Using Spatial Validity and Uncertainty Metrics to Determine the Relative Suitability of Alternative Suites of Oceanographic Data for Seabed Biotope Prediction. A Case Study from the Barents Sea, Norway

Abstract

The use of habitat distribution models (HDMs) has become common in benthic habitat mapping for combining limited seabed observations with full-coverage environmental data to produce classified maps showing predicted habitat distribution for an entire study area. However, relatively few HDMs include oceanographic predictors, or present spatial validity or uncertainty analyses to support the classified predictions. Without reference studies it can be challenging to assess which type of oceanographic model data should be used, or developed, for this purpose. In this study, we compare biotope maps built using predictor variable suites from three different oceanographic models with differing levels of detail on near-bottom conditions. These results are compared with a baseline model without oceanographic predictors. We use associated spatial validity and uncertainty analyses to assess which oceanographic data may be best suited to biotope mapping. Our results show how spatial validity and uncertainty metrics capture differences between HDM outputs which are otherwise not apparent from standard non-spatial accuracy assessments or the classified maps themselves. We conclude that biotope HDMs incorporating high-resolution, preferably bottom-optimised, oceanography data can best minimise spatial uncertainty and maximise spatial validity. Furthermore, our results suggest that incorporating coarser oceanographic data may lead to more uncertainty than omitting such data.

Fluoride contamination of groundwater resources in Ghana: Country-wide hazard modeling and estimated population at risk

“…For example, Pohjankukka et al, 2017 has shown that the metrics produced by non-spatial cross validation can be up to 40% more optimistic than those of spatial cross-validation. Similar results were reported by (Dolan et al, 2021;Ploton et al, 2020;Roberts et al, 2017).…”

Importance of Spatial Autocorrelation in Machine Learning Modeling of Polymetallic Nodules, Model Uncertainty and Transferability at Local Scale

“…Spatial autocorrelation (SAC) can help to address the issues mentioned above, at the same time not considering spatial autocorrelation may result in over-optimistic CV predictions. Similar to recent studies [17,[19][20][21]23,24,139,140], we showed that the conventional and commonly applied random 10-fold CV could not assess the model performance when spatial autocorrelation is present. In such cases, spatial-and cluster-blocking perform better.…”

“…Recent studies in terrestrial and marine spatial ML modeling showed that if the commonly used, non-spatial, random k-fold CV is used, the prediction performance is over-optimistic when SAC exists in the data. The magnitude of the spatial overfitting varies based on the degree of SAC among the training points, the environmental similarity among the regions, and the ML method used [17][18][19][20][21][22][23][24][25]. In order to address the influence of SAC, different CV schemes have been proposed; the most common are buffer distances among training locations [18,26,27], successive distances [28], leaving one training location out [19,29], and spatial block training.…”

Diversity, structure and spatial distribution of megabenthic communities in Cap de Creus continental shelf and submarine canyon (NW Mediterranean)

“…RF is a modelling approach commonly used to predict the distribution of shelf and deep-sea benthic communities or biotopes (e.g. Du Preez et al, 2016;Cooper et al, 2019;Dolan et al, 2021;Goode et al, 2021;O'Brien et al, 2022), which has shown to perform as well as other predictive modelling techniques for benthic habitat mapping (Hasan et al, 2012;Robert et al, 2016). Random Forest builds a large number of trees using subsets of the data and the environmental variables, with each tree making a prediction and the final assignation being based on the majority (Breiman, 2001).…”