Machine learning for predicting climate change impacts on Pseudopithomyces chartarum spore counts: a risk indicator of facial eczema

Authors: Wada M, Sagarasaeraneea O, Cogger N, Marshall J, Cuttance E, Macara G, Sood A, Vallee E
Publication: New Zealand Veterinary Journal, Volume Ahead of Print, Issue Ahead of Print, Dec 2025
Publisher: Taylor and Francis

Animal type: Cattle, Sheep
Article class: Research Article

---

Abstract:

Aims: To determine the importance of 11 climate variables on pasture spore count of Pseudopithomyces chartarum, a risk indicator of facial eczema (FE), and to forecast spore counts in New Zealand until 2100, using longitudinal P. chartarum pasture spore count data.

Methods: Between 2010 and 2017, spore counts (n = 6,975) were collected from 862 paddocks spread over 102 farms in the North Island of New Zealand. Historical and projected climate data were obtained from the National Institute of Water and Atmospheric Research. The spore count dataset was merged with climate data from corresponding locations, incorporating time lags of 1–53 weeks. Linear regression models were fitted for describing crude associations, while random forest models were fitted for determining variable importance and predicting future spore counts.

Results: Mixed-effect linear regression models explained up to 11% of the variance of log-transformed spore counts by a single lagged climate covariate. The best-fit random forest model had a testing accuracy of 80% in classifying low or high FE risk (> 20,000 spores) with an R² value of 43%. The random forest models suggested time-dependent importance of soil temperature at 10 cm depth, solar radiation, potential evapotranspiration, vapour pressure, soil moisture and minimum temperature, while no or weak evidence of variable importance was found for maximum temperature, rainfall, mean sea level atmospheric pressure, relative humidity and wind speed. Over the next 80 years, our model predicted an increase in the seasonal mean spore counts in the study farms by a mean of 17% (min 6, max 30%) under the high-end greenhouse gas emission scenario (representative concentration pathways (RCP) 8.5). Every decade was associated with an increase in the probability of high-risk spore counts (> 20,000) by 14–22% for the moderate to high emission scenarios (RCP 4.5–8.5). The model indicated increased peak spore counts across most regions over the next 80 years. Specifically, the entire North Island and three districts in the South Island were projected to have high mean peak spore counts by 2100.

Conclusions and clinical relevance: These findings could be used to target high-risk areas to implement mitigation or adaptation measures for FE. In addition, the study highlights the value of ecological data for forecasting environmental disease risks to enhance preparedness for climate change.

KEYWORDS: Facial eczema, environmental health, livestock disease, climate change, sustainable agriculture, disease forecasting, fungal spores

---