Natural phenomena are difficult to model because there are many real-world complications. These complications in turn affect the sensor network experimental design which aims to answer a scientific question of interest. For example, a light field is affected by the position of the sun, the height of the tree canopy, and the strength of the wind blowing the leaves, among other things.Instead of using equations which relate all these variables to the field, we can use a statistical model which captures our beliefs of what the field will be like. We can take prior information, which may come from our knowledge or from previous experimental experience, and combine it with uncertanties in unknown quantities using the bayesian statistical framework. We learn the parameters of our models from measured data, and use these models to inform our experimental design.If we use a model for the field which is incorrect to design our sensor network experiements, for example to place the sensors, then our final result will not be what we desired. This project explores ways in which these model choices affect the performance of the learning process.