the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
A physico-chemical framework for explaining the generation mechanism of Nocturnal Soil CO2 Uptake in Deserts
Abstract. Nocturnal soil CO2 influx (Fs) has been widely documented in global deserts, while the underlying mechanisms remain unclear. Based on 563 observations from 30 published studies across five major desert regions in the world, we employed multiple linear mixed-effects models (LMMs), random forest (RF) regression, polynomial regression, and structural equation modeling (SEM) to quantify the effects of air‑soil temperature difference (ΔT = Ta - Ts), soil water content (θv), soil pH, and ln(clay/sand) on soil nocturnal CO2 influx (Fs), and to establish a physico-chemical framework for explaining the generation mechanism of Fs. The results displayed a global mean Fs of −0.195 μmol m-2 s-1, with the strongest CO2 uptake occurring in the Central Asian Desert Area and the weakest in the North American Desert Area. Fs showed a significantly positive correlation with ΔT and negative correlations with θv, pH, and ln(sand/clay). SEM revealed that ln(sand/clay) and θv influenced Fs both directly and indirectly. The relationships of ΔT, θv, pH, and ln(clay/sand) with Fs suggested exerted pronounced effects on nocturnal CO2 influx in desert soils. The correlation analysis demonstrated that these environmental factors may serve as important drivers in the generation of nocturnal soil CO2 influx. Integrating our results ad previous studies, we proposed a comprehensive physico‑chemical mechanism framework. In this framework, thermal convection driven by the temperature difference served as the physical driving force, continuously transporting atmospheric CO2 into soil. One fraction is temporarily stored as gas in soil pores, and the other fraction enters chemical reactions under alkaline conditions. The above physico‑chemical processes generated the nocturnal CO2 influx process in desert soils.
- Preprint
(1464 KB) - Metadata XML
-
Supplement
(1199 KB) - BibTeX
- EndNote
Status: open (extended)
-
CC1: 'Comment on egusphere-2026-4692', Wu Xiao, 22 Sep 2026
reply
-
CC4: 'Reply on CC1', Jia-Bin Liu, 08 Oct 2026
reply
Thank you for your valuable time and constructive comments on our manuscript.
Comment 1:Line 15,16 The ratio is written as both ln(clay/sand) and ln (sand/clay) in different places. Please standardize.
Re: The corresponding part has been modified.
We have standardized the soil texture index as ln(clay/sand) throughout the entire manuscript. Modification locations: Abstract, Section 2.2, Section 3.2 (Pearson correlation, RF variable importance, partial dependence plots, hierarchical partitioning), Section 3.4 (SEM), and all relevant figure captions.
Comment 2:Line 36-40, the second paragraph lists previous single factor studies but does not clearly summarizethe major research gap (lack of cross‑site, multi‑factor interaction analysis).
Re: The core research gap has been explicitly supplemented at the end of the second paragraph in Introduction.
“Furthermore, most hypotheses come from long-term monitoring at single sites, where soil texture and pH vary little and their effects cannot be reliably separated from site-specific conditions. As a result, they cannot fully explain how uptake responds to changing environmental conditions.” (Line 41-44)
Comment 3:Duplicate section heading (Section 2.2)– There are two sections labeled "2.2 Data Extraction." The second should be renumbered as "2.3 Statistical Analyses" (or similar), with subsequent sections renumbered accordingly (currently 2.3 → 4, etc.).
Re: The duplicate section heading has been corrected and the section numbering has been adjusted. The second data extraction section has been renamed as Statistical Analyses.
“2.3 Statistical Analyses” (Line 102)
Comment 4:line 83, the article emphasizes abiotic/nocturnal CO2uptake, but in the Methods section, it links texture with "supporting microbial activity". If the article does not have data such as microbial biomass, respiration, and enzyme activity, it is best to avoid giving readers the impression that microbial activity is the direct mechanism of Fs.
Re: The reference to microbial activity has been removed from the description of the soil texture index.
“We therefore used the log-ratio of clay to sand content, ln(clay/sand), as an index of soil texture fineness to represent the capacity of soils to retain moisture.” (Line 83-85)
Comment 5:Figure 2 presents the average Fs and their confidence intervals for the five regions. However, the main text does not clearly state how many observations, sites, and studies are included in each region. This is very important because regions such as the Antarctic Desert Area may have significantly smaller sample sizes than the Central Asian Desert Area. It is recommended to report in Figure 2 or the Supplementary Table as: n observations / n sites / n studies. This will make the comparison between regions more transparent.
