How many trees did you actually check before authorizing the last application you made, and what concrete number led you to decide that yes? If the honest answer is that it was the calendar, that it was last week's rain, or that you saw three damaged trees on the way home, you are in the same situation as the vast majority of farms in the region, and you are paying for it in two ways at once, with product applied that was unnecessary and with applications that arrived late where they were needed.
The season we closed left us with the tools to see an entire field, prioritize where to walk, and differentiate management by zones. However, seeing is not deciding. The decision to intervene against a pest has its own logic, supported by a number compared to another number, and that logic is the subject of this new season, which will cover specific pests of coffee, Hass avocado, and citrus, their prediction models, and the biological control tools that work today.
The economic threshold, the calculation that decides if it's worth applying
The idea is old and remains the most profitable in pest management. In 1959, a group of entomologists formalized two concepts that today support all integrated management. The economic injury level is the lowest pest density that causes damage whose value equals the cost of controlling it, and the economic threshold is the density, always lower, at which control should be initiated so that the population does not reach that level of damage. The distance between the two is deliberate; it exists to give you the time needed for a work order, a purchase, and an application to take effect.
What is almost never explained is that the threshold is not a property of the insect; it is a property of your business in this campaign. It depends on the price you will receive for your product, the cost of the control you will use, the expected efficacy of that control, and the potential yield of the field. When the price of your product rises, the threshold lowers, because each lost fruit is worth more, and it justifies intervening earlier; when the cost of the input rises, the threshold rises, because more avoided damage is needed to pay for the application. A threshold copied from a pamphlet from ten years ago, or from a farm with a different cost structure, is miscalibrated by definition.
In farm terms, this has an immediate consequence. If you have never done that calculation, you have no threshold, and without a threshold, sampling is useless even if you do it well, because there is nothing to compare the result against. That is why the correct order is to first set the number that triggers action and only then go out to count.
There are institutional references that save the initial work. For coffee borer, coffee research has established an economic injury threshold of around five percent infestation, with a standardized sampling method that evaluates thirty trees per hectare. That number is a sensible starting point, not a universal constant, and it is advisable to adjust it with your own price and your own cost.
Counting seriously, how many points, how often, and why the usual route doesn't work
The sampling done by most farms has a bias that invalidates it without anyone noticing; it is done by walking where one always walks. The route follows the main path, stops where access is comfortable, and ends near the house, so the sample represents the most trafficked part of the field and not the field itself. Any focus that establishes itself in the back, on the steep slope, or behind the live barrier will appear in the record when it is already large.
A sampling that works needs three things: a defined unit, a distribution that covers the field, and a fixed frequency. The unit is what you count exactly, and it must always be the same: the same branch in the same position of the canopy, the same number of fruits per tree, the same height of the trap. The distribution means that the points are spread throughout the field with a criterion, in a zigzag or grid pattern, and that they are defined before going out and not on the fly. The frequency means that it is measured with the same periodicity whether or not there is a visible problem, because the value of a pest data point lies in the series and not in the isolated measurement.
The practical objection is always the same: that it takes time. Applied entomology solved that objection decades ago with sequential sampling plans, where a number of samples is not fixed in advance but is counted, and it stops as soon as the accumulated data allows for a confident decision on whether one is above or below the threshold. When the population is very high or very low, the decision appears with few samples, and the route is shortened; only when the result is close to the threshold is it necessary to keep counting. Studies comparing both approaches report reductions in sampling time of nearly eighty percent compared to conventional fixed-size sampling, and binomial plans that resolve the decision with around twenty-seven to thirty-one trees.
There is also a simplification that almost always survives the field test: counting presence and absence instead of counting individuals. Recording whether a tree has or does not have the pest is much faster, much less dependent on the operator's skill, and, for the purposes of crossing a threshold, is usually sufficient. The practical takeaway is that the best sampling system is not the most refined; it is the one your team will repeat every week without anyone having to insist.
