Forest restoration scenarios of Brazil’s Atlantic Forest with constrained multi-objective optimisation

Brazil’s Atlantic Forest has lost over 85% of its original cover and harbours one of the world’s densest concentrations of endemic species. Designing restoration plans that simultaneously maximise landscape connectivity, satisfy legal obligations, and stay within budget remains an open challenge. Existing decision-support tools require planners to pre-specify trade-off weights, returning a single solution that cannot reveal the full connectivity–cost exchange. We present FOREMOST (FOrest Restoration with Evolutionary Multi-Objective STrategies), a constrained multi-objective evolutionary optimisation model that delivers the complete Pareto-optimal front for restoration planning under hard constraints. Applied to the Pernambuco Endemism Centre (PEC; ≈ 44,100 km 2 ; 780 restorable cells), FOREMOST maximises Effective Mesh Size (MESH) and Integral Index of Connectivity (IIC) while minimising cost under six simultaneous Brazilian Forest Code hard constraints. Among three evolutionary algorithms, C-TAEA returns 159–198 feasible Pareto-optimal plans per run (mean hypervolume H V ¯ = 90.15 ± 4.62 , a composite score summarising how large and well-spread the Pareto front is; higher is better), outperforming R-NSGA-III ( H V ¯ = 76.89 ) and NSGA-II ( H V ¯ = 52.85 ; one run infeasible). All returned plans satisfy all six hard constraints simultaneously — a guarantee unavailable from weighted single-solution tools. Connectivity objectives correlate with independently derived endemic species area-of-habitat targets ( r ≥ 0.85 , p < 0.05 , n = 6 scenarios), validating MESH and IIC as ecological proxies for biodiversity recovery. FOREMOST identifies a budget-efficiency threshold: 39% of maximum achievable connectivity is secured at only 40% of maximum budget. The model scales to N = 300 (6472 candidate cells, δ ≈ 2.2 km) and is freely available at https://github.com/laureberti/FOREMOST (archived: doi:10.5281/zenodo.20921037 ).