Data for

Abstract

Context. Squirrel species in Alaska generally lack thorough conservation management plans, while all species are actively hunted with no bag limits, closed seasons, or any other restrictions, and the current ‘management’ is based on ambiguous hand-drawn distribution maps. This indicates a laissez-faire approach to Alaskan squirrel conservation management. Aims. Here, we attempt to improve this current situation by assessing the effectiveness of ensemble machine-learning prediction models as proposed add-ons to the traditional components of conservation management plans toward a more state-of-the-art approach to species conservation. Methods. We combined the Machine Learning algorithms TreeNet, CART, Random Forest, and Maxent with over 200 environmental and socio-economic predictors for the ensemble Super Species Distribution Models. These ensemble models were carried out for all squirrel species individually occurring in Alaska and a 600 km buffer area and two assemblage models combined: a) all species currently occurring only in Alaska and b) all species occurring in Alaska and the 600km buffer area. Key results. Most predicted distribution hotspots for squirrels in Alaska and the 600 km buffer area were observed near road and river systems (close to human activities) and the last glacial maximum refugia. Conclusions & Implications. Applying a machine learning ensemble distribution modeling framework to conservation management plans can add valuable science-based insights to better understand the landscape and species to be managed. Such insights include more accurate guidance on, e.g., habitat protection, hunting regulations, or any collaborative management initiatives. This can also be highly valuable for lands not directly managed by conventional agencies, e.g., land managed by the military or Native communities throughout the Pacific Rim.