Cross-validation strategies for data with temporal, spatial, hierarchical, or phylogenetic structure

By D. Roberts, V. Bahn, Simone Ciuti, M. Boyce, Jane Elith, G. Guillera‐Arroita, Severin Hauenstein, J. Lahoz‐Monfort, B. Schröder, Wilfried Thuiller and 4 more. 2017

2,280 citations recorded by the source corpus.

Subfields: Ecological Modeling, Environmental Engineering, Nature and Landscape Conservation

View Paper: https://www.semanticscholar.org/paper/0a311eaa99c20a1c88625cea0249f5430b895932

How researchers see this paper

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

In Ecological Modeling

In Environmental Engineering

In Nature and Landscape Conservation

Citations in the corpus

A review of evidence about use and performance of species distribution modelling ensembles like BIOMOD (2019, methodology): “In cross‐validation, different strategies can be used to divide data into calibration and validation subsets (Roberts et al., 2017), but the vast majority in our sample (n = 198) use BIOMOD's default strategy of repeatedly and randomly resampling data into a calibra‐ tion set and a validation set…”
Conducting and interpreting fish telemetry studies: considerations for researchers and resource managers (2019, background): “However, sedation itself can be stressful (Cooke et al. 2016b), and there is ongoing debate about whether fish have the capacity to feel pain (Rose et al. 2014).”
Mapping dominant leaf type based on combined Sentinel-1/-2 data – Challenges for mountainous countries (2021, methodology): “Such a reference data stratification strategy enables to reduce bias in training, validation and testing and, thus, leads to a better evaluation of model performances (Roberts et al., 2017; Schratz et al., 2019).”
Without quality presence–absence data, discrimination metrics such as TSS can be misleading measures of model performance (2018, methodology): “Cross ‐ validation procedures can lead to overoptimistic evaluations because of data autocorrelation, and specific procedures can be applied to avoid this further bias (Roberts et al., 2017).”
Leakage and the reproducibility crisis in machine-learning-based science (2023, background): “overoptimism.(58,59) Handling nonindependence between the training and test sets in general (i.”
Using Stacking to Average Bayesian Predictive Distributions (with Discussion) (2017, background): “Roberts et al. (2017) review cross-validation strategies for data with temporal, spatial, hierarchical, and phylogenetic structure.”
ENMeval 2.0: Redesigned for customizable and reproducible modeling of species’ niches and distributions (2021, background): “Random partitions typically lead to low independence between training and validation datasets, which can be particularly problematic when biological sam - pling bias exists (Radosavljevic & Anderson, 2014), and in such situa - tions results in artifactually high performance (Roberts et al., 2017).”