How are species models reviewed

The ABMI Science Centre creates updated species distribution models on regular intervals. This updated process involves data cleaning, updating modeling frameworks, and reviewing the model outputs for statistical sensibility. However, as we create models for seven taxonomic groups (Amphibians, Birds, Mammals, Lichens, Bryophytes, Vascular Plants, and Soil mites) which total to ~1000 models, we are unable to know if the models make ecological sense.

Upon creating new models, we ask both the Processing Centre, and external experts that specialize in each taxonomic group to provide feedback. On this page, we describe each of the potential figures reviewers might see and the types of feedback that are useful.

Detection Maps

For each species, regardless of the number of detections, we create simple occurrence maps. As the protocols vary between sites, we have simplified the data to show if a species was detected at an ABMI site, including revisits, using the terrestrial field protocols. In addition, we combine off-grid sites with their nearest ABMI on-grid site to maintain a legible figure.

We are interested in hearing feedback about occurrences that are highly unlikely (e.g., Pronghorn in the Canadian Sheild).

Vegetation use-availability plots

For species that do not meet our occurrence threshold within the modeling region (50 sites for birds, 20 for all other taxa), we create simple use-availability figures. Because these species are often data deficient, we acknowledge that the use-availability results may change drastically with new data. If a species model is created, we don’t expect any comments about these plots since we won’t display them.

We are interested in hearing feedback about the general patterns of these outputs with two important caveats. 1) Don’t review the use-availability figure a full species distribution model is present. 2) The full species models are more important than the use-availability figures. So only provide feedback if time and resources permit.

Vegetation Coefficient plots

For species that meet our occurrence threshold within the modeling region (50 sites for birds, 20 for all other taxa), we create species distribution models. Species models with poor fit (AUC < 0.7) will be excluded for submission to the Information Centre. However, we still may choose to display them here.

We want feedback about the general patterns habitat coefficients. Do we predict an old forest specialist in old forests, are footprint tolerant species found in footprint, etc. In addition, let us know if the general patterns are reasonable but a single coefficient value doesn’t align with your ecological knowledge.

Forested Sector Effects plots

If a model was created, we estimate the impacts of different sectors of human footprint on their relative abundance. These figures are derived using both the species distribution model and our landcover layers.

We would like to hear feedback about sector effects that don’t match the coefficient plots. For example, provided a comment if a species not found in agriculture footprint (e.g., crops, pasture) has a highly positive agriculture sector effect.

Forested and Prairie Linear Features Effects

If a model was created, we estimate the impacts of linear features on a species relative abundance. Linear features make up a small proportion of the landbase which can make estimating their impacts challenging. In addition, our models don’t account for things such as edge effects. In general, these can be interpreted similar to the Forested and Prairie sector effect plots, specifically the under footprint results.

We would like to hear feedback about linear effects that seem unreasonable based on the ecology of the species. For example, a species with a known negative response to linear features showing a positive response. However, since we are still working how best to report on these features, other figures such as the model coefficients and prediction maps are more important to review.

Soil use-availability plots

For species that do not meet our occurrence threshold within the modeling region (50 sites for birds, 20 for all other taxa), we create simple use-availability figures. Because these species are often data deficient, we acknowledge that the use-availability results may change drastically with new data. If a species model is created, we don’t expect any comments about these plots since we won’t display them.

We want feedback about the general patterns of these outputs with two important caveats. 1) Don’t review the use-availability figure a full species distribution model is present. 2) The full species models are more important than the use-availability figures. So only provide feedback if time and resources permit.

Soil Coefficients plots

For species that meet our occurrence threshold within the modeling region (50 sites for birds, 20 for all other taxa), we create species distribution models. Species models with poor fit (AUC < 0.7) will be excluded for submission to the Information Centre. However, we still may choose to display them here.

We are interested in hearing feedback about the general patterns habitat coefficients. Do we predict an old forest specialist in old forests, are footprint tolerant species found in footprint, etc. In addition, let us know if the general patterns are reasonable but a single coefficient value doesn’t align with your ecological knowledge.

Prairie Sector Effects

If a model was created, we estimate the impacts of different sectors of human footprint on their relative abundance. These figures are derived using both the species distribution model and our landcover layers.

We would like to hear feedback about sector effects that don’t match the coefficient plots. For example, provided a comment if a species not found in agriculture footprint (e.g., crops, pasture) has a highly positive agriculture sector effect.

Prediction Maps

If a model was created, we calculate their predicted relative abundance under current condition (2018), the reference condition (backfilled vegetation), and the difference between them.

We want feedback about if the current distribution is generally matches the reviewers ecological knowledge (e.g., higher abundance in Rocky Mountains versus Boreal, etc).