| 06 Jul 2026
Early leaf senescence reveals drought stress thresholds and mortality risk across temperate forests
Abstract. Drought-induced early leaf senescence is increasingly widespread in temperate forests, yet the thresholds triggering this response and whether it reflects adaptive drought avoidance or progressive stress remain poorly constrained. Combining six years (2018–2023) of Sentinel-2 observations with atmospheric and soil drought indicators and canopy mortality maps across German beech (Fagus sylvatica) and oak (Quercus robur, Q. petraea) forests, we show that early senescence emerges once cumulative drought stress exceeds species-specific hydraulic thresholds. Sustained atmospheric vapour pressure deficit (VPD) followed by late-summer soil water depletion sharply increases senescence probability, with critical thresholds occurring after six consecutive weeks of daily maximum VPD above 1.9 kPa in beech and 2.1 kPa in oak, and after two weeks of root-zone soil water potential (ψsoil) below −0.8 MPa (beech) and −0.9 MPa (oak). Sensitivity differs between forest types: beech-dominated stands respond more strongly to soil drought, whereas oak stands are more sensitive to atmospheric drought. Elevated spring VPD further amplifies early senescence risk in both species, pointing to cross-seasonal legacy effects in canopy stress responses. Critically, early senescence – and particularly its recurrence across consecutive years – is associated with significantly elevated canopy mortality, indicating progressive hydraulic stress rather than protective drought avoidance. Clustering forests by senescence dynamics and mortality outcomes reveals distinct drought-response types ranging from resistant stands buffered by favourable soil and climate conditions to vulnerable lowland forests with elevated mortality risk. Together, these results establish early leaf senescence as a landscape-scale indicator of drought stress severity and emerging mortality risk, providing a new framework for identifying forest vulnerability under intensifying climate extremes.
This manuscript investigates premature leaf senescence in European beech and oak forests, associating it with drought stress and canopy mortality. By integrating Sentinel-2 data, field phenology, drought indicators, and mortality records, it offers insights into drought-related forest responses across large spatial scales. However, I have some major concerns about the validation of the senescence metric, the handling of uncertainty in subsequent analyses, and the interpretation of statistical associations as drought thresholds or drivers. Methodological choices, validation results, and interpretations are not clearly separated between the Methods and Results sections, making it hard to follow how processing and modeling steps lead to the findings.
The key analyses heavily rely on satellite-derived senescence onset and early senescence classification, while agreement with field observations is only moderate, and uncertainty is insufficiently incorporated into drought and mortality analyses. Methodological choices regarding spatial aggregation, early senescence definition, predictor selection, and statistical interpretation require further justification and sensitivity analyses to evaluate the robustness of the conclusions.
See comments below:
Line 1: The title should be formulated more cautiously due to the substantial uncertainty in the analysis and the consequent limitations on the strength of the conclusions.
Line 10-25: The Abstract provides limited details on the study design and methodology. Specifically, it lacks clarity on how senescence onset was derived from Sentinel-2 data, defined, validated against field observations, and how its relationships with atmospheric and soil drought and subsequent mortality were quantified. Since the reported drought thresholds and mortality associations depend on these methods, I suggest briefly outlining the key analysis (senescence retrieval and validation, definition of premature senescence, spatial scale, and statistical analyses) in the Abstract. This would help readers better understand the scope and robustness of the findings.
Line 43-44 & 47-48: When these tree species are first mentioned, I suggest providing their full common and scientific names, for example European beech (also known as common beech; Fagus sylvatica L.) and pedunculate oak (Quercus robur L.). This would help avoid potential confusion, particularly for readers in the field of forestry.
Line 54-55: Any citations for “However, more recent evidence indicates that early senescence is increasingly associated with lasting hydraulic damage”? The term “lasting hydraulic damage” is vague. Please clarify which specific physiological processes or impairments are being referred to, such as drought-induced xylem embolism, persistent loss of hydraulic conductivity, or other types of hydraulic dysfunction.
Line 63-66: Please provide appropriate references to support this statement, particularly since the use of remote sensing data is emphasized in the subsequent text.
Line 110-114: This information should be presented in the Results section. The estimation of senescence onset (SO) begins by evaluating Sentinel-2-derived vegetation indices against ICP Forests field observations of coloration and litterfall onset. The comparison of vegetation indices and amplitude thresholds (which led to the selection of NDRE with an 80% threshold) represents an empirical finding rather than a methodological procedure.
