Using optimization to analyze the effect of mitigation decisions in humanitarian relief logistics

Natural disasters are gradually increasing due to population growth, expansion in the areas open to human usage on earth, urbanization and the effects of global warming. Natural disasters can result in massive human casualties as well as devastating economic consequences. Since natural disasters cannot be predicted when or how severe they will be, and because they cannot be prevented; effective disaster management studies should be conducted in order to prevent serious damage, civilian and property losses before and after they occur. An effective disaster management study should cover all the requirements before, during and after the disaster. In advance of the disaster occurrance buildings and transportation networks must be strengthened, immediately after the disaster occurrance humanitarian aid must be delivered to catastrophe victims and much later the disaster occurrance damaged buildings and transportation network sould be recovered. However, in the literature, there is no such a model that includes all of these disaster management decisions together. Within the scope of this thesis, a two-stage stochastic integer programming model has been developed which takes into consideration the interactions of all pre-disaster and post-disaster decisions as a whole. In the first stage of the developed model, in order to reduce possible life, property and economic losses when an earthquake occurs, the selection of buildings and transportation roads to be strengthened before the disaster and the level of reinforcement are decided. In the second stage of the model, distribution decisions regarding the humanitarian aid material demand after the earthquake and decisions to repair the destroyed buildings and transportation networks are made. In the model, the possibility of the earthquake to occur with different probabilities in different time periods is also taken into account. The Integer L-Shaped algorithm, which is one of the decomposition strategies that allows us to obtain effective results in the solution of two-stage stochastic programming models, is implemented to solve the given mathematical model. Developed test problems are solved in two hours by using both the proposed Integer L-Shaped method and the commercial solver CPLEX. The results are compared and the efficiency of the proposed methodology is shown. CPLEX was able to obtain a feasible solution for 228 (28% of problems) from the total 840 problem examples, while the proposed method obtained a feasible solution for 600 problem (71% of problems). In particular, the proposed method significantly outperforms CPLEX for tackling reasonably large problems.