Roads reshape forests: distance, access, and deforestation in Tanzania

Understanding how roads affect tropical forest loss is essential for evidence-based conservation planning. The distance-decay theory proposes that road impacts are strongest near roads, leading to higher deforestation rates close to transport networks. We analyzed forest cover loss across Tanzania between 2001 and 2020 using spatial district-year observations across 195 districts, focusing primarily on non-protected forests. Forest loss was evaluated across three accessibility zones: near-road (0–2.5 km), intermediate (2.5–5 km), and remote (>5 km). Tanzania’s forests experienced substantial loss over the study period. Non-protected areas recorded a cumulative loss of 3288,343 ha, representing 5.03% of near-road, 5.05% of intermediate, and 2.60% of remote zone land areas. Protected areas lost 684,400 ha in total, with loss concentrated in intermediate zones (51%), confirming that road accessibility shapes deforestation pressure regardless of protection status. Contrary to the distance-decay expectation, deforestation in both protected and non-protected areas was highest in the intermediate accessibility zone (2.5–5 km from roads). This zone had the largest normalized forest loss rate (12.96 m²/ha, over-representation index ORI = 1.364, where ORI is defined as the ratio of a zone’s share of total forest loss to its share of total study area; values >1.0 indicate disproportionately high loss), compared to near-road zones (0–2.5 km: 12.89 m²/ha, ORI = 1.357) and remote areas (>5 km: 6.70 m²/ha, ORI = 0.701) 48% lower than the intermediate zone underscoring that road proximity does not operate as a simple linear gradient but generates a spatially heterogeneous deforestation landscape. The measured drivers fire, cropland expansion, and settlement explained only 13.2% of deforestation variance, indicating fundamental limitations of satellite-based models. Other pressures including charcoal production, timber extraction, informal tenure systems, and socioeconomic conditions are likely to contribute significantly to the unexplained variance.