Environment reconstruction through sampling is a difficult task and usually requires a large amount of resources. In some applications such as the sun light field reconstruction, the sensors are very expensive. Hence, reducing the number of samples while meeting the distortion requirement becomes very important in such applications. Many phenomena can be measured by different types of sensors, which provide information at different scales and accuracy. Prior knowledge and external information may serve to provide the global information. By exploiting the underlying models of the phenomenon, we efficiently deploy the sensors and combine the information from different levels. In this way, we can approach the reconstruction accuracy reached by exhaustive sensing in one level with much less number of sensors.