Urbanisation driven spectral–structural divergence in forest extent: threshold-based validation using GEDI LiDAR in Ekiti State, Nigeria

Oluwafemi David Bejide , Kunle David Emiola, Ojo Davies Ajewole, Hezekiah Daramola Olaniran

Forest degradation in rapidly urbanising tropical regions is often underestimated due to reliance on two-dimensional spectral indicators that inadequately capture vertical forest structure. This study evaluates the reliability of satellite-derived forest extent in Ekiti State, Nigeria, by integrating multi-sensor remote sensing data with spaceborne LiDAR from the Global Ecosystem Dynamics Investigation.

Land use/land cover (LULC) dynamics between 2007 and 2024 were mapped using a fusion of Synthetic Aperture Radar (Sentinel-1 and ALOS PALSAR) and optical datasets (Landsat and Sentinel-2). Results indicate a decline in forest cover of 540.10 km² (−25.34%) alongside a 266.42% increase in built-up area, corresponding to an Urban Land Consumption Ratio (ULCR) of 3.12, reflecting spatially inefficient urban expansion and increasing anthropogenic pressure on forest landscapes.

To assess classification reliability, LULC-derived forest extent was validated against GEDI-derived canopy height using two structural thresholds (≥5 m and ≥10 m). At the 5 m threshold, agreement was consistently low (F1 = 0.581 in 2020; 0.497 in 2025), driven by very low recall (<0.41), indicating substantial omission of structurally defined forest. In contrast, the 10 m threshold yielded higher agreement (F1 = 0.745 in 2020; 0.662 in 2025), suggesting improved alignment with spectrally derived forest extent. However, a consistent decline in F1-score across both thresholds demonstrates increasing divergence between spectral classification and structural forest definition over time.

A moderate reduction in mean canopy height (10.6 m in 2020 to 9.34 m in 2025) further indicates ongoing structural degradation. Analysis of anthropogenic drivers reveals that urbanisation intensity, rather than population growth, is the dominant driver of both vegetation decline and structural change. These findings provide empirical evidence of threshold-dependent and temporally evolving spectral–structural decoupling, highlighting the need to integrate LiDAR-derived structural metrics into forest monitoring frameworks.