Leaf area density (LAD) vertical profiles are crucial for characterizing forest ecosystems. Unmanned aerial vehicle (UAV) light detection and ranging (LiDAR) systems offer cost-effective tools for assessing spatiotemporal LAD variations. We evaluated the applicability of a stochastic radiative transfer model, originally developed for airborne LiDAR, to cost-effective UAV-LiDAR observations in a deciduous broadleaf forest during leaf fall. The model describes beam penetration and multiple LiDAR returns within complex canopies, using either all available returns (multi-return model) or first returns only (first-return model). The multi-return model accounts for non-random foliage clumping via higher-order returns, whereas the first-return model forgoes this capability, being mathematically equivalent to a conventional gap-fraction approach. Both models were compared against litterfall-derived LAI estimates. The multi-return model markedly underestimated the LAI and failed to track its seasonal variation, because it neglects severe missing detection of higher-order returns, a critical issue under cost-effective UAV-LiDAR conditions likely attributable to small footprints and sensor constraints. Conversely, the first-return model tracked the seasonal decline more closely, though its quantitative accuracy remains limited. These results suggest that missing higher-order returns render the multi-return model inapplicable under current cost-effective UAV-LiDAR conditions, making the first-return model a more practical alternative despite its quantitative limitations. Future studies should address these limitations by accounting for missing higher-order returns or, more promisingly, extending the first-return model to better incorporate higher-order return information.