Regridding using lat/lon data

I am working with runoff data on a high resolution regional grid. The grid is 560x560 in size and I need to regrid the data to be on a lower resolution grid of size 384x320. The high res data has latitude/longitude coordinates and one dimensional values for each grid cell. The low res grid has lat/lon locations. I need to map each grid cell from the high res grid onto a grid cell in the low res grid so that none of the runoff data is lost and the runoff total over the whole space will be preserved. I have been unable to find a nearest neighbor search function that can use lat/lon to map data across grids. Does anyone have any suggestions? Thank you!

[Hitesh Bugata]

Use griddedinterpolant(oldlat,oldlon,variable data, newlat,newlon).

Answer 1

[KSSV]

doc interp2.

Answer 2

[SR]

Unfortunately interp2 does not work in this case because the query points are specified by the grid the data is being moved to so it loses a ton of data when going from a high res grid to a low res one as it doesn't look at the majority of cells within the high res grid. The matlab interpolation functions in general won't work here since the problem is essentially the opposite of interpolation.

[KSSV]

Have you tried interp2 and plot the result? When you use interp2 the values definitely match with your original grid.

[SR]

Yes. It does look accurate in a plot, however the summed runoff over the whole field is off by an order of magnitude due to the issues I listed above.

Answer 3

[Walter Roberson]

Let first_lat be the latitude for runoff_data(1,:), last_lat be the latitude for runoff_data(end,:), first_long be the longitude for runoff_data(:,1), last_long be the longitude for runoff_data(:,end)

row_vector_of_old_latitudes = linspace(first_lat, last_lat, size(runoff_data,1));
column_vector_of_old_longitudes = linspace(first_long, last_long, size(runoff_data,2)).';

row_vector_of_new_latitudes = linspace(first_lat, last_lat, 384);
column_vector_of_new_longitudes = linspace(first_long, last_long, 320).';

new_runoff_data = interp2(row_vector_of_old_latitudes, column_vector_of_old_longitudes, runoff_data, row_vector_of_new_latitudes, column_vector_of_new_longitudes, 'linear');

[Walter Roberson]

Afterwards,

nrd = new_runoff_data * size(runoff_data,1)/size(new_runoff_data,1) * size(runoff_data,2)/size(new_runoff_data,2)

This will not be an exact preservation of the total. For example with my random sample data the total of nrd was 1725168.80255951 compared to an actual total of 1724560

Getting an exact preservation without simple scaling by the ratio is tricky.