Forest inventory and ecological monitoring require dense, georeferenced maps of individual trees at scale, but forests present compounding challenges for mobile mapping systems. Heterogeneous terrain demands deployment across carrier types, dense canopy degrades GNSS and causes SLAM drift over long traversals, and the viewpoint gap between above-canopy and below-canopy sensing means no single platform can capture complete forest structure.

This thesis presents MapForest, a modular field robotics system that converts multi-modal sensor data (LiDAR, IMU, GNSS, and RGB) into georeferenced 3D reconstructions and GIS-compatible outputs. A compact, platform-agnostic payload deploys across five carrier types (handheld, bicycle, ATV, UAV, and the Freefly Alta-X) without hardware modification. The mapping backbone extends GLIM, a LiDAR-inertial SLAM framework, with covariance-aware GNSS priors and a Huber robust loss that suppress the influence of degraded fixes and multipath outliers. To fuse the complementary above-canopy and below-canopy reconstructions into a unified forest map, we develop two aerial-terrestrial alignment approaches: Tensor-MI, an analytical method that maximizes mutual information between tree-likelihood fields, and CRAF, a learned model combining modality-specific encoders with a cross-attention registration transformer. As a concrete ecological application, the system localizes invasive Tree-of-Heaven from onboard imagery using a fine-tuned YOLOv8 detector, projects image-space detections into the 3D map, and exports georeferenced GeoTIFF layers.

MapForest is evaluated across six field sites spanning approximately 30 km of multi-modal traversal data. Covariance-aware GNSS integration reduces trajectory error by 67.3% relative to a no-GNSS baseline. Tensor-MI achieves sub-meter alignment error, outperforming ICP and feature-based baselines. The Tree-of-Heaven detector achieves an F1 score of 0.77, demonstrating the system's utility for actionable ecological monitoring at the granularity unavailable to satellite or conventional aerial methods.