I have been wanting to learn TopoToolbox for a while, and last week I had a chance to get started. Spending time at the Universität Potsdam, Institut für Erd und Umweltwissenschaften, I have met with Wolfgang Schwanghart and Dirk Scherler occasionally. They are coauthors of TopoToolbox.
My ASU colleagues Adam Forte and Kelin Whipple have been using TopoToolbox increasingly over the last few years. Recently Kelin had a question as to why some DEMs grids were causing an error in TopoToolbox and why some were not. I dug into it and here is what I found. Projected DEMs should have EXACTLY the same x and y cell sizes and expressed in precise values (nearest meter or 10th or 100th of a meter).
DEMs (such as SRTM 30 meter) can be downloaded as geotiff formats and with geographic coordinate systems from sources such as OpenTopography: select Global Data tab.
For example, I have chosen a piece of data from Java along the Cimandiri fault (see Marliyani, et al., 2016 for example).
To be useful for most geomorphic analyses, the data should be projected to UTM so that the horizontal and vertical units are the same. Often, the projection is done in ArcGIS. And, usually it seems fine to let Arc determine the cell size automatically.
Here I set the cell size to exactly 30 m
Turning to TopoToolbox, as we compute drainage network properties, including contributing area, there is a check that the cell sizes are the same (inside the GRIDobj.m function):
if abs(abs(dx)abs(dy))>1e9; error('TopoToolbox:GRIDobj',... 'The resolution in x and ydirection must be the same'); endThe 1e9 is a somewhat arbitrarily small number which one would think would not cause a problem.
I looked into this problem tracking along with the TopoToolbox processing and some of the MATLAB built in tools.

Read the geotiff using MATLAB's geotiffread:
[demdata, R] = geotiffread(filename); geotiffreaddifference = R.CellExtentInWorldXR.CellExtentInWorldY;
The R object has a number of values including the cell size of the geotiff:
R.CellExtentInWorldX= 30.79776426908749800 R.CellExtentInWorldY= 30.79776426908791000 R.CellExtentInWorldXR.CellExtentInWorldY= 4.12115e13

Read the geotiff into the DEM object in TopoToolbox and do a similar check as above (refmat is similar to R):
DEM = GRIDobj(filename); refmatdifference = abs(DEM.refmat(2,1))abs(DEM.refmat(1,2));
Again we see the same very small grid size difference:
abs(DEM.refmat(2,1))abs(DEM.refmat(1,2))= 4.12115e13
 The problem comes in a calculation that builds out the x and y vectors that includes a cumulative multiplication from the grid sizes inside a TopoToolbox function refmat2XY:
nrrows = siz(1); nrcols = siz(2); x = [ones(nrcols,1) (1:nrcols)' ones(nrcols,1)]*R; x = x(:,1)'; y = [(1:nrrows)' ones(nrrows,2)]*R; y = y(:,2);
R is the refmat object. But, here we get the error:
dx=x(1)x(2)= 30.79776426916942000 dy=y(1)y(2)= 30.79776426777243600 abs(dx)abs(dy)= 1.39698e09
Which then would fail the GRIDobj test.
If I run the same set of checks on the file I projected with exactly 30 m grid cell size (recall above), I don't get the problem:
geotiffread: R.CellExtentInWorldX= 30.00000000000000000 R.CellExtentInWorldY= 30.00000000000000000 R.CellExtentInWorldXR.CellExtentInWorldY= 0 refmat difference as produced from GRIDobj: abs(DEM.refmat(2,1))abs(DEM.refmat(1,2))= 0 Difference after GRIDobj2mat: dx=x(1)x(2)= 30.00000000000000000 dy=y(1)y(2)= 30.00000000000000000 abs(dx)abs(dy)= 0
I think what is happening is a bit of roundoff error (see this link for detailed discussion: What Every Computer Scientist Should Know About FloatingPoint Arithmetic, by David Goldberg): "Therefore the result of a floatingpoint calculation must often be rounded in order to fit back into its finite representation." MATLAB does all of its calculations by default in double precision, so we should have access to 16 decimal digits (e.g., https://en.wikipedia.org/wiki/IEEE_7541985). So, why we loose precision up to 10^{9} at times is a little surprising to me, but it can obviously happen.
Thus there are a couple of workarounds:
 Comment out the dx and dy check in GRIDobj:
%if abs(abs(dx)abs(dy))>1e9; % error('TopoToolbox:GRIDobj',... % 'The resolution in x and ydirection must be the same'); % end
Or make the threshold larger.  Explicitly set the grid resolution to a round number when projecting in ArcGIS or other software.
 Use the reproject2utm.m command to project dems before using other TopoToolbox operations. It sets the resolution to be exactly the same for x and y
Here is a script to run these calculations: DEM_cell_size_Script.m
Thanks to Chris Crosby, Benjamin Gross, Wolfgang Schwanghart, Kelin Whipple, and Mike Zoldak for discussions.
Hi Ramón, thanks for pointing me to this problem with the function crop and numerical issues with floating point arithmetics. I have updated the function crop so that this error should not turn up again after cropping. https://github.com/wschwanghart/topotoolbox/blob/master/%40GRIDobj/crop.m
ReplyDeleteCheers, Wolfgang
Testimony of financial breakthrough from GOD through the help of Funding Circle Loan INC. (fundingloanplc@yahoo.com OR Call/Text +14067326622)...
ReplyDeleteHi I'am Evelyn Russell resident at 808 NE 19th St Oklahoma City, I am a single mother blessed with 2 daughters. For a while now I have been searching for a genuine loan lender who could help me with a loan as I no longer have a job, all I got were hoodlums who made me trust them and at the end they took my money without giving me any loan, my hope was lost, I got confused and frustrated, it became difficult for my family to feed with a good meal, I never wanted to have anything to do with any loan lending companies on the internet again. Not until I met a Godsent loan lender that changed my life and that of my family Through the help of a fellowship member "a lender with the fear of God in him Mr JASON RAYMOND, he was the man God sent to elevate my family from suffering. At first I thought it wasn’t going to be possible due to my previous experience until I received my loan worth $135,000.00 USD in less then 24hours. So my advise to anyone out there genuinely in need of a loan is to contact Funding Circle Loan INC through this official email: fundingloanplc@yahoo.com OR Call/Text +14067326622 and be financial lifted.