The pest runs on a different clock, degree days, and why the insect doesn't follow your calendar
An insect does not age with the days of the calendar; it ages with accumulated heat. Below a certain temperature, its development stops; above it, it accelerates, and the time it takes to complete a cycle depends on how much heat it has accumulated, not on how many times the sun has shone. This leads to the concept of degree days, which consists of accumulating each day the difference between the average temperature and that lower development threshold.
The coffee borer is the best-documented case; laboratory studies place its lower development threshold between 13.9 and 14.9 degrees Celsius and the upper around 32, with a requirement of about 299 heat units to complete the transition from egg to adult. That alone explains something that every coffee grower observes without naming it: that the pest behaves differently depending on the altitude of the field, because at higher altitudes there is less accumulated heat per day, and the cycle stretches.
The valuable part came when someone took that laboratory model to real commercial farms. A study that followed natural populations along an altitudinal gradient found an average development time from egg to adult of 38.5 days with a deviation of 3.46 days and estimated an average requirement of 332 degree days using one threshold and 386 using another, from the infestation of the fruit to the start of a new reproductive cycle. However, the finding with the most operational consequence was another. Estimates obtained with broad-scale temperature data were equivalent to those obtained with sensors placed within the crop, indicating that regional temperature monitoring is sufficient to predict the cycle. In farm terms, this means you do not need to buy a station for each field to start using degree days; a reliable climate source and a spreadsheet are enough. Keep that detail in mind, because in the next section we will see that this same broad scale, which works well here, fails when what is wanted to predict is a disease.
There are even developments that calculate thermal accumulation for this pest from satellite images, directly linking last season of this series with the current one. The concrete and significant utility is simple: if you know how many degree days have passed since flowering, you know at what stage of the cycle the population is and when it makes sense to sample more intensively, instead of discovering it when the damage is already on the fruit.
A pest record that survives the change of manager
Where does the historical pest record of your farm live today, and what happens to it when the person who manages it changes? In most operations, that knowledge is oral, it is lost with personnel rotation, and makes it impossible to answer the only question that matters at the end of the campaign: whether what you applied worked. Harvis helps you set up the monitoring record by field and by date, to compare each reading against the threshold you defined, to keep the historical record of applications with product, dosage, and responsible party, and to have the complete series available when it comes time to decide next year. Write to us, and we will set up that record with the structure of fields you already manage.
Climate risk models, what they predict well and what they don't
Climate variable prediction models are the most marketed promise of the moment and also the most misunderstood, so it is advisable to be precise about what scale they resolve.
For foliar diseases, the available evidence is solid at the regional and retrospective level. A coffee rust risk model built with climate reanalysis data, which estimated leaf wetness duration and canopy temperature, found no general trend of increased risk between 1990 and 2015 in the coffee-growing areas of the country, thus ruling out the hypothesis that the outbreak from 2008 to 2011 could be explained by climate change, although it did identify a significant elevation in the predicted risk of infection right in those years. It is a valuable result, and it is important to understand what type of result it is: an explanation at the regional and decade scale, not an alert for your field next week.
Distribution models built with machine learning aim at an even broader scale. A recent study modeled the spatiotemporal distribution of the coffee rust fungus by combining eight different algorithms and projected risk maps of re-emergence towards 2060 and 2100 under different climate scenarios. That type of study guides public policy, varietal renewal decisions, and long-term planning, and does not answer the question of whether you should apply this month.
What is getting closer to the farm scale are models fed with real-time climate variability, which already report the ability to anticipate disease severity with updated meteorological data, and epidemiological models adjusted to a specific phase of the crop, such as those developed for nurseries, where conditions are more homogeneous and therefore more predictable. The rule that emerges is quite clear: the more the environment where the model was adjusted resembles the environment where you are going to use it, the more confidence it deserves, and a model fed with a climate grid of dozens of kilometers cannot know that your field is in a hollow where the dew lasts three hours longer.
