Assessing the robustness of Random Forests to map land cover with high resolution satellite image time series over large areas

By Charlotte Pelletier, Silvia Valero, Jordi Inglada, Nicolas Champion, Gérard Dedieu. 2016

555 citations recorded by the source corpus.

Subfields: Atmospheric Science, Ecology, Media Technology

View Paper: https://www.semanticscholar.org/paper/2e37205d9d699622a505d4f02cb3a07365a73d53

How researchers see this paper

The researchers whose trusted networks rank it highest, strongest advocate first. Every researcher gets a different order.

In Atmospheric Science

In Ecology

In Media Technology

Citations in the corpus

Nominal 30-m Cropland Extent Map of Continental Africa by Integrating Pixel-Based and Object-Based Algorithms Using Sentinel-2 and Landsat-8 Data on Google Earth Engine (2017, background): “It is more robust, relatively faster in speed of classification, and easier to implement than many other classifiers [77].”
Operational High Resolution Land Cover Map Production at the Country Scale Using Satellite Image Time Series (2017, background): “They are also stable with respect to the choice of parameters [46] which makes them excellent candidates for operational processing chains.”
A 30-m landsat-derived cropland extent product of Australia and China using random forest machine learning algorithm on Google Earth Engine cloud computing platform (2018, unknown): “The RF classi ﬁ er is more robust than single decision tree (Chan and Paelinckx, 2008) and easier to be implemented than many other advanced classi ﬁ ers such as Support Vector Mission (SVM) (Pelletier et al., 2016) as recently established by Xiong et al. (2017a).”
Recent Applications of Landsat 8/OLI and Sentinel-2/MSI for Land Use and Land Cover Mapping: A Systematic Review (2020, background): “It is reasonable to suppose that better classification accuracies can be achieved if all bands throughout the SITS are used as features, instead of a greenness-related spectral index formulated from only two bands per image [71] because the full optical spectrum comprises wavelengths which are sensi”
Sentinel-2 cropland mapping using pixel-based and object-based time-weighted dynamic time warping analysis (2018, methodology): “Previous studies have also reported that RF achieves less accurate re-sults when a small number of training samples is used (Millard and Richardson, 2015; Valero et al., 2016).”
Land cover 2.0 (2018, methodology): “All non-parametric classi ﬁ ers require appropriate con ﬁ guration prior to their application and the optimal con ﬁ guration of these classi ﬁ ers has itself become the subject of investigation (e.g., Pelletier et al. 2016; Paci ﬁ ci et al. 2008).”
Detecting Mammals in UAV Images: Best Practices to address a substantially Imbalanced Dataset with Deep Learning (2018, methodology): “We use the same object proposals and feature extraction methods proposed in the original work, but replace the original exemplar SVM with a random forest classiﬁer, which is a solid baseline in remote sensing, providing nonlinear response at no extra computational expense (Pelletier et al., 2016).”
Temporal Convolutional Neural Network for the Classification of Satellite Image Time Series (2018, unknown): “To complete the experiments of Section IV, the RF implementation from Scikit-Learn has been used with standard parameter settings [13]: 500 trees at the maximum depth, and a number of randomly selected variables per node equals to the square root of the total number of features.”
GLC_FCS30: global land-cover product with fine classification system at 30 m using time-series Landsat imagery (2020, methodology): “…gives a higher classification accuracy, and is less sensitive to noise and feature selection than other widely used classifiers such as the support vector machine, artificial neural network, and classification and regression tree (Belgiu and Drăguţ, 2016; Du et al., 2015; Pelletier et al., 2016).”
Self-Supervised Pretraining of Transformers for Satellite Image Time Series Classification (2020, methodology): “some advantages such as short training time, easy parameterization, and high robustness to high-dimensional input features [13, 64].”