Classification and regression with random forests as a standard method for presence-only data SDMs: A future conservation example using China tree species

Zhang, Lei; Huettmann, Falk; Liu, Shirong; Park, Sun; Zhang, Xudong; Mi, Chunrong

doi:10.1016/j.ecoinf.2019.05.003

Cited by 58 publications

(45 citation statements)

References 54 publications

Supporting

Mentioning

Contrasting

Unclassified

Order By: Relevance

“…1General framework for species distribution modeling by random forests (classification tree (CT) and regression tree (RT) algorithms) and R functions used in this study. Adopted from Zhang et al [9]; * recommended methods.…”

Section: Methods Detailsmentioning

confidence: 99%

“…When converting numerical predictions into binary predictions, the optimal threshold varies with the choice of threshold-setting method. However, the choice of thresholds has practical consequences for estimating of RF model performance and species range shifts under climate change [9]. Hence, the use of an appropriate threshold appears to be a better choice for binary conversions for RF.…”

Section: Methods Detailsmentioning

confidence: 99%

“…Hence, the use of an appropriate threshold appears to be a better choice for binary conversions for RF. Zhang et al [9] demonstrated that the four threshold methods (MaxKappa, MaxOA, MinROCdist, and MaxTSS) based on the composite model accuracy measures (Kappa, TSS, ROC, OA) are promising objective methods for binary conversions of continuous predictions when presence-only data are available. These four methods can also produce the same threshold using either presence-only data or presence/absence data for CT and RT models.…”

Section: Methods Detailsmentioning

confidence: 99%

See 2 more Smart Citations

The use of classification and regression algorithms using the random forests method with presence-only data to model species’ distribution

et al. 2019

Self Cite

View full text Add to dashboard Cite

show abstract

Section: Methods Detailsmentioning

confidence: 99%

Section: Methods Detailsmentioning

confidence: 99%

Section: Methods Detailsmentioning

confidence: 99%

See 1 more Smart Citation

The use of classification and regression algorithms using the random forests method with presence-only data to model species’ distribution

et al. 2019

Self Cite

View full text Add to dashboard Cite

show abstract

“…SDM models were built with maxent [58], and Random Forest [59] methods using presence/background data and noncollinear predictor variables ( Table 2). Random forest algorithm has gained popularity in ecological niche modelling due to its good performance as reported in Zhang et al [60] and Shabani, Kumar, and Ahmadi [61] while the Maxent algorithm is popular due to its high predictive power than the traditional logistic regression [14] regardless of available sample size [17]. Species presence data were randomly split into two sub-sets; 70% for model fitting and 30% for testing community level predictions.…”

Section: Current and Future Species Distribution Modelling Model Fittmentioning

confidence: 99%

Assessing distribution changes of selected native and alien invasive plant species under changing climatic conditions in Nyeri County, Kenya

Sichangi

Mundia

Waititu

2020

Preprint

View full text Add to dashboard Cite

Changes in climatic conditions increases the risks of native and alien taxa expanding in geographical range and causing habitat transformations. The role of climate change in enhancing bio-invasions in local natural environments need to be assessed to guide on effective species management policy formulations. In this present study, we used species presence records, predictor variables and an ensemble of General Circulation Models data to predict suitable ecological niches for five of the selected invasive plant species within Nyeri County, Kenya. We predicted species distributions under RCP2.6, RCP4.5, and RCP8.5 emission scenarios for the years 2050 and 2070. We analysed species distribution changes to identify invasive species requiring immediate management action. Our analysis indicated that three of the five study species were suitable in ~50% of the study area while the other two were suitable in ~30% under the current climate. Lantana camara L. and Solanum campylacanthum Hochst. ex A.Rich species would experience the largest range shift distance of ~6 – 10km and the largest habitat gain of ~12 – 33% in the future. Caesalpinia decapetala (Roth) Alston, Opuntia stricta (Haw.) Haw. and Senna didymobotrya (Fresen.) H.S. Irwin & Barneby species on the other hand would have a decline in habitat range under future climate change scenarios. Although, S. didymobotrya is considered a native species, it would lose half of its current suitable habitat in the future. Range shift analysis showed all study species would generally shift to the north west direction or towards the Aberdare ranges. From this study we conclude that i nvasive species management programs for smaller geographical areas ought to consider projecting species distributions under climate change scenarios to identify areas with high possible biodiversity changes. This would be important to conservationists when prioritizing management actions of invasive species in the region where data on invasive species is still limited.

