Temperature is a critical metric for accurate harvest forecasting. Previously, we asked you to manually define the relationship between radiation and temperature using four different inputs. While this gave you control, it was complex to maintain, especially when weather conditions fluctuated or during warmer periods.
Why predict & our model performance
We analyzed our historical methods against the predictive model and found that the old linear calculation struggled when outside temperatures rose above 20°C. Greenhouse dynamics change in the heat, and the linear relationship breaks down.
Our new model accounts for this non linearity, resulting in massive accuracy gains:
Results
Condition | Absolute error improvement | Outliers (>1.5°C error) |
Outside Temp < 20°C | Reduced by 0.55°C | Reduced from 30% to 8% |
Outside Temp > 20°C | Reduced by 1.68°C | Reduced from 70% to 15% |
We use absolute errors instead of percentage errors because of how temperature scales. For example, a prediction of 3°C when the actual value is 1°C is just as inaccurate as predicting 23°C when the real value is 21°C.
Note: Analysis based on same day weather predictions.
Defining temperature with Cultivation Strategy and Harvest Forecast
You no longer need to tweak coefficients. Simply go to the Cultivation Strategy page and define the minimum temperature you expect for the upcoming weeks (similar to how you define targets for stem density).
Under the hood, we write the prediction to a target on your crop plan called Temperature (24-hour, predicted). This target replaces your old target Temperature (24-hour, calculated).
This also means that your KPI tile for calculated temperature is deprecated. Instead, we show a tile for predicted temperature. We've also deprecated the other temperature target which was used to keep track of manual temperature values entered by the user because this target is not used anywhere in our products anymore.
But what about climate strategies?
But what about climate strategies?
Growers define radiation and temperature strategies to steer plant development, which means their temperature targets can differ from the temperatures they actually achieve. A solid radiation and temperature strategy revolves around the RTR. This balances the energy a plant receives from radiation with the energy it consumes through temperature. The aim is to match the creation of assimilates with the energy required for maintenance and growth.
We based our initial KPI tiles on this principle. Growers could set temperature targets along the RTR curve by combining weather forecasts with our temperature calculation. This would then determine whether the 24-hour temperature tile showed green or red based on realised temperatures. On paper, this seemed like a clean way to monitor climate strategy and check whether greenhouse operations were aligned with the grower’s intent.
In practice, the tile wasn't used as intended. Users would look at it to get a quick sense of the realised temperature and then applied their own logic. The timing also worked against us. When Harvest Forecast launched in April 2025, it introduced automatic temperature calculations using a similar method but replaced the target that this tile relied on. Because Harvest Forecast depends on accurate temperature predictions, the tile shifted from representing a climate strategy to representing a accurate temperature prediction.
At the time of writing, we are moving even further away from the idea of grower defined climate strategies. Temperature is no longer determined primarily by what growers set in their cultivation plan. It now depends on a predictive model that aims to represent the greenhouse accurately and replaces the manual coefficient tweaking that support and growers used to do. This makes Harvest Forecast easier to use, but it also means the tile now reflects how well our model performs. In other words, the Temperature (24-hour, predicted) tile is green when our prediction is accurate and red when it is not. As a result, the tile can be green during a heatwave if the model correctly predicts the high temperature, even though this may contradict your intended RTR strategy.
We want to reduce the amount of manual input needed to use our product correctly. This feature moves us in that direction, and we accept that it comes at the cost of the original value the tile provided. The concept behind the tile was solid, but in practice many users did not use it as intended. With that in mind, we should explore a different approach that gives you a way to define a temperature strategy
Defining temperature in our Crop Plan Simulator
The Simulator now uses this same predictive engine. We have cleaned up the interface to remove unused columns.
Removed columns:
Max 24hr temperature
Min/Max radiation sum
Temperature (24 hour, calculated)
Remaining columns:
Min 24hr temperature: your input
Temperature (24 hour, predicted): the model output
By default, the simulator loads with the Temperature (24 hour, predicted) in green. This uses weather forecasts and your Min Temp input.
If you want to test a "What-If" scenario, you can overwrite the predicted temperature by typing a value directly into the cell.
To revert to the model's prediction, simply click the "Use suggestion" button on the cell or above the column.
Important: Because the predictions are live and weather dependent, temperature targets are no longer saved statically to your Crop Plan when you press "Save Strategy." We update this daily to ensure your forecast always uses the latest weather data.