Today, I finally installed MATLAB 2018a. As always, I am quite excited about the software releases shipped every 6 months. One of the release notes on performance particularly caught my interest.
Loops that contain mainly indexing and scalar math operations execute faster due to execution engine optimizations. (see here)
TopoToolbox contains a number of tight loops with scalar math and thus I expect that this release should make TopoToolbox generally faster. I tested this using the function flowacc with a large DEM (SRTM-3, 90 m resolution, about 239 Millionen pixels). flowacc has been coded to be fast and thus there is also a C-coded MEX-file available for benchmarking. However, the code can also be run as pure MATLAB code which I have done in 2017b and 2018a. I measured run-times with tic-toc using 20 repetitions. The results are shown below.
M-code in 2018a takes about 77% of the time required in 2017b. That is a major improvement. Moreover, run-times are close to the C-code benchmark (118%). Can it get much better than this?
Now let’s put this in relation to other software. What we have looked at are flow accumulation run-times for the entire Himalayan range at a resolution of 90 m. 4-6 seconds is quite fast, I guess. In ArcGIS, the same operation would take hours*. But I am a bit reluctant to test this in detail.
*Note that ArcGIS doesn’t keep data in the main memory. Thus, much of the time may be used to swap data between the main memory and the hard drive. A direct comparison is thus not fair.
Alexandru T. Codilean, Henry Munack and their colleagues have just released their Open and Global Database of Cosmogenic Radionuclide and Luminescence measurements in fluvial sediment (OCTOPUS). This is a great open and accessible resource of data!
The cosmogenic radionuclide (CRN) part of the database consists of Be-10 and Al-26 measurements in fluvial sediment samples along with ancillary geospatial vector and raster layers, including sample site, basin outline, digital elevation model, gradient raster, flow direction and flow accumulation rasters, atmospheric pressure raster, and CRN production scaling and topographic shielding factor rasters. The database also includes a comprehensive metadata and all necessary information and input files for the recalculation of denudation rates using CAIRN, an open source program for calculating basin-wide denudation rates from Be-10 and Al-26 data.
The luminescence part of the database consists of thermoluminescence (TL) and optically stimulated luminescence (OSL) measurements in fluvial sediment samples from stratigraphic sections and sediment cores from across the Australian continent and includes ancillary vector and raster geospatial data.
OCTOPUS can be accessed at: https://earth.uow.edu.au
The developers of OCTOPUS also submitted a manuscript describing the database in detail to the open access journal Earth System Science Data (Discussions). The paper is now accessible and open for interactive public discussion until 01 May 2018 at:
You are invited to download the data and take part in the discussion.
I am increasingly using MATLAB to create maps. I simply find the workflow cumbersome that combines TopoToolbox, ArcGIS or QGIS. This is among the reasons why I implemented a new function in TopoToolbox that helps you coloring DEMs, particularly those that include topography and bathymetry. The function is ttcmap and you’ll find it in the colormaps folder.
Here is an example application of this function. I’ll use a DEM that covers one of the world’s steepest topographic gradients from the Andes down to the Peru-Chile deep-sea trench.
Before you run the function, choose a colormap. Calling ttcmap without input arguments displays a table that shows the different colormaps and their respective elevation ranges.
ttcmap Colormap_Name Min_Elevation Max_Elevation _____________ _____________ _____________ 'gmtrelief' -8000 7000 'france' -5000 4000 'mby' -8000 4000 'gmtglobe' -10000 9500
Now here are three colormaps that span the elevation range in our DEM. Note that ttcmap outputs the array zlimits which is required as input to imageschs so that the land-sea boundary is exactly at zero. ttcmap is similar to the mapping toolbox function demcmap but more precisely sets the land-sea transition to the zero value.
[cmap,zlimits] = ttcmap(DEM,'cmap','gmtrelief'); imageschs(DEM,,'caxis',zlimits,'colormap',cmap)
[cmap,zlimits] = ttcmap(DEM,'cmap','mby'); imageschs(DEM,,'caxis',zlimits,'colormap',cmap)
[cmap,zlimits] = ttcmap(DEM,'cmap','gmtglobe'); imageschs(DEM,,'caxis',zlimits,'colormap',cmap)
I downloaded the colormaps from cpt-city, a great resource for colormaps.
