Diminishing snowpack intensifies crop yields sensitivity to soil moisture in the northern hemisphere

Liu, ﻿Han; Xiao, Pengfeng; Zhang, Xueliang; Wen, Jingyu; Liu, Yantao; Liu, Hao; Wang, Haikun

doi:10.5194/egusphere-2026-617

Preprints

https://doi.org/10.5194/egusphere-2026-617

Preprints

15 Apr 2026

| 15 Apr 2026

Diminishing snowpack intensifies crop yields sensitivity to soil moisture in the northern hemisphere

Han Liu, Pengfeng Xiao, Xueliang Zhang, Jingyu Wen, Yantao Liu, Hao Liu, and Haikun Wang

Abstract. Crops face increasing threats from drought due to global warming, causing declines in yield. Snowpack, as a distinct seasonal water resource, contributes to soil moisture that can supply additional water during the growing season, thereby reducing exposure to soil moisture deficit and stabilizing yield. However, the response of crop yield to snowpack, and the degree to which snowpack influence crop yield sensitivity to soil moisture, remain insufficiently understood. In this study, we combined snowpack, climate, soil, and crop yield datasets in the Northern Hemisphere from 2000 to 2022 to assess how snowpack affects the relation between soil moisture and yield of maize, spring wheat, and soybean. Our analysis reveals that crop yields respond positively to variations in snowpack, with the area showing significant positive correlation accounting for 52.99 % for maize, 72.75 % for spring wheat, and 82.66 % for soybean. Snowpack plays a predominantly negative regulatory role, with greater snowpack exerting a stronger buffering effect against drought impacts. In particular, late-spring snowmelt is the most influential contributor to this buffering effect. Moreover, an earlier shift of snowmelt peaks from April to March intensifies the temporal mismatch between water availability and water demand of crops, aggravating yield reductions. Among different crops, maize and soybean exhibit the strongest sensitivity to snowpack changes. These findings highlight the crucial role of snowpack in regulating soil moisture-yield relation, informing strategies to safeguard food production under climate change.

Received: 02 Feb 2026 – Discussion started: 15 Apr 2026

Publisher's note: Copernicus Publications remains neutral with regard to jurisdictional claims made in the text, published maps, institutional affiliations, or any other geographical representation in this paper. While Copernicus Publications makes every effort to include appropriate place names, the final responsibility lies with the authors. Views expressed in the text are those of the authors and do not necessarily reflect the views of the publisher.

Download & links

Preprint (PDF, 1340 KB)

Supplement (901 KB)

Download & links

Han Liu, Pengfeng Xiao, Xueliang Zhang, Jingyu Wen, Yantao Liu, Hao Liu, and Haikun Wang

Status: final response (author comments only)

Subscribe to comment alert

RC1:
'Comment on egusphere-2026-617', Anonymous Referee #1, 04 Sep 2026
The sensitivity of crop yields to soil moisture is an important topic because of its relevance to global food security. This study investigates how reduced snowpack affects this sensitivity in the Northern Hemisphere using the ERA5-Land reanalysis dataset and gridded annual crop-yield datasets. Partial correlation analysis and a separate dynamic linear model are applied at each pixel to demonstrate how diminishing snowpack intensifies crop-yield sensitivity to soil moisture. I have two major concerns regarding the quality of the crop-yield dataset and the reliability of the statistical model. The details are as follows.
The crop-yield dataset used in this study does not appear to be a commonly used or well-established dataset. Although the authors mention that it was validated against 2,823 administrative units across 43 countries and showed good accuracy, they provide only a link to the dataset, without citing a relevant peer-reviewed paper presenting the validation results. Therefore, the reliability of the crop-yield dataset cannot be adequately assessed.

I did not find any results regarding model evaluation or uncertainty quantification for the statistical model used in this study. Although Figure S3 presents the regression coefficients, the authors do not explain how the model performance was evaluated. In addition, “R” is used in the caption of Figure S3, whereas the legend refers to the “regression coefficient.” The statistical model is also developed separately for each pixel using only 23 annual observations, which may introduce substantial uncertainty. Collinearity also warrants further investigation because the model includes five strongly interrelated snow variables. I therefore suggest carefully describing how the statistical model was evaluated, how collinearity was accounted for, and how uncertainty was quantified.