Re: We have added the sample size information for each of the five desert regions in the text of Section 3.1. We have also updated the caption of Figure 2 to explain this sample size notation.
“Fs among five major desert regions. Points represent means with 95% confidence intervals. Values in parentheses indicate the number of studies for each region. Different letters indicate the obvious discrepancy across different areas.” (Figure 2 caption)
Comment 6:Partial dependence plots illustrate the marginal predictive relationship under the model conditions and cannot directly prove causal relationships. Therefore, expressions such as "confirm" or "driving" in the text can be appropriately weakened and replaced with "suggest" or "indicate", especially in the research framework of observational synthesis. For example, line 266 could be changed to "favourable for the progress of the reaction".
Re: Overstated causal expressions have been weakened throughout the text.
“These patterns indicated that the effects of ΔT and θv on Fs are predominantly nonlinear or threshold-dependent, which linear models cannot adequately capture.” (Line 178-179)
“These results suggested that nocturnal CO2 uptake in dryland soils was a multi-factor synergistically regulated process, whose core driving forces can be attributed to two main mechanisms: physical and chemical.” (Line 258-259)
Comment 7:The manuscript proposes that the air-soil temperature difference drives thermal convection, and further regards it as the main driving force in the physical process. However, the current data mainly come from statistical relationships and do not directly measure the pressure gradient within the soil, the gas velocity, or the CO2concentration gradient. Therefore, it is suggested to clearly define this mechanism as "supported hypothesis/conceptual mechanism", and further emphasize in Discussion 4.3 that there is currently a lack of direct process observation evidence.
Re: The proposed mechanism has been explicitly defined as a physico-chemical conceptual mechanism framework. We emphasized the lack of direct in-situ process observation evidence and proposed future validation approaches.
“By integrating the results of our analysis with previous findings, we proposed a comprehensive conceptual physico‑chemical mechanism framework to elucidate the possible generation processes of nocturnal atmospheric CO2 uptake by dryland soils (Fig. 10).” (Line 297-299)
“Direct in-situ measurements of soil pore pressure gradients, gas flow velocity, and vertical CO2 concentration profiles are still lacking, which are critical to directly verify the physico- chemical process. Future research could combine high-frequency pore gas monitoring and 13C isotopic labeling to test this framework under controlled conditions of temperature difference, moisture, pH, and texture.” (Line 309-312)
Comment 8:Line 207, the text states that SEM is "based on a priori hypotheses informed by the RF and VPA results", but in the Methods section, HP is actually introduced, while VPA is not.
Re: The terminology inconsistency has been corrected. "VPA" has been replaced with "hierarchical partitioning (HP)" to be consistent with the method description in the Methods section.
“To disentangle the causal pathways through which environmental factors influence Fs, we constructed a structural equation model (SEM) based on a priori hypotheses informed by the RF and hierarchical partitioning (HP) results.” (Line 207-208)
Citation: https://doi.org/10.5194/egusphere-2026-4692-CC4
-
CC4: 'Reply on CC1', Jia-Bin Liu, 08 Oct 2026
reply
-
CC2: 'Comment on egusphere-2026-4692', Zhen Liu, 23 Sep 2026
reply
This manuscript compiles 563 observations of nocturnal soil CO₂ flux (Fs) across global deserts and proposes a physico-chemical framework involving physical transport and chemical fixation. The topic is relevant, the dataset is valuable, and the proposed framework is potentially novel. However, several issues concerning variable definitions, sign conventions, statistical reporting, figure captions, and terminology need to be clarified. I recommend minor revision.
1. The Methods define ln(clay/sand)ln(clay/sand), whereas the Abstract, Results, and Fig. 8d often use ln(sand/clay) and ln(sand/clay). This inconsistency makes the direction of the correlation difficult to interpret. Please use one index consistently throughout the manuscript and clearly state whether an increase represents finer texture or higher sand content. Also check that the signs in the Pearson correlation, RF, HP, and SEM results are consistent with this definition.
2. The overall research objective does not pertain to any specific global region. Therefore, if there are no data for South America and Australia, it is recommended to remove the colors.
3. Line 209, soil pH had a direct effect of -0.44 on Fs and an indirect effect of -0.092 via soil water content. The indirect effect here does not match that shown in Figure 9.
4.Line 253, Variable importance in the random forest model (Fig. 5), but the RF variable importance graph is Fig. 4, and Fig. 5 is a partial dependence plot. The citation here should be changed to Fig. 4.
