I need to improve a bit the way I manage raster data so, I decided to make extensive use of mosaic datasets whenever appropriate. I'd need a bit of criticism to signal if I'm doing something wrong or not exactly optimal. I'm making my own tests but I do not feel 100% sure in this area of expertize.
I have the following scenario.
RASTER DATA. I have a collection of around 500 raster datasets (tif) wich are tiles of 10 deg covering the entire world. Spatial reference is GCS_WGS_1984, resolution is 0.00025 , 8 bit unsigned integer
OBJECTIVES. Provide a mosaic for the entire world to be used for processing as well as an European clip projected in ETRS_89_LAEA snapped to a standard extent. This mosaic would be used for processing as well
My first attempt to put down a workflow is as follows:
1. I create a mosaic dataset (named Step1) in GCS_WGS_1984 loading all the data and building overviews. Input rasters have stats and pyramids.
2. I create a second mosaic dataset called Step2 in GCS_WGS_1984 loading all rasters that are covering European extent. Then I apply a reproject raster function which would turn the mosaic in ETRS_89_LAEA setting the cell size at 25 meters and the registration point in order to align to my output extent. Finally, on top of reproject, I apply a clip raster function to matched my standard extent in LAEA
REGARDING STEP 1
1. Imaging I need to ingest the all world mosaic to a geoproc function (for example to calculate zonal statistics). How about size of request ? I assume I need to follow guidelines in this help page and set rows and columns properties in max size of request to the max num of columns and rows of the mosaic and the "maximum number of rasters per mosaic" = to the total number of rasters in the mosaic. Is this correct ?
REGARDING STEP 2
1. An alternative would be to use a Step1 derived mosaic dataset (referenced) and set the coordinate system to ETRS_1989_LAEA. However I cannot control the registration point to allign to my extent in LAEA. Correct ?
2. Instead of loading the rasters covering Europe I could have loaded all of them and then apply a chain of reproject and clip (or maybe clip -reproject - clip). Comments ? The advantage of this approach is that I can avoid a manual selection of rasters based on names (I think I cannot filter Add Rasters by workspace and extent....)
3. After applying the reproject function I think I have to modify size of request (cols and rows) because of different cols and rows in the projected space. Correct.
4. Which interpolation method the reproject is using ? Can I control that ? Is it controlled by Default resampling method
An important goal is being sure that the projected mosaic is reliable for geoprocessing (particularly zonal stats at 100 m in LAEA) and I'm not missing some important points that can cause wrong results.