A relatively permissive threshold of p < 0.1 is used in this study. Would the main results and conclusions remain robust if the more conventional threshold of p < 0.05 were used?

Title: It might be better to use “crop-yield sensitivity”, and “Northern Hemisphere” should be capitalised.

Lines 20 and 185-190: The percentages 52.99%, 72.75%, and 82.66% appear to represent the proportion of positively correlated pixels among pixels with statistically significant correlations, rather than the proportion of the entire crop area. Figure 1 indicates that only approximately 5.32%, 9.00%, and 10.07% of the respective crop areas have significant positive correlations. The denominator must be stated explicitly; otherwise, the abstract substantially overstates the spatial extent of the relationship.

Although Text S1 in the Supplementary Material describes how the snowpack metrics were calculated, the main manuscript does not direct readers to it. It would be helpful to add a statement such as “Detailed calculations are provided in Supplementary Text S1.”

Growing-season definition: A fixed June–September growing season and February–May cold season are applied across all crops and Northern Hemisphere locations. This overlooks large spatial and crop-specific differences in planting, emergence, and harvest dates. It also conflicts with the later interpretation that soybeans respond differently because it is sown later. The analysis should use crop- and location-specific calendars or, at minimum, provide sensitivity tests using alternative seasonal windows.

Line 159-162 and Figure 2: The bin boundaries or centres, binning method, sample sizes, error-bar definition, and sensitivity units are missing.

Figure 3: How are the “dominant snowpack indicators” defined and identified? An explicit definition is needed.

The reference list requires careful checking, and the literature review could be improved. For example, the references in lines 532–534 and 478–480 appear to be duplicates, and the capitalisation of journal titles is inconsistent. The reference attributed to Fuyou et al. (2024) appears to concern agricultural-field segmentation rather than the gridded crop-yield dataset described in the Methods and should therefore be verified. It would also be helpful to review recent publications on snow, crop yield, and food security to ensure that important relevant studies have not been overlooked.
Citation: https://doi.org/10.5194/egusphere-2026-617-RC1
RC2: 'Comment on egusphere-2026-617', Anonymous Referee #2, 16 Sep 2026

This manuscript addresses an important and timely topic: the sensitivity of crop yields to snowpack-modulated soil moisture across the Northern Hemisphere. Integrating snowpack, climate, soil moisture, and crop yield datasets for maize, spring wheat, and soybean over 2000–2022 carries potential value. The research question is highly relevant to food security under climate change, and the multi-crop, spatially explicit analytical framework is, in principle, appropriate. Nevertheless, the manuscript has weaknesses in statistical inference, causal interpretation, and methodological transparency. Several key conclusions appear overstated relative to the supporting analyses.
Specific comments:

1.My primary concern lies with the statistical method used to quantify sensitivity, given the difficulty of confounding control. The authors first estimate pixel-level yield sensitivity to soil moisture, then use this estimated sensitivity as a response variable in a second-stage regression with snowpack metrics. This creates a generated regressor problem: first-stage coefficient estimates contain measurement error, and ignoring this uncertainty in the second stage can produce biased estimates and inflated significance.

