Forest management increasingly requires decision-support approaches capable of addressing trade-offs between production, ecological resilience, economic feasibility and long-term sustainability. This is relevant in Mediterranean landscapes, where land fragmentation, recurrent fires, climate change and landscape-scale planning increase uncertainty in silvicultural decisions. In this context, this work presents a Web-oriented decision-support approach for multi-objective forest management planning, based on a customised version of the Non-Dominated Sorting Genetic Algorithm II (NSGA-II).

The approach was developed to help forest managers explore alternative management compositions before implementing a specific plan. The system includes two modules: one simulates management options for each stand, considering composition, species proportions, yield class, age class, harvest-age intervals and silvicultural system; the other applies a customised NSGA-II algorithm to identify solutions that address conflicting objectives. In the case analysed, the optimisation problem aimed to maximise total harvested timber volume over the planning horizon and minimise the standard deviation of harvest volumes between periods, while ensuring that all solutions complied with validity constraints.

The methodological innovation lies in adapting NSGA-II to the structure of forest planning problems. The algorithm incorporates a customised initial population and a specialised mutation operator designed to reduce invalid solutions and improve the exploration of feasible combinations of stands, species compositions and harvest ages. It was implemented in Python using the PyMoo framework and tested through a semi-hypothetical case study in the Coimbra region, Portugal. The study area was organised according to landscape-planning principles, although the optimisation focused on the forest matrix. Fourteen stands were included, resulting in a high-dimensional chromosome with 437 genes.

The customised NSGA-II was compared with the standard NSGA-II using different combinations of population size, number of generations, crossover probability and mutation probability. Performance was assessed through non-dominated solutions, spacing, computational time and hypervolume. The results show that the customised algorithm consistently generated a larger and more diverse set of valid solutions. With 200 individuals and 500 generations, it produced 123 non-dominated solutions in approximately 50 seconds, compared with 98 obtained by the standard algorithm in a similar runtime. It also showed a more stable hypervolume convergence pattern and a denser Pareto front, indicating better exploration of the solution space and improved representation of trade-offs between timber production and temporal regularity of harvests.

The results demonstrate the potential of customised evolutionary algorithms to support forest planning in complex and uncertain contexts. By providing a set of Pareto-optimal alternatives rather than a single prescriptive solution, the proposed approach enables managers and decision-makers to compare trade-offs, assess silvicultural pathways and integrate technical analysis into planning. Its integration into Web-based systems may increase accessibility for practitioners and small landowners. Overall, this work contributes to more transparent, flexible and sustainability-oriented tools for adaptive forest management at stand and landscape levels.