the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
| 28 Jan 2026
Phenology-modulated Crop Responses to the 2024 Spring-Early Summer Compound Dry-Hot Event in the North China Plain
Abstract. Compound dry-hot events are intensifying under climate change and pose growing risks to agricultural production. From April to June 2024, the North China Plain (NCP) experienced an extreme compound dry-hot event. Using satellite-based normalized difference vegetation index (NDVI), gross primary productivity (GPP), and crop yield statistics, this study quantified crop growth responses and identified the dominant climatic drivers during this event. The climate anomaly was characterized by pronounced warming in April and June, a continuous decline in precipitation and soil water from April onward, and a record-high vapor pressure deficit (VPD) in June, forming a persistent dry-hot stress. NDVI and GPP increased markedly in April and remained slightly positive in May, but both collapsed to their lowest levels since 2000 in June. Consistent with these vegetation signals, provincial yield statistics and experimental plot observations showed increased winter wheat yields but reduced summer maize yields. Sensitivity and contribution analyses revealed distinct phenology-modulated mechanisms: in April, elevated temperatures and vegetation carryover effects comparably enhanced vegetation activity in winter-wheat-dominated croplands; in May, vegetation dynamics were controlled almost entirely by the previous-month carryover effect, reflecting the growing influence of accumulated vegetation state; and in June, as winter wheat reached maturity and newly sown maize entered early establishment, VPD emerged as the primary limiting factor, strongly suppressing photosynthetic activity and seedling establishment. These findings demonstrate how phenological transitions modulate crop vulnerability to compound dry-hot events and provide useful insights for agricultural early warning, crop management, and climate adaptation strategies in the NCP.
- Preprint
(2424 KB) - Metadata XML
-
Supplement
(2789 KB) - BibTeX
- EndNote
Status: open (until 21 Mar 2026)
- RC1: 'Comment on egusphere-2026-142', Anonymous Referee #1, 26 Feb 2026 reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 94 | 47 | 13 | 154 | 21 | 11 | 24 |
- HTML: 94
- PDF: 47
- XML: 13
- Total: 154
- Supplement: 21
- BibTeX: 11
- EndNote: 24
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This manuscript investigates crop responses to the 2024 spring–early summer compound dry-hot event in the North China Plain (NCP) using NDVI, FluxSat GPP, climate reanalysis data, and yield statistics. The topic is timely and relevant given increasing climate extremes and food security concerns.
However, major methodological limitations and interpretational weaknesses severely compromise the reliability of the results and conclusions. In particular, the coarse land-cover masking, mixed-pixel contamination, lack of crop-type discrimination, and limited model rigor undermine the attribution of NDVI/GPP anomalies to crop growth dynamics. Given these fundamental issues, I cannot recommend publication in its current form.
Major concerns
1.Cropland mask is too coarse and does not isolate crop signals
The study masks non-cropland grids using the MODIS MCD12C1 land-cover product (0.05° resolution). This spatial resolution is insufficient for isolating crop vegetation signals.
Within “cropland” pixels in the NCP, substantial non-crop components remain, including shelterbelts and windbreak trees, orchards and agroforestry vegetation, grasses and weeds, rural settlements and impervious surfaces, bare soil and fallow land, etc.
Because NDVI and GPP respond to all photosynthetically active vegetation, these components may contribute significantly to the observed anomalies. Consequently, the reported NDVI/GPP signals cannot be confidently attributed to crop growth.
Higher-resolution land-cover products (10–30 m) are available for NCP and should be used to better isolate cropland signals.
Moreover, all datasets were resampled to 0.1° (~10 km) resolution to match the climate data. This coarse aggregation is not strictly necessary. NDVI and productivity metrics could be derived from higher-resolution sensors (e.g., Landsat or Sentinel), which would better represent field-scale variability.
At the 0.1° scale, even when retaining only cropland pixels (from the 0.05° mask), each grid cell still contains heterogeneous land cover and management conditions. The vegetation signal therefore represents landscape-level vegetation activity, rather than crop-specific responses.
2.Lack of discrimination among winter wheat, maize, and other crops
The NCP follows a winter wheat–summer maize rotation, but this study does not distinguish crop types within pixels.
At any time, pixels may include: wheat at heading or maturity stage; harvested wheat fields; newly sown maize; other crops (vegetables, peanuts, cotton, orchards); fallow land.
Because NDVI and GPP vary strongly among crop types and phenological stages, the vegetation signal cannot be uniquely attributed to wheat or maize dynamics.
3.Harvest transition and bare soil exposure confound June signals
The manuscript acknowledges that winter wheat harvest and maize sowing occur in early June. During this transition, canopy removal reduces NDVI, bare soil exposure dominates reflectance, and early maize emergence contributes little greenness. Thus, the June NDVI/GPP collapse may partly reflect land surface transition rather than physiological drought stress. This confounding effect is not adequately quantified.
4.Uncertainty of FluxSat GPP over croplands is insufficiently addressed
The FluxSat GPP product is derived from MODIS reflectance and flux tower calibration. While useful, it carries uncertainties in intensively managed croplands, especially under irrigation and rapid phenological transitions. I would expect the author to quantify expected GPP uncertainty over croplands in NCP; performance in irrigated agroecosystems as NCP region is heavily irrigated.
