Articles
Determining the level of water stress of almond trees from weather, irrigation and soil moisture data using machine learning
Article number
1468_42
Pages
321 – 326
Language
English
Abstract
Water tension within the plant is recognized as a sensitive indicator of plant water status.
For many crops, long-term research has established stem water potential (SWP) thresholds that can detrimentally affect yield.
This information is useful to schedule irrigation targeting desired levels of water stress at each crop stage.
Stem water potential has been traditionally measured periodically using a pressure chamber.
Currently, microtensiometer sensors that are embedded into the woody tissue of trees can be used for continuous SWP monitoring in some species.
The large data set obtained with the automatic measurements of SWP, provides the opportunity for exploring the capabilities of machine learning (ML) models in estimating SWP from more accessible data sources.
With this objective in mind, the performance of a Gradient Boosting Trees (GBT) model in discriminating among water stress levels was evaluated during the 2024/25 irrigation season on almond trees using daily water applied, weather and soil tension data (348 records in total). The stress level was determined by the difference between the lowest SWP recorded on a day and the baseline for that day.
Results showed that the precision and recall parameters differed among each stress level with the no stress group showing the lowest values.
Using weather and water applied as input data, the model had a F1 score (relation between the precision and recall parameters) of 0.51, 0.73 and 0.76 for the no stress, mild stress and moderate stress levels, respectively.
These values improved to 0.67, 0.86 and 0.86 in each of these groups when soil water tension data was also considered.
This case-study showed that a GBT model allowed determining the level of water stress of almond trees in an irrigation sector based on weather, water applied and soil moisture data with high precision.
Larger data sets could improve results specially for non- and severe stress conditions.
For many crops, long-term research has established stem water potential (SWP) thresholds that can detrimentally affect yield.
This information is useful to schedule irrigation targeting desired levels of water stress at each crop stage.
Stem water potential has been traditionally measured periodically using a pressure chamber.
Currently, microtensiometer sensors that are embedded into the woody tissue of trees can be used for continuous SWP monitoring in some species.
The large data set obtained with the automatic measurements of SWP, provides the opportunity for exploring the capabilities of machine learning (ML) models in estimating SWP from more accessible data sources.
With this objective in mind, the performance of a Gradient Boosting Trees (GBT) model in discriminating among water stress levels was evaluated during the 2024/25 irrigation season on almond trees using daily water applied, weather and soil tension data (348 records in total). The stress level was determined by the difference between the lowest SWP recorded on a day and the baseline for that day.
Results showed that the precision and recall parameters differed among each stress level with the no stress group showing the lowest values.
Using weather and water applied as input data, the model had a F1 score (relation between the precision and recall parameters) of 0.51, 0.73 and 0.76 for the no stress, mild stress and moderate stress levels, respectively.
These values improved to 0.67, 0.86 and 0.86 in each of these groups when soil water tension data was also considered.
This case-study showed that a GBT model allowed determining the level of water stress of almond trees in an irrigation sector based on weather, water applied and soil moisture data with high precision.
Larger data sets could improve results specially for non- and severe stress conditions.
Authors
R.F. Maia, J. Hornbuckle, C. Ballester
Keywords
Gradient Boosting Trees (GBT), soil moisture, stem water potential
Groups involved
Online Articles (70)