blockCV: an R package for generating spatially or environmentally separated folds for k-fold cross-validation of species distribution models

Valavi, Roozbeh; Elith, Jane; Lahoz‐Monfort, José J.; Guillera‐Arroita, Gurutzeta

doi:10.1101/357798

Cited by 135 publications

(183 citation statements)

References 17 publications

Supporting

Mentioning

180

Contrasting

Unclassified

Order By: Relevance

“…In our first strategy for blocking, we divided our study region into a ‘checkerboard’ pattern with 53 equal‐sized square blocks, each with a size of ~83 × 83 km, using the ‘blockCV’ package (Valavi et al ) within the R ver. 3.4.1 statistical language environment (<http://www.r-project.org>).…”

Section: Methodsmentioning

confidence: 99%

Testing whether ensemble modelling is advantageous for maximising predictive performance of species distribution models

et al. 2020

Self Cite

View full text Add to dashboard Cite

Predictive performance is important to many applications of species distribution models (SDMs). The SDM ‘ensemble’ approach, which combines predictions across different modelling methods, is believed to improve predictive performance, and is used in many recent SDM studies. Here, we aim to compare the predictive performance of ensemble species distribution models to that of individual models, using a large presence–absence dataset of eucalypt tree species. To test model performance, we divided our dataset into calibration and evaluation folds using two spatial blocking strategies (checkerboard‐pattern and latitudinal slicing). We calibrated and cross‐validated all models within the calibration folds, using both repeated random division of data (a common approach) and spatial blocking. Ensembles were built using the software package ‘biomod2’, with standard (‘untuned’) settings. Boosted regression tree (BRT) models were also fitted to the same data, tuned according to published procedures. We then used evaluation folds to compare ensembles against both their component untuned individual models, and against the BRTs. We used area under the receiver‐operating characteristic curve (AUC) and log‐likelihood for assessing model performance. In all our tests, ensemble models performed well, but not consistently better than their component untuned individual models or tuned BRTs across all tests. Moreover, choosing untuned individual models with best cross‐validation performance also yielded good external performance, with blocked cross‐validation proving better suited for this choice, in this study, than repeated random cross‐validation. The latitudinal slice test was only possible for four species; this showed some individual models, and particularly the tuned one, performing better than ensembles. This study shows no particular benefit to using ensembles over individual tuned models. It also suggests that further robust testing of performance is required for situations where models are used to predict to distant places or environments.

show abstract

Section: Methodsmentioning

confidence: 99%

Testing whether ensemble modelling is advantageous for maximising predictive performance of species distribution models

et al. 2020

Self Cite

View full text Add to dashboard Cite

show abstract

“…To avoid this, we selected pseudo‐absence and background points within a convex hull around the presence records, extended with a buffer of 10% of the longest distance between presence records. We cross‐validated the models with presence and absence data arranged in spatially independent blocks, using the blockCV package (Valavi ). Presence and absence data were divided into 100 km wide squared blocks arranged in 8 cross‐validation folds.…”

Section: Methodsmentioning

confidence: 99%

Vanilla distribution modeling for conservation and sustainable cultivation in a joint land sparing/sharing concept

et al. 2020

View full text Add to dashboard Cite

Vanilla, an expensive but popular spice used in many industries, faces problems related to its supply. Some of these problems are due to the fact that vanilla cultivation is based on clonal material of a single species (Vanilla planifolia) and is dominated by just a few countries located outside the native growing areas of aromatic vanilla species, which is the neotropics. Despite the economic importance of this crop, relatively little attention has been paid to its wild relatives, in particular with respect to their biology, ecology, and potential use. We hypothesized that species distribution models (SDMs) can identify suitable areas for both the conservation and cultivation of vanilla crop wild relatives (CWRs), following a joint land sparing/land sharing (SPASHA) approach, thus offering alternative sourcing areas and production methods. This is the first study that explored the use of ensemble SDMs to provide applicable land use maps related to the conservation and sustainable cultivation of wild vanilla species in Costa Rica, contributing to a solution for the problems related to current vanilla production systems. We focused on four aromatic vanilla CWRs, native to Costa Rica, to make land use policy recommendations for this country, and more specifically for the biological corridor Osa and its surroundings within our study region Area de Conservaci on Osa (ACOSA). The resulting distribution maps, with a mean AUC of 0.89, reflected their current potential distribution (ranging from unsuitable to suitable) in Costa Rica. Combining them with recent land use and conservation area maps of our study region, we defined (1) areas for vanilla conservation and (2) areas for sustainable vanilla cultivation within agroforestry systems. These land use recommendations can now be integrated within the National Bio-Corridor Program (PNCB) that aims at making biological corridors more productive by proposing alternative income generation for local communities living within these areas. Our approach can be applied to identify priority areas for implementing the SPASHA approach on other vanilla CWRs and in more regions across its native growing ranges, given the availability of land use maps and enough occurrence records to build accurate SDMs.

show abstract

“…We generated spatial blocks of training and testing data with the R package “blockCV” v.2.0.0. (Valavi, Elith, Lahoz‐Monfort, Guillera‐Arroita, & Warton, 2018). We created fivefold and set the size of the spatial blocks to the median of the spatial autocorrelation range across the input environmental variables, which were sampled at 5,000 random locations.…”

Section: Methodsmentioning

confidence: 99%

Unraveling the habitat preferences of two closely related bumble bee species in Eastern Europe

Geue

Thomassen

2020

Ecology and Evolution

View full text Add to dashboard Cite

Co‐occurrence of closely related species is often explained through resource partitioning, where key morphological or life‐history traits evolve under strong divergent selection. In bumble bees (genus Bombus), differences in tongue lengths, nest sites, and several life‐history traits are the principal factors in resource partitioning. However, the buff‐tailed and white‐tailed bumble bee (Bombus terrestris and B. lucorum respectively) are very similar in morphology and life history, but their ranges nevertheless partly overlap, raising the question how they are ecologically divergent. What little is known about the environmental factors determining their distributions stems from studies in Central and Western Europe, but even less information is available about their distributions in Eastern Europe, where different subspecies occur. Here, we aimed to disentangle the broad habitat requirements and associated distributions of these species in Romania and Bulgaria. First, we genetically identified sampled individuals from many sites across the study area. We then not only computed species distributions based on presence‐only data, but also expanded on these models using relative abundance data. We found that B. terrestris is a more generalist species than previously thought, but that B. lucorum is restricted to forested areas with colder and wetter climates, which in our study area are primarily found at higher elevations. Both vegetation parameters such as annual mean Leaf Area Index and canopy height, as well as climatic conditions, were important in explaining their distributions. Although our models based on presence‐only data suggest a large overlap in their respective distributions, results on their relative abundance suggest that the two species replace one another across an environmental gradient correlated to elevation. The inclusion of abundance enhances our understanding of the distribution of these species, supporting the emerging recognition of the importance of abundance data in species distribution modeling.

show abstract

blockCV: an R package for generating spatially or environmentally separated folds for k-fold cross-validation of species distribution models

Cited by 135 publications

References 17 publications

Testing whether ensemble modelling is advantageous for maximising predictive performance of species distribution models

Testing whether ensemble modelling is advantageous for maximising predictive performance of species distribution models

Vanilla distribution modeling for conservation and sustainable cultivation in a joint land sparing/sharing concept

Unraveling the habitat preferences of two closely related bumble bee species in Eastern Europe

Contact Info

Product

Resources

About