From machine learning to citizen science: methods for estimating pest disease prevalence

Abstract

EFSA is requested to provide guidelines for estimation of pest prevalence surveys. In line with these guidelines this study performs a literature review, data extraction and general research to investigate the prevalence (monitoring) survey methods and tools that might already be available in the literature and other sources. Pest disease surveillance is required for informed assessment and management. Quantifying disease prevalence is a key factor for the disease impacts on ecosystems, one health, and food security. Surveys involve systematic or opportunistic sampling of a sub‐set of the target population in space and time. With the rapid development of smart sensors, high accuracy images, digital maps, land cover as well as climatic data, environmental data are more ubiquitous than ever before. In the light of more data novel statistical methods are developed or traditional statistical methods can be calibrated and used differently, advancing their scope. This work summarizes recent developments of quantitative methods for estimating invasive alien species and pest disease prevalence. The study includes statistical and computational methods, elements of experimental design, linear and non‐linear methods, machine learning, non‐invasive sampling, capture recapture, remote sensing via satellite or unmanned aerial vehicles (drones), and citizen science. In addition, plant image databases of pests, diseases, and invasive species are listed. Mobile phone or web applications and their potential use for disease detection are presented. Computational software tools and libraries are listed. Interdisciplinary combinations between the methods are also discussed. Such approaches may offer novel insights, rapid assessment, and cost‐efficient surveillance prevalence estimates. These should be seen as complementary to current field survey methods for enhancing surveillance and estimating prevalence.