Can we trust our habitat maps? Assessing map sensitivity for decision-making
Bibliographic record
Abstract
Marine habitat maps are increasingly used to inform decisions in marine spatial planning. Despite their ability to efficiently communicate relevant information about habitats to decision-makers and stakeholders, habitat maps and species distribution models (SDMs) are only simplified representations of real habitats. While the interpretation of the outcomes and the method used to generate them should always be guided by expert advice, we still do not fully understand how the many decisions involved in map production impact the outcome. The objective of this study was to explore the sensitivity of habitat maps and SDMs to choices made when producing them. We used a dataset of German Bank, Canada, which includes multibeam bathymetric data, backscatter data, sea scallops (Placopecten magellanicus) presence data, and ground-truthing video data of potential habitats. To assess the impact of spatial scale, bathymetric and backscatter data were computed at five spatial resolutions. To assess the impact of data selection, 24 terrain attributes were derived from these bathymetric data. To assess the impact of data quality, different types and levels of artefacts were artificially introduced in the bathymetric data. Different combinations of these data were then iteratively used to generate habitat maps and SDMs. First, 644 maps of potential habitats based on biophysical characteristics of the area were produced using unsupervised classifications. Secondly, 644 SDMs of sea scallops were generated using maximum entropy (MaxEnt). The performance of the maps and SDMs were quantified and compared, and the spatial distribution of potential habitats and predicted sea scallop distribution were examined. Results indicate that variations in scale, data selection and data quality can produce very different outcomes. When selecting different terrain attributes and data of different spatial resolutions, the accuracy of habitat maps and the performance of SDMs varied, and differences up to 58% in the spatial distribution of habitats and predicted species presence were observed. Introducing artefacts also directly impacted the quality of terrain attributes at all scales, and propagated through to the habitat maps and SDMs. However, the impacts of these artefacts on habitat maps and SDMs were very unpredictable. While introducing artefacts sometimes decreased map accuracy and SDM performance, in other cases it artificially increased them. Differences in spatial distribution of potential habitats of up to 35% were observed. Local differences in habitat suitability for the SDMs reached as high as 75%, therefore changing from non-suitable to suitable area, or vice-versa. Results also show that small differences in measures of map accuracy (e.g. kappa coefficient of agreement) can translate into big differences in spatial distribution of habitats. Our study showed that it is essential to remain critical of the outcomes of mapping methodologies and to recognize the limitations of the data and techniques used. Understanding the different trade-offs involved in the map production is essential when maps are to be used to inform decision-making. We recommend using multiple maps and SDMs - for instance through ensemble mapping techniques - when trying to inform spatial planning efforts. This will enable the quantification of maps and SDMs variability to different factors and account for that variability when making decisions. We should also aim to develop specific tools for assessing error and uncertainty propagation in the habitat mapping workflow. These will result in mapping products with a higher degree of confidence and lower uncertainty.
Fetched live from OpenAlex and de-inverted. Abstracts are not stored in this database: the inverted indexes are 8.6 GB of the frame’s 9.3 GB of text, and the host has 13 GB free.
How this classification was reachedexpand
Full frame machine prediction
Teacher imitationNot calibrated prevalence, not ground truth. Human validation pending. The Gemma side is a direct model label for every work in the frame, read from the title-only record. The Codex side is a classifier learned from the 10,348 direct Codex labels and calibrated to design-weighted sample rates; fields without enough sample support carry no Codex call. Candidate is the union of the two sides; consensus is their intersection. These outputs are machine_predicted_unvalidated and are not human labels.
Distilled classifier scores by category (both heads)
| Category | Codex | Gemma |
|---|---|---|
| Metaresearch | 0.109 | 0.531 |
| Meta-epidemiology (narrow) | 0.001 | 0.001 |
| Meta-epidemiology (broad) | 0.001 | 0.002 |
| Bibliometrics | 0.005 | 0.005 |
| Science and technology studies | 0.001 | 0.005 |
| Scholarly communication | 0.008 | 0.011 |
| Open science | 0.002 | 0.004 |
| Research integrity | 0.003 | 0.003 |
| Insufficient payload (model declined to judge) | 0.003 | 0.001 |
Machine scores (provisional)
The two teacher heads of the student model, read on this work. A score orders the frame for review; it never asserts a category, and the validation status ships verbatim with every row.
Baseline scores from an immature model (maturity gate not passed, 7 training rounds). Scores rank; they never assert a category.
score_only:v0-immature-baseline · verbatim from the scoring run: score_only means the number may rank works, and no category label ships from itClassification
machine, unvalidatedMachine predicted; a candidate call from one source (direct Gemma or distilled Codex), not a consensus.
How this classification was reached, model by model and score by score, is at the end of the page under "How this classification was reached".