5. Line 221, Please check the superscript on the entire text.
6. The method mentions using parameters such as AIC to evaluate SEM, but no relevant results are provided in the outcome.
7. Line 254, It states its independent contribution in VPA analysis, but in the method, hierarchical partitioning (HP) is used instead of VPA. Please unify it to HP, or explain that VPA is a different name for the same analysis.Citation: https://doi.org/10.5194/egusphere-2026-4692-CC2 -
AC1: 'Reply on CC2', Jia-Bin Liu, 29 Oct 2026
reply
Thank you for your valuable time and constructive comments on our manuscript.
Comment 1: The Methods define ln(clay/sand), whereas the Abstract, Results, and Fig. 8d often use ln(sand/clay) and ln(sand/clay). This inconsistency makes the direction of the correlation difficult to interpret. Please use one index consistently throughout the manuscript and clearly state whether an increase represents finer texture or higher sand content. Also check that the signs in the Pearson correlation, RF, HP, and SEM results are consistent with this definition.
Re: The soil texture index has been used consistently throughout the entire manuscript as ln(clay/sand) defined as an index of soil texture fineness. We have also rechecked the sign conventions across all relevant analyses—Pearson correlation, RF variable importance, hierarchical partitioning (HP), and SEM path coefficients—and confirmed that all signs are consistent with this ln(clay/sand) definition.
Comment 2: The overall research objective does not pertain to any specific global region. Therefore, if there are no data for South America and Australia, it is recommended to remove the colors.
Re: Figure 1 has been revised to remove highlighting colors from South American and Australian deserts where no datasets were included in the synthesis. Only the five desert regions with included study sites are highlighted.
Comment 3: Line 209, soil pH had a direct effect of -0.44 on Fs and an indirect effect of -0.092 via soil water content. The indirect effect here does not match that shown in Figure 9.
Re: The SEM path coefficients have been rechecked and the calculation logic of the indirect effect has been clarified. The original value of −0.092 was the pH → soil water content path coefficient, which was mistakenly reported as the indirect effect. The indirect effect of pH on Fs via soil water content is calculated as the product of the pH → θv path coefficient (-0.092) and the θv → Fs path coefficient (−0.304), which equals 0.028.
“Soil pH had a direct effect of −0.44 on Fs and an indirect effect of 0.028 via soil water content. ln(clay/sand) directly affected ΔT (0.183) and θv (0.163), and exerted a direct effect of 0.265 on Fs. θv directly affected ΔT (−0.363), and had a direct effect of −0.304 on Fs.” (Line 214-216)
Comment 4: Line 253, Variable importance in the random forest model (Fig. 5), but the RF variable importance graph is Fig. 4, and Fig. 5 is a partial dependence plot. The citation here should be changed to Fig. 4.
Re: The figure citation error has been corrected. The citation has been updated from Fig. 5 to Fig. 4.
“Soil pH was identified as the predictor affecting Fs, with its variable importance in the random forest model (Fig. 4) and its independent contribution in HP analysis both reaching significant levels (Fig.6).” (Line 257-258)
Comment 5: Line 221, Please check the superscript on the entire text.
Re: The superscript and subscript formats throughout the manuscript have been thoroughly checked and standardized, covering chemical formulas, physical units, and statistical metrics.
“Across 563 observations from global deserts, the mean nocturnal soil CO2 influx (Fs) was −0.195 μmol m-2 s-1 (−72.84 g C m-2 yr-1).” (Line 224-225)
Comment 6: The method mentions using parameters such as AIC to evaluate SEM, but no relevant results are provided in the outcome.
Re: Model fit statistics for SEM have been explicitly reported in the revised manuscript. The SEM fit results have been added in Section 3.4, including Fisher’s C statistic, P value, degrees of freedom, and the actual AIC value..
“Model fit statistics indicated adequate fit (Fisher’s C = 2.397, P = 0.302, df = 2). The Akaike Information Criterion (AIC=4422.8) was used for comparative model selection, consistent with the methods described above.” (Line 210-212)
Comment 7: Line 254, It states its independent contribution in VPA analysis, but in the method, hierarchical partitioning (HP) is used instead of VPA. Please unify it to HP, or explain that VPA is a different name for the same analysis.
Re: The terminology inconsistency has been corrected. The term has been unified as hierarchical partitioning (HP) throughout the manuscript, and all references to “VPA” have been removed.
