Stranding data can significantly bias marine mammal habitat suitability models

• Stranding data overestimate manatee habitat suitability by over three times. • Sightings and telemetry data provide the most reliable habitat suitability models. • First study showing stranding data bias habitat modeling. • Carcass drift influences stranding locations, misaligning them with habitat use. • Stranding data should be cautiously used and combined with drift models when possible. In the absence of sightings, stranding records can be used to parameterize habitat models for marine mammal conservation. However, their reliability for identifying suitable habitat remains uncertain. We assessed how stranding data influence the habitat predictions for American manatee ( Trichechus manatus ) in the Potiguar Basin, Brazil. Using MaxEnt, we compared models built using: (1) sightings and telemetry data, (2) stranding records alone, and (3) all three data sources combined. We found that the first model based solely on sightings and telemetry produced the most accurate and ecologically meaningful predictions. In contrast, the stranding-only model overestimated suitable habitat by more than threefold, while the combined model overpredicted it by more than twofold. These differences between model predictions are best explained by carcass drift and detection biases and indicate that stranding locations do not reliably reflect areas of actual habitat use. This quantitative assessment provides new insights into the significant biases stranding data can introduce into habitat suitability models. Our findings also highlight the need to prioritize direct observations, and to apply drift modeling and validation against direct observations when using stranding data, to ensure accurate and actionable conservation planning.