Climate change threatens food safety: three computational tools could help manage the risk

Climate change is no longer just a story about melting ice sheets and shifting weather patterns. According to the sixth assessment report of the Intergovernmental Panel on Climate Change, average global surface temperature is expected to rise between 2.2 and 3.5 degrees Celsius by the end of the twenty-first century. Beyond that warming trend, the world faces more floods, droughts, hurricanes, heavy precipitation events, water acidification and potential sea-level rise. These disruptions will strain food supply chains and favour the proliferation of certain pathogens, the contamination of food by some chemicals and the emergence of foodborne diseases. The European Food Safety Authority’s Emerging Risks Unit has already recognised climate change as a significant factor contributing to emerging risks in food and feed safety over the medium to long term. Now, a new Perspective article argues that the food safety sector needs computational tools to adapt, and that three in particular could make the difference.

The article, published in the journal Food Safety and Risk, examines how three in-silico tools — Multi-Criteria Decision Analysis, predictive microbiology and machine learning — can help manage microbiological food safety in a warming world. The term in-silico is used as an umbrella expression for computational and quantitative approaches that use mathematical, statistical or algorithmic methods to analyse information, simulate systems or support decision-making. Each of the three approaches performs a distinct but complementary function. Multi-Criteria Decision Analysis supports the comparison and ranking of management options according to multiple criteria. Predictive microbiology uses mathematical models to simulate microbial behaviour under environmental and process conditions. Machine learning uses data-driven algorithms to identify patterns in large datasets and support prediction and classification. Together, they can contribute to decision support at different stages of food safety management and be adopted by different stakeholders, from farmers to regulators.

The stakes are high. Increased temperatures enhance the risk of salmonellosis and campylobacteriosis. Almost 93 million cases of gastroenteritis due to Salmonella are confirmed annually worldwide, of which approximately 155,000 infections are fatal, with the foodborne route established in approximately 85 percent of the cases. Seasonal patterns in salmonellosis are well established, with higher incidence during warmer periods, and associations between short-term variations in ambient temperature and salmonellosis incidence have been reported, with increases ranging from 1 to 2 percent in Canada to 4 to 10 percent in Australia and some European countries for each 1 degree Celsius increase. Campylobacteriosis, with over 168,000 human cases annually, is the most frequently reported zoonosis in the European Union, and higher occurrence of Campylobacter in chicken flocks and human cases has been reported during summer months and under humid conditions. Yet the picture is not simple: despite generally higher summer incidence, heatwaves have been associated with reduced Campylobacter incidence, illustrating that short-term responses to extreme events may differ from broader seasonal patterns and projected long-term climate-change effects.

The first tool, Multi-Criteria Decision Analysis, is illustrated by a case study on a dairy farm operating under current hot-weather conditions. After a pre-selection step among possible Good Agricultural Practices and Good Hygienic Practices measures, four food safety measures were ranked: sand bedding renewal frequency, chitosan supplementation, cooling mister operation and phage spray. The evaluation considered ten criteria classified into four supra-criteria: stakeholder acceptance, food safety effectiveness, environmental impact and impact on milk properties. The analysis determined the best compromise among the selected measures: an increase in the frequency of changing the sand bedding ranked first and an increase in the operation of cooling misters ranked last. This flexibility allows heterogeneous information to be considered jointly without necessarily requiring large quantitative datasets or substantial computational resources. However, relatively low computational demand should not be interpreted as methodological simplicity, as the approach requires careful formulation of the decision problem and involves context- and value-dependent methodological choices. Outranking methods can also limit compensation between criteria, so that poor performance on one criterion is not necessarily offset by good performance on another.

The second tool, predictive microbiology, is demonstrated through a study focusing on Geobacillus stearothermophilus, a spoilage microorganism producing highly heat-resistant spores. In UHT products, surviving spores may germinate and grow when exposed for sufficient time to temperatures above their minimum growth temperature, potentially resulting in product spoilage and loss of commercial sterility. The study evaluated the thermal treatment effect on spores during UHT processing of a milk alternative product and the temperature effect on spore germination and outgrowth during distribution and storage, considering Poland and Greece as two different current regional climatic contexts in Europe. The baseline scenario used recorded temperature profiles from 2022, with a uniform 2 degree Celsius increase applied to the hourly baseline profiles for both countries. Under this scenario, the probability of loss of commercial sterility increased from zero to 1.0 × 10−4 in Poland and by a factor of two or three in Greece, depending on air conditioning use at the consumer stage. Increasing thermal-treatment intensity and using insulated trucks during distribution substantially reduced this probability.