5.Attribution risks conflating correlation with causation
The analysis demonstrates statistical associations between climate anomalies and vegetation indices but does not establish causal mechanisms. Given mixed pixels, heterogeneous crop composition, and non-crop vegetation contributions, attributing anomalies directly to crop growth responses remains inferential rather than mechanistic.
6.Multiple Linear Regression framework is inadequate for extreme events
The study applies a linear regression model to characterize responses under an extreme compound event. However, the vegetation responses to heat and drought are nonlinear; Thresholds and tipping points are well documented; Interactions among VPD, soil moisture, and temperature are nonlinear; Phenological transitions introduce structural discontinuities. A linear model is therefore unlikely to capture system behavior under extreme stress.
7.Model performance reporting and interpretation are insufficient
The manuscript reports R² ≈ 0.6 only for GPP, while performance metrics for NDVI are not provided.
The interpretation over such metrics are unclear. In the regression formulation, is the response variable and is included as a predictor. Normally the R2 is to quantify the predicted Y and the observed Y, but in the Xs, is also there, which is a prediction from the model, I am not sure how this R2 is calculated and what the R2 represents, this requires additional elaboration.
Moreover, multicollinearity assessment need to be done and show how this could affect the model explanation power under extreme conditions.
8.Definition issue for anomaly
Another very important issue is the definition of anomaly in this study, for example L154 (formula 2), the anomaly is defined as the yield in 2024 minus the mean yield within 2001-2023 and then divide by mean yield, this is more of the deviation from the mean condition.
Within the years 2001-2023, the yields are varying too, there will be a "normal" range of the variation, the part for the 2024 deviation outside this "normal" range is the real anomaly.
I would ask the author to redo the calculation by defining a true anomaly signal - Standardized anomaly (z-score).
This should apply for all the variables: yield, GPP, NDVI, etc.
Other concerns
Irrigation is a dominant control on crop productivity in the NCP, yet it is not explicitly represented.
Crop stages are inferred rather than validated using phenological datasets.
Same as above, for the crop species over the land, they are based on the assumptions (maize-wheat rotation).
Specific comments:
L30-32, NDVI and GPP are still increasing markedly in April and May during the dry-hot year? Which crops is this? Maize or wheat or any other crops? How do you know the decrease in June is solely due to the dry-hot event? Not because of harvesting, tillage, or other management practices.
For NDVI and GPP, as they are from satellite, it is also very likely that the signal in this month is different from other months or other years due to image saturations or local weather conditions. How do you know it is not the satellite issue?
L36-37, any proof of this?
L97-98, Land cover data with 0.05x0.05 degree is not acceptable as you are using it to keep only the cropland pixels.
L101-104, this is a big problem, you cannot simply assume that there are only maize and wheat on the field at a certain month, please use crop species masks at fine resolution. From your current setting, inside a 0.05x0.05 degree pixel or 0.1x0.1 degree pixel, there are too many fields, they can be covered with different crops other than maize/wheat, and even for maize/wheat, they might be in different growing stages, with different management (different sowing time, irrigation, weed/pest control, tillage, harvest), which could lead to a very different optical characteristics that were captured by the satellite, and all of them are mixed together, you cannot simply attribute them to a simple maize/wheat at the same growing stage.
L128-129, for NDVI, why not derive from landsat or sentinel with 10/30m resolution.
L129-131, need to validate the GPP data over croplands in NCP, with/without irrigation.
L136, with such coarse resolution, I doubt the model validity under phonological transitions, vegetation indices during harvest transitions may reflect canopy removal, bare soil exposure, land surface changes, rather than climate-driven physiological stress. How do you know if the trend you find in your data and model really reflects climate-driven physiological stress, rather than canopy removal, bare soil exposure, or land surface changes.
L139-140, only 1 site?
L145-156, I am not sure if you should call this anomaly, as I also mentioned before, this is more of the derivation from the mean condition. Please revise it with z-score. This should apply for GPP and NDVI too, as you are discussing the anomaly of them too.
L157-180, I strongly doubt the capability of a linear model can capture these relations under extreme weather conditions. Please compare with other machine learning algorithms.
L172, sensitivities represent statistical associations rather than direct physiological responses.
L171-174, not clear how this sensitivity is defined and calculated, please elaborate more.
L175-180, about the contribution, as in the model, previous-month vegetation anomalies are also included in the model as the predictor, can you clarify the ecological meaning of this term and discuss its potential influence on attribution results.
L184, For figure 2a, i, m, q, the shading area is the condition within 1σ? The lines indicated the condition for different years? From what we can observe, there are many years that outside this 1σ, why not included all those years for analysis the extreme condition's impact on crop growth and yield?? Otherwise, you should exclude those years in your baseline condition.
L221-222, there might be correlations, but not the causations?
L232-233, How do you separate the impact of "practice" from the results? June includes wheat harvest; bare soil exposure; maize sowing period. The remote sensing signals may reflect land surface change; canopy removal; crop rotation timing and not purely physiological stress. This is the most important interpretational risk.
L240-242, the seeding density in different years are the same? This is the dry weight or what? What is the water content when balancing for different years? Is there irrigation? What is this error bar indicating? Error from what? If this is the site observation, why is there this error bar for a single site?
L298, what are these R2 showed in figure 6a-c? The predicted GPP vs. the observed GPP? Please plot the scatterplot with density as the color, instead of the maps. In your formula, you have Y_t as the response variable, and Y_t-1 as the predicter, how do you calculate this R2? And how to interpret it?