Rwanda Gatare (Nordic approach)


One of the coffees I have at home and will be roasting shortly (its a different lot to the one I had already tried, hopefully a more simple to solve one :)) so … if you find the “over my shoulder” approach that I am trying here interesting, I will be posting as much as I can about the process here …


Cool, looking forward to it! I’ve had lots of problems with underdevelopment in my Rwandan Kivu from Sweet Maria’s, and was kind of pleased to taste the same metallic, astringent, green flavor from a new local specialty roaster’s Rwandan😏. It’s hard stuff, literally! I’ve seen the sweet Maria’s bloggers use it as an example of a bean that takes a lot of heat, and from what I’ve seen it really does. The Ikawa espresso roasts are what got me in the right neighborhood. I just tried a slightly heightened and lengthened version of ikawa espresso roast 1 and that seems like the best yet in color and aroma, but I’m still not sure I’ve purged the metallic flavor. We’ll see. Happy roasting!


Ok everyone - Here is my first test of recording a colour development of the roast.
If anyone is interested I can give detailed instructions, but I did it in a professional postproduction app, so it might take some time to recreate in another cheaper software.
@Geoff_IKAWA - I Think, this could be implemented in the Ikawa app actually, together with some holder to have good height, and some passive HW things to help along, this could be recorded in realtime (I think the modern smartphones are strong enough to average the data and plot the colored line, recording a small square video along with it for reference. Also its very probably possible to set all exposition parameters manually, and use a led light used as a flash on them (I did) … which is not bad … so I could be quite neutral actually … -small grey square can be used to neutralize even better :))

The time of the video, colour square under the video, and the roast profile and charted colours is not well aligned, sorry for that, will have to think a bit more about how I put that together to match perfectly … just a test now.

(btw - found out I can not upload video … so … uploading to my Youtube and will add the link soon)

Image if roasted beans with some adata here

and here is a first version of the video … I will reupload once fixed all issues.

How many of us here interested in these graphs do have smartphone with a led light capable of capturing the colours? :slight_smile:


Hey really nice post! But I don’t think your are extending your development, rather, it is going to fast and the “chameleon’s eyes” aren’t lining up. Maybe this will help.


Sure … I did not actually try to extend development … just tried to record colour changes. I think this might be quite an objective measure of the different changes in the profile … if I create such plots and line them side by side, it should be clearly visible which part was longer or shorter. I will try to look at the plot with different color analysis viewing tools we use for colour grading … there may be even better more precise way how to show the changes, because the second half where it is brown is quite hard to see a change in there.


My short term plan now is, to choose one relatively simple profile (non spike one) and with this tool at hand try to keep all the rest of the profile be the same, but extend different parts to see how that changes the tastes - like extending only the yellow, or only the caramel …


Amazing, very cool @pavel! Looking forward to your next videos.
(also, awesome coffee, one of my favourites :slight_smile: )


Will test this on another batch of this Rwanda. She is HAAAAARD … so far I dont think I get even close to her potential.


Thank you @Geoff_IKAWA - I think it would be perfect if more of us can produce similar plots. Not sure if it will be easilly doable with some free software though. The capturing part is just making use of the smartphones good sensor and quite good led light, and running it in full manual so that it does not compensate and shift whiteballance or exposure. The averaging part is so far just a blur on a cropped part (I have some plans how to exclude shadows and chaff from averaging, but that will be quite a bit more complicated :)) ) I then made a sample in a form of small band every 1s and had the software plot it horizontaly to get the colour changes.
I think if built into the app, it should be able to do the - collect 1s worth of frames, average them, blur the outcome and sample colour - on the fly, and just plot it as a data into the graph … making it much less data than 10 minutes of HD video I recorded :smiley:


Hmmm … yeah … intensity waveform, that is the correct way how to aid readability :smiley:


sounds out of my wavelength, but any help I can give creating a study etc etc lemme know.


