Wednesday, April 16, 2014

D3 map Styling tutorial IV: Drawing gradient paths

After creating the last D3js example, I was unsatisfied with the color of the path. It changed with the typhoon class at every moment, but it wasn't possible to see the class at every position. When I saw this example by Mike Bostock, I found the solution.

Understanding the gradient along a stroke example

First, how to adapt the Mike Bostock's Gradient Along Stroke example to a map.
The map is drawn using the example Simple path on a map, from this post. The only change is that the dashed path is changed with the gradient.
You can see the result here.
The differences from drawing a simple path start at the line 100:
var line = d3.svg.line()
      .x(function(d) { return projection([d.lon,])[0]; })
      .y(function(d) { return projection([d.lon,])[1]; });

      .data(quad(sample(line(track), 8)))
      .style("fill", function(d) { return color(d.t); })
      .style("stroke", function(d) { return color(d.t); })
      .attr("d", function(d) { return lineJoin(d[0], d[1], d[2], d[3], trackWidth); });

  •  The line definition remains the same. From every element it gets, it takes the lat and lon attributes, projecting them, and assigning them to the x and y path properties
  • A color function is defined at line 41, which will interpolate the color value from green to red:
    var color = d3.interpolateLab("#008000", "#c83a22");
  • The data is not the line(track) directly, as in the former example, but passed through the functinos sample and quad.
  • The sample function assigns a property t with values between 0 and 1, which is used to get the color at every point.
  • Finally, the function lineJoin is used to draw a polygon for the sampled area.
The functions used in the Mike Bostock's example aren't explained, I'll try to do it a little:
  • sample takes a line (the data applied to a line function), and iterates with the precision parameter as increment along the string, creating an array with all the calculated points.
  • quad takes the points calculated by the sample function and returns an array with the adjacent points (i-1, i, i+1, i+2).
  • lineJoin takes the four points generated by quad, and draws the polygon, with the help of lineItersect and perp functions.

Drawing the typhoon track with the colors according to the typhoon class

The final example draws the typhoon path changing smoothly the color according to the typhoon class.
The animation of the path, and the rotating icon are explained in the third part of the tutorial. In this case, the way to animate the path will change.
For each position of the typhoon, a gradient path is drawn, because the gradient is always between two colors. So the part of the code that changes is:
      //Draw the path, only when i > 0 in otder to have two points
      if (i>0){
        color0 = color_scale(track[i-1].class);
        color1 = color_scale(track[i].class);

        var activatedTrack = new Array();

        var color = d3.interpolateLab(color0, color1);
        .data(quad(sample(line(activatedTrack), 1)))
          .style("fill", function(d) { return color(d.t);})
          .style("stroke", function(d) { return color(d.t); })
          .attr("d", function(d) { return lineJoin(d[0], d[1], d[2], d[3], trackWidth); });

      i = i + 1;
          if (i==track.length)

  • Inside the animation interval (line 145), the gradient path is create for each position (starting with the second one to have two points)
  • The two colors are taken from the point information
  • An array with the two points is created, with the name activatedTrack. I tried using more points, but the result is very similar.
  • The color interpolation is calculated (line 172)
  • The gradient colored path is created (line 173). Note that the name is path+i, to make different paths each iteration, and not to overwrite them. The method is the same as the one used in the first section.
Besides, an invisible path with all the positions is created, so the typhoon icon can be moved as it was in the third part of the tutorial.


Monday, March 31, 2014

Slides for the workshop "Introduction to Python for geospatial uses"

Last 26th, 27th and 28th of March, the 8as Jornadas SIG Libre were held in Girona, where I had the opportunity to give a workshop about Python for geospatial uses.

The slides in Spanish:

The Slides in English:

The example files in both languages:

The meeting was awesome, if you have the opportunity and understand Spanish, come next year!

Monday, March 24, 2014

Shaded relief images using GDAL python

After showing how to colour a DEM file, classifying it, and calculating its isobands, this post shows how to create a shaded relief image from it.
The resulting image
A shaded relief image simulates the shadow thrown upon a relief map. This shadow is usually blended with some colouring, related to the altitude, a terrain classification, etc.
The shadow is usually drawn considering that the sun is at 315 degrees of azimuth and 45 degrees over the horizon, which never happens at the north hemisphere. This values avoid strange perceptions, such as seeing the mountain tops as the bottom of a valley.

