Modeling Species Distribution and Change Using Random Forest

Evans, Jeffrey S.; Murphy, Melanie A.; Holden, Zachary A.; Cushman, Samuel A.

doi:10.1007/978-1-4419-7390-0_8

Cited by 270 publications

(259 citation statements)

References 55 publications

Supporting

Mentioning

238

Contrasting

Unclassified

Order By: Relevance

“…Default values (square root of the number of environmental variables for classification models, 1/3 the number of environmental variables for regression models), consistently produced the best results and were used in all models. Classification trees are highly sensitive to imbalances in the response variable, as the model will minimize the overall error rate at the expense of the error rate of the minority class (Chen et al, 2004;Evans et al, 2011). As our taxa data were skewed (>10:1 ratio of absences to presences for all species), we set the "cutoff " parameter in "randomForest" equal to species prevalence for each presence-absence model (Guo et al, 2004), which greatly reduced omission errors without affecting overall model performance.…”

Section: Random Forest (Rf)mentioning

confidence: 99%

High-Resolution Habitat Suitability Models for the Conservation and Management of Vulnerable Marine Ecosystems on the Louisville Seamount Chain, South Pacific Ocean

et al. 2017

View full text Add to dashboard Cite

Vulnerable marine ecosystems (VMEs) are ecosystems at risk from the effects of fishing or other kinds of disturbance, as determined by the vulnerability of their components (e.g., habitats, communities, or species). Habitat suitability modeling is being used increasingly to predict distribution patterns of VME indicator taxa in the deep sea, where data are particularly sparse, and the models are considered useful for marine ecosystem management. The Louisville Seamount Chain is located within the South Pacific Regional Fishery Management Organization (SPRFMO) Convention Area, and some seamounts are the subject of bottom trawling for orange roughy by the New Zealand fishery. The aim of the present study was to produce high-resolution habitat suitability maps for VME indicator taxa and VME habitat on these seamounts, in order to evaluate the feasibility of designing within-seamount spatial closures to protect VMEs. We used a multi-model habitat suitability mapping approach, based on bathymetric and backscatter data collected by multibeam echo sounder survey, and data collected by towed underwater camera for the stony coral and habitatforming VME indicator species Solenosmilia variabilis, as well as two taxa indicative of stony coral habitat (Brisingida, Crinoidea). Model performance varied among the different model types used (Boosted Regression Tree, Random Forest, Generalized Additive Models), but abundance-based models consistently out-performed models based on presence-absence data. Uncertainty for ensemble models (combination of all models) was lower overall compared to the other models. Maps resulting from our models showed that suitable habitat for S. variabilis is distributed around the summitslope break of seamounts, and along ridges that extend down the seamount flanks. Only the flat, soft sediment summits are predicted to be unsuitable habitat for this stony coral species. We translated a definition for stony coral-reef habitat into a S. variabilis abundance-based threshold in order to use our models to map this VME habitat. These maps showed that coral-reef occurred in small and isolated patches, and that most of the seabed on these seamounts is predicted to be unsuitable Rowden et al.High-Resolution VME Habitat Suitability Maps habitat for this VME. We discuss the implications of these results for spatial management closures on the Louisville Seamount Chain seamounts and the wider SPRFMO area, and future modeling improvements that could aid efforts to use habitat suitability maps for managing the impact of fishing on VMEs.

show abstract

Section: Random Forest (Rf)mentioning

confidence: 99%

High-Resolution Habitat Suitability Models for the Conservation and Management of Vulnerable Marine Ecosystems on the Louisville Seamount Chain, South Pacific Ocean

et al. 2017

View full text Add to dashboard Cite

show abstract

“…Given the minor differences in the outcomes of the correlation and the random forest analysis for the UK, both methods appear recommendable. Generally, the strength of the random forest algorithm is that it can handle interactions and nonlinearities among variables, and thus identify non-intuitive relationships (Evans et al, 2011;Hastie et al, 2009). Furthermore, random forests are robust to noise (Breiman, 2001;Hastie et al, 2009), and the bootstrap sampling provides a way to account for the uncertainty of the impact data.…”

Section: Performance Of Drought Indicatorsmentioning

confidence: 99%

A quantitative analysis to objectively appraise drought indicators and model drought impacts

Bachmair

Svensson

Hannaford

et al. 2016

Hydrol. Earth Syst. Sci.

