Integration of machine learning into process-based modelling to improve simulation of complex crop responses

Abstract

Machine learning (ML) is the most advanced field of predictive modelling and incorporating it into process-based crop modelling is a highly promising avenue for accurate predictions of plant growth, development and yield. Here, we embed ML algorithms into a process-based crop model. ML is used within GLAM-Parti for daily predictions of radiation use efficiency, the rate of change of harvest index and the days to anthesis and maturity. The GLAM-Parti-ML framework exhibited high skill for wheat growth and development in a wide range of temperature, solar radiation and atmospheric humidity conditions, including various levels of heat stress. The model exhibited less than 20% error in simulating the above-ground biomass, grain yield and the days to anthesis and maturity of three wheat cultivars in six countries (USA, Mexico, Egypt, India, the Sudan and Bangladesh). Moreover, GLAM-Parti reproduced around three quarters of the observed variance in wheat biomass and yield. Existing process-based crop models rely on empirical stress factors to limit growth potential in simulations of crop response to unfavourable environmental conditions. The incorporation of ML into GLAM-Parti eliminated all stress factors under high temperature environments and reduced the physiological model parameters down to four. We conclude that the combination of process-based crop modelling with the predictive capacity of ML makes GLAM-Parti a highly promising framework for the next generation of crop models.

A new framework for predicting and understanding flowering time for crop breeding

“…The methodology demonstrated in this paper is modular and could be imported into existing crop models to improve their ability to simulate G ∗ E. The inclusion of modular ML classification modules into existing crop models has been successfully demonstrated by Droutsas et al. (2022), and a similar approach could be taken for the genotype‐specific classification algorithm presented in this paper. As part of rapid modernisation of breeding programmes targeted at increasing genetic gain, software is being built to simulate the breeding pipelines themselves, and a modular phenological model such as this one could also feed into these optimisation procedures.…”

“…Feature engineering is ML terminology for describing the process of finding and developing the variables that will be used to predict flowering time. Following Droutsas et al (2022), in order to ensure that the models could provide dynamic predictions, we converted each day of the growing season to a binary value-0 if the plant was in the vegetative stage and 1 if the plant was in the reproductive stage. This binary feature was our target label (the variable we predicted).…”

“…For example, an a priori classification of genotypes into those likely to flower early under heat can help breeders to understand the contribution of avoidance vs. true tolerance in their germplasm. Further, combining the ability to predict flowering time with an understanding of pest dynamics can help breeders to ensure that experiments are sown to avoid flowering coinciding with severe attacks of insect or fungi.The methodology demonstrated in this paper is modular and could be imported into existing crop models to improve their ability to simulate G * E. The inclusion of modular ML classification modules into existing crop models has been successfully demonstrated byDroutsas et al (2022), and a similar approach could be taken for the genotype-specific classification algorithm presented in this paper. As part of rapid modernisation of breeding programmes targeted at increasing genetic gain, software is being built to simulate the breeding pipelines themselves, and a modular phenological model such as this one could also feed into these optimisation procedures.…”

Synergizing Crop Growth Models and Digital Phenotyping: The Design of a Cost-Effective Internet of Things-Based Sensing Network

“…For example, one can reinforce the Wi-Fi connection or create a local database that stores the data in the event of Wi-Fi interruptions. In line with Droutsas et al [24], who proposed the integration of machine learning models into a process-based model, the described network aims to enhance actual data analysis systems and reduce modeling fine-tuning processes. Although further tests are needed, the proposed sensing network has the potential to overcome the phenotyping pitfalls identified by Saint-Cast et al [10], namely the lack of common semantics and thorough data exchange platforms.…”