An assessment of the effectiveness of a random forest classifier for land-cover classification

Rodríguez-Galiano, Víctor; Ghimire, Bardan; Rogan, John; Chica-Olmo, Mario; Rigol-Sánchez, J.P.

doi:10.1016/j.isprsjprs.2011.11.002

Cited by 2,166 publications

(1,283 citation statements)

References 64 publications

Supporting

Mentioning

1,076

Contrasting

Unclassified

Order By: Relevance

“…Our results clearly show this actually introduces bias and we advocate for future users of random forest to carefully select training data that are representative and proportional to the composition of actual landscape. Second, following other researchers who found that random forest outperforms other methods for prediction and classification (e.g., Cushman et al 2010;Evans et al 2011;Rodriguez-Galiano et al 2012;Schneider 2012), we expected that random forest would outperform logistic regression and the naïve model based on distance to forest edge. Consistent with this hypothesis, random forest was the highest performing method in all three nations comprising Borneo.…”

Section: Discussionmentioning

confidence: 76%

Multiple-scale prediction of forest loss risk across Borneo

et al. 2017

View full text Add to dashboard Cite

Context The forests of Borneo have among the highest biodiversity and also the highest forest loss rates on the planet. Objectives Our objectives were to: (1) compare multiple modelling approaches, (2) evaluate the utility of landscape composition and configuration as predictors, (3) assess the influence of the ratio of forest loss and persistence points in the training sample, (4) identify the multiple-scale drivers of recent forest loss and (5) predict future forest loss risk across Borneo. Methods We compared random forest machine learning and logistic regression in a multi-scale approach to model forest loss risk between 2000 and 2010 as a function of topographical variables and landscape structure, and applied the highest performing model to predict the spatial pattern of forest loss risk between 2010 and 2020. We utilized a naïve model as a null comparison and used the total operating characteristic AUC to assess model performance. Results Our analysis produced five main results. We found that: (1) random forest consistently outperformed logistic regression and the naïve model; (2) including landscape structure variables substantially improved predictions; (3) a ratio of occurrence to nonoccurrence points in the training dataset that does not match the actual ratio in the landscape biases the predictions of both random forest and logistic regression; (4) forest loss risk differed between the three nations that comprise Borneo, with patterns in Kalimantan highly related to distance from the edge of the previous frontier of forest loss, while Malaysian Borneo showed a more diffuse pattern related to the structure of the landscape; (5) we predicted continuing 123Landscape Ecol (2017) 32:1581-1598 DOI 10.1007 very high rates of forest loss in the 2010-2020 period, and produced maps of the expected risk of forest loss across the full extent of Borneo. Conclusions These results confirm that multiplescale modelling using landscape metrics as predictors in a random forest modelling framework is a powerful approach to landscape change modelling. There is immense immanent risk to Borneo's forests, with clear spatial patterns of risk related to topography and landscape structure that differ between the three nations that comprise Borneo.

show abstract

Section: Discussionmentioning

confidence: 76%

Multiple-scale prediction of forest loss risk across Borneo

et al. 2017

View full text Add to dashboard Cite

show abstract

“…The principle of ensemble classifiers is that a large collection of weaker classifiers (individual CTs in this case) can be used to create a strong classifier. RF involves the construction of large number of individual trees from the training data (Rodriguez‐Galiano, Ghimire, Rogan, Chica‐Olmo, & Rigol‐Sanchez, 2012). How the trees are constructed differs from CT in that a random selection of training data is used for each tree so that each tree is trained on a different set of data.…”

Section: Methodsmentioning

confidence: 99%

Impact of ecological redundancy on the performance of machine learning classifiers in vegetation mapping