120

101

View full text Add to dashboard Cite

Abstract. Drought monitoring and early warning is an important measure to enhance resilience towards drought. While there are numerous operational systems using different drought indicators, there is no consensus on which indicator best represents drought impact occurrence for any given sector. Furthermore, thresholds are widely applied in these indicators but, to date, little empirical evidence exists as to which indicator thresholds trigger impacts on society, the economy, and ecosystems. The main obstacle for evaluating commonly used drought indicators is a lack of information on drought impacts. Our aim was therefore to exploit text-based data from the European Drought Impact report Inventory (EDII) to identify indicators that are meaningful for region-, sector-, and season-specific impact occurrence, and to empirically determine indicator thresholds. In addition, we tested the predictability of impact occurrence based on the best-performing indicators. To achieve these aims we applied a correlation analysis and an ensemble regression tree approach, using Germany and the UK (the most data-rich countries in the EDII) as test beds. As candidate indicators we chose two meteorological indicators (Standardized Precipitation Index, SPI, and Standardized Precipitation Evaporation Index, SPEI) and two hydrological indicators (streamflow and groundwater level percentiles). The analysis revealed that accumulation periods of SPI and SPEI best linked to impact occurrence are longer for the UK compared with Germany, but there is variability within each country, among impact categories and, to some degree, seasons. The median of regression tree splitting values, which we regard as estimates of thresholds of impact occurrence, was around −1 for SPI and SPEI in the UK; distinct differences between northern/northeastern vs. southern/central regions were found for Germany. Predictions with the ensemble regression tree approach yielded reasonable results for regions with good impact data coverage. The predictions also provided insights into the EDII, in particular highlighting drought events where missing impact reports may reflect a lack of recording rather than true absence of impacts. Overall, the presented quantitative framework proved to be a useful tool for evaluating drought indicators, and to model impact occurrence. In summary, this study demonstrates the information gain for drought monitoring and early warning through impact data collection and analysis. It highlights the important role that quantitative analysis with impact data can have in providing "ground truth" for drought indicators, alongside more traditional stakeholder-led approaches.

show abstract

“…Furthermore, the lack of distributional assumptions and the use of a subset of variables in each splitting node result in several advantages for classification trees over traditional methods such as discriminant function analyses or linear discriminant analyses (see for instance [62] for a discussion). RF has been applied in ecology [62,63], bioinformatics [64], the health sciences [65] and recently in statistical phylogeography [66].…”

Section: (E) Random Forest Analysesmentioning

confidence: 99%

Identifying cryptic diversity with predictive phylogeography

et al. 2016

View full text Add to dashboard Cite

Identifying units of biological diversity is a major goal of organismal biology. An increasing literature has focused on the importance of cryptic diversity, defined as the presence of deeply diverged lineages within a single species. While most discoveries of cryptic lineages proceed on a taxon-by-taxon basis, rapid assessments of biodiversity are needed to inform conservation policy and decision-making. Here, we introduce a predictive framework for phylogeography that allows rapidly identifying cryptic diversity. Our approach proceeds by collecting environmental, taxonomic and genetic data from codistributed taxa with known phylogeographic histories. We define these taxa as a reference set, and categorize them as either harbouring or lacking cryptic diversity. We then build a random forest classifier that allows us to predict which other taxa endemic to the same biome are likely to contain cryptic diversity. We apply this framework to data from two sets of disjunct ecosystems known to harbour taxa with cryptic diversity: the mesic temperate forests of the Pacific Northwest of North America and the arid lands of Southwestern North America. The predictive approach presented here is accurate, with prediction accuracies placed between 65% and 98.79% depending of the ecosystem. This seems to indicate that our method can be successfully used to address ecosystemlevel questions about cryptic diversity. Further, our application for the prediction of the cryptic/non-cryptic nature of unknown species is easily applicable and provides results that agree with recent discoveries from those systems. Our results demonstrate that the transition of phylogeography from a descriptive to a predictive discipline is possible and effective.

show abstract

Modeling Species Distribution and Change Using Random Forest

Cited by 270 publications

References 55 publications

High-Resolution Habitat Suitability Models for the Conservation and Management of Vulnerable Marine Ecosystems on the Louisville Seamount Chain, South Pacific Ocean

High-Resolution Habitat Suitability Models for the Conservation and Management of Vulnerable Marine Ecosystems on the Louisville Seamount Chain, South Pacific Ocean

A quantitative analysis to objectively appraise drought indicators and model drought impacts

Identifying cryptic diversity with predictive phylogeography

Contact Info

Product

Resources

About