Deep Learning Prediction of Drought-Induced Foraging Changes in Wild Herbivores

Abstract

Introduction: Drought is an environmental stressor, which may change the vegetation productivity, the water availability, the habitat condition, and the foraging behavior of wild herbivores. It is difficult to predict the changes in the foraging behavior because the behavior of the herbivores is affected by several interacting environmental factors. This paper proposes a framework that uses deep learning for predicting the drought effect on wild herbivore foraging behavior.

Materials & Methods: The proposed framework uses drought indices, precipitation, temperature, vegetation status, water availability, availability of forage, herbivores' density, and foraging information. Data preprocessing consists of dealing with missing values, data normalization, data encoding, and temporal synchronization of environmental and behavioral data. Feature fusion is applied to merge drought and habitat features with foraging features prior to predicting them through a deep learning approach. The proposed model is compared to Random Forest, Support Vector Machine, and Decision Tree models using accuracy, precision, recall, and F1-score measures.

Results: The deep learning algorithm had an accuracy score of 83.7%, a precision of 82.9%, a recall of 81.8%, and an F1 score of 82.3% compared to other models such as Random Forest, Support Vector Machine, and Decision Trees. From the above findings, it is clear that incorporation of environmental and behavioral factors will enable better identification of drought-induced foraging behavior patterns. The results further demonstrate the potential of deep learning to capture nonlinear relationships between drought severity, resource availability, and herbivore feeding responses.

Conclusion: The constructed framework offers an effective predictive technique for evaluating drought-related modifications in the foraging behavior of wild herbivores. Application of the proposed model that integrates environmental characteristics and behavioral information may be useful in ecological monitoring, habitat evaluation, and wildlife management. The predictive model can be applied to different types of herbivorous animals in various environments by utilizing satellite imagery, GPS tracking, camera traps, and other ecological data.