That is why the combination that works is model plus sampling. The model tells you when to increase vigilance, and the sampling tells you if action is needed. Using the model as a trigger for applications, skipping the counting, is repeating the calendar error with a layer of technology on top.
Applying without a threshold costs you twice
The visible cost of an unnecessary application is the product and the labor, and that is the lesser of the two.
The invisible cost has two components that accumulate over the years. The first is the selection pressure on the pest population, because each application leaves the less susceptible individuals alive and, repeated without criteria and with the same chemical family, builds resistance in your own farm. The second is the elimination of natural enemies that were working for free, which tend to be more sensitive to broad-spectrum products than the target pest, and whose disappearance produces the population rebound that is later interpreted as proof that more applications were needed.
As we already addressed in our blog Pesticides, modes of action, and resistance management, rotation between modes of action and reducing the number of applications are the two levers that preserve the useful life of available molecules; the threshold is precisely the instrument that reduces that number without increasing risk, because it eliminates unnecessary applications and leaves intact those that are needed.
A field, four weeks, and a spreadsheet
All of this can be tested in a single field and in a month, without buying anything.
Start with the threshold calculation. Take the price you expect to receive per unit of product, the total cost of an application per hectare including labor, and the expected yield of the field, and calculate how much damage that application would need to avoid to pay for itself. That percentage is your economic injury level, and your action threshold goes one step below, to give you time to react. If the exercise seems rough, do it roughly; an approximate threshold calculated with your numbers is worth more than a precise one calculated with someone else's.
Then define the sampling unit and write it in a sentence so that two different people count the same thing. Mark the points on a sketch of the field before going out, distributed and not concentrated, and deliberately include the uncomfortable places where no one walks. Set the day of the week and stick to it.
For four weeks, count and record without making any decisions. That point requires discipline and is what makes the exercise work, because four consecutive readings show you the trend, and the trend is what allows you to anticipate instead of react. Record the average temperature of each day in the same spreadsheet to start accumulating degree days, even if you take it from a regional climate source.
At the end of the month, you will have three things that you do not have today: your own threshold, a series of four points with their direction, and an idea of how much real time monitoring costs, which almost always turns out to be less than feared. From there, the decision to apply stops being an opinion and becomes a comparison between two numbers.
Counting is cheaper than spraying
Pest management by threshold is not a sophisticated or expensive technique; it is a change in the order of operations. Today, the usual sequence is to suspect, apply, and wait; the sequence that works is to measure, compare against a number defined in advance, and only then decide. The cost difference between the two, sustained over several campaigns, is the difference between a farm that manages its pests and one that chases them.
What you can do starting Monday fits into five steps. Calculate the threshold with your own price and your own cost, because a borrowed threshold is miscalibrated, and you won't notice it. Write the sampling unit in a sentence so that two people count the same thing and mark the points on the sketch before going out, including the places where no one walks. Count presence and absence instead of individuals if that is what your team will maintain every week, because the system that is abandoned is worth nothing. Start accumulating degree days with a regional climate source, as the evidence shows it is sufficient to predict the pest cycle. And treat climate risk models as a signal to monitor more, never as an order to apply, because none of them knows your field. A producer with four weeks of counting and a written threshold makes decisions that their neighbor, with the same pest and the same climate, can only guess.
Do you want to set up threshold monitoring for your farm and stop deciding by calendar? Write to us on WhatsApp or schedule a demo, and we will show you how Harvis turns your routes into a series that you can consult when it comes time to decide.
