Pareto Local Search For Multi-Objective Land-Use Allocation

Abstract

Multi-objective land-use allocation (MOLA) is a multi-combinatorial optimization problem, in which land-use types are allocated to units of land to optimize multiple objectives while satisfying imposed constraints. MOLA techniques are among others used by researchers and land planners to assist in planning sustainable land-use. As land-use change is currently an important human driver of environmental degradation, the demand for MOLA techniques has increased. Main challenges in this field are the scalability of the spatial optimisation process to enable the optimisation for larger areas and multiple objectives and to develop realistic solutions. This research examines the potential of a local search based algorithm Pareto Local Search (PLS) for MOLA to address these issues. In this research, a modular (iterated) PLS algorithm for MOLA, named (I)PLS-MOLA, was designed and implemented. Together with this algorithm, three search operators for MOLA, two repair operators and a new data structure for storing MOLA solutions were set up. The proposed algorithm, operators and data structure were applied to a case study that concerns land-use optimization in Brazil, which could provide insight in how Brazil could plan its future land-use to meet the future food, feed and biofuel demands in a sustainable manner. Eventually, the performance of the PLS-MOLA and IPLS-MOLA algorithm were compared to the performance of NSGA-II, which is currently the most used optimization technique for MOLA. The results show that PLS-MOLA and IPLS-MOLA prove to be more scalable than NSGA-II, with a more efficient and better converging optimization process. Especially with larger MOLA problems, the solution quality of PLS is significantly higher than the solution quality of NSGA-II. Combined with the fact that PLS can also search more efficiently for solutions that can emerge from the current situation, assuring more realistic solutions, the algorithm proves to be a promising technique for future MOLA research.