TopoToolbox version 2.3 will have a better documentation. With the latest commits to our github repository, you are now able to take a glance at our work-in-progress. Just download it, follow the instruction in the readme file, and then type
in the command window. MATLAB’s documentation browser will pop up and the landing page will have a new entry in the section Supplemental Software. The link will bring you to the TopoToolbox documentation.
The documentation features a Getting Started section, numerous guides, and a function reference (which is, however, not linking to the individual help functions).
Moreover, you can now search the documentation which will hopefully make it much easier to find the right tools.
The documentation is not yet complete. We are still working on including the help sections of each function into the documentation, which will make the functionality even more accessible and browsable.
**** addendum January 15, 2018 ****
After permanently setting the paths, you might need to restart MATLAB in order to be able to view the documentation in the help browser.
The TopoToolbox blog has now been online since October 2014. Since then, the blog had ~22k visitors. With a total of 8.7k, 2017 was the year with the most visitors. Yet, monthly statistics show quite some variability throughout the years. The blog tends to have fewer visitors in December and during the summer months July and August than during the remaining months. In addition, there are strong fluctuations between usage statistics of months that have seen many posts and those when I had no time to post. This variability superimposes the general trend. We may thus ask the question, whether we truely experience an increase in visitor counts.
To answer this question, I took monthly visitor counts during the last two years (unfortunately, I can only access the last two years). I saved them in an Excel file which I then imported to MATLAB. My aim was then to do a regression analysis to be able to identify possible trends. Since the dependent data are counts, a standard least squares regression is clearly not advisable. Rather, count data are modelled using Poisson statistics. I thus chose a generalized linear regression model with counts as a Poisson response, and time as independent variable. I used ‘log’ as a link function so that the response variable is Y~Poisson(exp(b1 + b2(X))).
The function fitglm (which is part of the Statistics and Machine Learning Toolbox) is perfectly suited to solve this regression problem. However, the function takes a frequentist approach which I didn’t want to follow here. Rather, I wanted to learn how to tackle this problem with Bayesian inference. Gladly, MATLAB increasingly features tools for Bayesian analysis such as the Hamiltonian Monte Carlo (HMC) sampler. The only thing I needed was some guidance of how to derive the prior distributions and how to calculate the posterior probability density function. For this, I found this paper by Doss and Narasimhan quite useful.
Writing tools for Bayesian analysis is not easy and the code quite difficult to debug. But finally, I managed to get my analysis working, and to return following figure:
Clearly, the posterior distribution of beta_1, the slope of the regression, does not include zero so I can assume with high certainty, that we truly observe a trend to higher visitor numbers. Can this finding be extrapolated? Well, I don’t know. Just keep on visiting this blog frequently, and we’ll see.
If you are interested in the technical details, here is the function:
function bayesianpoissonregress(t,y) %Bayesianpoissonregression % % t --> datetime vector (equally spaced) % y --> vector with counts t = t(:); y = y(:); % Numeric month indices X = (1:size(t,1))'; % add offset to design matrix X = [ones(size(X,1),1) X]; % link function gamma = @(beta) exp(beta(:)'*X')'; % Prior distributions of the parameters are gaussian a = [0 0]; % prior means b = [20 20]; % prior std d = a./b + sum(X.*y); %% MCMC sampling hmc = hmcSampler(@(beta) logpdf(beta),[0 0],'UseNumericalGradient',true); % Estimate max. a posteriori map = estimateMAP(hmc); % tune sampler hmc = tuneSampler(hmc,'StartPoint',map); % draw chain chain = drawSamples(hmc,'StartPoint',map); %% Plot results figure subplot(1,2,1) [~,ax] = plotmatrix(chain); ylabel(ax(1,1),'\beta_0'); ylabel(ax(2,1),'\beta_1'); xlabel(ax(2,1),'\beta_0'); xlabel(ax(2,2),'\beta_1'); subplot(1,2,2) % random set of parameters from chain sam = randperm(size(chain,1),100); sam = chain(sam,:); hold on for r = 1:size(sam,1) chat = gamma(sam(r,:)); plot(t,chat,'color',[.7 .7 .7]); end plot(t,y,'--s') ylabel('Visitors') box on % -- end of function %% Log pdf function p = logpdf(beta) % posterior distribution according to % http://citeseerx.ist.psu.edu/viewdoc/download?doi=10.1.1.39.9684&rep=rep1&type=pdf % (Eq. 2.2) % posterior p = - sum(1./(b.^2) .* beta.^2) + sum(d .* beta) - sum(gamma(beta)); end end
Before this year ends, here is an announcement for an event next year: The Central European Conference on Geomorphology and Quaternary Sciences will take place in Giessen in September 2018. The event is jointly organized by the German Working Group on Geomorphology (AK Geomorphologie) and the German Quaternary Association (DEUQUA – Deutsche Quartärvereinigung). Here are the key dates:
|Start of Registration|
|End of Early Registration|
|Deadline of Abstract Submission|
|End of Late Registration Period|
|September 23 – 27
You’ll find more information >>here<<.
