An approach is presented to efficiently produce high quality gridded data records from the large, global point-based dataset returned by the Diviner Lunar Radiometer Experiment aboard NASA’s Lunar Reconnaissance Orbiter. The need to minimize data volume and processing time in production of science-ready map products is increasingly important with the growth in data volume of planetary datasets. Diviner makes on average >1400 observations per second of radiance that is reflected and emitted from the lunar surface, using 189 detectors divided into 9 spectral channels. Data management and processing bottlenecks are amplified by modeling every observation as a probability distribution function over the field of view, which can increase the required processing time by 2–3 orders of magnitude. Geometric corrections, such as projection of data points onto a digital elevation model, are numerically intensive and therefore it is desirable to perform them only once. Our approach reduces bottlenecks through parallel binning and efficient storage of a pre-processed database of observations. Database construction is via subdivision of a geodesic icosahedral grid, with a spatial resolution that can be tailored to suit the field of view of the observing instrument. Global geodesic grids with high spatial resolution are normally impractically memory intensive. We therefore demonstrate a minimum storage and highly parallel method to bin very large numbers of data points onto such a grid. A database of the pre-processed and binned points is then used for production of mapped data products that is significantly faster than if unprocessed points were used. We explore quality controls in the production of gridded data records by conditional interpolation, allowed only where data density is sufficient. The resultant effects on the spatial continuity and uncertainty in maps of lunar brightness temperatures is illustrated. We identify four binning regimes based on trades between the spatial resolution of the grid, the size of the FOV and the on-target spacing of observations. Our approach may be applicable and beneficial for many existing and future point-based planetary datasets.