In environmental science, where information from sensor devices are sparse, data fusion for mapping purposes is often based on geostatistical approaches. We propose a methodology called adaptive distance attention that enables us to fuse sparse, heterogeneous, and mobile sensor devices and predict values at locations with no previous measurement. The approach allows for automatically weighting the measurements according to a priori quality information about the sensor device without using complex and resource-demanding data assimilation techniques. Both ordinary kriging and the general regression neural network (GRNN) are integrated into this attention with their learnable parameters based on deep learning architectures. We evaluate this method using three static phenomena with different complexities: a case related to a simplistic phenomenon, topography over an area of 196 $ {km}^2 $ and to the annual hourly $ {NO}_2 $ concentration in 2019 over the Oslo metropolitan region (1026 $ {km}^2 $). We simulate networks of 100 synthetic sensor devices with six characteristics related to measurement quality and measurement spatial resolution. Generally, outcomes are promising: we significantly improve the metrics from baseline geostatistical models. Besides, distance attention using the Nadaraya–Watson kernel provides as good metrics as the attention based on the kriging system enabling the possibility to alleviate the processing cost for fusion of sparse data. The encouraging results motivate us in keeping adapting distance attention to space-time phenomena evolving in complex and isolated areas.