurate measurement individual tree each plot, task that cannot effectively be achieved by existing globally covered, discrete footprint-based satellite platforms. To address this gap, study was conducted using over 1117 km2 close-range light detection ranging (lidar) data, which enables heights plots with high precision. Apart from lidar incorporated climatic, edaphic, topographic, vegetative, synthetic aperture radar data explanatory variables map tree-based arithmetic (ha) weighted (hw) at 30 m resolution across China. Due limitations obtaining basal area within uncrewed aerial vehicle (UAV) calculated through weighting an square its height. In addition, overcome potential influence different vegetation divisions large spatial scale, we also developed machine-learning-based mixed-effects (MLME) model The results showed average ha hw China were 11.3 13.3 standard deviations 2.9 3.3 m, respectively. accuracy mapped products validated utilizing field data. correlation coefficient (r) for ranged 0.603 0.906 0.634 0.889, while root error (RMSE) 2.6 4.1 4.3 Comparing canopy maps derived area-based approach, it found our performed better aligned more closely natural definition methods presented provide solid foundation estimating carbon storage, changes structure, managing assessing availability. dataset constructed publicly available https://doi.org/10.5281/zenodo.12697784 (Chen et al., 2024).

Язык: Английский

---