MacIntyre

Niekerk

Dobrowolski

et al. 2018

Ecology and Evolution

View full text Add to dashboard Cite

Vegetation maps are models of the real vegetation patterns and are considered important tools in conservation and management planning. Maps created through traditional methods can be expensive and time‐consuming, thus, new more efficient approaches are needed. The prediction of vegetation patterns using machine learning shows promise, but many factors may impact on its performance. One important factor is the nature of the vegetation–environment relationship assessed and ecological redundancy. We used two datasets with known ecological redundancy levels (strength of the vegetation–environment relationship) to evaluate the performance of four machine learning (ML) classifiers (classification trees, random forests, support vector machines, and nearest neighbor). These models used climatic and soil variables as environmental predictors with pretreatment of the datasets (principal component analysis and feature selection) and involved three spatial scales. We show that the ML classifiers produced more reliable results in regions where the vegetation–environment relationship is stronger as opposed to regions characterized by redundant vegetation patterns. The pretreatment of datasets and reduction in prediction scale had a substantial influence on the predictive performance of the classifiers. The use of ML classifiers to create potential vegetation maps shows promise as a more efficient way of vegetation modeling. The difference in performance between areas with poorly versus well‐structured vegetation–environment relationships shows that some level of understanding of the ecology of the target region is required prior to their application. Even in areas with poorly structured vegetation–environment relationships, it is possible to improve classifier performance by either pretreating the dataset or reducing the spatial scale of the predictions.

show abstract

“…RF was selected for this study because it generally outperforms conventional classifiers such as the Gaussian maximum likelihood classifier [61,62], while performing favorably, or equally well, to other non-parametric approaches; e.g., CART [63,64], Support Vector Machines [32,65,66], Artificial Neural Networks [67], and K-Nearest Neighbor [68]. It is a powerful non-linear and non-parametric classifier that allows for fusion and aggregation of high-dimensional data from various sources (e.g., optical, SAR, and topography [30,69,70]; SAR and topography [21,58,71]; and optical and topography [72][73][74]).…”

Section: Image Classificationmentioning

confidence: 99%

Mapping the Dabus Wetlands, Ethiopia, Using Random Forest Classification of Landsat, PALSAR and Topographic Data

et al. 2017

View full text Add to dashboard Cite

Abstract:The Dabus Wetland complex in the highlands of Ethiopia is within the headwaters of the Nile Basin and is home to significant ecological communities and rare or endangered species. Its many interrelated wetland types undergo seasonal and longer-term changes due to weather and climate variations as well as anthropogenic land use such as grazing and burning. Mapping and monitoring of these wetlands has not been previously undertaken due primarily to their relative isolation and lack of resources. This study investigated the potential of remote sensing based classification for mapping the primary vegetation groups in the Dabus Wetlands using a combination of dry and wet season data, including optical (Landsat spectral bands and derived vegetation and wetness indices), radar (ALOS PALSAR L-band backscatter), and elevation (SRTM derived DEM and other terrain metrics) as inputs to the non-parametric Random Forest (RF) classifier. Eight wetland types and three terrestrial/upland classes were mapped using field samples of observed plant community composition and structure groupings as reference information. Various tests to compare results using different RF input parameters and data types were conducted. A combination of multispectral optical, radar and topographic variables provided the best overall classification accuracy, 94.4% and 92.9% for the dry and wet season, respectively. Spectral and topographic data (radar data excluded) performed nearly as well, while accuracies using only radar and topographic data were 82-89%. Relatively homogeneous classes such as Papyrus Swamps, Forested Wetland, and Wet Meadow yielded the highest accuracies while spatially complex classes such as Emergent Marsh were more difficult to accurately classify. The methods and results presented in this paper can serve as a basis for development of long-term mapping and monitoring of these and other non-forested wetlands in Ethiopia and other similar environmental settings.

show abstract

An assessment of the effectiveness of a random forest classifier for land-cover classification

Cited by 2,166 publications

References 64 publications

Multiple-scale prediction of forest loss risk across Borneo

Multiple-scale prediction of forest loss risk across Borneo

Impact of ecological redundancy on the performance of machine learning classifiers in vegetation mapping

Mapping the Dabus Wetlands, Ethiopia, Using Random Forest Classification of Landsat, PALSAR and Topographic Data

Contact Info

Product

Resources

About