Sustainable land-use optimization using NSGA-II: theoretical and experimental comparisons of improved algorithms

Context United Nations outlined 17 Sustainable Development Goals (SDGs), but at the current rate of progress most will not be achieved within the desired timeframe. Since a third of SDGs are directly related to land resources, it is crucial to improve the effectiveness and efficiency of land-use planning. In that regard, there is particular value in algorithmically optimizing land-use planning to better support sustainability. An ideal tool for such optimizations is the nondominated sorting genetic algorithm II (NSGA-II). Objectives Improved versions of NSGA-II have been actively developed for land-use problems, but no thorough evaluations and very few comparative studies have been performed. Thus, the objective is to conduct a thorough evaluation of and a systematic comparison between improved NSGA-II algorithms for sustainable land-use optimization. Methods We identified both the most popular and the latest improved algorithms. A theoretical comparison was first made between them in terms of initialization, crossover, mutation, and archiving strategy. Then, a framework consisting of four hierarchal levels (principle, macro-criteria, micro-criteria, and indicators) was developed and applied to make a comprehensive comparison through experiments. Results The most popular algorithm was demonstrated to produce high-quality results and be computationally efficient, whereas the other performs better in the diversity of results, space efficiency, and the degree of optimization. Both algorithms exhibited excellent performance in handling constraints. Conclusions Possible approaches to further improve the algorithms include borrowing ideas of scale optimization and gene flow. The proposed framework is capable of guiding further improvement by developers and algorithm selection by users.

Automated summary

Evaluated and compared improved versions of NSGA-II algorithms for sustainable land-use optimization, finding one with high-quality results and computational efficiency, while another excelled in result diversity, space efficiency, and optimization degree. Both algorithms demonstrated excellence in handling constraints.