LED Spectrum Analysis: Digitize an SPD Chart from a Datasheet

When designing lighting systems — especially LED grow lights or LED lamps for domestic and industrial lighting — engineers and DIY enthusiasts often run into the same problem: component manufacturers publish spectral charts in their datasheets with the whole family of white LEDs on a single plot, for example from 2700 K to 6500 K.

Spectrum digitization is the process of converting the graphical image of the spectrum of an LED or other light source into a digital data array (a set of "wavelength–intensity" values) for subsequent processing.

Saved digitized spectrograms of light sources in .json or .csv format can subsequently be loaded into the tool for simulating the spectrum of LEDs and other light sources^↗, which allows you to obtain the resulting spectrum when mixing different LEDs (and more), and evaluate the final CCT, CRI, and CQS. The tool also calculates the total PPE and includes many other useful metrics that are of paramount importance in the development of various lighting fixtures.

Below is an example of the spectral charts for the Bridgelux Thrive series, high colour rendering LEDs with a linear spectrum (a competitor to SunLike LEDs). A screenshot of part of page 6 of the official document:

It is not unusual to need a spectral analysis of one specific LED from the family, or even to compare the spectra of two different LEDs from different manufacturers. That is exactly what our tool is for. If you need the spectrum of a single LED, you no longer have to strain your eyes or draw lines from the peaks down to the wavelength axis X.

Our tool also lets you derive the parameters of an LED or a lamp from its spectrum when you have the spectrum of the light source but none of the other figures, for example the colour rendering index by CQS or TM-30.

The spectrum digitizing tool performs a basic LED spectral analysis, particularly with regard to how efficiently a given white LED can be used for plant lighting.

You can now take any PDF from a manufacturer, screenshot the spectral chart and within a couple of minutes get a finished report with the spectral fractions across the ePAR range, an estimate of the 650–670 nm band and a measurement of the 660 nm peak. CCT, CQS and TM-30 are calculated as well.

The tool also lets you download the numerical array (the spectrum) for further engineering work.

The tool we built estimates the spectrum of a light source with reasonably high accuracy, though not laboratory accuracy. In most cases that is enough for engineering assessment and analysis. To obtain the full set of light source parameters with laboratory accuracy, a calibrated integrating sphere^↗ is used.

How to use the spectrum digitizing tool: step-by-step instructions

The interface of our widget is designed so that digitizing takes as little time as possible.

To guard against DDoS attacks, the tool is available to signed-in users only. Please register or sign in with your Google account.

For the same reason there is a 15-second delay on the digitize and auto-calibrate commands.

For an accurate result, follow these instructions:

Step 1. Screenshot the spectral chart from the datasheet

Download the datasheet for the LED or lamp you need and find the page with the spectral chart. Take a sharp screenshot of it. Below is a screenshot of the spectral chart for the SunLike series from Seoul Semiconductor at colour temperatures of 3000, 4000 and 5000 K.

Step 2. Upload the screenshot into the tool

Select the screenshot of the spectral chart and upload it into our tool:

Step 3. Calibrate the chart axes

After step two the tool displays the uploaded chart inside its working area.

To digitize the uploaded chart and obtain accurate figures, the starting coordinates of the chart on the X and Y scales have to be set as precisely as possible. Our tool has an automated calibration function: to start, simply press the “Auto-calibrate” button located to the right above the uploaded chart.

Once auto-calibration finishes, coloured markers appear at the start and the end of the X and Y axes:

Make sure the zero-point markers line up exactly with the origin of the chart, and that the maximum points sit strictly at the end of the axis scale.

Make sure the Max Y field contains a number matching the Max Y marker: the marker equals the highest peak of the chart. Usually this is 1 or 100 %.

When you need to compare charts taken from two different images, bring the value to a single format manually — it can be either 1 or 100.

The zero point of the chart must be placed correctly at the origin. The Max wavelength (X) field must match the scale of the chart: incorrect values will shift the digitized curve.

Below is a screenshot showing correct calibration:

If auto-calibration fails, or if you need to set a different maximum on the Y axis, press the “Reset calibration” button and place the markers by hand: manual calibration mode starts automatically as soon as “Reset calibration” is pressed. Hovering the cursor over the chart activates a magnifier for precise marker placement.

The order for placing the calibration markers manually is: zero point, Max X, Max Y.

Step 4. Pick the curve colour with the eyedropper

Press the standard “Eyedropper” tool to select the colour of the curve you want to digitize. Once the “Eyedropper” is active, a zoom view appears to make it easier to hit the thin line of the curve.

The “Colour tolerance 10–250” field controls the spread around the selected shade. If digitizing errors occur, you can tighten the value down to 10–20, or conversely allow computer vision a wider spread around the shades of the selected curve by raising the value to 75–100 or more.

Step 5. Digitize the chart

Once the colour has been picked with the “Eyedropper”, the “Digitize chart” button becomes active automatically.

To get the digitized curve, press “Digitize chart”.

