in response to some comments: yes, this will have all of the same issues as the model it was trained on. e.g., precipitation is notoriously hard to get right. you just have to treat this as you would any model output.