After a disaster, the power grid helps support infrastructure systems that are essential to the recovery effort in addition to the critical services it provides every day. This paper provides an approach for optimal power grid restoration in disaster contexts with environmental hazards. These hazards, such as unstable structures, smoke, chemical hazards, and abnormal radioactivity, may pose acute and accumulating risks to repair workers and impede the restoration process. We therefore formulate a mixed-integer linear program (MILP) that generates an optimal restoration plan with constraints imposed by acute and accumulating environmental hazards. We also develop a heuristic inspired by trends that we observe in optimal restoration strategies, and we compare its performance to that of an optimal restoration strategy. For our case study, we model a stylized disaster that approximates the patterns of a number of disasters including earthquakes, fires, industrial facility explosions, or nuclear reactor incidents. We analyze the performance of the heuristic and optimal restoration strategies on a modified IEEE 123-bus test network. We find that the optimal restoration strategy is able to restore power service more quickly than the heuristic strategy while also exposing repair workers to less acute and cumulative environmental hazards. We also find that as disaster severity increases, the performance difference between the heuristic and optimal restoration strategies grows. Finally, our results show that both the optimal and heuristic algorithms can be useful tools for identifying vulnerable regions of a power grid.