The authors are careful to note that this simplified temperature-perturbation scenario was not associated with a specific RCP or SSP pathway or future time horizon and was not derived from a specific climate model or model ensemble. It should therefore be interpreted as a way of exploring the potential effects of warmer conditions rather than as a location- and time-specific climate projection. Moreover, temperature represents only one dimension of climate change. Changes in precipitation, relative humidity, climatic variability and extreme events may also affect microbial behaviour and food-chain conditions through context-dependent and potentially interacting pathways. An important consideration when applying predictive microbiology to climate-change scenarios is the domain of applicability of the models. Predictions are most reliable within the temperature, pH or water activity ranges used for model development and validation. Extrapolation beyond those ranges, or to different food matrices or microbial strains, may increase uncertainty and potentially produce biologically unrealistic responses.

The third tool, machine learning, is illustrated by a study using data from the French national poultry control and surveillance program. In that work, 3,784 cecal sample results collected in 2018 and 2020 across 18 French departments were combined with twelve derived meteorological variables capturing temperature, precipitation, humidity and their delayed effects. Several machine learning models were developed and compared to predict Salmonella occurrence while accounting for class imbalance. Meteorological conditions occurring four weeks prior to sampling were the most informative predictors. Warm temperatures between 20 and 30 degrees Celsius and moderate relative humidity between 70 and 90 percent were associated with an increased Salmonella occurrence, whereas frequent rainfall was associated with a reduced occurrence, likely through dilution and environmental washout mechanisms. These findings highlight the nonlinear and delayed associations between meteorological conditions and Salmonella contamination.

The authors stress that this analysis cannot characterise long-term climatic variability or persistent climate trends and should not be interpreted as direct evidence of climate-change effects, because the dataset covered only two years. Rather, it represents a proof of concept for using machine learning to investigate associations between observed meteorological conditions and Salmonella occurrence. The limited temporal coverage also constrained model validation: a random stratified train–test split was used, and model performance was not evaluated on an independent year, geographical region or external dataset. Longer time series and independent temporal and geographical validation are therefore needed to assess model generalizability. A further limitation concerns extrapolation beyond the historical training domain, since future combinations or ranges of meteorological conditions may differ from those represented in the training data, resulting in distribution shifts and potentially reduced model performance.

Independent studies reinforce the potential of these approaches. Stathas and colleagues used Multi-Criteria Decision Analysis to prioritise emerging microbial and chemical hazards associated with fresh produce and cereal grains in the European Union under climate change, combining expert knowledge, predictive microbiology, outbreak data and literature to rank hazards according to climate-related drivers. Another application analysed the lentil food system in France using 16 criteria classified under four supra-criteria — food safety, nutrition, economic impact and environmental impact — with 2050 scenarios considering increased imports or local production to meet an assumed increase in consumer demand associated with the protein transition and climate change. In the machine learning domain, Golden and colleagues compared random forest and gradient boosting models to predict Listeria spp. prevalence in pastured poultry farm environments using farm-management and meteorological variables, with weather-related factors including temperature, humidity and wind speed contributing to the predictions.

The Perspective article concludes with a sober assessment of maturity. Multi-Criteria Decision Analysis is an established decision-support methodology, but the climate-related examples presented primarily represent research applications. Predictive microbiology is methodologically well established with long-standing applications in food safety, yet its explicit integration with future climate scenarios remains mainly research-based. Machine learning is widely established as a data-analytic methodology, but its use for climate-informed microbiological food safety surveillance remains largely at the research or proof-of-concept stage in the applications discussed. Whatever the tool, its added value depends on the quality and representativeness of the available data, model validity, appropriate consideration of uncertainty, and the ability of users to interpret the outputs within their intended context. The authors emphasise that these approaches should be regarded as aids to management and decision-making rather than substitutes for decision-makers, and that in-silico tools cannot, on their own, guarantee food safety. As the climate continues to change, the food safety sector will need every aid it can get — but the tools will only be as good as the data, validation and judgement behind them.

Subject of Research: The role of in-silico tools in microbiological food safety management under climate change

Article Title: Climate change and microbiological food safety management: the role of in-silico tools

Article References: Feliciano, R. J., Sylla, A., Membré, J.-M., & Delaunay, L. (2026). Climate change and microbiological food safety management: the role of in-silico tools. Food Safety and Risk, 13(1), Article 47. https://doi.org/10.1186/s40550-026-00170-0

Image Credits: AI Generated

DOI: 10.1186/s40550-026-00170-0

Keywords: climate change, food safety, in-silico tools, Multi-Criteria Decision Analysis, predictive microbiology, machine learning, Salmonella, Campylobacter, Geobacillus stearothermophilus, risk management, foodborne disease, surveillance