Well … simply put … intensity histogram is a plot that has on a horizontal axis intensity (“lightness”) and on vertical axis amount of pixels of that level … thats a traditional histogram seen on many image editing apps. intensity waveform puts it in another way, horizontal axis is a horizontal dimension of the picture, vertical is intensity at that collumn of pixels (or more often the pixels with their colour are plotted on a place on vertical axis that fits their intensity – its quite an interesting view one can see in profesional colour grading suites) … but, what that means for me is, that I will plot the colours horizontally, but not cut them in a rectangular way as before, but cut the top part depending in the intensity at that point, which will make it a coloured band with a curve at the top trimming it. Once I use median to get rid of the noise it makes a very nice curve :smiley:


Ok … here it cometh:

Latest edition of the colour plot, with above mentioned lightness as a top cutting curve.
What do you think?

edit: there is a little wave in the beginning and curved end … dont take those into consideration, as they are produced by median smoothing the values, and forgetting that I should not leave the ends on black - so there they smoothed in a non ideal way :smiley:

Tanzania Iyela NA

So if I’m understanding this graph, so as the roast progresses, it peaks at bean color eveness at gold and then seperates as the roast progresses with the end of the roast the most uneven in color. This would make sense for 99% of the roasts I’ve done…


Nonoo … the y axis is lightness … the lower the column the darker.
The green beans are darker, as they dry and go towards yellow they get lighter for a moment, and from there darker they go …


oh I see, can you change the y-axis to time (duration) of bean color? I think that would help create a link to color and its duration in the roast. Maybe that could help link the time of a bean color and the significance to taste and things like average color at 1C aka pyrolysis.


Since we can then see that color duration, the color of the bean can also be correlated (roughly to the temperature of the bean) thus we can see the bean temperature curve.

In order to clearly define color to temp, you could do a roast for 12 minutes at incremental points set to one specific temperature, with a batch size of an average weight and volume. Reaching one specific temperature by accelerating in the first minute and then holding that temperature for 11 minutes.

Since that would require. 477 roasts for each degree (150-527F), every 10 would be about 48 roasts. me thinks help would be required…


Not sure if I understood you correctly …

as it is, the current plot has horizontal axis being time (aligned with the roast profile, so it should match quite well with the roast profile plot) … so the colours are placed horizontaly in time into the roast profile curves time.
the vertical (y) axis visualizes the lightness of the colour (because I have noticed that once the colours go brown its not very visible how much change is there, so I tried to make it more visible with colour and lightness in the same plot)
So you can see the colour duration as it is … at least I think so :slight_smile:

The idea of establishing the colour-temperature correlation is great … but wouldnt the colour change in time even at constant temperature? I am not sure exactly … but I guess once some of the reactions start to happen, they will only happen faster or slower. So there might be no exact temperature to colour, but maybe there could be visible at what temperatures some reactions begin, helping define the stages. So it is definitely something to try.

Btw … do you have an iphone? If you do I think we may be able to match those pretty nicely, so some bigger experiments could be spread wider …
(I am now thinking of not using a video, but maybe a photo every 10s, taken raw … because its necessary also to eliminate the postprocessing that happens to the data on the phone … and also, the shape of the curve as it darkens is not correct now, because the data used for it are gamma corrected, so under quite a curve themselves … so the absolute shape is not correct)


Ok now that I’ve had my cuppa and I’m fully awake…

Forgive me for my presumptions: I think the color difference problem is tied directly to the lighting would of course have to be full spectrum (but I think that is an easy $20-40(bulb and lamp) fix.

Taking pictures and analyzing them under a “photoshop” like program, though I think more like every 2-3 sec rather than 10s. where the change would be massive and not gradual…(?)

after many many many roasts that never got darker, I can say yes. If you roast a bean at 300F, sure it goes through changes but…the color it ends at will not change, And it will be consistent for a LONG time (minutes). I do believe that there will be gray areas where a color reading will look the same, but I can’t say for certain. And I have no doubt that a steady 150F will be pale greenish/yellow throughout (untested).

(as i stroke my non-existent beard) yes, yes… indubitably

I think an I phone would be a great tool to use, yes I have one but an old iPhone 5. Photo would be the way to go…i think too.


Ah, sure I ment one picture every 1s not 10 … that would be too large step.

Not sure about the led in iphone5 they changed it at one point to have better colours of skin when using it, so it should be more spread around the spectrum (cheap leds are very much narrow peaks with big holes between them, and usually very much of blue peak). But otherwise should be fairly close when taking pickers in raw dng, I think if we find a way how to standardize the procedure and calibrate the colours and ranges (which is quite possible though not absolutely precise, but precise enough will do ). I have some ideas that I have to test first, though … and make sure I am using the iphones camera and light to its full potential.