Machine learning models perform better than traditional empirical models for stomatal conductance when applied to multiple tree species across different forest biomes
Stomatal closure decreases water loss and is one of the main mechanisms that trees can use to mitigate drought-induced physiological stress. The adaptability of trees to drought is likely to be of increasing importance as climate changes occur around the world. Modelling stomatal regulation can help...
Guardado en:
| Autores principales: | , , |
|---|---|
| Formato: | article |
| Lenguaje: | EN |
| Publicado: |
Elsevier
2021
|
| Materias: | |
| Acceso en línea: | https://doaj.org/article/47652b4c8d7a4f499a9e4b26f0d7ccb1 |
| Etiquetas: |
Agregar Etiqueta
Sin Etiquetas, Sea el primero en etiquetar este registro!
|
| Sumario: | Stomatal closure decreases water loss and is one of the main mechanisms that trees can use to mitigate drought-induced physiological stress. The adaptability of trees to drought is likely to be of increasing importance as climate changes occur around the world. Modelling stomatal regulation can help improve our understanding of how forests respond to their environment. Traditionally, empirical models have been used to model stomatal responses, however these models cannot always capture nonlinear responses and their parameters are often difficult to measure. In this study various machine learning (ML) models were able to capture stomatal responses of multiple tree species. We showed that ML can be a useful tool for predicting stomatal response based on climate variables and species traits. A random forest model performed the best with an R2 of 75 %, compared to the empirical Ball-Berry stomatal conductance model (BWB) (R2 = 41 %). In this study, the use of a combined dataset consisting out of data from multiple studies were successfully used, showcasing the use of data across studies. This also allowed for an ML model to be trained on 36 tree species from 5 forest biomes, from measurements taken across 6 continents, instead of being limited to one species, increasing the versatility of the model. The importance of species-specific stomatal responses was highlighted, with the drought strategies used by plants significantly influencing stomatal responses and predictions. ML models were able to capture these trends parsimoniously without prior knowledge of the underlying physiology of the tree species. The quality of combined datasets are however still not desirable, and long-term data collection using standardized measuring protocols are required to increase the strength of ML models. Data taken across different climatic regions and vegetation types can also help improve the adaptability of ML models. Regardless of limitations on data accessibility, ML shows promise in modelling plant responses to changes in climate. Focus on the use of ML together with traditional models can help give further insight into various ecological mechanisms. |
|---|