Line 115-121: The spatial and observational matching between the ICP Forests measurements and satellite-derived SO is not sufficiently clear. ICP observations are made at the level of individual trees within field plots, whereas Sentinel-2 provides spatially integrated canopy signals at the pixel level. You should explicitly describe how individual tree observations and their associated DoYs were converted into a field reference SO for comparison with the satellite estimates. For example, were tree-level observations averaged at the plot level? Or was each individual observation independently paired with the satellite-derived SO? This becomes particularly important for validation at different spatial aggregation scales (50 m, 1 km, and 10 km; Fig. S2). When multiple field observations or plots occurred within an aggregated satellite grid, how were these observations handled, and what constituted a single independent field–satellite pair used to calculate the reported correlations and RMSE? And in Fig. S1, “…Note that the remotely sensed SO was here averaged over an area of 40x40 km around the ICP Forests plots”, I am confused here: Which specific scale is being validated? 40x40 km?
Line 230: The first main result presents the spatial and temporal patterns of senescence onset and classifies early senescence based on the derived SO product. However, the uncertainty associated with SO retrieval is not addressed at this stage. Validation results in the Supplement indicate only moderate agreement between satellite-derived and field-observed SO, with temporal errors on the order of several weeks (RMSE). This uncertainty is particularly important given the use of a fixed DoY 244 threshold to distinguish early from normal senescence: observations near this threshold may easily change classification within the uncertainty range of the SO estimate. Figure 1 however presents the resulting senescence patterns as deterministic, which may give an overly certain impression of the spatial extent and frequency of early senescence.
Line 235: This sentence should be placed in the Discussion section.
Line 240–244: The thresholds for VPD and ψsoil are presented to two decimal places and were derived from logistic models using SOearly as a binary outcome. However, the satellite-derived SO shows only moderate agreement with field observations, with RMSEs on the order of several weeks. This raises an important question: can an SO product with temporal uncertainty of this magnitude reliably support a strict binary classification of early versus normal senescence based on a single cutoff (DoY 244)?
Line 261: This statement implies a causal relationship and a mechanistic explanation, but the observational model does not appear to directly support this interpretation. Including spring drought conditions might reveal additional statistical associations with SOearly, but this alone does not prove that drought stress accumulated across seasons caused the observed senescence response. Please consider presenting the statistical results more neutrally in the “Results” section (e.g., in terms of associations or model performance) and reserve the interpretation of cumulative drought effects for the “Discussion” section.
Line 289-290: This statement extends beyond the statistical results presented here. The observed link between early senescence and subsequent mortality does not alone prove that early senescence “serves as a landscape-scale indicator” of declining forest vitality or emerging mortality risk. Such a claim requires further evidence, particularly that early senescence predicts mortality beyond concurrent or prior drought stress and other factors. Please present the association more neutrally in the Results section and keep broader interpretation in the Discussion.
Line 343-344: The claim that these patterns “closely mirror independent reports of observed drought impacts reported across Germany during 2018–2023” requires references. Please specify the independent reports or datasets referenced and provide proper citations.
Line 370: The exact drought thresholds should be interpreted cautiously. These model-derived values depend on a binary SOearly response, which is uncertain due to satellite-derived SO data and the selection of the DoY 244 cutoff. They should not be considered precise physiological or ecological limits, and their uncertainty and sensitivity to methodological choices must be acknowledged.
Line 377-379: I am not convinced that agreement with independent field measurements alone justifies the “broader generality” of the identified threshold across Central Europe. Due to uncertainties in SO retrieval, binary classification, and the model-derived threshold, I suggest revising this conclusion unless more direct and systematic external validation is provided.
Line 390: The claim appears stronger than supported by the observational analyses. Although the models show associations between SOearly and soil/atmospheric drought across seasonal periods, this does not establish a causal “driver” relationship or confirm a cumulative physiological effect. Please distinguish statistical associations from mechanistic interpretations of cumulative drought stress and moderate the wording accordingly.
Line 451-462: The discussion of the clustering analysis is hard to follow because the three clusters identified in the Results are not explicitly named or referenced here. It would be helpful to clearly distinguish between the variables used to define the clusters and the independent environmental characteristics used to interpret them. Several characteristics discussed, such as senescence, drought sensitivity, and mortality response, seem directly related to the clustering variables. Please clarify how the response types assigned to the clusters were derived and which findings independently support these interpretations.
Line 471-476: This sentence is too long and should be split for clarity. More importantly, the “sensitivity patterns identified here” do not provide a “mechanistic basis” for the subsequent interpretation. The analyses show statistical associations between drought variables and early senescence but do not directly examine the physiological mechanisms behind these relationships. Please differentiate the observed statistical patterns from mechanistic evidence and revise the wording accordingly.