Fine-Scale Mangrove Species Classification Based on UAV Multispectral and Hyperspectral Remote Sensing Using Machine Learning

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(16), P. 3093 - 3093

Published: Aug. 22, 2024

Mangrove ecosystems play an irreplaceable role in coastal environments by providing essential ecosystem services. Diverse mangrove species have different functions due to their morphological and physiological characteristics. A precise spatial distribution map of is therefore crucial for biodiversity maintenance environmental conservation ecosystems. Traditional satellite data are limited fine-scale classification low resolution less spectral information. This study employed unmanned aerial vehicle (UAV) technology acquire high-resolution multispectral hyperspectral forest imagery Guangxi, China. We leveraged advanced algorithms, including RFE-RF feature selection machine learning models (Adaptive Boosting (AdaBoost), eXtreme Gradient (XGBoost), Random Forest (RF), Light Machine (LightGBM)), achieve mapping with high accuracy. The assessed the performance these four two types image (UAV imagery), respectively. results demonstrated that had superiority over offering enhanced noise reduction performance. Hyperspectral produced overall accuracy (OA) higher than 91% across models. LightGBM achieved highest OA 97.15% kappa coefficient (Kappa) 0.97 based on imagery. Dimensionality extraction techniques were effectively applied UAV data, vegetation indices proving be particularly valuable classification. present research underscored effectiveness images using approach has potential significantly improve ecological management strategies, a robust framework monitoring safeguarding habitats.

Language: Английский

---

A Comparative Analysis of Remote Sensing Estimation of Aboveground Biomass in Boreal Forests Using Machine Learning Modeling and Environmental Data

Sustainability, Journal Year: 2024, Volume and Issue: 16(16), P. 7232 - 7232

Published: Aug. 22, 2024

It is crucial to have precise and current maps of aboveground biomass (AGB) in boreal forests accurately track global carbon levels develop effective plans for addressing climate change. Remote sensing as a cost-effective tool offers the potential update AGB real time. This study evaluates different machine learning algorithms, namely Light Gradient Boosting Machine (LightGBM), Extreme (XGBoost), Random Forest (RF), Support Vector Regression (SVR), predicting forests. Conducted Qilian Mountains, northwest China, integrated field measurements, space-borne LiDAR, optical remote sensing, environmental data training dataset. Among 34 variables, 22 were selected estimation modeling. Our findings revealed that LightGBM model had highest level accuracy (R2 = 0.84, RMSE 15.32 Mg/ha), outperforming XGBoost, RF, SVR models. Notably, effectively addressed issues underestimation overestimation. We also observed disparity among models widens with increasing altitude. Remarkably, consistently demonstrates optimal performance across all elevation gradients, residuals generally below 25 Mg/ha low-value overestimation −38 high-value underestimation. The developed this presents viable alternative approach enhancing based on technology.

Language: Английский

---

A Machine Learning Algorithm Using Texture Features for Nighttime Cloud Detection from FY-3D MERSI L1 Imagery

Remote Sensing, Journal Year: 2025, Volume and Issue: 17(6), P. 1083 - 1083

Published: March 19, 2025

Accurate cloud detection is critical for quantitative applications of satellite-based advanced imager observations, yet nighttime presents challenges due to the lack visible and near-infrared spectral information. Nighttime using infrared (IR)-only information needs be improved. Based on a collocated dataset from Fengyun-3D Medium Resolution Spectral Imager (FY-3D MERSI) Level 1 data CALIPSO CALIOP lidar 2 product, this study proposes novel framework leveraging Light Gradient-Boosting Machine (LGBM), integrated with grey level co-occurrence matrix (GLCM) features extracted IR bands, enhance capabilities. The LGBM model GLCM demonstrates significant improvements, achieving an overall accuracy (OA) exceeding 85% F1-Score (F1) nearly 0.9 when validated independent product. Compared threshold-based algorithm that has been used operationally, proposed exhibits superior more stable performance across varying solar zenith angles, surface types, altitudes. Notably, method produced over 82% OA cryosphere surface. Furthermore, compared models without inputs, enhanced effectively mitigates thermal stripe effect MERSI L1 data, yielding accurate masks. Further evaluation MODIS-Aqua mask product indicates delivers precise (OA: 90.30%, F1: 0.9397) MODIS 84.66%, 0.9006). This IR-alone advancement offers reliable tool detection, significantly enhancing satellite observations.

