Phenological Scales: How to Record What You See So It Can Serve as a Model

The most widespread belief about crop models is that the hard part is the model. It is assumed that observing the plot is the easy part, something anyone does by looking, and that the complication begins when it is time to calculate. It is exactly the other way around. Summing degree days is elementary arithmetic and takes ten minutes with a notebook. What ruins your own series, season after season, is the record: stages described with words that each person understands differently, trees that change from one walk to the next, blank weeks with no explanation, and a notebook that by the end of the year does not allow you to say on what date the plot was in what state.

That has an uncomfortable consequence for anyone who wants to build their own thermal constant. A rough model calibrated with your own observations predicts better on your farm than a sophisticated model taken from the literature, because local calibration absorbs the error in the constant. But that advantage depends entirely on your own observations being comparable to one another, and an inconsistent record is not a rough model, it is noise with dates. Two years of a poorly kept notebook cannot be fixed later with statistics.

“It already bloomed” is not a record

It is worth seeing what that phrase is missing. It is missing which tree, because if this week’s route looked at some trees and next week’s route looks at others, the series mixes individuals and variation among trees gets confused with progress over time. It is missing the degree, because a plot with the first open buds and a plot in full anthesis are at different moments and both are described the same way. It is missing the exact date on which it was observed, not the date on which it was written down. And it is missing a stage definition that does not depend on who is looking.

That last point is what decides whether the series is useful. Two people walking the same plot on the same day can record different stages if what separates them is their criterion and not a code. The problem is not honesty or training, it is definition: as long as the state is described with adjectives, each observer will place the threshold in a different place, and the difference between observers will be added into the difference between dates. When you later calculate a constant with those data, the constant will carry the disagreement of your crew.

A usable record is therefore much shorter than it seems and much stricter. It needs an identified tree, an observation date, a stage code taken from a published scale, and the name of the observer. Four columns. What it does not need is free-form description, which is precisely what most notebooks have.

The BBCH scale, two digits that make what each person sees comparable

The tool that solves this has existed for decades and is free. The BBCH scale codes plant development stages with two digits, the first for ten main stages numbered from 0 to 9, from germination to senescence, and the second to subdivide each main stage into ten secondary stages, also from 0 to 9. The code does not describe, it defines. A number corresponds to a state with verifiable criteria, and that is why two observers trained on the same scale converge, while two observers describing with words do not.

For perennial tropical crops there is a variant that matters and that almost nobody uses. Species with rhythmic growth, which have more than one growth cycle per year, are coded with an extended three-digit BBCH scale, where the middle digit identifies the shoot or intermediate cycle to which the observation belongs. Keep that detail of the third digit in mind, because it is what decides whether your coffee or avocado series can be summed or whether you are summing what should not be summed. Coffee, avocado, and citrus bloom more than once a year, so the two-digit scale is enough to say what state a fruit is in, but not to say which flowering it came from.

The two scales that are directly useful to you are already published. For coffee there is an application of the extended BBCH scale that describes growth stages from germination and vegetative propagation, through leaf development, branch formation and elongation, inflorescence development, flowering, fruit development, fruit and seed ripening, all the way to senescence. For avocado there is another that details the stages from vegetative dormancy to commercial harvest maturity, and that was built explicitly to help standardize observations made in different environments or under different experimental conditions. As we already covered in our blog What Do Machines Really See When They Look at Your Crops?, which you can read here, a sensor delivers a signal but not a diagnosis; the scale is what turns what a human eye sees into data that another human, and later a model, can read without reinterpreting it.

How many trees, how many branches, and how often

The next practical question is how much effort this requires, and the published answer is more modest than people fear. In orchard phenology studies, a common protocol observes three trees per plot, characterizing on each one a branch facing north and one facing south, according to the BBCH scale. Three trees and two branches. Orientation is not an aesthetic detail: the exposed branch and the shaded branch advance at different rates, and averaging them is part of what makes the data represent the tree and not one side of the tree.

Intensity is concentrated where development is fast. The work that built the avocado BBCH scale marked 79 buds distributed across 15 branches and measured once a week during the first months and twice a week in the flowering window, over two consecutive seasons. That is the pattern worth copying: low frequency in long stages, high frequency in stages where the state changes in days. Recording every fifteen days during flowering is a guarantee that you will miss the event you wanted to date.

There is a design decision that matters more than frequency, and that is permanently marking the trees. A numbered tape on the trunk turns the series into a series of individuals followed over time, which makes it possible to calculate how much the same tree varies from one week to the next and how much trees differ from one another. Without marked trees you have a sequence of photos of different plots, and no constant comes out of that.

Knowing that everyone on the farm records the same stage

If tomorrow the foreman and the agronomist walk the same plot separately, will they record the same stage code on the same trees? On most farms the honest answer is that nobody knows, because there is no way to compare, and that invisible disagreement is what later makes two years of records useless for calibrating anything. Harvis helps you record the walk from the mobile phone with the crop scale already loaded, so the observer chooses a code instead of writing a phrase, each marked tree has its own history, the spread between observers and between trees can be seen before it contaminates the series, and the observed stage can be crossed with the plot’s thermal accumulation so the constant comes from your data. Write to us and we will set it up on the plots and routes you already use.