“Soil pH was identified as the predictor affecting Fs, with its variable importance in the random forest model (Fig. 4) and its independent contribution in HP analysis both reaching significant levels (Fig.6).” (Line 257-258)
Citation: https://doi.org/10.5194/egusphere-2026-4692-AC1
-
AC1: 'Reply on CC2', Jia-Bin Liu, 29 Oct 2026
reply
-
CC3: 'Comment on egusphere-2026-4692', Xue Yang, 28 Sep 2026
reply
Major comment
The topic is relevant to the Earth and environmental science community served by EGUsphere because nocturnal CO₂ uptake remains poorly constrained in dryland carbon budgets, and a cross-site synthesis could make a useful contribution. The manuscript also assembles a potentially valuable dataset and attempts to connect physical and chemical controls.
The present analysis, however, does not yet support the title’s claim that it explains a generation mechanism. The evidence consists of observational data extracted from heterogeneous studies, many predictors vary mainly between studies or sites, and several analyses appear to treat clustered observations as independent. Contradictory definitions and sign interpretations for ΔT, Fs, and the texture ratio further prevent a stable interpretation of the reported effects. The annual carbon calculation also appears to apply a full-year, 24-hour multiplier to a nocturnal flux. These issues affect the principal conclusions and require substantial reanalysis and reframing.
- lines 8–23; Discussion, lines 243–307: Could the authors describe the proposed physicochemical mechanism as a conceptual framework supported by cross-study associations, rather than as a demonstrated causal mechanism, and revise the title, abstract, and conclusions accordingly?
- lines 65–126: Could the authors add the search date, screening procedure, data-selection rules, study/site hierarchy, and complete model settings so that readers can understand how the 563 observations were selected and analysed?
- lines 145–211 and, lines 243–285; Figs. 3, 5, 8, and 9: Could the authors use one texture variable throughout, either `ln(clay/sand)` or `ln(sand/clay)`, and check all signs and interpretations of ΔT and negative Fs so that “stronger uptake” has one consistent meaning?
- lines 220–236: Could the authors correct the annual conversion of nocturnal Fs, state the assumed number of nighttime hours and uptake days, and present the NEE and soil-respiration comparisons as illustrative estimates unless the quantities are directly comparable?
- Lines 310–314; Supplement, Table I: Could the authors provide a machine-readable dataset and analysis code, reconcile the supplement with the reference list, and use one consistent Data and code availability statement?
Minor comment
- Lines 18–20,Revise “suggested exerted pronounced effects” and “results ad previous studies”; both phrases are grammatically incomplete.
- Line 100,Rename Section 2.3 because the heading repeats “Data Extraction”; “Statistical analysis” would be more accurate.
- Lines 124–126,Correct “Fisher ‘s C” and list and cite all R packages used for the LMM, RF, hierarchical partitioning, and SEM.
- Lines 139–140,Replace “obvious discrepancy” with “statistically significant differences.”
- Lines 153–156,Correct the Fig. 3 caption because the displayed legend uses red for positive correlations and blue for negative correlations.
- Lines 164–168,Describe `%IncMSE` as predictive variable importance rather than “explanatory capacity,” and state whether it was calculated from out-of-bag predictions.
- Lines 180–182,Correct the Fig. 5 caption to match the panel order and use the same soil-texture ratio as the figure.
- Lines 193–195,Remove the stray “(c)” after “subsequent analyses” and identify the test used for EC.
- Lines 209–213 and 279–285,Reconcile the ΔT path coefficient, which is 0.249 in the Results and Fig. 9 but 0.2741 in the Discussion.
- Lines 223–225,Correct “NEE acrossed” and use either `yr⁻¹` or `a⁻¹` consistently.
- Lines 228–230,Correct the soil-respiration unit from `s⁻¹¹` to `s⁻¹`.
- Lines 257–266,Standardize the chemical notation, charges, subscripts, equilibrium arrows, and reaction numbering.
- Lines 306–309,Correct “labelingto” and specify which carbon pools the proposed ¹³C experiment would measure.
- After line 309,Add a concise Conclusion that answers the stated objectives and distinguishes observed associations from proposed mechanisms.
- Lines 316–319,Reconcile “Yao Gao” in the author list with “Yang Gao” in the contribution statement.
-
AC2: 'Reply on CC3', Jia-Bin Liu, 04 Nov 2026
reply
We sincerely thank the reviewer for the constructive comments. We have carefully revised the manuscript according to each comment. Below we provide a point-by-point response.
