Location: Application Technology Research
Project Number: 5082-30500-001-070-S
Project Type: Non-Assistance Cooperative Agreement
Start Date: Sep 15, 2025
End Date: Sep 14, 2026
Objective:
This work will focus on: (1) the collection of a dataset to quantify variability in container substrate temperature as a function of ambient environmental conditions, surface temperatures and vegetation phenological state; and (2) develop data-driven modelling approaches to predict container substrate temperatures to better understand fertilizer release and improve management of rootzone temperatures and root health.
Approach:
Data collection will focus on the use of weather data and proximal remote sensing. Previous work has identified several key factors influencing pot media temperature. Solar 2 radiation is a primary driver of media temperature, often causing internal media temperatures to exceed ambient air temperatures. Variations in solar exposure due to positioning within potting arrangements can significantly alter media temperatures relative to adjacent pots. Additionally,
daily air temperature fluctuations have been consistently reported to correlate with media temperature dynamics. Further studies have examined the extent to which media temperature fluctuates in response to air temperature. Other atmospheric variables, such as wind speed and humidity, have been proposed to exert lesser effects and are generally considered insignificant relative to air temperature and solar forcing. Similarly, energy transfer through conduction and
evaporation has been reported to have minimal influence on media temperatures. Media properties also play a crucial role in modulating temperature responses to external forcings. Differences in media composition and moisture content have been shown to affect temperature response times and amplitudes. Pot size and material further contribute to media temperature variability, with smaller, darker pots experiencing greater heating due to solar absorption.
Here we will explore standard weather variables and their ability to predict root zone media temperature, including variables such as: air temperature, incident shortwave radiation, cover treatment surface temperature, wind speed and atmospheric vapor pressure deficit. Potentially confounding factors such as media composition, moisture, and pot size will be kept uniform to mitigate additional sources of variability. Additionally we will examine the use of proximal remote sensing observations of surface temperature and normalized difference vegetation index to improve model performance.
Neural networks and other machine learning methods are widely used for predictive tasks and have consistently demonstrated their effectiveness in environmental time-series applications, effectively able to optimally utilize the information in predictor variables for prediction tasks. These machine learning approaches are well-suited to capturing potential nonlinearities in these horticultural datasets and may yield notable improvements over linear models. We will develop and evaluate rigorously cross-validated machine learning models that integrate weather and/or proximal remote sensing variables to predict root zone media temperatures across diurnal to seasonal variability.