Monitoring forest biodiversity is a critical concern for countries worldwide. In recent years, technological advancements have led to the miniaturization and simplification of commercially available consumer drones (i.e., UAVs) while simultaneously enhancing image quality, positioning accuracy, automation, and safety. These advancements offer a promising solution for automated and long-term forest monitoring. UAV-captured images can be leveraged to identify species, analyze canopy structure, and assess biodiversity and ecosystem health. To facilitate these analyses, 3D models of the surveyed forest should be generated. This paper introduces a framework for constructing 3D models and orthomosaic images of forests using UAV-captured images. The proposed framework relies on 3D Gaussian Splatting and its variant, Scaffold Gaussian Splatting, for high-quality 3D reconstruction. Additionally, we incorporate the Multi-Scale Structural Similarity Index (MS-SSIM) loss into the loss function of both 3D Gaussian Splatting and Scaffold Gaussian Splatting. This loss function improves similarity evaluation across multiple scales, capturing both fine details and broader structural patterns. To evaluate the performance of our framework, we collected RGB images from UAVs in two forested areas: Ben En National Park and Vietnam National University of Forestry (VNUF). Experimental results demonstrate that incorporating MS-SSIM loss enhances novel-view synthesis performance. The generated 3D models and orthomosaic images successfully preserve structural details while enabling artifact-free rendering of complex and dense foliage.