Major Comments
Comment 1: lines 8–23; Discussion, lines 243–307: Could the authors describe the proposed physicochemical mechanism as a conceptual framework supported by cross-study associations, rather than as a demonstrated causal mechanism, and revise the title, abstract, and conclusions accordingly?
Re: The proposed framework should be presented as a conceptual model supported by cross-study associations rather than a demonstrated causal mechanism. We have revised the title, abstract, Introduction objectives, Discussion Section 4.3, and added a new Conclusion section to reflect this framing.
Title: “A physico-chemical conceptual framework for explaining the generation mechanism of Nocturnal Soil CO2 Uptake in Deserts”
Abstract: “...we employed multiple linear mixed-effects models (LMMs), random forest (RF) regression, polynomial regression, and structural equation modeling (SEM) to quantify the effects of airsoil temperature difference (ΔT = Ta - Ts), soil water content (θv), soil pH, and ln(clay/sand) on soil nocturnal CO2 influx (Fs), and to propose a physico-chemical conceptual framework for explaining the generation mechanism of Fs. ”(Line 9-13)
“Correlation analysis indicated that these environmental factors may serve as important drivers in the generation of nocturnal soil CO2 influx. Integrating our results and previous studies, we proposed a conceptual comprehensive physico‑chemical mechanism framework.”(Line 18-20)
Introduction: “(3) to propose an integrated conceptual physico-chemical framework for nocturnal soil CO2 uptake in deserts.”(Line 63-64)
Section 4.3: “A conceptual physico-chemical framework for nocturnal soil CO2 uptake in deserts.” (Line 302)
Comment 2: lines 65–126: Could the authors add the search date, screening procedure, data-selection rules, study/site hierarchy, and complete model settings so that readers can understand how the 563 observations were selected and analysed?
Re: We have expanded the Methods section as follows: (1) Search date (2) Screening procedure: we have added a three-step description (3) Study hierarchy: in the LMM description, we specified that “site” nested within “study” as a random intercept, making the hierarchical structure explicit. (4) Complete model settings: we have listed all R packages used with their corresponding analyses: lme4 (LMMs), randomForest (RF, ntree = 1000), rdacca.hp (HP), piecewiseSEM (SEM), and ggplot2 (figures). We also clarified that %IncMSE was calculated from out-of-bag (OOB) predictions.
Section 2.1: “Literature searches were finalized on 30 September 2025. The screening procedure followed a three-step process: initial retrieval identified 147 records from three databases; 112 records were excluded after title and abstract screening based on the eligibility criteria; the full texts of the remaining 35 records were assessed, and 30 studies finally met all inclusion criteria.” (Line 73-76)
Section 2.3: “The final dataset contained 563 observation points, nested within 16 sites across 30 published studies, covering five major desert regions worldwide.” (Line 108-109)
“we used linear mixed-effects models (LMMs) for all regression analyses, with ‘site’ nested within ‘study’ as a random intercept.” (Line 114-115)
“The RF model was built using the randomForest package in R, with 1000 trees (ntree = 1000). Variable importance was quantified by the percentage increase in mean squared error (%IncMSE), calculated from out-of-bag (OOB) predictions, as a measure of predictive variable importance.” (Line 122-124)
“HP was performed using the rdacca.hp package in R.” (Line 129-130)
“SEM was performed using the piecewiseSEM package in R.” (Line 133-134)
“All statistical analyses were performed in the R 4.5.1 (R Core Team, 2025) with RStudio (Posit Team, 2025). Packages used included lme4 (LMMs), randomForest (RF), rdacca.hp package (HP), piecewiseSEM (SEM) and ggplot2 (Wickham, 2016). ” (Line 136-137)
Comment 3. Lines 145–211 and 243–285; Figs. 3, 5, 8, and 9: Could the authors use one texture variable throughout, either ln(clay/sand) or ln(sand/clay), and check all signs and interpretations of ΔT and negative Fs so that "stronger uptake" has one consistent meaning?
Re: We have standardized the soil texture index as ln(clay/sand) throughout the entire manuscript. We also corrected the Fig. 3 caption (red = positive, blue = negative) and the Fig. 5 caption panel order. All interpretations of ΔT and negative Fs now consistently refer to "stronger uptake" as more negative Fs values.
