I work with a lot of high resolution raster files and using mosaic datasets to view the data has become an integral part of my work flow. However, running any time of analysis with mosaic datasets seems to be extremely time consuming. For example, running a conditional using raster calculator seem to turn into a never ending process. The mosaic dataset I'm currently working with has 147268 columns and 136493 rows with a 0.5 m resolution cell size. The cell values are 8 bit unsigned. This is rather large dataset, however, if I convert the mosaic dataset into just a single large raster file, running the same same calculation takes significantly less time.
Is this a lost cause? Should I just be running the steps on the smaller individual raster tiles and then assembling them as new mosaic dataset? Or does anyone have any tricks for running analyses with mosaic datasets? Are there any best practices I should be aware of?
Any help/thoughts are much appreciated.
20 billion cells is a large dataset! What's the underlying data format? If they are all ASCII files for example I'm guessing there is the need for some sort of internal conversion? I would exploring ensuring the data that your mosaic is built upon are all raster datasets (e.g. TIF with compression turned on), smallest bit-depth possible, have statistics built and an attribute table if the underlying data is integer).