Deep Learning Analysis of Snow Cover Effects on Wild Ungulate Habitat Selection

Abstract

Introduction: Snow cover is one of the key environmental factors that impact the habitat choice, movement, food access, and winter survival of wild ungulates. Differences in the snow conditions such as snow density, vegetation presence, topography, and altitude could significantly affect habitat suitability. Traditional habitat-choice models could have certain drawbacks in depicting nonlinear relationships between different environmental variables. The current research focuses on the use of deep learning in analyzing the impacts of snow cover on wild ungulate habitat choice. Methodology: Spatial framework for deep learning based on spatial integration of ungulate distribution data with snow cover and other environmental factors such as vegetation, altitude, slope, aspect, and season was created. Environmental datasets were first aligned, preprocessed, normalized, and separated into training, validation, and testing sets. Convolutional neural network was used to analyze the spatial information and learn about the relationships between environmental conditions and habitat suitability. Model performance was evaluated through accuracy, precision, recall, F1 score, and area under the receiver operating characteristic curve (AUC). Results: The suggested deep learning algorithm had the following metrics: accuracy was 87.9%, precision 87.3%, recall 86.8%, F1 87.0%, and AUC 0.91. These metrics were better compared to the conventional algorithms tested. Suitability decreased along with the intensity of snow cover, while low-snow habitats demonstrated better prediction suitability. It can be said that taking snow into account, along with terrain and vegetation, enhances the pattern of habitat selection. Conclusion: This research highlights the use of deep learning techniques and remote sensing data about environmental factors in habitat selection analysis for wild ungulates. The approach is suitable for winter habitat evaluation and wildlife management due to its ability to find regions that are characterized by good environmental conditions. Future work should focus on using GPS data, multisession satellite images, and live snow data.