Figure 3 caption: “Pearson correlation heatmap illustrating the relationships between Fs and environmental factors across global desert ecosystems. Variables include Fs, ΔT, θv, soil pH, ln(clay/sand), Ta, and Ts. The color gradient corresponds to the Pearson correlation coefficient (r), with red tones indicating positive correlations and blue tones indicating negative correlations, spanning from -1 (perfect negative correlation) to 1 (perfect positive correlation). Numerical values within each cell represent the specific correlation coefficients between the paired variables.” (Line 163-167)
Figure 5 caption: “Partial dependence plots from random forest model illustrating the marginal effects of key environmental variables on nocturnal soil CO2 flux (Fs) in desert ecosystems. The panels show: (a) Effect of ΔT (air–soil temperature difference); (b) Effect of pH (soil pH); (c) Effect of ln(clay/sand); (d) Effect of θv (soil volumetric water content). The horizontal axes represent standardized values of environmental predictors, and the vertical axes denote the predicted response of soil CO2 flux driven by each single variable." (Line 188-192)
Comment 4. Lines 220–236: Could the authors correct the annual conversion of nocturnal Fs, state the assumed number of nighttime hours and uptake days, and present the NEE and soil-respiration comparisons as illustrative estimates unless the quantities are directly comparable?
Re: We have corrected the annual conversion. The original calculation applied a full-year, 24-hour multiplier to a nocturnal flux. We now explicitly assume 12 hours of nighttime uptake per night over 365 days, yielding an annual estimate of approximately −36.9 g C m−2 a−1. We also explicitly label the NEE and soil-respiration comparisons as illustrative estimates and explain why they are not directly comparable.
Section 4.1: “In the present study, mean annual nocturnal soil CO2 influx across global deserts was estimated as follows: assuming nocturnal soil CO2 uptake occurs for 12 hours per night over 365 days per year, the annual uptake was −36.9 g C m-2 a-1. This illustrative estimate represents roughly 17%–28% of reported desert ecosystem NEE, suggesting that nocturnal soil CO2 influx may constitute an important component of carbon absorption in desert ecosystems. Note that this comparison uses only nocturnal soil CO2 uptake and does not consider daytime soil CO2 release. ” (Line 236-241)
Comment 5. Lines 310–314; Supplement, Table I: Could the authors provide a machine-readable dataset and analysis code, reconcile the supplement with the reference list, and use one consistent Data and code availability statement?
Re: We have rewritten the Data and code availability statement. The machine-readable dataset, all analysis code, and figure-preparation code have been deposited in Zenodo and are publicly available. We have also reconciled the supplement table with the reference list.
Code and data availability: “The machine-readable dataset, all analysis code and the code for figure preparation in this study are available at https://doi.org/10.5281/zenodo.23754106.”
Minor Comments
Minor 1. Lines 18–20: Revise "suggested exerted pronounced effects" and "results ad previous studies"; both phrases are grammatically incomplete.
Re: Both grammatical errors have been corrected.
Abstract: “The relationships of ΔT, θv, pH, and ln(clay/sand) with Fs suggested that these factors exerted pronounced effects on nocturnal CO2 influx in desert soils. Correlation analysis indicated that these environmental factors may serve as important drivers in the generation of nocturnal soil CO2 influx. Integrating our results and previous studies, we proposed a conceptual comprehensive physico‑chemical mechanism framework.” (Line 16-20)
Minor 2. Line 100: Rename Section 2.3 because the heading repeats "Data Extraction"; "Statistical analysis" would be more accurate.
Re: The duplicate section heading has been corrected.
Section 2.3: "Statistical analysis"
Minor 3. Lines 124–126: Correct "Fisher ‘s C" and list and cite all R packages used for the LMM, RF, hierarchical partitioning, and SEM.
Re: We corrected the typographical error and now list all R packages with their respective functions.
Section 2.3: “All statistical analyses were performed in the R 4.5.1 (R Core Team, 2025) with RStudio (Posit Team, 2025). Packages used included lme4 (LMMs), randomForest (RF), rdacca.hp package (HP), piecewiseSEM (SEM) and ggplot2 (Wickham, 2016).” (Line 136-137)
Minor 4. Lines 139–140: Replace "obvious discrepancy" with "statistically significant differences."
Re: The wording has been corrected in both the main text and the Figure 2 caption.
Minor 5. Lines 153–156: Correct the Fig. 3 caption because the displayed legend uses red for positive correlations and blue for negative correlations.
Re: The color direction in the caption has been reversed to match the figure.
