In many environment sensing applications, we can observe the field by multiple types of sensors in multiple levels. We propose that we can achieve the accuracy of exhaustive sensing at one level by sparse sensing at multiple levels. Our previous results proved that we do achieve the same reconstruction accuracy as exhaustive sensing with reduced number of sensors. But by fusing measurement from different sensors directly in our previous approach, the improvement is limited. In our current work, we established field models to incorporate more information. The distribution models help assign uncertainty to the measurements. By applying correlation models in reconstruction process, we can probe the structure in the data and further improve reconstruction accuracy.