Combining GF-2 and RapidEye satellite data for mapping mangrove species using ensemble machine-learning methods

Peng, Liheng; Liu, Kai; Cao, Jingjing; Zhu, Yuanhui; Li, Fengshou; Liu, Lin

doi:10.1080/01431161.2019.1648907

Cited by 30 publications

(11 citation statements)

References 82 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…Two barrel‐mounted panchromatic cameras and two sets of multispectral cameras are assembled on this China‐made satellite, which are collectively referred to as the Panchromatic and Multispectral Sensor (PMS). One 0.8‐m‐resolution panchromatic band and four 3.2‐m‐resolution multispectral bands are provided in the PMS data (Peng et al, 2019). Four GF‐2 raw images acquired on February 27, 2019, two images obtained on April 7, 2019, two images captured on February 2, 2018, one image recorded on February 17, 2018, two images acquired on April 8, 2017, and one image obtained on February 2, 2017 were adopted in the study.…”

Section: Case Study—puge County Southwest Chinamentioning

confidence: 99%

Automatic identification of buildings vulnerable to debris flows in Sichuan Province, China, by GIS analysis and Deep Encoding Network methods

Liu

2022

J Flood Risk Management

View full text Add to dashboard Cite

Debris flows commonly cause tremendous damage to buildings in mountainous areas. The identification of buildings susceptible to debris flows is vital for settlement risk management. The efficient identification method is a major issue limiting the targeted regional policy setting. By combining geographic information system (GIS) and Deep Encoding Network (DE-Net) methods, we proposed an automatic identification method for buildings highly susceptible to debris flows with large-scale digital elevation data and high-resolution remote sensing imagery. The judgment criteria were based on a vulnerability matrix containing different threshold values of the horizontal distance (HD) and vertical distance (VD) between buildings and channels obtained from the statistics of 362 buildings destroyed by 23 debris flows and the maximum debris flow depths of 26 events, respectively. Five steps, which are debris flow channel extraction, building extraction, building cluster segmentation, distance calculation, and building classification, were implemented in the method. A case study in Puge County, Sichuan Province, demonstrated the high identification potential of the method for buildings susceptible to debris flows in large areas with only scarce information available. The identification results provide valuable information regarding high-risk residential areas to governments and facilitate targeted measure design in these areas in the initial planning stage.

show abstract

Section: Case Study—puge County Southwest Chinamentioning

confidence: 99%

Automatic identification of buildings vulnerable to debris flows in Sichuan Province, China, by GIS analysis and Deep Encoding Network methods

Liu

2022

J Flood Risk Management

View full text Add to dashboard Cite

show abstract

“…AdaBoost combines single-node decision trees named stumps, in series: once a tree is trained, the subsequent tree tries to correct the errors in the previous one, by adapting the weights associated with each tree in order to take the final decision. AdaBoost has been recently applied to different ecological applications [50,51], while few works are also focused on microalgae [16,26,52].…”

Section: Introductionmentioning

confidence: 99%

Machine Learning Methods Applied to the Prediction of Pseudo-nitzschia spp. Blooms in the Galician Rias Baixas (NW Spain)

Aláez

Palenzuela

Spyrakos

et al. 2021

IJGI

View full text Add to dashboard Cite

This work presents new prediction models based on recent developments in machine learning methods, such as Random Forest (RF) and AdaBoost, and compares them with more classical approaches, i.e., support vector machines (SVMs) and neural networks (NNs). The models predict Pseudo-nitzschia spp. blooms in the Galician Rias Baixas. This work builds on a previous study by the authors (doi.org/10.1016/j.pocean.2014.03.003) but uses an extended database (from 2002 to 2012) and new algorithms. Our results show that RF and AdaBoost provide better prediction results compared to SVMs and NNs, as they show improved performance metrics and a better balance between sensitivity and specificity. Classical machine learning approaches show higher sensitivities, but at a cost of lower specificity and higher percentages of false alarms (lower precision). These results seem to indicate a greater adaptation of new algorithms (RF and AdaBoost) to unbalanced datasets. Our models could be operationally implemented to establish a short-term prediction system.

show abstract

“…Como método de classificação, o Ensemble demonstrou ser eficaz, com maior acurácia em relação aos modelos independentes (MINOWA, 2008;ENGLER et al, 2013;BUI et al, 2019;GIGOVIĆ et al, 2019;PENG et al, 2020). Com base no potencial dessa técnica, é imprescindível testá-la na estimação de variáveis florestais, a fim de verificar se há ganhos em precisão e acurácia no Ensemble em relação aos modelos que utilizam uma única técnica preditiva.…”

Section: Introductionunclassified

APRENDIZADO DE MÁQUINA E REGRESSÃO LINEAR NA ESTIMATIVA DO VOLUME DE Eucalyptus NA AMAZÔNIA ORIENTAL

Neto

Moura

Araujo

et al. 2021

BIOFIX

View full text Add to dashboard Cite

O crescimento da área florestada trouxe a necessidade de aprimorar técnicas que proporcionem estimativas mais precisas e acuradas do volume de árvores individuais. Neste contexto, técnicas de aprendizagem de máquina foram aplicadas para este fim, contudo, ainda não foi testado o uso combinado dessas técnicas (Ensemble). Assim, a hipótese deste estudo é que o Ensemble proporciona estimativas mais precisas e acuradas do volume que os métodos convencionais de estimativa e aprendizado de máquina isoladamente. Com isso, objetivou-se avaliar a estimativa do volume individual do Eucalyptus spp. utilizando técnicas isoladas e conjuntas de aprendizado de máquina. O estudo foi realizado em duas fazendas em Paragominas, Pará. Foram cubadas 90 árvores, distribuídas em 4 clones, com 6 anos idade e 1.111 indivíduos.ha-1. Utilizaram-se 70% das árvores para o ajuste e 30% para a validação. Aplicaram-se quatro métodos distintos de estimação: regressão linear (Modelo de Schumacher-Hall), máquinas de vetores de suporte (SVM), redes neurais artificiais (RNA) e Ensemble (SVM, RNA e modelos lineares generalizados - GLM) com distribuição gaussiana). O Ensemble resultou em maior coeficiente de determinação, enquanto a RNA no menor. O erro padrão da estimativa variou entre 8,21 % e 12,99 %, para as técnicas Ensemble e RNA, respectivamente. Os modelos preditivos apresentaram generalização acurada e precisa, com RMSE entre 7,01 % e 13,17 %. Schumacher-Hall e Ensemble apresentaram leve tendência em subestimar o volume, já SVM e RNA superestimaram. Os modelos de Schumacher-Hall e Ensemble foram os mais precisos e acurados à predição do volume para os dados utilizados.

show abstract

Combining GF-2 and RapidEye satellite data for mapping mangrove species using ensemble machine-learning methods

Cited by 30 publications

References 82 publications

Automatic identification of buildings vulnerable to debris flows in Sichuan Province, China, by GIS analysis and Deep Encoding Network methods

Automatic identification of buildings vulnerable to debris flows in Sichuan Province, China, by GIS analysis and Deep Encoding Network methods

Machine Learning Methods Applied to the Prediction of Pseudo-nitzschia spp. Blooms in the Galician Rias Baixas (NW Spain)

APRENDIZADO DE MÁQUINA E REGRESSÃO LINEAR NA ESTIMATIVA DO VOLUME DE Eucalyptus NA AMAZÔNIA ORIENTAL

Contact Info

Product

Resources

About