Since the emergence of sensor data streams, increasing amounts of observations have to be transmitted, stored and retrieved. Performing these tasks at the granularity of single points would mean an inappropriate waste of resources. Thus, we propose a concept that performs a partitioning of observations by spatial, temporal or other criteria (or a combination of them) into data segments. We exploit the resulting proximity (according to the partitioning dimension(s)) within each data segment for compression and efficient data retrieval. While in principle allowing lossless compression, it can also be used for progressive transmission with increasing accuracy wherever incremental data transfer is reasonable. In a first feasibility study, we apply the proposed method to a dataset of ARGO drifting buoys covering large spatio-temporal regions of the world´s oceans and compare the achieved compression ratio to other formats.