In this example, the script calculates the hillshade image, a coloured image, and blends them into the shaded relief image.

As usual, all the code, plus the sample DEM file, can be found at GitHub.

The hillshade image

I didn't know how to create a shaded relief image using numpy. Eric Gayer helped me with some samples, and I found some other information here.
The script is:
Creates a shaded relief file from a DEM.

from osgeo import gdal
from numpy import gradient
from numpy import pi
from numpy import arctan
from numpy import arctan2
from numpy import sin
from numpy import cos
from numpy import sqrt
from numpy import zeros
from numpy import uint8
import matplotlib.pyplot as plt

def hillshade(array, azimuth, angle_altitude):
    x, y = gradient(array)
    slope = pi/2. - arctan(sqrt(x*x + y*y))
    aspect = arctan2(-x, y)
    azimuthrad = azimuth*pi / 180.
    altituderad = angle_altitude*pi / 180.
    shaded = sin(altituderad) * sin(slope)\
     + cos(altituderad) * cos(slope)\
     * cos(azimuthrad - aspect)
    return 255*(shaded + 1)/2

ds = gdal.Open('w001001.tiff')  
band = ds.GetRasterBand(1)  
arr = band.ReadAsArray()

hs_array = hillshade(arr,315, 45)

  • The script draws the image using matplotlib, to make it easy
  • The hillshade function starts calculating the gradient for the x and y directions using the numpy.gradient function. The result are two matrices of the same size than the original, one for each direction.
  • From the gradient, the aspect and slope can be calculated. The aspect will give the mountain orientation, which will be illuminated depending on the azimuth angle. The slopewill change the illumination depending on the altitude angle.
  • Finally, the hillshade is calculated.

 The shaded relief image is calculated using the algorithm explained in the post Colorize PNG from a raster file and the hillshade.
As in the coloring post, the image is read by blocks to improve the performance, because it uses a lot of arrays, and doing it at once with a big image can take a lot of resources.
I will coment the code block by block, to make it easier. The full code is here.

The main function, called shaded_relief, is the most important, and calls the different algorithms:
def shaded_relief(in_file, raster_band, color_file, out_file_name,
    azimuth=315, angle_altitude=45):
    The main function. Reads the input image block by block to improve the performance, and calculates the shaded relief image

    if exists(in_file) is False:
            raise Exception('[Errno 2] No such file or directory: \'' + in_file + '\'')    
    dataset = gdal.Open(in_file, GA_ReadOnly )
    if dataset == None:
        raise Exception("Unable to read the data file")
    band = dataset.GetRasterBand(raster_band)

    block_sizes = band.GetBlockSize()
    x_block_size = block_sizes[0]
    y_block_size = block_sizes[1]

    #If the block y size is 1, as in a GeoTIFF image, the gradient can't be calculated, 
    #so more than one block is used. In this case, using8 lines gives a similar 
    #result as taking the whole array.
    if y_block_size < 8:
        y_block_size = 8

    xsize = band.XSize
    ysize = band.YSize

    max_value = band.GetMaximum()
    min_value = band.GetMinimum()

    #Reading the color table
    color_table = readColorTable(color_file)
    #Adding an extra value to avoid problems with the last & first entry
    if sorted(color_table.keys())[0] > min_value:
        color_table[min_value - 1] = color_table[sorted(color_table.keys())[0]]

    if sorted(color_table.keys())[-1] < max_value:
        color_table[max_value + 1] = color_table[sorted(color_table.keys())[-1]]
    #Preparing the color table
    classification_values = color_table.keys()

    max_value = band.GetMaximum()
    min_value = band.GetMinimum()

    if max_value == None or min_value == None:
        stats = band.GetStatistics(0, 1)
        max_value = stats[1]
        min_value = stats[0]

    out_array = zeros((3, ysize, xsize), 'uint8')