As soon as digitizing starts you immediately get the digitized curve in exactly the same colour as in the original screenshot.

A report with the basic calculated data appears directly below the digitized spectrum:

The screenshot shows the result of the spectral analysis for the SunLike 5000 K LED.

Step 6. Export the spectral data

After digitizing, the tool lets you download the spectral data array in JSON or CSV format, and save a screenshot of the chart as a PNG, either as an outline or with colour fill.

Step 7. Compare spectra from two datasheets

What if you need to compare spectra from two different datasheets?

1. Without reloading the page, press “Select chart” and choose the screenshot of the second spectral chart.
2. Repeat steps 3–5 of the instructions. It is important to set the same Y-axis maximum for both charts.
3. Both spectra are now displayed on a single plot for convenient comparison.

The boxes above the plot show the colour of each spectrum and the default label “Spectrum 1”; for convenience you can give the spectra your own names. In this example the blue spectrum belongs to the SunLike 5000 K LED and the yellow one to the Thrive 5000 K LED.

Step 8. Additional spectrum analysis: comparison with the McCree curve and the solar spectrum

To maximize the convenience of spectrum analysis, I have added the ability to display the McCree curve as a green dashed line and the solar spectrum as a yellow dashed line on the graph. 

These two options allow you to evaluate LEDs in terms of their application for plants, as well as visually compare the spectrum of a Sunlike or Thrive LED directly on the graph, free from any marketing influence.

Our tool allows you to digitize the spectrum of an LED grow light or other light source from an image and perform a basic analysis. This includes not only dividing the spectrum into separate bands and evaluating the color rendering index, but also provides the ability to compare the LED spectrum with the actual spectrum of sunlight.

Technical note

Up to ten curves can be digitized in one session for comparison. The tool has no other general limits on digitizing.

Digitizing runs on computer vision built from two Python libraries: OpenCV handles the vision side and NumPy is the mathematical core. The tool has a built-in filter that strips the chart of its own gridlines and labels, which makes it possible to digitize black spectral curves.

IMPORTANT! Our spectrum digitizing tool is not a precision instrument and is intended for general spectral analysis. If you need data with laboratory accuracy, use professional laboratory equipment.

Digitizing accuracy: compared with the datasheet figures

Below is a comparison of the values obtained by digitizing the spectrum against the figures from the official documentation for the SunLike 5000 K LED:

Where the digitizing error comes from

There are three sources of error, and none of them has anything to do with the tool itself.

The first is the quality of the source image. A spectral chart inside a PDF is almost always lossily compressed, and taking a screenshot adds rasterization on top: where the manufacturer drew a single line, the file contains a band several pixels wide with blurred edges. Computer vision takes the centre of that band, but on the steep flanks of a peak a shift of even two pixels vertically turns into a noticeable difference in intensity.

The second source is axis calibration. If the zero-point marker misses the origin by even one pixel, the entire wavelength scale shifts as a whole, and a 450 nm peak becomes, say, 452 nm. Calibration therefore deserves more attention than the rest of the process: the tool does every other step by itself, but the accuracy of the reference points is down to you.

The third source is the most important one, and it lies outside digitizing altogether. A datasheet shows a typical spectrum, not a measured sample. The manufacturer publishes an averaged curve for the family, whereas real LEDs are spread across chromaticity bins: two diodes from the same batch carrying the same 5000 K marking can differ in chromaticity coordinates within a MacAdam ellipse, and their spectra will not coincide. How far apart depends on the bin, which is covered in detail in the article on the MacAdam ellipse and SDCM^↗. The practical conclusion follows: a digitized spectrum is valid for comparing LEDs against each other, for estimating band fractions and for design estimates, but it does not replace a measurement of the actual sample. If you need the parameters of your own batch, the spectrum is measured with a spectrophotometer and the total luminous flux with a calibrated integrating sphere.

LED spectral power distribution (SPD): what it is and why it matters

LED SPD (Spectral Power Distribution) — is a fundamental characteristic of any light source. It shows the optical radiant power as a function of wavelength across the visible range.

With a digitized LED spectrum in hand, you can analyse specific bands. Our widget has a “Spectral range analysis” block with convenient sliders (from 350 to 800 nm). You can instantly read off the peak of the spectrum (472.6 nm, for example).

The spectrum digitizing tool is well suited to assessing white LED spectra from the point of view of their use in LED grow lights. When designing plant lighting it is important to understand how the power is distributed across the ePAR range.

Spectral fractions across the ePAR range: what the percentages mean

The digitizing tool automatically splits the spectrum into five bands and reports the share of photon flux in each one, not the share of power. That distinction matters: a photon at 660 nm carries less energy than one at 450 nm, so in energy terms the red part always looks smaller than it does in photon terms — and a plant counts photons.

