Urban green infrastructure (UGI) is an essential approach to mitigate the impact of urbanization and ameliorate the microclimates in cities. Knowing the optimal spatial distribution of green infrastructure is vital to launch effective greening projects and formulate the promotion strategy for sustainable urban transition. But few studies have explored spatial organization of UGI on an urban scale to maximize its benefits for urban renewal. In this study, we developed a spatial optimization method that combines an urban cooling model and genetic algorithm to study the optimal geographical arrangement of roof greening in Kowloon, Hong Kong. Two objectives for roof greening implementation are compared in terms of the optimized spatial pattern and cooling effects. One is to achieve the maximum reduction of the mean air temperature. The other is to minimize the population-weighted temperature for the lower heat exposure. We found that spatial optimization of roof greening can reduce more air temperature than random distribution when the target of roof greening coverage ranges between 30% and 70%. Marginal reduction of mean air temperature significantly increases when the roof greening coverage is over 40%. The two objectives have very close results of mean air temperature but the objective of minimizing heat exposure can create more health benefits. The optimization shows that as the target of roof greening coverage increases, the geographical locations of the buildings with superior roof greening proportions (i.e., the prioritized buildings) move from the near-forest area to the inner city, while the lower greening proportions are mostly found on dense and small-rooftop buildings. The difference of prioritized buildings between the two objectives becomes significant when the target of roof greening coverage increases, which shows the trade-offs between the cooling potential and population exposure among regions. Our analysis proposed an analytic approach to transform strategies into effective planning solutions in the process toward a large-scale implementation, which can be extended to evaluate spatial planning of other UGIs.