    #The iteration over the blocks starts here
    for i in range(0, ysize, y_block_size):
        if i + y_block_size < ysize:
            rows = y_block_size
            rows = ysize - i
        for j in range(0, xsize, x_block_size):
            if j + x_block_size < xsize:
                cols = x_block_size
                cols = xsize - j

            dem_array = band.ReadAsArray(j, i, cols, rows)
            hs_array = hillshade(dem_array, azimuth, 

            rgb_array = values2rgba(dem_array, color_table, 
                classification_values, max_value, min_value)

            hsv_array = rgb_to_hsv(rgb_array[:, :, 0], 
                rgb_array[:, :, 1], rgb_array[:, :, 2]) 

            hsv_adjusted = asarray( [hsv_array[0], 
                hsv_array[1], hs_array] )          

            shaded_array = hsv_to_rgb( hsv_adjusted )
            out_array[:,i:i+rows,j:j+cols] = shaded_array
    #Saving the image using the PIL library
    im = fromarray(transpose(out_array, (1,2,0)), mode='RGB')
  • After opening the file, at line 20 comes the first interesting point. If the image is read block by block, some times the blocks will have only one line, as in the GeoTIFF images. With this situation, the y gradient won't be calculated, so the hillshade function will fail. I've seen that taking only two lines gives coarse results, and with lines the result is more or less the same as taking the whole array. The performance won't be as good as using only one block, but works faster anyway.
  • Lines 32 to 51 read the color table and file maximim and minumum. This has to be outside the values2rgba function, since is needed only once.
  • Lines 54 to 66 control the block reading. For each iteration, a small array will be read (line 67). This is what will be processed. The result will be written in the output array defined at line 52, that has the final size.
  • Now the calculations start:
    • At line 69, the hillshade is calculated
    • At line 72, the color array is calculated
    • At line 75, the color array is changed from rgb values to hsv. 
    • At line 78, the value (the v in hsv) is changed to the hillshade value. This will blend both images. I took the idea from this post.
    • Then the image is transformed to rgb again (line 81) and written into the output array (line 83)
  • Finally, the array is transformed to a png image using the PIL library. The numpy.transpose function is used to re-order the matrix, since the original values are with the shape (3, height, width), and the Image.fromarray function needs (height, width, 3). An other way to do this is using scipy.misc.imsave (that would need scipy installed just for that), or the Image.merge function.

The colouring funcion is taken from the post  Colorize PNG from a raster file, but modifying it so the colors are only continuous, since the discrete option doesn't give nice results in this case:
def values2rgba(array, color_table, classification_values, max_value, min_value):
    This function calculates a the color of an array given a color table. 
    The color is interpolated from the color table values.
    rgba = zeros((array.shape[0], array.shape[1], 4), dtype = uint8)

    for k in range(len(classification_values) - 1):
        if classification_values[k] < max_value and (classification_values[k + 1] > min_value ):
            mask = logical_and(array >= classification_values[k], array < classification_values[k + 1])

            v0 = float(classification_values[k])
            v1 = float(classification_values[k + 1])

            rgba[:,:,0] = rgba[:,:,0] + mask * (color_table[classification_values[k]][0] + (array - v0)*(color_table[classification_values[k + 1]][0] - color_table[classification_values[k]][0])/(v1-v0) )
            rgba[:,:,1] = rgba[:,:,1] + mask * (color_table[classification_values[k]][1] + (array - v0)*(color_table[classification_values[k + 1]][1] - color_table[classification_values[k]][1])/(v1-v0) )
            rgba[:,:,2] = rgba[:,:,2] + mask * (color_table[classification_values[k]][2] + (array - v0)*(color_table[classification_values[k + 1]][2] - color_table[classification_values[k]][2])/(v1-v0) )
            rgba[:,:,3] = rgba[:,:,3] + mask * (color_table[classification_values[k]][3] + (array - v0)*(color_table[classification_values[k + 1]][3] - color_table[classification_values[k]][3])/(v1-v0) )
    return rgba
The hillshade function is the same explained at the first point
The functions rgb_to_hsv and hsv_to_rgb are taken from this post, and change the image mode from rgb to hsv and hsv to rgb.


Tuesday, February 25, 2014

3D terrain visualization with python and Mayavi2

I have always wanted to draw these 3D terrains like those in, which are amazing. But the examples were all using software I don't use, so I tried to do it with python.
The final result

As usual, you can get all the source code and data at my GitHub page.