Classic PAR under ANSI/ASABE S640 is the 400–700 nm band. The extended ePAR range adds far-red, 700–750 nm, and starts at 380 nm. The split in the report follows that logic:

• 380–400 nm — the edge of UV-A. Outside PAR, contributes almost nothing to photosynthesis, but affects secondary metabolite synthesis and plant compactness.
• 400–500 nm — blue. Governs compactness, leaf thickness and stomatal behaviour.
• 500–600 nm — green and yellow. In white LEDs this is always the largest share, because this is where the phosphor emits. Green photons penetrate deeper into the leaf and into the canopy than red or blue ones, and work where red no longer reaches.
• 600–700 nm — red. The main driver of photosynthesis and the cheapest photons per joule.
• 700–750 nm — far-red. Outside PPF, counted separately, and it drives stem elongation, leaf size and flowering through phytochrome. This is covered in detail in the article on far-red 730 nm in grow lights^↗.

For the digitized spectrum of a white SunLike LED the report gave this picture: 0.1 % in UV-A, 23.5 % in blue, 37.0 % in green, 35.1 % in red and 4.5 % in far-red.

Keep in mind that the extended ePAR range is not universally accepted: the DesignLights Consortium still keeps only 400–700 nm in its efficacy calculations. Fractions across ePAR are therefore convenient for analysing the shape of a spectrum, but when comparing fixtures by their datasheet figures, check which range the manufacturer had in mind.

The McCree curve and the solar spectrum: what to compare a digitized spectrum against

The band fractions on their own do not yet tell you whether a spectrum is good or bad. A baseline for comparison is needed, and in practice there are two of them.

The McCree curve (McCree, 1972) shows the relative quantum yield of photosynthesis: how efficiently a photon of each wavelength is converted into assimilated CO₂. The curve was obtained by averaging across 22 crop species and has two maxima — near 440 and 620 nm — with the blue peak reaching roughly 70 % of the height of the red one. Between them, in the green region, lies a shallow trough: a green photon does work, but less well than a red one.

The practical value of the curve is that it lets you weight a digitized spectrum and obtain not merely a number of photons, but their usefulness. It comes with hard limits of applicability, however, and they are usually forgotten. The measurements were made on a single leaf rather than a whole plant, at a low light level of roughly 15–150 µmol·m⁻²·s⁻¹, and in monochromatic light. In a dense canopy green light penetrates deeper than red and performs noticeably better than the curve predicts, while the Emerson effect — photons of different wavelengths working together — is not accounted for at all. And above all: the McCree curve describes instantaneous photosynthesis, not yield, plant shape or flowering time.

The solar spectrum serves as the second reference point, and a far more intuitive one. For the AM1.5G reference spectrum under ASTM G173-03, the photon flux fractions in the same five bands come out as follows: 1.9 % in UV-A, 22.2 % in blue, 29.1 % in green, 31.6 % in red and 15.2 % in far-red. Midday sun delivers about 1980 µmol·m⁻²·s⁻¹ across the PAR range at 430 W/m² in that same band.

Comparing a white LED with the sun shows two things at once. The blue and green fractions of a good white LED are close to the solar ones — which is precisely why LEDs with a linear spectrum such as SunLike and Thrive look natural to the eye. Far-red, on the other hand, is three times scarcer in the LED: 4.5 % against 15.2 % for the sun. Under open sky the ratio of far-red to PAR is about 18 %, and a plant under a white LED with no 730 nm added receives a completely different phytochrome signal than it would in the field.

To compare shapes rather than fractions, it is convenient to take the normalized solar spectrum or the D65 standard illuminant and overlay it on the digitized curve as a second plot. The dip in the cyan-green region around 480 nm, characteristic of phosphor-converted white LEDs, becomes visible immediately.

LED colour rendering and modern metrics: TM-30 and CQS

When designing domestic or industrial lighting, the colour rendering index matters just as much as CCT. Manufacturers sometimes quote colour rendering under the older CRI standard, which is a poor fit for judging the light quality of modern LED devices. If you have a spectral chart, you can digitize it and get a report and a plot for the modern TM-30 standard and for CQS. Read more about the CRI and CQS colour rendering indices and the TM-30 standard^↗.

Using the resulting spectral data array, the algorithm automatically computes the basic photometric parameters and the modern colorimetric metrics, including the distance from the Planckian locus (Duv)^↗ but with a significant margin of error.

The spectrum digitizing tool will be especially useful to DIY enthusiasts, students and designers of LED luminaires and grow lights.

The spectral analysis parameters the digitizing tool provides

Below are the main metrics and data you receive automatically once the curve has been digitized:

Conclusion

Our digitizer turns imprecise marketing pictures into numbers that can be processed mathematically. A thorough LED spectral analysis no longer requires complex software or transferring points from a chart by hand. The tool saves developers time, letting them compare the real characteristics of LEDs quickly at the luminaire design stage and pick the LEDs with the right spectral distribution for the job.

I would add that the main advantage of our tool is how simple it is to use, together with the built-in mathematical core that computes the basic parameters. Plus a basic spectral analysis aimed at using LEDs in horticulture.