Deep learning meets tree phenology modelling: PhenoFormer versus process‐based models DOI Creative Commons
Vivien Sainte Fare Garnot, Lynsay Spafford, J. Jelle Lever

и другие.

Methods in Ecology and Evolution, Год журнала: 2025, Номер unknown

Опубликована: Май 26, 2025

Abstract Predicting phenology, that is the timing of seasonal events plant life such as leaf emergence and colouration in relation to climate fluctuations, essential for anticipating future changes carbon sequestration tree vitality temperate forest ecosystems. Existing approaches typically rely on either hypothesis‐driven process models or data‐driven statistical methods. Several studies have shown outperform methods when predicting under climatic conditions differ from those training data, change scenarios. However, terms methods, deep learning remain underexplored species‐level phenology modelling. We present a neural architecture, PhenoFormer, prediction using meteorological time series. Our experiments utilise country‐scale data set comprising 70 years approximately 70,000 phenological observations nine woody species, focussing Switzerland. extensively compare PhenoFormer 18 different process‐based traditional machine including Random Forest (RF) Gradient Boosted Machine (GBM). results demonstrate outperforms while achieving significant improvements comparable performance best models. When similar our model improved over by 6% normalised root‐mean‐squared error (nRMSE) spring 7% nRMSE autumn phenology. Under involving substantial shifts between testing (+1.21°C), reduced an average 8% across species compared RF GBM, performed par with These findings highlight potential modelling call further research this direction, particularly projections. Meanwhile, advancements achieved can provide valuable insights species‐specific responses change.

Язык: Английский

Dealing With Two Stresses: Impact of a Damaging Spring Frost Followed by a Summer Drought on Saplings of Four Temperate Tree Species DOI Open Access
Na Luo, Yann Vitasse, Arthur Geßler

и другие.

Plant Cell & Environment, Год журнала: 2025, Номер unknown

Опубликована: Март 30, 2025

ABSTRACT Global warming increases the likelihood that temperate tree species will face damaging late spring frost (LSF) and severe summer drought during same growing season. However, interactive effects of these two stresses are barely explored. We investigated physiological growth responses Acer campestre , Fagus sylvatica Quercus robur petraea saplings to artificially induced LSF drought, focusing on stomatal gas exchange, carbon partitioning, nonstructural carbohydrates (NSCs), phenology growth. depleted NSCs changed allocation patterns 1 month after event. Additionally, decreased diameter increment root A. F. in current year. Drought affected exchange all species, reduced biomass Q. exacerbated detrimental effect robur's NSCs. Our findings indicate prioritized canopy restoration immediately LSF, favored reserve replenishment before until end Furthermore, we highlight risk year could push beyond their limits emphasize importance studying multiple stressors' interactions better understand threshold profoundly alter forest ecosystems.

Язык: Английский

Процитировано

0
Deep learning meets tree phenology modelling: PhenoFormer versus process‐based models DOI Creative Commons
Vivien Sainte Fare Garnot, Lynsay Spafford, J. Jelle Lever

и другие.

Methods in Ecology and Evolution, Год журнала: 2025, Номер unknown

Опубликована: Май 26, 2025

Abstract Predicting phenology, that is the timing of seasonal events plant life such as leaf emergence and colouration in relation to climate fluctuations, essential for anticipating future changes carbon sequestration tree vitality temperate forest ecosystems. Existing approaches typically rely on either hypothesis‐driven process models or data‐driven statistical methods. Several studies have shown outperform methods when predicting under climatic conditions differ from those training data, change scenarios. However, terms methods, deep learning remain underexplored species‐level phenology modelling. We present a neural architecture, PhenoFormer, prediction using meteorological time series. Our experiments utilise country‐scale data set comprising 70 years approximately 70,000 phenological observations nine woody species, focussing Switzerland. extensively compare PhenoFormer 18 different process‐based traditional machine including Random Forest (RF) Gradient Boosted Machine (GBM). results demonstrate outperforms while achieving significant improvements comparable performance best models. When similar our model improved over by 6% normalised root‐mean‐squared error (nRMSE) spring 7% nRMSE autumn phenology. Under involving substantial shifts between testing (+1.21°C), reduced an average 8% across species compared RF GBM, performed par with These findings highlight potential modelling call further research this direction, particularly projections. Meanwhile, advancements achieved can provide valuable insights species‐specific responses change.

Язык: Английский

Процитировано

0