Understanding and responding to climate-related disasters like floods requires more than just raw satellite images—it needs tools that can interpret and explain what’s happening on the ground. While satellite images are powerful for monitoring such events, they can be difficult to understand without expert knowledge. Our work uses artificial intelligence to make this easier. We developed a system that lets people ask questions about satellite images in natural language and get clear, useful answers. What sets our system apart is its ability to 'think through' complex questions. For example, if asked, “Are all residential buildings affected by flooding?”, the system doesn’t just answer yes or no. It counts the buildings, checks which ones are flooded, and explains that, say, 5 out of 7 are underwater while the others are on higher ground. This step-by-step reasoning helps emergency teams act faster and smarter, and improves learning by training the model to explain its thinking. Our system achieved 33% higher accuracy than models trained only on direct answers.