A Machine Learning-Based Model for Flight Turbulence Identification Using LiDAR Data DOI Creative Commons
Zibo Zhuang, Hui Zhang, Pak Wai Chan

et al.

Atmosphere, Journal Year: 2023, Volume and Issue: 14(5), P. 797 - 797

Published: April 27, 2023

By addressing the imbalanced proportions of data category samples in velocity structure function LiDAR turbulence identification model, we propose a flight model utilizing both conditional generative adversarial network (CGAN) and extreme gradient boosting (XGBoost). This can fully learn small- medium-sized samples, reduce false alarm rate, improve robustness, maintain stability. Model training involves constructing balanced dataset by generating that conform to original distribution via CGAN. Subsequently, XGBoost is iteratively trained on sample set obtain classification level. Experiments show recognition accuracy achieved CGAN-generated augmented improves 15%. Additionally, when incorporating LiDAR-obtained wind field data, performance surpasses traditional algorithms such as K-nearest neighbours, support vector machines, random forests 14%, 8%, 5%, respectively, affirming excellence for classification. Moreover, comparative analysis conducted Zhongchuan Airport crew report showed 78% accuracy, indicating enhanced ability under data-imbalanced conditions. In conclusion, our CGAN/XGBoost effectively addresses proportion imbalance issue.

Language: Английский

U-Net-LSTM: Time Series-Enhanced Lake Boundary Prediction Model DOI Creative Commons
Lirong Yin, Lei Wang,

Tingqiao Li

et al.

Land, Journal Year: 2023, Volume and Issue: 12(10), P. 1859 - 1859

Published: Sept. 29, 2023

Change detection of natural lake boundaries is one the important tasks in remote sensing image interpretation. In an ordinary fully connected network, or CNN, signal neurons each layer can only be propagated to upper layer, and processing samples independent at moment. However, for time-series data with transferability, learned change information needs recorded utilized. To solve above problems, we propose a boundary prediction model combining U-Net LSTM. The ensemble LSTMs helps improve overall accuracy robustness by capturing spatial temporal nuances data, resulting more precise predictions. This study selected Lake Urmia as research area used annual panoramic images from 1996 2014 (Lat: 37°00′ N 38°15′ N, Lon: 46°10′ E 44°50′ E) obtained Google Earth Professional Edition 7.3 software set. uses network extract multi-level features analyze trend boundaries. LSTM module introduced after optimize predictive using historical storage forgetting well current input data. method enables automatically fit time series mine deep changes. Through experimental verification, model’s changes training reach 89.43%. Comparative experiments existing U-Net-STN show that U-Net-LSTM this has higher lower mean square error.

Language: Английский

Citations

116
A Parallel-Cascaded Ensemble of Machine Learning Models for Crop Type Classification in Google Earth Engine Using Multi-Temporal Sentinel-1/2 and Landsat-8/9 Remote Sensing Data DOI Creative Commons

Esmaeil Abdali,

Mohammad Javad Valadan Zoej, Alireza Taheri Dehkordi

et al.

Remote Sensing, Journal Year: 2023, Volume and Issue: 16(1), P. 127 - 127

Published: Dec. 28, 2023

The accurate mapping of crop types is crucial for ensuring food security. Remote Sensing (RS) satellite data have emerged as a promising tool in this field, offering broad spatial coverage and high temporal frequency. However, there still growing need type classification methods using RS due to the intra- inter-class variability crops. In vein, current study proposed novel Parallel-Cascaded ensemble structure (Pa-PCA-Ca) with seven target classes Google Earth Engine (GEE). Pa section consisted five parallel branches, each generating Probability Maps (PMs) different multi-temporal Sentinel-1/2 Landsat-8/9 images, along Machine Learning (ML) models. PMs exhibited correlation within class, necessitating use most relevant information reduce input dimensionality Ca part. Thereby, Principal Component Analysis (PCA) was employed extract top uncorrelated components. These components were then utilized structure, final performed another ML model referred Meta-model. Pa-PCA-Ca evaluated in-situ collected from extensive field surveys northwest part Iran. results demonstrated superior performance achieving an Overall Accuracy (OA) 96.25% Kappa coefficient 0.955. incorporation PCA led OA improvement over 6%. Furthermore, significantly outperformed conventional approaches, which simply stack sources feed them single model, resulting 10% increase OA.