 Getting the data

After trying different locations, I decided to use the mountain of Montserrat, close to Barcelona, since it has nice stone towers that are a good test for the DEM and the 3D visualization. An actual picture of the zone used is this one:
Montserrat monastery
The building is a good reference, since the stone only areas make the result testing much more difficult.
All the data has been downloaded from the ICGC servers:
  • The DEM data was downloaded from the Vissir3 service, going to the section catàleg i descàrregues -> MDE 5x5. The file is named met5v10as0f0392Amr1r020.txt, but I cut a small part of it, to make mayavi2 work smoother using:

    gdalwarp -te 401620 4604246 403462 4605867 -s_srs EPSG:25831 -t_srs EPSG:25831 met5v10as0f0392Amr1r020.txt dem.tiff
  • The picture to drap over the dem file can be downloaded using the WMS service given by the ICGC:,4604246.0,403462.0,4605867.0&WIDTH=1403&HEIGHT=1146&LAYERS=orto5m&STYLES=&FORMAT=JPEG&BGCOLOR=0xFFFFFF&TRANSPARENT=TRUE&EXCEPTION=INIMAGE
 It's not as automatic as I would like, but if it's possible to download a DEM and the corresponding image, it's possible to create the 3D image.

Creating the image

First, let's plot the DEM file in 3D using mayavi2:
Plotting the terrain DEM with Mayavi2

from osgeo import gdal
from mayavi import mlab

ds = gdal.Open('dem.tiff')
data = ds.ReadAsArray()

mlab.figure(size=(640, 800), bgcolor=(0.16, 0.28, 0.46)), warp_scale=0.2)
  • In first place, we import as usual, gdal and numpy. Also the mlab library from mayavi, which lets set the mayavi canvas.
  • The data is read, as usual, with the gdal ReadAsArray method. 
  • The figure is created. This works like creating the Image object in the PIL library, creating the canvas where the data wil be drawn. In this case, the size is 640 x 800 pixels, making the figure bigger can affect the performance in some old computers. bgcolor sets the blue color as the background.
  • The surf method will plot the surface. The input has to be a 2D numpy array, which is what we have.  
    • The warp_scale argument sets the vertical scale. In this case, letting the default value (1?) creates a really exaggerated effect so its better to play a little to get a more realistic effect.
    • The colors depend of the Z value at each point, and can be changed using the color or colormap option.
  • The show() method makes the image to stay visible when running the example from a script. If you use ipython, you don't need this step.
  • If you want to save the figure as a png, you can either use the icon in the mayavi window or call the method mlab.savefig('image_name')
  • If you want to move the camera (change the prespective), you can use the roll/yaw/pitch methods:
    f = mlab.gcf()
    camera =

The plotted DEM
Now, let's put an aerial image over the 3D visualization:
Draping an image over a terrain surface
from osgeo import gdal
from tvtk.api import tvtk
from mayavi import mlab
import Image

ds = gdal.Open('dem.tiff')
data = ds.ReadAsArray()
im1 ="ortofoto.jpg")
im2 = im1.rotate(90)"/tmp/ortofoto90.jpg")
bmp1 = tvtk.JPEGReader()
bmp1.file_name="/tmp/ortofoto90.jpg" #any jpeg file


mlab.figure(size=(640, 800), bgcolor=(0.16, 0.28, 0.46))

surf =, color=(1,1,1), warp_scale=0.2) = True = 'plane' = my_texture

  • The most important new import is tvtk. TVTK is a python api that allows to work with VTK objects. Actually, my knowledge of Mayavi2 is very limited, but I see TVTK as an extension.
  • The DEM data is read the same way, using the ReadAsArray method.
  • The aerial image, named ortofoto.jpg, is not in the correct orientation. It took me a lot of time to get what was happening. I rotate it opening the image with the PIL library and using the rotate method (lines 11 to 13)
  • Then, the tvtk object with the texture is created, loading the image with a JPEGReader object, and assigning it to the Texture object (lines 14 to 19). 
  • The figure and the 3D surface is created as in the other example (lines 22 and 24)
  • Then, the surface is modified to show the image over it (lines 25 to 28). 
The final result

The result is correct, but the aerial image is a little moved from the place it should be so, since the terrain is really steep, part of the buildings are drawn on the stone walls! I edited the WMS coordinates a little so the result is slightly better. Anyway, the method is correct.