Figure 3 caption: “...Variables include Fs, ΔT, θv, soil pH, ln(clay/sand), Ta, and Ts. The color gradient corresponds to the Pearson correlation coefficient (r), with red tones indicating positive correlations and blue tones indicating negative correlations, spanning from -1 (perfect negative correlation) to 1 (perfect positive correlation). Numerical values within each cell represent the specific correlation coefficients between the paired variables.”
Minor 6. Lines 164–168: Describe %IncMSE as predictive variable importance rather than "explanatory capacity," and state whether it was calculated from out-of-bag predictions.
Re: We have rephrased %IncMSE as a measure of predictive variable importance and explicitly stated it was calculated from out-of-bag (OOB) predictions.
Section 2.3: “Variable importance was quantified by the percentage increase in mean squared error (%IncMSE), calculated from out-of-bag (OOB) predictions, as a measure of predictive variable importance. ” (Line 123-124)
Minor 7. Lines 180–182: Correct the Fig. 5 caption to match the panel order and use the same soil-texture ratio as the figure.
Re: The panel order and texture variable in the caption now match the figure.
Figure 5 caption: “...The panels show: (a) Effect of ΔT (air–soil temperature difference); (b) Effect of pH (soil pH); (c) Effect of ln(clay/sand); (d) Effect of θv (soil volumetric water content). The horizontal axes represent standardized values of environmental predictors, and the vertical axes denote the predicted response of soil CO2 flux driven by each single variable”
Minor 8. Lines 193–195: Remove the stray "(c)" after "subsequent analyses" and identify the test used for EC.
Re: The stray "(c)" has been removed and the EC test has been identified.
Figure 7 caption: “Scatter plots of soil CO2 flux (Fs) versus air temperature (Ta) (a), soil temperature (Ts) (b), and soil electrical conductivity (EC) (c) in global desert regions. EC was tested by Pearson correlation and showed no significant effect (p > 0.05).”
Minor 9. Lines 209–213 and 279–285: Reconcile the ΔT path coefficient, which is 0.249 in the Results and Fig. 9 but 0.2741 in the Discussion.
Re: The ΔT path coefficient has been unified to 0.249 throughout the manuscript. The Discussion value (previously 0.2741) has been corrected to 0.249 to match the Results and Fig. 9.
Minor 10. Lines 223–225: Correct "NEE acrossed" and use either yr⁻¹ or a⁻¹ consistently.
Re: The typo has been fixed and units are now consistently expressed as a⁻¹.
Minor 11. Lines 228–230: Correct the soil-respiration unit from s⁻¹¹ to s⁻¹.
Re: The unit typo has been corrected.
Minor 12. Lines 257–266: Standardize the chemical notation, charges, subscripts, equilibrium arrows, and reaction numbering.
Re: All chemical equations have been standardized with correct subscripts, charges, equilibrium arrows, and sequential numbering.
Minor 13. Lines 306–309: Correct "labelingto" and specify which carbon pools the proposed ¹³C experiment would measure.
Re: The typo has been fixed, and we explicitly stated the carbon pools that the ¹³C experiment would trace.
Section 4.3: “Future research could combine high-frequency pore gas monitoring and 13C isotopic labeling to test this framework by tracing key carbon pools — 13CO2 into soil solid and vapor phases under controlled conditions of temperature difference, moisture, pH, and texture. ” (Line 322-324)
Minor 14. After line 309: Add a concise Conclusion that answers the stated objectives and distinguishes observed associations from proposed mechanisms.
Re: We have added a new Section 5 (Conclusion) that addresses all three research objectives and explicitly distinguishes observed associations from the proposed conceptual framework.
“Based on a global synthesis of 563 observations from 30 studies, we found a mean nocturnal soil CO2 uptake of −0.195 μmol m−2 s−1 across deserts, with significant regional variation. ΔT, soil pH, θv, and ln(clay/sand) were all significantly associated with nocturnal flux. We proposed a physico‑chemical conceptual framework linking physical transport and chemical fixation to the observed uptake. This mechanism remains a hypothesis inferred from cross-site correlations rather than directly measured, and awaits in-situ verification through 13C isotopic labelling in the future.” (Line 326-330)
Minor 15. Lines 316–319: Reconcile "Yao Gao" in the author list with "Yang Gao" in the contribution statement.
Re: The author name has been unified to "Yang Gao" in both the author list and the contribution statement.
