Mondrian's color palette, 1908 to 1943, as pie charts

Most people picture Mondrian as red, blue and yellow blocks with black lines. It took him more than a decade to get there. I took 5 of his paintings from 1908 to 1943, measured the 7 dominant colors in each, and turned them into pie charts to see what the road to those blocks looks like as data.

Five pie charts of the dominant colors in Mondrian paintings from 1908 to 1943, going from blues to grays to red, white and yellow — All five paintings side by side, in the order he painted them.

The five paintings

1908-10: Evening; Red Tree

Evening; Red Tree by Piet Mondrian next to a pie chart of its 7 dominant colors, mostly blues — The red tree is 2.8% red.

The painting is named after a red tree, and the tree gets 2.8% of the canvas (Sienna). Four blue and near-black clusters take 85%. The red is there, but it sits in thin branches on top of a blue field, and a pixel count doesn't care what you look at first.

1911: Gray Tree

Gray Tree by Piet Mondrian next to a pie chart of its 7 dominant colors, all shades of gray — Seven clusters, seven grays.

Three years later the color is gone. I asked the script for 7 colors and it returned 7 grays, from Silver to Black. The biggest one, Dark Gray, is 22.1%, so no single tone dominates either.

1915: Composition No. 10 (Pier and Ocean)

Composition No. 10 (Pier and Ocean) by Piet Mondrian next to a pie chart of its colors, almost half White Smoke — Almost half the canvas is background.

The tree is gone, and the pier and the sea are short black and white dashes. White Smoke takes 47.3%, and with Light Gray and Silver the light tones add up to 78%. The dashes themselves are thin, so they end up in the small dark slices (Black is 3.4%).

1930: Composition II in Red, Blue and Yellow

Composition II in Red, Blue and Yellow by Piet Mondrian next to a pie chart of its colors, half of it red — Red takes half the canvas.

This is the one everyone knows, and the split is less even than the name suggests. Red is 50% (Tomato 28.8% plus Crimson 21.2%, two clusters for one big square). Blue is 4.9% and yellow is under 1%. The grid of black lines is 7.6%.

1942-43: Broadway Boogie Woogie

Broadway Boogie Woogie by Piet Mondrian next to a pie chart of its colors, white and gold with small red and blue slices — Yellow replaces black.

Mondrian painted this in New York, a year before he died. There's no black in the chart at all. The lines that used to be black are yellow now (Gold, 20.6%), broken up by small squares of red (Brown, 7.7%) and blue (Dark Slate Blue, 5.7%).

All the numbers

The biggest slice in each painting, and how much of the canvas the colors people associate with Mondrian take up:

Year	Painting	Biggest color	Red	Blue	Yellow	Black
1908-10	Evening; Red Tree	Dark Slate Blue 28.4%	2.8%	66.9%	0%	18.1%
1911	Gray Tree	Dark Gray 22.1%	0%	0%	0%	8.1%
1915	Pier and Ocean	White Smoke 47.3%	0%	0%	0%	3.4%
1930	Composition II	Gainsboro 34%	50.0%	4.9%	0.7%	7.6%
1942-43	Broadway Boogie Woogie	White Smoke 53.2%	7.7%	5.7%	22.9%	0%

Red, blue, yellow and black are my grouping of the CSS color names. For example, the 1908 blue is Dark Slate Blue, Steel Blue and Midnight Blue added together, and the 1943 yellow is Gold plus Dark Khaki.

How I made it

Same method as my famous paintings as pie charts post, where the full Python code is:

Public-domain image from Wikimedia Commons
  → Downscale to 800px
  → Convert pixels to CIELAB color space
  → K-means clustering (7 clusters, scikit-learn)
  → Name each cluster after the nearest CSS color
  → Slice size = share of pixels in that cluster

Then I put each painting's 7 colors and percentages into the pie chart maker, used the extracted hex colors for the slices, and exported JPEGs.

What this method gets wrong

Downscaling to 800px blurs thin lines into the colors next to them. That hurts Mondrian more than most painters, because his black grid lines are a few pixels wide. The 7.6% black in the 1930 painting is a lower bound, and some of the "Gainsboro" around the lines is likely black and white blurred together.

Fixed at 7 clusters, K-means also splits large flat areas. The 1930 red square came out as two clusters (Tomato and Crimson) probably because of shading and canvas texture in the photo. On the canvas it reads as one red.