TY - JOUR
T1 - Integrating machine learning with agroecosystem modelling
T2 - Current state and future challenges
AU - Aderele, Meshach Ojo
AU - Srivastava, Amit Kumar
AU - Butterbach-Bahl, Klaus
AU - Rahimi, Jaber
N1 - Publisher Copyright:
© 2025 The Authors
PY - 2025/7
Y1 - 2025/7
N2 - Machine learning (ML), especially deep learning (DL), is gaining popularity in the agroecosystem modelling community due to its ability to improve the efficiency of computationally intensive tasks. By reviewing previous modelling studies using the PRISMA technique, we present several examples of ML applications in this domain. The potential of using such models is highligthed. The different types of integration and model-building methods are categorized into process-based modelling (PBMs) and data-driven modelling (DDMs), which simulate different aspects of agroecosystem dynamics. While PBMs excel at capturing complex biophysical and biogeochemical processes, they are computationally intensive and may not always be solvable using analytical methods. To address these challenges, machine learning (ML) techniques, including deep learning (DL), are increasingly being integrated into agroecosystem modelling. This integration involves replacing PBMs with data-driven models, using hybrid models that combine PBMs and ML, or constructing simplified versions of PBMs through meta-modelling. ML-based meta-models offer computational efficiency and can capture intricate patterns and non-linear relationships in complex agricultural systems. However, challenges such as interpretability and data requirements remain. This review highlights the importance of addressing gaps and challenges to fully realize the potential of ML to identify the most promising ways of field management in promoting sustainable agricultural systems. It also highlights specific considerations such as data requirements, interpretability, model validation, and scalability for the successful integration of ML with PBMs in agriculture and the transformative potential of combining ML with PBMs, particularly in extending simulations from field to global scales and streamlining data collection processes through advanced sensor technologies based on their applications.
AB - Machine learning (ML), especially deep learning (DL), is gaining popularity in the agroecosystem modelling community due to its ability to improve the efficiency of computationally intensive tasks. By reviewing previous modelling studies using the PRISMA technique, we present several examples of ML applications in this domain. The potential of using such models is highligthed. The different types of integration and model-building methods are categorized into process-based modelling (PBMs) and data-driven modelling (DDMs), which simulate different aspects of agroecosystem dynamics. While PBMs excel at capturing complex biophysical and biogeochemical processes, they are computationally intensive and may not always be solvable using analytical methods. To address these challenges, machine learning (ML) techniques, including deep learning (DL), are increasingly being integrated into agroecosystem modelling. This integration involves replacing PBMs with data-driven models, using hybrid models that combine PBMs and ML, or constructing simplified versions of PBMs through meta-modelling. ML-based meta-models offer computational efficiency and can capture intricate patterns and non-linear relationships in complex agricultural systems. However, challenges such as interpretability and data requirements remain. This review highlights the importance of addressing gaps and challenges to fully realize the potential of ML to identify the most promising ways of field management in promoting sustainable agricultural systems. It also highlights specific considerations such as data requirements, interpretability, model validation, and scalability for the successful integration of ML with PBMs in agriculture and the transformative potential of combining ML with PBMs, particularly in extending simulations from field to global scales and streamlining data collection processes through advanced sensor technologies based on their applications.
KW - Agroecosystem
KW - Data-driven models(DDM)
KW - Machine learning (ML)
KW - Process-based models (PBM)
UR - http://www.scopus.com/inward/record.url?scp=105000064414&partnerID=8YFLogxK
U2 - 10.1016/j.eja.2025.127610
DO - 10.1016/j.eja.2025.127610
M3 - Journal article
AN - SCOPUS:105000064414
SN - 1161-0301
VL - 168
JO - European Journal of Agronomy
JF - European Journal of Agronomy
M1 - 127610
ER -