Search

• Leaf senescence is a dynamic process starting from peak greenness timing (PGT). • Leaf senescence date (LSD) depends on PGT and senescence duration (SD). • Responses of LSD to late-season drought are quantified over Mongolian grassland. • Changes in PGT and SD contribute differently to LSD's response to drought. • Drought-associated reduction in annual maximum EVI decreases late-season GPP. Leaf senescence is a key trait for vegetation in temperate regions to overcome winter temperatures, and extreme drought events have profoundly affected its timing. The effects of drought on autumn phenology have been investigated by correlating leaf senescence date (LSD) to multiple drought indices. However, leaf senescence is a dynamic process starting from peak greenness timing (PGT) and lasting several months (senescence duration, SD). How PGT and SD changes affected the LSD's response to drought is still poorly understood. Here, we investigated LSD's response to drought in the Mongolian grassland based on PGT and SD. We first extracted the LSD, PGT, and SD from the MODIS Enhanced Vegetation Index (EVI) during 2001–2020 and determined the late-season drought events (LDE) using the Standardized Precipitation Evapotranspiration Index. Subsequently, we quantified the response of LSD to LDE as the difference between drought and normal years (ΔLSD), and then quantified the contributions of changes in PGT and SD on ΔLSD. We also analyzed the effects of drought on late-season productivity. The results showed that LSD may advance or delay under LDE, but the earlier LSD (in 68.4 % of areas) was mainly distributed in warmer and drier regions. Earlier LSD under drought was more controlled by earlier PGT, while later LSD under drought was more controlled by extended SD. Changes in late-season productivity under LDE were more correlated with changes in annual maximum EVI (EVI max) than ΔLSD. These results enhanced our understanding of how LSD of temperate grassland responds to drought events and emphasized the importance of investigating the response of autumn phenology to drought from a process-based perspective. [Display omitted] [ABSTRACT FROM AUTHOR]

• We used three machine learning (ML) algorithms to simulate leaf senescence date. • The ML-based models outperformed process-based models for six tree species. • Interpretable ML algorithms could decipher the effects of multiple driving factors. Predicting tree phenology accurately is essential for assessing the impact of climate change on ecosystems. However, the current process-based models are still hard to fully explain and quantify the effects of multiple biotic and environmental factors (e.g., spring phenology, local adaptation, productivity, climatic variables) on autumn phenology. Using leaf senescence data (1980-2015) of 232766 site-year records (3020 sites and 6 deciduous tree species) in Europe, we first calibrated and evaluated 4 process-based models developed by previous studies. Subsequently, by using different machine learning (ML) algorithms (RF, EBM, and GAMI-Net), we developed 3 ML-based models for predicting the leaf senescence date of the same species and quantifying the importance and response function of 63 biotic or environmental variables. We found that the root mean square error (RMSE) of process-based models (averaged from all species) for the test dataset ranged from 11.97 to 12.91 days. The ML-based models outperformed process-based models for all species, with RMSE ranging from 10.01 to 10.58 days. For most species, the recently developed ML algorithms (EBM and GAMI-Net) are more effective than the classical RF algorithm developed in the early 21th century. Besides the temperature and photoperiod in autumn, the geographic factors (especially elevation, longitude, and latitude) were identified as the most important variables in the ML-based models, implying that leaf senescence date is an adaptive trait. Furthermore, for most species investigated, earlier leaf-out dates tended to advance the leaf senescence date. Our results highlight that the ML algorithms not only could effectively improve the performance of the process-based models for predicting the leaf senescence date, but also help to understand the nonlinear and interactive effects of multiple driving factors on autumn phenology. [ABSTRACT FROM AUTHOR]

Autumn phenology can regulate climate-biosphere interactions and net primary production within the ecosystem. However, studies modeling spatiotemporal variations in leaf coloring date (LCD) remain limited, especially for species-specific phenology on a continental scale. Aiming to simulate spatiotemporal variations in LCD in three widespread tree species ( Ulmus pumila , Fraxinus chinensis, and Robinia pseudoacacia ) across China, we used phenological observation records acquired from the China Phenology Observation Network (CPON) during 1963–2010 to establish and compare three LCD models (multiple regression (MA), temperature-photoperiod (TP), spring-influenced autumn (SIA)). Subsequently, we simulated the mean LCD of the three species using the most effective model and examined the effect of geographical factors (i.e., latitude, longitude, and altitude) on LCD through multiple regression analysis. Empirical Orthogonal Function (EOF) analysis was applied to identifying the most extensive and influential spatial modes of LCD variability and how they changed with time. The results showed that: (1) The LCD of F. chinensis was fitted better with the statistical model using monthly temperature as the independent variables (MR model). The LCD of F. chinensis was delayed by a temperature rise in August and September, but advanced by a temperature rise in May and June. The LCD of U. pumila and R. pseudoacacia was fitted better with the TP and SIA models, in which the photoperiod determined the date when the cold temperature started to accumulate. (2) The simulated mean LCD of U. pumila , F. chinensis, and R. pseudoacacia was October 6, October 16 and October 22, respectively. Latitude, longitude, and altitude had a significant influence on mean LCD of the three tree species. With increasing latitude and altitude, the LCD of all three species became earlier. However, the impact of longitude on the mean LCD varied among species. (3) For all the three species, the first EOF mode presented a consistent pattern of LCD variability across space, suggesting that an earlier or later LCD occurred simultaneously in the whole China. Meanwhile, the second EOF mode exhibited contrary signals of LCD variability in the north and south for F. chinensis and R. pseudoacacia . Over the past 50 years, the LCD of all the three species has delayed. The delaying trend revealed by the first EOF mode was 1.25 ( p < 0.01), 0.21 ( p < 0.01), and 0.53 days/decade (not significant) for U. pumila , F. chinensis and R. pseudoacacia , respectively. These results provide the basis for a better understanding of the phenology process in autumn and how it responds to climate change. [ABSTRACT FROM AUTHOR]

• Chilling accumulation (CA) and heat requirement (HR) are determined for 14 species. • An inconsistent CA-HR relationship between years was found. • Using one-year experimental data to simulate spring phenology is inappropriate. • Current chilling models need to be improved by performing more experiments. The controlled experiments are increasingly used to examine the effect of chilling on the spring phenology of temperate woody plants. Previous studies demonstrated a delaying effect of lack of chilling on spring events due to a negative relationship between chilling accumulation (CA) and heat requirement (HR). However, most experiments only lasted for one year, and limited studies examined the consistency of the CA-HR relationship between years. Here, we conducted a twig experiment with 14 temperate woody species in two winter seasons (2018/2019 and 2019/2020) to assess the between-year difference in the impact of chilling on HR of flowering or leaf-out. The CA and HR for each species in each year were estimated based on the common-used chilling and forcing model. The results showed that 8 of 14 species exhibited a significantly different slope or intercept of the linear function of HR against CA between two winter seasons. Using the first-year CA-HR function to simulate the second-year spring events would result in large uncertainties with root-mean-square errors ranging from 1.3 to 9.9 days. In addition to the original chilling model, we further selected an optimal one for each species from another 8 chilling models. The optimal chilling models varied among species and could reduce the RMSE of phenological simulation of 10 species by 0.1–3.7 days. Our results suggest that using one-year experimental data to develop the phenological model or estimate the chilling and heat requirement of spring events is unsuitable. Further studies are still needed to better quantify the rate of chilling under different temperature conditions for various species. [ABSTRACT FROM AUTHOR]