Today, I read about the Tanaka method, an approach to depicting relief. The method relies on filled contours and coloring contour lines according to their direction and light origin. I have seen solutions in QGIS (here and here) which made me thinking whether the Tanaka method can also be implemented in MATLAB.
One difficulty with Tanaka’s method is that coloring of contour lines requires knowledge about whether higher values are found to the left or the right of a contour line. GDAL’s contouring algorithm has consistent behavior in this respect. However, MATLAB’s algorithm doesn’t. My trick is to calculate surface normals and their angle to the light source, and then interpolate the gridded directions to the vertex locations of the contour lines.
Now here is my approach which uses some undocumented features of the HG2 graphics. An excellent resource for undocumented features in MATLAB is Yair Altman’s blog.
function tanakacontour(DEM,nlevels) %TANAKACONTOUR Relief depiction using Tanaka contours % % Syntax % % tanakacontour(DEM,nlevels) % % See also: contourf, hillshade % % Author: Wolfgang Schwanghart (w.schwanghart[at]geo.uni-potsdam.de) % Date: 8. December, 2017 if nargin == 1; nlevels = 10; end % calculate solar vector azimuth = 315; azid = azimuth-90; azisource = azid/180*pi; [sx,sy,~] = sph2cart(azisource,0.1,1); % get coordinate matrices and z values [Z,X,Y] = GRIDobj2mat(DEM); % calculate surface normals [Nx,Ny] = surfnorm(Z); N = hypot(Nx,Ny); Nx = Nx./N; Ny = Ny./N; H = GRIDobj(DEM); H.Z = Nx*sx + Ny*sy; % Get coordinate matrices and Z [~,h] = contourf(X,Y,Z,nlevels); axis image drawnow % colormap cmap = gray(255)*255; cval = linspace(0,1,size(cmap,1))'; % the undocumented part for r = 1:numel(h.EdgePrims) % the EdgePrims property contains vertex data x = h.EdgePrims(r).VertexData(1,:); y = h.EdgePrims(r).VertexData(2,:); % interpolate hillshade grayscale values to contour vertices c = interp(H,double(x(:)),double(y(:))); % interpolate to colormap clr = interp1(cval,cmap,c); % adjust color data to conform with EdgePrims ColorData property clr = clr'; clr = uint8([clr;repmat(255,1,size(clr,2))]); % set all properties at once set(h.EdgePrims(r),'ColorBinding','interpolated','ColorData',clr,'LineWidth',1.2); end
Ok, now let’s try this on some data. In this code, I have used GRIDobj, the class for gridded, georeferenced data.
[X,Y,Z] = peaks(500); tanakacontour(GRIDobj(X,Y,Z),10)
And now for the Big Tujunga DEM:
DEM = GRIDobj('srtm_bigtujunga30m_utm11.tif'); tanakacontour(DEM,5)
Now that looks ok, but I am not totally convinced by these unrealistic terraces, to be honest. Moreover, my approach so far doesn’t include changing line widths. Contour lines should actually be thicker for slopes that face both towards and away from the light source. However, I didn’t figure out how to do this so far. Any ideas?