The project activities were divided in three main Work Packages (WPs) related to the specific objectives of the project and were supported by an additional WP dedicated to data collection and analysis.
The first WP explored different modelling aspects related to the biology of terrestrial arthropods and their interaction with the external environment and with the other organisms living within the agroecosystem. I have introduced a general physiologically-based model which provide, as a solution, the number of individuals over time in each life stage. This general model has been tested on the two target species in both open field and controlled environments, leading also to set up an operating protocol to parameterise and validate the models. This piece of work will support further applications of the methodology introduced to other case studies of agricultural and ecological interest.
In addition, PestFinder revised the data collection process and analysis of life tables, one of the most common experiments carried out in insect science to explore the biology of the species and to retrieve quantitative information that further support model parameterisation. This part of the work was not foreseen in the initial planning and was inspired by a thorough revision of the literature which identified a gap in knowledge that should have been filled.
The general model was also crucial to describe spatial aspects and multitrophic relations, two aspects which were still poorly explored by the current literature. The introduction of general multitrophic models led to a more realistic description of the interaction between different organisms living the agroecosystems (e.g. plant-pest and pest-enemy).
The second WP substantially contributed to my training activities as well, as I had to learn the most common state-estimation and optimisation techniques from control engineering before their implementation to the case of terrestrial arthropods. This WP led to apply estimators, or state-observers, to insect populations for the first time. These algorithms "find the good compromise" between the simulated output of the system, provided by the model, and the real output of the system, provided by measurements, leading to an improved knowledge of the infestation level. As measurements is an important component of the problem, PestFinder led to a thorough revision of the most common measurement techniques applied in insect science and to their mathematical translation. Besides bioecological models, accordingly, this project provided a set of sensing models that can explain, from a mathematical point of view, the measurement process. This result led to connect, for the first time, insect science with metrology, building a bridge between the two scientific communities. WP2 also explored optimal sensing, namely the model-driven data collection process which ensures the minimal uncertainty associated with the knowledge of the infestation level within the field. Although not initially foreseen by the project, I have explored optimal sensing problems and the use of estimators in plant pathology as well, using some recent results obtained by the SAAS unit on human epidemics. This additional activity introduced the use of estimators for the first time on the second pillar of plant protection, providing helpful hints for further developments.
The third WP revised the concept of decision-making in agriculture by introducing general guidelines to set up optimisation problems. The dynamic model developed in WP1, complemented with the estimators developed in WP2, are a component of a more general problem which reflects the principia of IPM. Control actions affect the dynamic of the system, but at the same time they have an associated cost. The IPM problem, accordingly, is a cost-benefit analysis: PestFinder showed mathematically that the optimal control strategy (the number of treatments and the timing of application) does not correspond to eradicate the pest from the field, but to keep the population "low enough such that the yield is maximised". In this rationale, multitrophic models play a fundamental role, as the cost (control action) acts directly on the pest population and the gain (the yield) comes from the number of fruits/biomass which are "safe" at the harvest time. This part of PestFinder also highlighted the importance of connecting multiple trophic levels together with a mathematical description of the effects of the control action on the overall dynamics.