In addition, snowpack metrics such as SWEmax, SCED, and spring snowmelt are likely highly collinear. Comparisons identifying their “dominant” roles are therefore unstable. I suggest testing whether the identity of the dominant indicator remains robust across regions and crop types.
2.The manuscript relies heavily on pixel-wise correlation and regression analyses, adopting a significance threshold of p<0.1. This threshold is relatively lenient, and the authors do not state whether multiple testing correction was implemented. Gridded datasets typically exhibit strong spatial autocorrelation, which can severely inflate the effective sample size and produce overly optimistic p-values.
3.There is a inconsistency between the abstract and the results. The abstract states that “the area showing significant positive correlation accounting for 52.99% for maize, 72.75% for spring wheat, and 82.66% for soybean.” However, the results indicate that these percentages refer to the proportion of positive correlations among significantly correlated pixels, not the proportion of the entire study area.
4.The actual significant positive area proportions are much smaller: 5.32% for maize, 9.00% for spring wheat, and 10.07% for soybean. For maize, the significant positive area is only slightly larger than the significant negative area (4.72%). This weakens the claim that snowpack has a consistently dominant positive effect on maize yields.
5.The study defines June–September as the average growing season for spring crops and February–May as the snow accumulation and melt period. However, maize, spring wheat, and soybean have very different sowing dates, growth stages, and peak water demand periods. These also vary strongly across the Northern Hemisphere.
6.The study defines rainfed grid cells as those with irrigated area fraction less than 30%. This threshold is quite high and may include substantial irrigated agriculture. I recommend using a stricter threshold, such as <10%, for the main analysis, and testing sensitivity to <10%, and <20%.
7.Moreover, the Global Map of Irrigation Areas v5 represents conditions circa 2005, while the study period is 2000–2022. Irrigation expansion over this period could bias the results. This limitation should be explicitly discussed, and dynamic irrigation data should be used if possible.
8.There are inconsistencies in how “sensitivity” is discussed. Figure 2 suggests that spring wheat has numerically higher soil moisture sensitivity than maize and soybean. Yet the abstract states that “maize and soybean exhibit the strongest sensitivity to snowpack changes.” These are different concepts: sensitivity to soil moisture versus sensitivity to snowpack. The manuscript must define these terms clearly and use them consistently.
9.Furthermore, the relationships in Figure 2 are not consistently monotonic. Some intervals have large error bars, and the pattern varies by crop and snowpack metric. The statement that sensitivity “declines with higher SWEmax, later SCED, and greater spring snowmelt volume” is therefore too general and should be qualified.
10.The title and abstract contain grammatical errors: “crop yields sensitivity” should be “crop yield sensitivity”; “crop yields responses” should be “crop yield responses.”
11.“accounted by climate change” should be “attributed to climate change.”
12.The statement “significant decline in SWEmax (7.66%)” is ambiguous. Is this an area proportion or a decline magnitude? If it is an area proportion, write “7.66% of the study area.”
13.Reference formatting is not fully consistent. Some entries lack journal names, volume/issue numbers, or DOIs.

Citation: https://doi.org/10.5194/egusphere-2026-617-RC2

Han Liu, Pengfeng Xiao, Xueliang Zhang, Jingyu Wen, Yantao Liu, Hao Liu, and Haikun Wang

Supplement

https://doi.org/10.5194/egusphere-2026-617-supplement

Han Liu, Pengfeng Xiao, Xueliang Zhang, Jingyu Wen, Yantao Liu, Hao Liu, and Haikun Wang

Viewed

Total article views: 469 (including HTML, PDF, and XML)

HTML	PDF	XML	Total	Supplement	BibTeX	EndNote
297	142	30	469	54	27	26

HTML: 297
PDF: 142
XML: 30
Total: 469
Supplement: 54
BibTeX: 27
EndNote: 26

Views and downloads (calculated since 15 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	137	35	8	180
May 2026	90	40	4	134
Jun 2026	12	5	2	19
Jul 2026	16	20	4	40
Aug 2026	12	21	2	35
Sep 2026	30	21	10	61

Cumulative views and downloads (calculated since 15 Apr 2026)

Month	HTML	PDF	XML	Total
Apr 2026	137	35	8	180
May 2026	90	40	4	134
Jun 2026	12	5	2	19
Jul 2026	16	20	4	40
Aug 2026	12	21	2	35
Sep 2026	30	21	10	61

Viewed (geographical distribution)

Total article views: 448 (including HTML, PDF, and XML) Thereof 448 with geography defined and 0 with unknown origin.

Country	#	Views	%

Latest update: 16 Sep 2026

Short summary

Climate change is intensifying drought and threatening food production, yet the role of snowpack remains unclear. We find that greater snowpack benefits crop yields by supplying additional soil moisture. Especially late-spring snowmelt plays the most role, supplying effective moisture when crop water demand is high. However, snowmelt peaks shifting from April to March creates mismatches between water supply and demand, causing yield losses. Maize and soybean are more sensitive to these changes.


Total:	0
HTML:	0
PDF:	0
XML:	0