Language: Английский

---

Integrating Multi-Source Remote Sensing Data and Machine Learning for Predicting Tree Density and Cover in Argania spinosa

Smart Agricultural Technology, Journal Year: 2025, Volume and Issue: unknown, P. 100911 - 100911

Published: March 1, 2025

Language: Английский

---

Machine learning models for solar forecasting and impact on green hydrogen production costs

International Journal of Hydrogen Energy, Journal Year: 2025, Volume and Issue: 132, P. 225 - 238

Published: May 1, 2025

Language: Английский

---

Modeling forest structural variables of Eucalyptus dunnii Maiden stands under short-rotation management using SAR, multispectral, soil-derived, and field-based data

Forest Ecology and Management, Journal Year: 2025, Volume and Issue: 588, P. 122759 - 122759

Published: May 9, 2025

Language: Английский

---

Comparing the potential of tree-based and area-based forest height metrics for aboveground biomass estimation in complex forest landscapes

Ecological Indicators, Journal Year: 2025, Volume and Issue: 176, P. 113610 - 113610

Published: May 23, 2025

Language: Английский

---

Land Surface Longwave Radiation Retrieval from ASTER Clear-Sky Observations

Remote Sensing, Journal Year: 2024, Volume and Issue: 16(13), P. 2406 - 2406

Published: June 30, 2024

Surface longwave radiation (SLR) plays a pivotal role in the Earth’s energy balance, influencing range of environmental processes and climate dynamics. As demand for high spatial resolution remote sensing products grows, there is an increasing need accurate SLR retrieval with enhanced detail. This study focuses on development validation models to estimate using measurements from Advanced Spaceborne Thermal Emission Reflection Radiometer (ASTER) sensor. Given limitations posed by fewer spectral bands data ASTER compared moderate-resolution sensors, proposed approach combines atmospheric radiative transfer model MODerate TRANsmission (MODTRAN) Light Gradient Boosting Machine algorithm SLR. The MODTRAN simulations were performed construct representative training dataset based comprehensive global profiles surface emissivity spectra data. Global sensitivity analyses reveal that key inputs accuracy retrievals should reflect thermal signals near-surface conditions. Validated against ground-based measurements, upward (SULR) downward (SDLR) infrared elevation estimations resulted root mean square errors 17.76 W/m2 25.36 W/m2, biases 3.42 3.92 respectively. Retrievals show systematic related extreme temperature moisture conditions, e.g., causing overestimation SULR hot humid conditions underestimation SDLR arid While challenges persist, particularly addressing variables cloud masking, this work lays foundation sensors like ASTER. potential applications extend upcoming satellite missions, such as Landsat Next, contribute advancing high-resolution capabilities improved understanding

Language: Английский

---

A Novel Workflow for Mapping Forest Canopy Height by Synergizing ICESat-2 and Multi-Sensor Data

Forests, Journal Year: 2024, Volume and Issue: 15(12), P. 2139 - 2139

Published: Dec. 4, 2024

Precise information on forest canopy height (FCH) is critical for carbon stocks estimation and management, but mapping continuous FCH with satellite data at regional scale still a challenge. By fusing ICESat-2, Sentinel-1/2 images ancillary data, this study aimed to develop workflow obtain an map using machine learning algorithm over large areas. The vegetation-type was initially produced by phenology-based spectral feature selection method. A characteristic-based model then proposed spatially after multivariate quality control. Our results show that the overall accuracy (OA) average F1 Score (F1) eight main vegetation types were more than 90% 89%, respectively, agreed well census demonstrated greater potential in prediction, R-value 60.47% traditional single model, suggesting addition of control structure characteristics could positively contribute prediction FCH. We generated 30 m evaluated product about 35 km2 airborne laser scanning (ALS) validation (R = 0.73, RMSE 2.99 m), which 45.34% precise China FCH, 2019. These findings demonstrate our monitoring will greatly benefit accurate resources assessment.

Language: Английский

---