In this paper we introduce a statistical method tobuild two-dimensional gas distribution maps (Kernel DM+V/Walgorithm). In addition to gas sensor measurements, the proposedmethod also takes into account wind information by modelingthe information content of the gas sensor measurements as abivariate Gaussian kernel whose shape depends on the measuredwind vector. We evaluate the method based on real measurementsin an outdoor environment obtained with a mobile robot thatwas equipped with gas sensors and an ultrasonic anemometerfor wind measurements. As a measure of the model quality wecompute how well unseen measurements are predicted in termsof the data likelihood. The initial results are encouraging andshow a clear improvement of the proposed method compared tothe case where wind is not considered.