Language: Английский

Citations

35
Monitoring inland water via Sentinel satellite constellation: A review and perspective DOI

Fanxuan Zeng,

Chunqiao Song, Zhigang Cao

et al.

ISPRS Journal of Photogrammetry and Remote Sensing, Journal Year: 2023, Volume and Issue: 204, P. 340 - 361

Published: Sept. 28, 2023

Language: Английский

Citations

34
Opportunities and Challenges of Spaceborne Sensors in Delineating Land Surface Temperature Trends: A Review DOI Creative Commons
M. Razu Ahmed, Ebrahim Ghaderpour, Anil Gupta

et al.

IEEE Sensors Journal, Journal Year: 2023, Volume and Issue: 23(7), P. 6460 - 6472

Published: Feb. 24, 2023

Understanding the land surface temperature (LST) trends is crucial for policymakers and stakeholders to develop adaptation mitigation strategies suitable a sustainable environment coping in face of climate change. This article presents systematic review studies related delineating spaceborne sensor-based LST trends, including information on instruments constellations satellites (missions) that provide thermal infrared (TIR) passive microwave (PMW) observations. About 99% used TIR, where 76% were Moderate Resolution Imaging Spectroradiometer (MODIS, onboard Terra/Aqua) Opportunities, challenges, research gaps using TIR PMW observations also explored, with either polar-orbiting or geostationary satellites. We identified calibrated dataset (e.g., processed, harmonized, standardized) extremely limited each constellation, multiple instruments, make it fully useful entire mission period. A few problematic methodological concepts identified, images longer time series. Using only images, acquired different calendar months years, would not true annual over study period because they can be influenced by seasonal variations. To estimate warming cooling daytime, nighttime, diurnal use MODIS could useful, even though does acquire during maximum minimum daily cycle. indicated further investigations into those recommended directions overcome most these limitations.

Language: Английский

Citations

25
A New Deep Learning Restricted Boltzmann Machine for Energy Consumption Forecasting DOI Open Access
Aoqi Xu, Man‐Wen Tian,

Behnam Firouzi

et al.

Sustainability, Journal Year: 2022, Volume and Issue: 14(16), P. 10081 - 10081

Published: Aug. 15, 2022

A key issue in the desired operation and development of power networks is knowledge load growth electricity demand coming years. Mid-term forecasting (MTLF) has an important rule planning optimal use systems. However, MTLF a complicated problem, lot uncertain factors variables disturb consumption pattern. This paper presents practical approach for MTLF. new deep learning restricted Boltzmann machine (RBM) proposed modelling energy consumption. The contrastive divergence algorithm presented tuning parameters. All parameters RBMs, number input variables, type inputs, also layer neuron numbers are optimized. statistical suggested to determine effective variables. In addition climate such as temperature humidity, effects other economic investigated. Finally, using simulated real-world data examples, it shown that one year ahead, mean absolute percentage error (MAPE) peak less than 5%. Moreover, 24-h pattern forecasting, MAPE all days

Language: Английский

Citations

33
Comparison of machine learning models for flood forecasting in the Mahanadi River Basin, India DOI Creative Commons
Sanjeev Sharma, Sangeeta Kumari

Journal of Water and Climate Change, Journal Year: 2024, Volume and Issue: 15(4), P. 1629 - 1652

Published: March 16, 2024

ABSTRACT Developing accurate flood forecasting models is necessary for control, water resources and management in the Mahanadi River Basin. In this study, convolutional neural network (CNN) integrated with random forest (RF) support vector regression (SVR) making a hybrid model (CNN–RF CNN–SVR) where CNN used as feature extraction technique while RF SVR are models. These compared RF, SVR, artificial (ANN). The influence of training–testing data division on performance has been tested. Hyperparameter sensitivity analyses performed to select best value hyperparameters exclude nonsensitive hyperparameters. Two hydrological stations (Kantamal Kesinga) selected case studies. Results indicated that CNN–RF performs better than other both stations. addition, it found improved accuracy forecasting. results show models’ at 50–50% division. Validation not overfitting or underfitting. demonstrate can be potential river basins.

