Ascolese, Roberta (2025) Smart Pest Management through AI-Driven and Model-Based Approaches: A Case Study on the Fruit Flies Ceratitis capitata and Bactrocera dorsalis in the Campania Region. [Tesi di dottorato]

Documento PDF ascolese_roberta_38_COMPLETO.pdf Visibile a [TBR] Amministratori dell'archivio Download (9MB) | Richiedi una copia

Abstract

Pest management planning has become increasingly challenging due to climate change and the rise in anthropogenic activities. Global warming has affected the life cycle of locally adapted insect species: the developmental and flight periods of different pests no longer align with previously recorded biological data. The timing of first emergence and the abundance of adults of various pests have become difficult to predict. The rise in average temperatures led to an increase in overwintering insect populations and to an expansion of their geographic distribution. Direct consequences include population surges, increased damage to cultivated fields, and economic losses. Furthermore, global trade and travel have also promoted the introduction and spread of exotic species into new regions, which can often establish up in areas where they were previously absent, affecting local biodiversity. The Campania region in southern Italy represents a real-world case of these phytosanitary issues, due to the increasing damage caused by the Mediterranean fruit fly, Ceratitis capitata Wiedemann (Diptera: Tephritidae), and the recent detection of the Oriental fruit fly, Bactrocera dorsalis Hendel (Diptera: Tephritidae), which is ranked among the top 20 EU priority quarantine pests due to its broad polyphagy, high ecological adaptability, and invasive behaviour. In this scenario, the present study aims to update the phenological data of Ceratitis capitata and to evaluate the potential establishment of Bactrocera dorsalis in the Campania region. By integrating Artificial Intelligence and modelling techniques, the research investigates the population dynamics of these species within local agroecosystems. Furthermore, it seeks to determine whether the presence of Bactrocera dorsalis is the result of acclimatization to current microclimatic conditions or of repeated introductions. A monitoring network carried out combining traditional and electronic traps (e-traps), enhanced with Deep Learning algorithms for species recognition, has been set up. Additionally, smart control units have been installed to collect fundamental agrometeorological variables. Phenologically-based and Machine Learning models, including degree-day analysis and Artificial Neural Networks (ANNs), are evaluated by integrating biological, entomological, and agrometeorological data to study the suitable climatic conditions for the pests and to predict the adult emergence. Seasonal trends were analysed for both species, and a Deep Learning–based recognition system was developed for automated specimen identification. E-traps also enabled the study of circadian rhythms of both species, providing insights into the most suitable time of day for applying targeted and more effective treatments. Physiologically-based models employing delay differential equations (DDE models), which describe stage-development dynamics by incorporating key biological parameters such as development, mortality, biological delays, and fertility rates, are validated with field monitoring data to simulate temperature-dependent population dynamics and describe seasonal fluctuations under varying agroclimatic conditions. To further investigate, field trials have been conducted to determine whether Bactrocera dorsalis can overwinter or whether the climatic conditions of the Campania region are lethal for pupae. Results revealed a strong predominance and adaptability of C. capitata, whose activity extended throughout much of the year, indicating phenological shifts likely driven by milder winters. In contrast, B. dorsalis displayed discontinuous adult activity, suggesting that its persistence in Campania is still limited by cold tolerance and possibly reliant on repeated introductions. E-trap monitoring data revealed the daily activity patterns of both species, identifying the times of day when adult populations are most concentrated and active. In particular, Ceratitis capitata showed two main peaks of activity, one in the morning and another in the afternoon, whereas Bactrocera dorsalis appeared to be most active during the early morning hours. Degree-day analyses were carried out for all years of the study, where Ceratitis capitata consistently showed an earlier first field emergence compared to literature data, suggesting a shift in its life cycle driven by climate change and increasingly warmer temperatures. This trend is further supported by the species’ current near-year-round presence in the region. In contrast, Bactrocera dorsalis spring generation would have been expected based on thermal requirements, yet the first annual captures always occurred several months after the last detections of the previous year (approximately six months in 2023), indicating an unrealistically long diapause period. These findings, together with haplotype analyses, support the hypothesis that B. dorsalis populations might result from repeated new introductions rather than successful overwintering. ANN's predictions highlighted the following key agroclimatic variables as significant inputs: air and soil temperature, soil moisture, wind speed, and rainfall. ANN models highlighted air and soil temperature, soil moisture, wind speed, and rainfall as key predictors of pest occurrence. These variables enabled the prediction of population density under simulated agroclimatic scenarios, allowing estimates to be linked to minimum intervention thresholds for more effective management strategies. DDE simulations successfully reproduced population peaks consistent with field observations. In particular, the simulations provided a detailed representation of the population dynamics of both species throughout their entire life cycle. For each day of the modelled period, the system generated curves describing the abundance of each developmental stage (from eggs to adults, further divided by sex), thus allowing a comprehensive estimation of total population density over time. The simulated trends closely matched the temporal pattern of male captures obtained from field monitoring, validating the model’s reliability. Moreover, the integration of local mean temperature data ensured that the simulated phenological events reflected the actual microclimatic conditions of the study area, strengthening the consistency between observed and predicted population fluctuations. Altogether, these results demonstrate that combining continuous, high-resolution monitoring with data-driven and mechanistic modelling can substantially enhance forecasting accuracy and decision-making in Integrated Pest Management (IPM). The developed framework provides a scalable, adaptive, and climate-aware approach for protecting Mediterranean crops and supports the transition toward more intelligent and sustainable pest control strategies in agriculture. These outcomes represent essential tools for planning and implementing targeted phytosanitary interventions, contributing to the development of a reliable Decision Support System (DSS) for sustainable pest management in the Campania region. By merging AI-driven predictions with field-based monitoring, the study strengthens the understanding of fruit fly population dynamics and fosters more efficient and environmentally responsible IPM practices.

Downloads

Actions (login required)

Modifica documento