  • GitHub: The source code and example data
  • Mayavi2: 3D scientific data visualization and plotting 
  • This entry to a mailing list gave me the tip to create the example
  • If you want to do more or less the same, but using JavaScript, Bjørn Sandvik posted this excellent example.
  • ShadedRelief: Cool 3D and shaded relief examples.

Saturday, January 18, 2014

D3 map Styling tutorial III: Drawing animated paths

The example La Belle France, or the original La Bella Italia by Gregor Aisch, have a nice ferry sailing along a path to connect the islands with the continent.
Drawing a path in a map, and animating some icon on it can be a nice tool to show information about routes, storm tracks, and other dynamic situations.

This example shows how to draw the Haiyan typhoon track on the map drawn in the last post.

The working examples are here:
Creating a geopath
Animating an object on the path

Getting the data

Both the base map data and the typhoon data are explained in the post  D3 map Styling tutorial I: Preparing the data

Creating a geopath

First, how to draw the path line on a map. The working example is here.

  • The base map is drawn in the simplest way, as shown in this example, so the script stays clearer.
  • The typhoon track is loaded from the json file generated in the first tutorial post (line 55)
  • The path is created and inserted from lines 68 to 75:
    • A d3.svg.line element is created. This will interpolate a line between the points. An other option is to draw segments from each point, so the line is not so smooth, but the actual points are more visible. 
      • The interpolate method sets the interpolation type to be used.
      • x and y methods, set the svg coordinates to be used. In our case, we will transform the geographical coordinates using the same projection function set for the map. The coordinates transformation is done twice, one for the x and another for the y. It would be nice to do it only once.
    • The path is added to the map, using the created d3.svg.line, passing the track object as a parameter to be used by the line function. The class is set to path, so is set to a dashed red line (line 20)
Drawing the paths is quite easy, taking only two steps.

Animating an object on the path


The typhoon position for every day is shown on the path, with an icon. The icon size and color change with the typhoon class. The working example is here.

This second example is more complex than the first one:
  • The base map has a shadow effect. See the second part of the tutorial for the source.
  • The map is animated:
    • Line 135 sets an interval, so the icon and line can change with the date.
    • A variable i is set, so the array elements are used in every interval.
    • When the dates have ended, the interval is removed, so everything stays quiet. Line 158.
  • An icon moves along the path indicating the position of the typhoon
    •  Line 128 created the icon. First, I created it using inkscape, and with the xml editor that comes with it, I copied the path definition. This method can get really complex with bigger shapes.
    • Line 136 finds the position of the typhoon. The length of the track is found at line 122 with the getTotalLength() method.
    • Line 137 moves the icon. A transition is set, so the movement is continuous even thought the points are separated. The duration is the same as the interval, so when the icon has arrived at the final point, a new transformation starts to the next one.
    • Line 141 has the transform operation that sets the position (translate), the size (scale) and rotation (rotate). The factors multiplying the scale and rotation are those only to adjust the size and rotation speed. They are completely arbitrary.
  • The path gets filled when the icon has passed. I made this example to learn how to do it. Everything happens at line 144. Basically, the trick is creating a dashed line, and playing with the stroke-dashoffset attribute to set where the path has to arrive.
  • The color of the path and icon change with the typhoon class
    • At line 54, a color scale is created using the method d3.scale.quantile
      • The colors are chosen with colorbrewer, which is a set of color scales for mapping, and has a handy javascript library to set the color scales just by choosing their name. I learned how to use it with this example by Mike Bostock.
    • The lines 156 and 142 change the track and icon colours.
  • Finally, at line 154, the date is changed, with the same color as the typhoon and track.


Simple path on a map - The first example
Haiyan typhoon track - The second example
D3 map Styling tutorial I: Preparing the data
D3 map Styling tutorial II: Giving style to the base map
Animated arabic kufic calligraphy with D3 - Animating paths using d3js
La Bella Italia - Kartograph example by Gregor Aisch
La Belle France - The same example as La Bella Italia, but using D3js
Every ColorBrewer Scale - An example to learn how to use ColorBrewer with D3js