Language: Английский

Citations

8
Reconstructing daytime and nighttime MODIS land surface temperature in desert areas using multi-channel singular spectrum analysis DOI Creative Commons
Fahime Arabi Aliabad, Mohammad Zare, Hamid Reza Ghafarian Malamiri

et al.

Ecological Informatics, Journal Year: 2024, Volume and Issue: unknown, P. 102830 - 102830

Published: Sept. 1, 2024

Language: Английский

Citations

6
Automated Mapping of Wetland Ecosystems: A Study Using Google Earth Engine and Machine Learning for Lotus Mapping in Central Vietnam DOI Open Access
Pham Huu Ty, Nguyen Hao Quang,

Khac-Phuc Le

et al.

Water, Journal Year: 2023, Volume and Issue: 15(5), P. 854 - 854

Published: Feb. 22, 2023

Wetlands are highly productive ecosystems with the capability of carbon sequestration, providing an effective solution for climate change. Recent advancements in remote sensing have improved accuracy mapping wetland types, but there remain challenges accurate and automatic mapping, additional requirements complex input data a number types natural habitats. Here, we propose approach using Google Earth Engine (GEE) to automate extraction water bodies growing lotus, type high economic cultural values central Vietnam. Sentinel-1 was used K-Means clustering, whilst Sentinel-2 combined machine learning smile Random Forest (sRF) Gradient Tree Boosting (sGTB) models map areas lotus. The derived from S-1 images confidence (F1 = 0.97 Kappa coefficient 0.94). sGTB outperformed sRF model deliver growth (overall 0.95, 0.92, Precision 0.93, F1 0.93). total lotus area estimated at 145 ha distributed low land study site. Our proposed framework is simple reliable technique, has scalable potential GEE, capable extension other large-scale worldwide.

Language: Английский

Citations

16
Fuzzy Similarity Analysis of Effective Training Samples to Improve Machine Learning Estimations of Water Quality Parameters Using Sentinel-2 Remote Sensing Data DOI Creative Commons
Alireza Taheri Dehkordi, Mohammad Javad Valadan Zoej, Ali Mehran

et al.

IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, Journal Year: 2024, Volume and Issue: 17, P. 5121 - 5136

Published: Jan. 1, 2024

Continuous monitoring of Water Quality Parameters (WQPs) is crucial due to the global degradation water quality, primarily caused by climate change and population growth. Typically, Machine Learning (ML) models are employed retrieve WQPs, but they require a large amount training samples accurately capture data relationships. Even with sufficient data, discrepancies still exist between values predicted in-situ WQPs. This study proposes Fuzzy Similarity Analysis (FSA) technique enhance ML estimates WQPs using prediction errors in Effective Training Samples (ETS). The method was successfully applied Turbidity (Turb) Specific Conductance (SC) Lake Houston, USA, Sentinel-2 remote sensing data. Three algorithms, namely Mixture Density Networks, Support Vector Regression, Partial Least Squares were tested evaluate method's effectiveness. results showed that FSA significantly improved accuracy all predictions. improvement resulted up 9.15% reduction Mean Absolute Percentage Error (MAPE) 12% increase R2 for Turb, while SC, improvements 5.47% MAPE 7% R2. adaptability proposed other various satellite different promising quality inland waters.

Language: Английский

Citations

6
Evaluating the impacts of anthropogenic, climate, and land use changes on streamflow DOI Creative Commons
Hossein Ruigar, Samad Emamgholizadeh, Saeid Gharechelou

et al.

Journal of Water and Climate Change, Journal Year: 2024, Volume and Issue: 15(4), P. 1885 - 1905

Published: March 15, 2024

ABSTRACT Several factors, including natural and human-induced, can affect river discharge. This study aims to examine the influence of land use changes climate change on monthly average streamflow time series in Talar River basin, northern Iran. To investigate impact human namely point source operations, streamflow, DBEST method was used detect any breakpoint caused by gradual climate. The SWAT model simulate basin at Kiakola Shirghah stations, between 2001 2020. maps were created for years 2019. Calibration validation station showed that Nash-Sutcliffe (NSE) had an efficiency 0.8 0.76, respectively, while station, same values 0.84 0.75. Findings revealed activities, specifically combined a 60% River. They further combination harvesting played most significant role basin's outflow scale.

Language: Английский

Citations

5