From notebook to constant: how your own base temperature emerges

With two aligned series, the observed stage and the daily temperature, it is possible to estimate the plot’s base temperature instead of copying it. The methods are established, and the most recent review on degree-day models from flowering to harvest names them precisely: the one with the lowest standard deviation in accumulated degree days, the one with the lowest coefficient of variation, and the regression coefficient method.

The logic of the first is the easiest to execute and the easiest to explain. You take several seasons or several plots in which you observed the same development interval, for example from flowering to physiological maturity. For a candidate base temperature, say 8 degrees, you calculate the accumulated degree days for that interval in each case and measure how dispersed those totals are from one another. You repeat with 9, with 10, with 11, with 12. The correct base is the one that leaves the totals most similar across cases, because if the constant is the true one, the same development interval should require the same accumulated heat regardless of the year or altitude. Minimum dispersion is the signal. The coefficient of variation method does the same thing by expressing dispersion as a proportion of the mean, which is preferable when comparing intervals of very different duration.

Those methods require repetitions, and that is where the only effort number that matters in this topic comes from. One flowering per year gives you one data point per year, so two years are two points and three years are three. If your crop has two usable flowerings per year, as happens in much of Colombian coffee growing, two years give you four points and the exercise becomes viable much sooner. Comparing several plots at different altitudes in the same season adds points without waiting, and that is the practical reason to record more than one plot from the beginning.

The review is explicit about a warning that is worth adopting before you begin. The estimated base temperature value varies considerably depending on the methodology used to determine the threshold and on the degree-day model chosen. There is no true base waiting to be discovered; there is a base that, with your method and your data, minimizes the dispersion of your observations. Several authors point out that these procedures have a statistical rather than physiological basis, and that the resulting parameter may not have direct biological interpretation. That does not invalidate it for prediction, it invalidates it for debate.

Why two years of notebook can perform worse than a published number

The first limit is the one we just named and it deserves to remain written as a usage warning. A base calibrated with few data and high dispersion can end up worse than the published value for the species. Local calibration wins when the series is consistent; when it is not, your record error matters more than the literature error, and at least the literature error is documented.

The second limit is the third digit we set aside above. In a crop with several flowerings per year, the stages observed in fruits from different cycles cannot be averaged or added in the same series, because they belong to clocks that started on different dates. A coffee plot with four flowerings has four fruit cohorts living together on the same branch, and a record that does not identify which one each observation belongs to will produce a development curve that corresponds to no real fruit.

The third limit is that the sampling method changes the result, not just its precision. Studies comparing phenological sampling methodologies have found that the indices obtained depend on the procedure used to collect the observations, and there is specific literature devoted to comparing methods for estimating seasonal development from BBCH-scale records. In farm terms, changing protocol halfway through the series, from three trees to ten, or from biweekly to weekly, breaks comparability just like changing observers. If you are going to change something, change it at the start of a season and write it down.

The fourth point is about transferring the figures. The three trees with two branches, the 79 buds on 15 branches, and the weekly or biweekly frequency correspond to concrete experimental designs, with their own objectives and crops. What transfers is the principle: marked and identified individuals, code instead of description, high frequency in fast stages, and a protocol that does not change within the series. Sample size transfers as a floor, not as a recipe, and if your plot is very heterogeneous, three trees will not represent it.

Three trees, two branches, and a two-digit code

The exercise does not require buying anything, and it is the only one in this series whose main input is discipline. Choose a plot and mark three representative trees with numbered tape, in the middle third of the plot and not on the edge. On each tree identify two branches, one facing north and one south, and mark them too.

Download the published scale for your crop and print a single sheet with the codes you will actually use, which are the ones for inflorescence development, flowering, fruit development, and ripening. Do not print the full scale. A sheet with eight or ten codes, each with its description in one line, is what fits in the pocket of the person doing the walk, and what does not fit is not used.

Do the walk once a week and twice a week when you see buds. On each visit record four things per branch: the date, the tree number, the stage code, and who observed. Nothing else. If a branch has fruits from two different flowerings, write two lines and indicate which cohort each one belongs to, because that is the detail that later saves the series.

Once a month, repeat the walk with two different people on the same day, without letting them consult each other, and compare the codes. The differences that appear are your observation error, and it is worth knowing before calculating any constant. If the two people disagree by more than one secondary code, the problem is in the printed sheet and not in the people.

At the end of the season, cross the stage column with the plot’s thermal accumulation and write a two-column table: stage code and accumulated degree days until it is reached. That table, with two seasons on top of it, is your model. It will not be elegant and it will have gaps, but it is made with your variety, your altitude, and your shade, and none of those three things appears in the published numbers.