show abstract

“…The main objective is to evaluate machine learning models as useful, universal, and accurate multi-hazard mapping products that can be applied by land use managers and planners. Based on a review of the literature, we have selected a set of machine learning models including; generalized linear model (GLM) 55 – 57 , random forest (RF) 17 , 58 , 59 , a support vector machine (SVM) 60 – 62 , boosted regression trees (BRT) 63 – 65 , mixture discriminate analysis (MDA) 66 , 56 , multivariate adaptive regression splines (MARS) 67 , 56 , 68 , and functional discriminant analysis (FDA) 17 , 66 for multi-hazard mapping. Finally, based on the accuracies of the models, available data, and the sources of models, the SVM, GLM and FDA algorithms were used to map hazards in the study area.…”

Section: Introductionmentioning

confidence: 99%

A machine learning framework for multi-hazards modeling and mapping in a mountainous area

Yousefi

Pourghasemi

Emami

et al. 2020

Sci Rep

View full text Add to dashboard Cite

this study sought to produce an accurate multi-hazard risk map for a mountainous region of iran. the study area is in southwestern iran. the region has experienced numerous extreme natural events in recent decades. This study models the probabilities of snow avalanches, landslides, wildfires, land subsidence, and floods using machine learning models that include support vector machine (SVM), boosted regression tree (BRT), and generalized linear model (GLM). Climatic, topographic, geological, social, and morphological factors were the main input variables used. the data were obtained from several sources. The accuracies of GLM, SVM, and functional discriminant analysis (FDA) models indicate that SVM is the most accurate for predicting landslides, land subsidence, and flood hazards in the study area. GLM is the best algorithm for wildfire mapping, and FDA is the most accurate model for predicting snow avalanche risk. The values of AUC (area under curve) for all five hazards using the best models are greater than 0.8, demonstrating that the model's predictive abilities are acceptable. A machine learning approach can prove to be very useful tool for hazard management and disaster mitigation, particularly for multi-hazard modeling. the predictive maps produce valuable baselines for risk management in the study area, providing evidence to manage future human interaction with hazards. Human interactions with natural extreme events, or hazards, are increasing globally 1. Natural disasters have affected people and natural environments generating vast economic losses around the world. However, in some developed counties disasters have been decreasing since 1900 2,3. Hazard is the probability of occurrence in a specified period and within a given area of a potentially damaging of a given magnitude 4,5. The definition incorporates the concepts of location (where?), time (when, or how frequently?) and magnitude (how large?). Total risk (R) means the expected number of lives lost, person injured, damage to property, or disruption of economic activity due to a particular natural phenomenon, and is therefore the product of specific risk (RS) and elements at risk (E) 6. In addition, RS is the expected degree of loss due to a natural phenomenon. Landscapes around the world are reflections of diverse natural processes. The probabilities of extreme natural events are typically greater in more natural areas and are, in fact, extensions of natural systems. Exposure of people to these extreme natural processes could be reduced and limited if predictive models based on new approaches and deeper knowledge of effective factors were employed 7. Mountainous areas are commonly sites of snow avalanches 8,9 , landslides 4,10 , floods 11,12 , mudflows 13 , ice avalanches 14 , soil erosion 15-17 , rock falls 18 , and wildfires 19-24. Most studies focus on a single hazard, even when there are multiple hazardous processes affecting the same landscapes 8,25-30. However, hazards sometimes interact with each other. Sometimes, the mitigation of one

show abstract

Classification and regression with random forests as a standard method for presence-only data SDMs: A future conservation example using China tree species

Cited by 58 publications

References 54 publications

The use of classification and regression algorithms using the random forests method with presence-only data to model species’ distribution

The use of classification and regression algorithms using the random forests method with presence-only data to model species’ distribution

Assessing distribution changes of selected native and alien invasive plant species under changing climatic conditions in Nyeri County, Kenya

A machine learning framework for multi-hazards modeling and mapping in a mountainous area

Contact Info

Product

Resources

About