Select to view content in your preferred language

Best Practices For Raster Resampling Edge Effects

651
9
Jump to solution
11-25-2024 09:38 AM
Labels (2)
Ed_
by MVP Regular Contributor
MVP Regular Contributor
  • Say i have a set of rasters from the USGS website
  • I first clip them them to HUC-12s using the Extract by Mask Tool
  • Reproject the rasters to projected
  • Use the raster resample to change their resolution from 1m to 3m. Using Bilinear as the resampling technique as setting No Data to Minimum (have to set no data values as Smooth DEM ArcHydro Tool requires it)
  • These DEMs on the HUC-12 level will then be mosaiced together into a raster dataset
  • My question is though would there be edge effects issues while performing the above steps? If so, then how can the be minimized?

 

 

Question | Analyze | Visualize
0 Kudos
1 Solution

Accepted Solutions
DanPatterson
MVP Esteemed Contributor

@Ed_ the only sampling methods appropriate for continuous data are the bilinear and cubic convolution.

Resample (Data Management)—ArcGIS Pro | Documentation

Now there is a cavaet in the cubic, beyond the "its takes longer", in that you may get calculated values beyond the range of data that actually exists in your area of interest.

IMHO I would take a look at bilinear as a first go.

Of course this is all predicated on how "accurate" your dem is in the first place (eg. not).

You are messing around with lots of processes so I wouldn't get too hung up on which is best.  Follow the cell projection guidelines then try one of the two resampling methods and examine the results in light of what you know.


... sort of retired...

View solution in original post

9 Replies
DanPatterson
MVP Esteemed Contributor

How the Cell Size Projection Method environment setting works—ArcGIS Pro | Documentation

should be looked at since you are projecting and resampling and are worried about extent/edges

 


... sort of retired...
Ed_
by MVP Regular Contributor
MVP Regular Contributor

Hi Dan, happy  Monday, hope all is well 🙂 and thank you for sharing the link as it is good to know since the DEMs will be reprojected from GCS to PCS (one that kinda works for the whole state). The reprojected and resampled DEMs will be clipped to either HUC12 boundaries or county boundaries and then mosaiced to state level.  

Question | Analyze | Visualize
0 Kudos
Ed_
by MVP Regular Contributor
MVP Regular Contributor

Also Dan the resampling is being done from 1m to 3m

Question | Analyze | Visualize
0 Kudos
Ed_
by MVP Regular Contributor
MVP Regular Contributor

Also which Resampling No Data method out of the available five is better for DEMs?

Ed__0-1732570095132.png

 

Question | Analyze | Visualize
0 Kudos
Ed_
by MVP Regular Contributor
MVP Regular Contributor

@SteveLynch can you also chime on this please? thank you 

just for context the study area is the state of IL

Question | Analyze | Visualize
0 Kudos
DanPatterson
MVP Esteemed Contributor

@Ed_ the only sampling methods appropriate for continuous data are the bilinear and cubic convolution.

Resample (Data Management)—ArcGIS Pro | Documentation

Now there is a cavaet in the cubic, beyond the "its takes longer", in that you may get calculated values beyond the range of data that actually exists in your area of interest.

IMHO I would take a look at bilinear as a first go.

Of course this is all predicated on how "accurate" your dem is in the first place (eg. not).

You are messing around with lots of processes so I wouldn't get too hung up on which is best.  Follow the cell projection guidelines then try one of the two resampling methods and examine the results in light of what you know.


... sort of retired...
Ed_
by MVP Regular Contributor
MVP Regular Contributor

Thank Dan 🙂 yes i have started with bilinear interpolation method as well

The DEM source is 1m USGS 3DEP


Now what method should I use for No Data Values because i can't leave them empty since i have to later use the `Smooth High Resolution DEM` tool in ArcHydro and this tool requires the DEM to not have empty No Data values

Ed__0-1732572004972.png

 

Question | Analyze | Visualize
0 Kudos
DanPatterson
MVP Esteemed Contributor

If you have nodata in your area, it is there for a reason, so I would stick with None to keep it so.  This doesn't fill in values, it only assigns a value to use for nodata.

I am not inclined to fill in nodata areas under any circumstances (eg via focal statistics or regiongroup and nibble depending on the data type) unless they are only a couple of cells in size


... sort of retired...
Ed_
by MVP Regular Contributor
MVP Regular Contributor

Sounds good Dan, thank you so much 🙂

Question | Analyze | Visualize
0 Kudos