📈 Our simulations revealed that certain key algorithm's parameters are good predictors for biases in density: so, early exploratory analyses can reveal if an automated approach can be used, and how much over/underestimation of the parameter is expected.

📊 We showed that CT-DS and REM both provide reliable estimates when an algorithm (#MegaDetector,
@sarameghanbeery.bsky.social @doctorofrock.bsky.social) is used to filter blank images, and CT-DS can even be used with an automated taxonomic classifier (#WildlifeInsights) with minimal bias.

Can AI improve #cameratrap studies?

Our study in @methodsinecoevol.bsky.social shows how #machinelearning can enhance wildlife surveys by automating species detection & classification and enabling unbiased density estimates 📷🦌.

Check it out! ⬇️
besjournals.onlinelibrary.wiley.com/doi/10.1111/...

📈 Our simulations revealed that certain key algorithm's parameters are good predictors for biases in density: so, early exploratory analyses can reveal if an automated approach can be used, and how much over/underestimation of the parameter is expected.

📊 We showed that CT-DS and REM both provide reliable estimates when an algorithm (#MegaDetector,
@sarameghanbeery.bsky.social @doctorofrock.bsky.social) is used to filter blank images, and CT-DS can even be used with an automated taxonomic classifier (#WildlifeInsights) with minimal bias.

🦡🦊🦌 We estimated density both on the field and in a simulation setting with two popular camera-trap models: Camera-Trap Distance Sampling (CT-DS) and Random Encounter Model (REM). Image classification was carried out both manually, and using two machine learning algorithms.