Consulted references
- Economic Thresholds and Economic Injury Levels. (n.d.). Radcliffe's IPM World Textbook, University of Minnesota. https://ipmworld.umn.edu/pedigo
- Utah State University Extension. (n.d.). Pest Management Decision-Making, The Economic-Injury Level Concept. https://extension.usu.edu/pests/research/eil-concept.php
- Conceptualisation of Economic Injury Level and Economic Threshold, Agricultural Management, Food Security, Climate Factors, and Value Chain. (n.d.). Agriculture, MDPI. https://www.mdpi.com/2077-0472/16/5/542
- Evaluación de un método de muestreo para broca del café. (n.d.). Cenicafé, 56(3). https://biblioteca.cenicafe.org/bitstream/10778/150/1/arc056(03)237-249.pdf237-249.pdf)
- Sequential and Binomial Sampling Plans to Estimate Thrips tabaci Population Density on Onion. (2021). Insects, MDPI. https://www.mdpi.com/2075-4450/12/4/331
- Development of Enumerative and Binomial Sequential Sampling Plans for Monitoring Lymantria dispar in Mediterranean Oak Forests. (2022). Agronomy, MDPI. https://www.mdpi.com/2073-4395/12/7/1501
- Spatial Distribution and Development of Sequential Sampling Plans for Diaphorina citri Kuwayama. (2021). Agronomy, MDPI. https://www.mdpi.com/2073-4395/11/7/1434
- Temperature-Dependent Biology and Population Performances of the Coffee Berry Borer Hypothenemus hampei on Artificial Diet. (2023). Insects, MDPI. https://doi.org/10.3390/insects14060499
- Temperature-dependent development and survival of immature stages of the coffee berry borer Hypothenemus hampei. (n.d.). Bulletin of Entomological Research, Cambridge University Press. https://www.cambridge.org/core/journals/bulletin-of-entomological-research/article/abs/temperaturedependent-development-and-survival-of-immature-stages-of-the-coffee-berry-borer-hypothenemus-hampei-coleoptera-curculionidae/87F18569C5977ECEB4158398A5D1DC13
- Coffee berry borer development across an elevational gradient, applying laboratory degree-day predictions to natural field populations. (n.d.). PLOS One. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0218321
- Calculation of degree days of Hypothenemus hampei through satellite images. (n.d.). SciELO. https://www.scielo.org.mx/scielo.php?pid=S2007-09342020000300543&script=sciarttextplus&tlng=en
- Modelling coffee leaf rust risk in Colombia with climate reanalysis data. (n.d.). Philosophical Transactions of the Royal Society B. https://royalsocietypublishing.org/rstb/article/371/1709/20150458/23115/Modelling-coffee-leaf-rust-risk-in-Colombia-with
- Spatiotemporal modeling of Hemileia vastatrix using multiple machine learning algorithms, implications for disease surveillance of coffee leaf rust. (2025). Journal of Plant Pathology, Springer. https://link.springer.com/article/10.1007/s42161-025-01958-3
- Predicting crop disease severity using real time weather variability through machine learning algorithms. (2025). Scientific Reports, Nature. https://www.nature.com/articles/s41598-025-18613-7
- Weather-driven epidemiology and forecasting of Myrothecium leaf spot in coffee seedlings, a machine learning approach. (2026). Frontiers in Climate. https://www.frontiersin.org/journals/climate/articles/10.3389/fclim.2026.1733203/full
- Thermal Tolerance of the Coffee Berry Borer Hypothenemus hampei, Predictions of Climate Change Impact on a Tropical Insect Pest. (n.d.). PLOS One. https://journals.plos.org/plosone/article?id=10.1371/journal.pone.0006487
- CTAHR, University of Hawai'i. (n.d.). Recomendaciones para el manejo integrado de plagas de la broca del café. https://www.ctahr.hawaii.edu/oc/freepubs/pdf/ip-33_spanish.pdf
- Penn State Extension. (n.d.). Orchard IPM, An Overview. https://extension.psu.edu/orchard-ipm-integrated-pest-management-an-overview
- UC Statewide IPM Program, University of California. (n.d.). Establishing Action Thresholds. https://ipm.ucanr.edu/agriculture/floriculture-and-ornamental-nurseries/establishing-treatment-thresholds/