Sustainable forest and land-use management in Tanzania: NAFORMA-based and machine learning forecasts to 2050 under Tanzania Dira 2050

Forests and land-use systems are vital for climate change mitigation and sustainable development in sub-Saharan Africa, particularly in Tanzania, where rapid land-use changes increasingly threatened ecosystems. This study assessed long-term forest and land-use dynamics in Tanzania up to 2050, aligned with the Tanzania Development Vision 2050, with the aim of informing climate-smart land-use planning. The study utilised data from the National Forest Resources Monitoring and Assessment (NAFORMA) 2014 and employed advanced machine learning and spatial modelling techniques. Future land-use dynamics were projected under three scenarios: Business-as-Usual (BAU), Sustainable Management (STB), and High-Pressure (HP). Scenario differences and model performance were evaluated to demonstrate the predictive advantages of machine learning approaches over traditional methods, thereby generating policy-relevant insights. The findings showed statistically significant divergence among scenarios. Descriptively, total forested land (forest + woodland) declined from 48.09 million ha in 2014 to 32.99 million ha by 2054 under the BAU scenario and to 28.30 million ha under the HP scenario, while the STB pathway moderated the decline to 43.85 million ha. Inferentially, these trends were highly significant (BAU: β = −0.410, F = 96.3, p < 0.001; HP: β = −0.553, F = 121.9, p < 0.001), whereas the STB trend was significantly smaller (β = −0.135, F = 14.8, p = 0.003), supporting Hypothesis H₃ that alternative management pathways generated statistically distinct outcomes. Cultivated land expanded significantly under BAU (β = +0.168, F = 88.4, p < 0.001) and HP (β = +0.251, F = 104.6, p < 0.001), but at a reduced rate under STB (β = +0.067, F = 12.2, p = 0.004). Machine learning models achieved high predictive performance (Random Forest: overall accuracy = 0.87, Kappa = 0.81, AUC = 0.89; Artificial Neural Network: overall accuracy = 0.84, Kappa = 0.78), confirming Hypothesis H₂. Translating land-use outcomes into carbon dynamics revealed that national forest carbon stocks declined from approximately 1,060.8 Mt C in 2014 to 707.8 Mt C under BAU and 590.8 Mt C under HP by 2054, compared with 963.8 Mt C under STB. Corresponding cumulative CO₂-equivalent emissions reached 2,596 MtCO₂-eq under BAU and 3,411 MtCO₂-eq under HP, while remaining limited to 723 MtCO₂-eq under STB, with scenario differences statistically significant (p < 0.01). These results confirmed Hypothesis H₃ at the climate-policy level. Generallyl, the study highlighted that Tanzania’s forest and land-use future to 2050 depended critically on policy decisions. Continuation of historical land-use trends posed substantial risks to climate mitigation objectives, whereas improved governance and sustainable land-use practices enhanced system stability and substantially reduced emissions. The study therefore recommended prioritising sustainable forest and land-use governance, institutionalising NAFORMA as a long-term national monitoring system, and integrating machine learning–based forecasting into climate and land-use policy processes.