Application of machine learning models to predict potential distribution and risk of potato cyst nematodes (Globodera rostochiensis and Globodera pallida)

Yitong He

Potato cyst nematodes (PCNs), including the golden (yellow) cyst nematode (Globodera rostochiensis, gPCN) and pale (white) cyst nematode (G. pallida, pPCN), are invasive pests that pose a significant threat to agriculture, especially the potato industry, causing yield and economic losses in many countries and regions. Effective biosecurity strategies and early elimination of PCNs depend on accurate prediction and identification of suitable habitats for these pests. However, the global distribution of PCNs has not been systematically reported, and their potential distribution under climate change conditions remains unknown. Therefore, this study aims to identify the potential global distribution and risk regions of gPCN and pPCN, both independently and integrally, and to reveal their distribution changes under climate change conditions using machine learning processes, including fuzzy logic and ensemble modelling. The Maximum Entropy Model (Maxent) associated with the Geographic Information System (GIS) was employed to reveal the global distribution of the gPCN and pPCN. The bioclimate variables and soil quality were included in the model, and the global arable land was used for the regional risk assessment. Results showed that 53% of the global land surface was suitable for gPCN or pPCN or both, and both species can colonise more than 75% of the global cultivated lands. The niche similarity between two PCN species was tested with a fuzzy generalised linear model. Then, an integrated PCNs dataset was employed to calibrate and evaluate the Maxent model. Results showed that the model constructed on the integrated dataset possessed higher accuracy in comparison with individual datasets. Taking China as an example, the prediction was in accord with real actual presence records, which amended the miss of the predecessor. In China, the high-risk regions occupied more than half of the arable lands, including 66% of the potato-producing areas. Ensemble modelling techniques were applied to predict the potential global distribution of potato cyst nematodes (PCNs) under changing climate conditions. Global Climate Models were used to generate uniform climate predictors to reduce variation among predictors. Then two types of ensembles were constructed, multi-algorithm ensembles (EMA) and single-algorithm ensembles (ESA), using five machine-learning models and comparing their performance. Our findings suggest that PCNs' range is likely to shift northward, with a reduction in tropical regions and an increase in northern latitudes. However, the total area of suitable regions will remain relatively stable, comprising approximately 16-20% of the total land surface (18% under current conditions). Notably, the ESA of Artificial Neural Network exhibited the best performance, while being cost-effective, outperforming the EMA in some cases. In summary, the results of this research, particularly the established multispecies-based model will provide practical support for decision-makers and practitioners to implement biosecurity strategies from a global and climate change perspective that incorporate prevention or promptly enforce control practices to limit the damage caused by future incursions.

Application of machine learning models to predict potential distribution and risk of potato cyst nematodes (Globodera rostochiensis and Globodera pallida)

Files and links (1)

Abstract

Details

Metrics