First post in hopefully a series of entries where I’m planning to share various OSM-related experiments I have conducted over the years.

Mapper asked if area should be mapped as grassland, scrub or heath. Since this kind of information is better to be extracted from infrared imagery, without knowing where the mapper is from, I pointed them to use global satellite dataset provided by European Space Agency (OSM wiki link) at Copernicus browser.

Two most commonly user IR imageries are using CIR-NRG and CIR-NGR styles. These acronyms essentially mean that when compared to regular RGB (red, green, blue) pictures, infrared images drop blue signal channel and instead use IR as red, and then original red and green as green and blue. For NRG, red becomes green and green turns blue; NGR is vice versa with green staying green and reds are blue.

Turned out that while Copernicus does have multiple infrared imagery layers (such as one simply called False Color is NGR), but because infrared channel is relatively overexposed compared to visible light, then default configuration showed nothing but red (IR) patches on black earth. Solution was building a custom rendering with linear adjustments.
In hindsight, Copernicus browser has button called “Effects and advanced options applied” where one could apply most of those transformations directly without custom layers.

Anyways, custom layer is basically javascript code that should execute on client’s browser for every single pixel on image and calculate RGB values for each. Here’s the tutorial.
I haven’t figured out most feasible way to add pictures to diary posts yet.

• Go to https://browser.dataspace.copernicus.eu/
• Click on green diagonal arrow (↗️) on sidepanel to see latest images
• Select Layers -> Custom (at bottom of list) -> Custom script (rightmost tab)
• Paste script and click apply below textbox.
• Colour balance and lightness can be tweaked using sliders under button “Effects and advanced options applied”

Credit for writing the function goes to the Large Language Model.

//VERSION=3
// Forestry false-color (NGR -> RGB) with percentile-based piecewise stretch
// ---- Paste percentile values from histograms here ----
const P = {
  B08: { p10: 0.03, p50: 0.30, p90: 0.64 }, // NIR
  B03: { p10: 0.02, p50: 0.063, p90: 0.15 }, // Green
  B04: { p10: 0.01, p50: 0.041, p90: 0.12 }  // Red
};

function setup() {
  return {
input: [{ bands: ["B08","B03","B04","dataMask"], units: "REFLECTANCE" }],
 output: [
  { id: "default", bands: 4 }, // RGB + alpha
  { id: "index",   bands: 1, sampleType: "FLOAT32" }
]};}

/**
 * Percentile stretch maps v so p10->0, p50->0.5, p90->1.
 */
let p = (v, p10, p50, p90) =>
  v <= p10 ? 0 :
  v >= p90 ? 1 :
  v <= p50 ? 0.5 * (v - p10) / (p50 - p10)
           : 0.5 + 0.5 * (v - p50) / (p90 - p50);

function evaluatePixel(s) {
  // NGR -> RGB mapping for forestry false color:
  // R <- NIR (B08), G <- Green (B03), B <- Red (B04)
  let r = p(s.B08, P.B08.p10, P.B08.p50, P.B08.p90);
  let g = p(s.B03, P.B03.p10, P.B03.p50, P.B03.p90);
  let b = p(s.B04, P.B04.p10, P.B04.p50, P.B04.p90);

  return {
 default: [r, g, b, s.dataMask],
 // Index is used for building historgram - helps to guess 10-50-90 cutoffs
 index:[
 //Math.min(70,Math.floor(s.B08*100)),
 //Math.min(16,Math.floor(s.B03*100)),
 //Math.min(16,Math.floor(s.B04*100)),
 r,g,b // Sanity check on output
 // Histogram does support multi-element array, but they are later simply concated to single array
]};}

The function will turn three input channels into adjusted RGB channels. Each channels is linearly adjusted so that value p10 becomes 0.0, p50 0.5 and p90 turn to 1.0. pX stands for percentile. There’s also 4th output series called index, which was used for manually determining percentile values as application supports calculating histogram for just single series at once.

Originally published on OSM World discord on 2025-09-14