Description du sujet de thèse

Part of the decline in species attributed to global warming results from the phenomenon of desynchronization of phenology (seasonality of life cycles) that can lead to a breakdown of trophic relationships. One of the major challenges in understanding these effects is our ability to acquire data in the wild on the phenology of individual species at high temporal resolution - a daunting task with current ecological monitoring methods. Recent advances in technology and artificial intelligence (AI) offer the possibility of developing low-cost sensors for monitoring biodiversity in the wild with high spatio-temporal resolution. This thesis will draw on the ecological expertise of SETE and the AI and network expertise of IRIT to build a low-cost system for automated monitoring of phenological desynchronization between plants, herbivorous insects, and insectivors.
Global changes associated with anthropogenic activity have triggered the decline of many species in recent decades (Parmesan 2006). Among these changes, global warming, in addition to altering the ranges of species, is causing deregulation of the timing of important biological events-a phenomenon known as phenology. Adjustment of phenology is an important mechanism for species to cope with changes such as rising temperatures: when spring arrives earlier, species can reproduce earlier to maintain conditions similar to historical levels. However, these changes are not always beneficial (Visser and Both 2005). Species are linked through competition, mutualism, and predation, so the success of one species often depends on another. If each species responds differently to global warming, their phenologies may no longer match, so that some species fare better while others suffer more. For example, many songbirds time their breeding so that the most demanding period of parental care - when chicks need food the most - coincides with peak prey resources such as lepidopteran caterpillars. As the climate warms, plants advance leafing; butterflies thereby advance their breeding so that caterpillars emerge with these new leaves; and songbirds advance their breeding so that peak chick demand coincides with peak caterpillar populations. However, slower evolution in the upper levels of the food chain means that, despite advancing phenology, there is a mismatch between trophic levels that can dramatically alter population trends for different species, resulting in declines for some species and increases for others (Radchuk et al 2019).
New technologies and artificial intelligence could democratize the ability to measure phenological lag and provide a major advance in our understanding of this key impact of climate change.
The goal of the proposed project is to develop (or adapt and optimize where they exist) a suite of low-cost sensors and associated AI algorithms to measure the phenology of key life events for a range of species at different trophic levels. The initial effort will focus on the classical budburst-insect-bird paradigm to develop and optimize the systems.
Insect abundance measurements, in particular, remain a challenge as current methods are largely destructive and require human intervention for capture and identification. In this thesis, we will develop AI algorithms (with Axel Carlier) to process images from each sensor and with small training sets to make measurements in new study sites possible. An important goal of this work would be to realize embeddable algorithms on the sensors, thus decreasing the need for storage, power consumption, and providing options for lightweight remote data transfer in the future (with Rahim Kacimi). To verify the field measurements, we will collect ecological data in the field using classical methods with Alexis Chaine's team.
The final results of this project will be 1) the development of new sensors to measure insect abundance 2) the adaptation of existing sensors for plant phenology and bird breeding phenology 3) the scientific validation of these new tools. We will aim to make the sensors inexpensive and provide plans for an open-source model to stimulate more research on phenological lag.
The project is ambitious as it proposes development in three different domains (ecology, AI, mechatronics) nevertheless a significant support budget will allow to outsource some components depending on the candidate's skills.
Role and profile of the PhD student: The student will play 3 essential roles: 1) he/she will be in charge of the development of AI algorithms for sensor image processing under the supervision of Carlier and Cauchoix and 2) he/she will coordinate and participate in all aspects of the project (ecology, mechatronics, AI) to ensure the development of a robust automated monitoring system, 3) he/she will analyze and publish the project results. The student will balance the constraints imposed by ecology (field) and natural history of organisms (seasonality) with those imposed by current technology (data science, electronics in the wild). Through this PhD, the student will also have the opportunity to interact with high-level researchers in our related networks. This dissertation will allow the candidate to gain skills at the forefront of interdisciplinary academic research while maintaining contact with innovative environmental start-ups. The ideal candidate will have a background in computer science and experience in deep learning as well as a strong interest in ecology. However, applications from ecologists/naturalists with regular Python practice and machine learning skills will be considered with the greatest interest.

Contexte de travail

This thesis project emerges from the collaboration between ecology teams (A. Chaine, M Cauchoix) and computer science teams (Axel Carlier, Rahim Kacimi) and is funded by an interdisciplinary grant from two CNRS institutes (INEE and INS2I). The main part of the thesis will focus on the development of electronic and software tools for ecological censusing with field deployment and ecological measurements to verify the automated metrics. As most of the work will be focused on computational tools (AI and electronics), the candidate will spend most of the time in Toulouse in Carlier's lab. Ecological measurements and prototype deployment will take place in Moulis and the candidate will spend some time each year doing field work with the Chaine lab in Ariège.

Contraintes et risques

Occaisional field work in a forest environment.