Simple models that capture the complexity of multi-species coexistence
Modelos simples que capturan la complejidad de la coexistencia
Principal investigator
Ignasi Bartomeus
Financial institution
MIN ECON, IND y COMPETITIVIDAD
Fecha de inicio
Fecha de fin
Code
CGL2017-92436-EXP
Department
Ecology and Evolution
Brief description
Comprender el mantenimiento de la biodiversidad es fundamental para la ecología, especialmente frente al cambio ambiental inducido por los humanas y las alarmantes tasas de perdida de biodiversidad. Hemos avanzado mucho en la creación de sólidos modelos matemáticos capaces de predecir la coexistencia entre especies que interactuan atraves de diferentes niveles tróficos. Estas avances incluyen herramientas conceptuales y matemáticas desarrolladas por nuestro grupo que permiten la evaluación simultánea de la coexistencia en comunidades completas, compuestas por varios niveIes tróficos, por ejemplo, entre plantas, polinizadores, herbívoros y parasitos. Sin embargo, la evalución empírica de este marco teórico ha demostrado ser más desafiante de lo esperado por dos razones relacionadas. En primer lugar, hay una escasez de datos sobre interacciones multitróflcas para comunidades completas integradas por varios tipos de interacciones. En segundo lugar: los modelos de coexistencia actuales son complejos y el número de parametros a estimar crece exponencialmente con el número de especies en la comunidad. Debemos encontrar nuevas formas de conciliar el poder de grandes conjuntos de datos con modelos arraigados en una teoría solida. El uso de las técnicas de Machine Learning ha revolucionado la capacidad predictiva de varios problemas complejos aprendiendo patrones a partir de datos, pero los algoritmos del aprendizaje automatico tradicionalmente no son interpretados y, por lo tanto, están desconectados de la teoria. Aqui proponemos usar algoritmos basados en reglas difusas para simplificar la estimación de parámetros sin perder la interpretabilldad. Además completaremos dos conjuntos de datos empíricos.que comprenden comunidades completas. Para unir datos y modelos, elegimos una pregunta clave: ¿pueden las técnicas informáticas ayudar a predecir la estructura de interacción de especies que mejora la coexistencía de las especies.
Understanding biodiversity maintenance is central to ecology, especially on the face of human induced environmental change and the alarming rates of biodiversity loss. We have made great progress in building solid mathematical models able to predict coexistence among interacting species across trophic levels. These advances include recent conceptual and mathematical toolboxes developed by our group allowing the simultaneous assessment of coexistence on complete communities, composed by several trophic levels, for example between plants, pollination, and herbivores. However, the empirical evaluation of this theoretical framework has proved to be more challenging than expected for two reasons. First, there is a paucity of data sets measuring multitrophic interactions for complete communities integrated by several types of interactions (e.g. including competition, predation, pollination or parasitism), Second, the current coexistence models are complex and the number of parameters estimate grows exponentially with the number species in the community, making them impractical for real world communities. To solve this conundrum, we nood to find new ways to reconcile the power of large data sets with models rooted in solid theory. The use of Machine Learning techniques has revolutionized the predictive ability of several complex problems by learning patterns from data, but Machine Learning algorithms are traditionally non-interpretable, and hence disconnected from theory. Here we propose to use in development rule-based algorithms to simplify parameter estimation without losing interpretability. In addition, we will complete two unique highly resolved empirical multi-trophic datasets comprising complete communities in Spain and Canada. To tight together data and models, we choose a key question at the forefront of coexistence theory: can computer techniques help predicting the species interaction structure that enhances multi-species coexistence?
Understanding biodiversity maintenance is central to ecology, especially on the face of human induced environmental change and the alarming rates of biodiversity loss. We have made great progress in building solid mathematical models able to predict coexistence among interacting species across trophic levels. These advances include recent conceptual and mathematical toolboxes developed by our group allowing the simultaneous assessment of coexistence on complete communities, composed by several trophic levels, for example between plants, pollination, and herbivores. However, the empirical evaluation of this theoretical framework has proved to be more challenging than expected for two reasons. First, there is a paucity of data sets measuring multitrophic interactions for complete communities integrated by several types of interactions (e.g. including competition, predation, pollination or parasitism), Second, the current coexistence models are complex and the number of parameters estimate grows exponentially with the number species in the community, making them impractical for real world communities. To solve this conundrum, we nood to find new ways to reconcile the power of large data sets with models rooted in solid theory. The use of Machine Learning techniques has revolutionized the predictive ability of several complex problems by learning patterns from data, but Machine Learning algorithms are traditionally non-interpretable, and hence disconnected from theory. Here we propose to use in development rule-based algorithms to simplify parameter estimation without losing interpretability. In addition, we will complete two unique highly resolved empirical multi-trophic datasets comprising complete communities in Spain and Canada. To tight together data and models, we choose a key question at the forefront of coexistence theory: can computer techniques help predicting the species interaction structure that enhances multi-species coexistence?