What you do not record, you cannot predict

There is an asymmetry that explains why so few farms have their own series. The cost of recording is paid every week and the benefit arrives two years later, so the exercise competes every Monday against tasks that pay off that same day and loses. But the data that was not taken cannot be recovered: an unrecorded flowering does not come back, and the season that passed without a notebook is a season that will never enter your calibration. It is the only input on the farm that cannot be bought later, not even for more money.

What you can do starting Monday fits into five decisions. Mark three trees with numbered tape in your main plot and identify two branches on each one, because without fixed individuals there is no series, only a succession of looks. Print a sheet with the eight or ten codes from your crop scale that you will actually use, because a full scale does not fit into a walk and a short sheet does. Replace free-form description with the code on the form, because that is the only thing that makes two observers record the same thing. Increase the frequency to twice a week when buds appear, because that is the stage where the state changes in days and where the dates you wanted are lost. And once a month send two people on the same walk separately to measure how much they disagree, because that is your real margin of error and nobody else is going to calculate it for you. The producer who finishes two seasons with marked trees, codes, and dated stages will be able to estimate their own base temperature and predict with it; the one who keeps writing “it already bloomed” will keep borrowing a constant whose origin they do not know.

Do you want your farm walks to be recorded with the crop scale and crossed with the temperature of each plot, so the constant comes from your data and not from an article? Write to us on WhatsApp or schedule a demo and we will show you how Harvis turns a weekly walk into the series you need to calibrate.

References consulted

1. Meier, U. and collaborators. The BBCH system to coding the phenological growth stages of plants, history and publications. Journal für Kulturpflanzen. https://ojs.openagrar.de/index.php/Kulturpflanzenjournal/article/view/12142
2. Arcila-Pulgarín, J. and collaborators. Application of the extended BBCH scale for the description of the growth stages of coffee. Annals of Applied Biology. https://doi.org/10.1111/j.1744-7348.2002.tb00151.x
3. Alcaraz, M. L., Thorp, T. G. and Hormaza, J. I. (2013). Phenological growth stages of avocado (Persea americana) according to the BBCH scale. Scientia Horticulturae, 164, 434–439. https://doi.org/10.1016/j.scienta.2013.09.051
4. Phenological and Biophysical Mediterranean Orchard Assessment Using Ground-Based Methods and Sentinel 2 Data. (2024). Remote Sensing, MDPI, 16(18), 3393. https://doi.org/10.3390/rs16183393
5. Advances in Growing Degree Days Models for Flowering to Harvest, Optimizing Crop Management with Methods of Precision Horticulture, A Review. (2025). Horticulturae, MDPI, 11(12), 1415. https://doi.org/10.3390/horticulturae11121415
6. Effects of sampling methodology on phenology indices, insights from sites across India and modelling. (2025). Journal of Biosciences. https://doi.org/10.1007/s12038-025-00537-z
7. A comparison of methods to estimate seasonal phenological development from BBCH scale recording. International Journal of Biometeorology. https://doi.org/10.1007/s00484-011-0421-x
8. Da Silva Angelo, P. C., Ferreira, I. B., De Carvalho, C. H. S., Matiello, J. B. and Sera, G. H. (2019). Arabica Coffee Fruits Phenology Assessed through Degree Days, Precipitation, and Solar Radiation Exposure on a Daily Basis. International Journal of Biometeorology, 63, 831–843. https://doi.org/10.1007/s00484-019-01693-2
9. Bonhomme, R. (2000). Bases and limits to using degree.day units. European Journal of Agronomy, 13(1), 1–10. https://www.sciencedirect.com/science/article/pii/S1161030100000587
10. Development of a precise thermal time model for grain filling in barley, a critical assessment of base temperature estimation methods from field-collected data. Field Crops Research. https://www.sciencedirect.com/science/article/abs/pii/S0378429020312879
11. Flowering and Fruiting of Coffea arabica L., A Comprehensive Perspective from Phenology. (2025). Plants, MDPI, 14(21), 3396. https://doi.org/10.3390/plants14213396
12. Quantification and Qualification of Floral Patterns of Coffea arabica L. in Colombia. (2023). Plants, MDPI, 12(18), 3332. https://doi.org/10.3390/plants12183332
13. Unigarro, C. A. and collaborators. Harvest-Dependent Variability in the Ripening and Fruit Characteristics of the Cenicafé 1 Variety of Coffee. Crops, MDPI, 6(4), 75. https://doi.org/10.3390/crops6040075
14. Registro de la floración del café. Avances Técnicos Cenicafé, 495. https://biblioteca.cenicafe.org/bitstream/10778/1105/1/AT0495.pdf
15. La fenología del café, una herramienta para apoyar la toma de decisiones. Avances Técnicos Cenicafé, 441. https://cenicafe.org/es/publications/avt0441.pdf
16. Development of a uniform phenology scale (BBCH) in hazelnuts. Scientia Horticulturae. https://www.sciencedirect.com/science/article/pii/S0304423821009444
17. University of California Statewide IPM Program. Degree-Days, how to calculate and use degree-day models. https://ipm.ucanr.edu/WEATHER/ddconcepts.html