Citation: https://doi.org/10.5194/egusphere-2026-4692-AC2
-
RC1: 'Comment on egusphere-2026-4692', Anonymous Referee #1, 19 Oct 2026
reply
This manuscript compiles 563 observations of nocturnal soil CO₂ flux (Fs) across global deserts and proposes a physico-chemical framework involving physical transport and chemical fixation. The topic is relevant, the dataset is valuable, and the proposed framework is potentially novel. However, several issues concerning variable definitions, sign conventions, statistical reporting, figure captions, and terminology need to be clarified. I recommend minor revision.
1. Line 18, there is a grammatical error.
2. The Methods define ln(clay/sand)ln(clay/sand), whereas the Abstract, Results, and Fig. 8d often use ln(sand/clay)ln(sand/clay). This inconsistency makes the direction of the correlation difficult to interpret. Please use one index consistently throughout the manuscript and clearly state whether an increase represents finer texture or higher sand content. Also check that the signs in the Pearson correlation, RF, HP, and SEM results are consistent with this definition.
3. The overall research objective does not pertain to any specific global region. Therefore, if there are no data for South America and Australia, it is recommended to remove the colors.
4. Line 209, soil pH had a direct effect of -0.44 on Fs and an indirect effect of -0.092 via soil water content. The indirect effect here does not match that shown in Figure 9.
5.Line 253, "Variable importance in the random forest model (Fig. 5)", but the RF variable importance graph is Fig. 4, and Fig. 5 is a partial dependence plot. The citation here should be changed to Fig. 4.
6. Line 221, Please check the superscript on the entire text.
7. The method mentions using parameters such as AIC to evaluate SEM, but no relevant results are provided in the outcome.
8. Line 237-238, "The mean Fs values for the Central Asian Desert Area, Circum Mediterranean Desert Area, North American Desert Area, Southern African Desert Area, and Antarctic Desert Area were ranged from −0.138 to −0.224 μmol m-2 s-1 (Fig. 2)", grammatical error.
9. Line 254, it states "its independent contribution in VPA analysis", but in the method, hierarchical partitioning (HP) is used instead of VPA. Please unify it to HP, or explain that VPA is a different name for the same analysis.
10. Line 289, "Previous studies in different desert regions had referred the various mechanisms underlying nocturnal soil CO2 influx.", grammatical error.Citation: https://doi.org/10.5194/egusphere-2026-4692-RC1 -
AC3: 'Reply on RC1', Jia-Bin Liu, 05 Nov 2026
reply
We sincerely thank the reviewer for the careful reading and constructive comments. The manuscript has been revised accordingly.
Comment 1: Line 18, there is a grammatical error.
Re: We apologize for the grammatical error in the Abstract. The phrase “suggested exerted” was ungrammatical and we have revised it.
“Correlation analysis indicated that these environmental factors may serve as important drivers in the generation of nocturnal soil CO2 influx.” (Line 18-19)
Comment 2: The Methods define ln(clay/sand), whereas the Abstract, Results, and Fig. 8d often use ln(sand/clay). This inconsistency makes the direction of the correlation difficult to interpret. Please use one index consistently throughout the manuscript and clearly state whether an increase represents finer texture or higher sand content. Also check that the signs in the Pearson correlation, RF, HP, and SEM results are consistent with this definition.
Re: The soil texture index has been used consistently throughout the entire manuscript as ln(clay/sand) defined as an index of soil texture fineness. We have also rechecked the sign conventions across all relevant analyses—Pearson correlation, RF variable importance, hierarchical partitioning (HP), and SEM path coefficients—and confirmed that all signs are consistent with this ln(clay/sand) definition.
Comment 3: The overall research objective does not pertain to any specific global region. Therefore, if there are no data for South America and Australia, it is recommended to remove the colors.
Re: Figure 1 has been revised to remove highlighting colors from South American and Australian deserts where no datasets were included in the synthesis. Only the five desert regions with included study sites are highlighted.
Comment 4: Line 209, soil pH had a direct effect of -0.44 on Fs and an indirect effect of -0.092 via soil water content. The indirect effect here does not match that shown in Figure 9.
Re: The SEM path coefficients have been rechecked and the calculation logic of the indirect effect has been clarified. The original value of −0.092 was the pH → soil water content path coefficient, which was mistakenly reported as the indirect effect. The indirect effect of pH on Fs via soil water content is calculated as the product of the pH → θv path coefficient (-0.092) and the θv → Fs path coefficient (−0.304), which equals 0.028.
“Soil pH had a direct effect of −0.44 on Fs and an indirect effect of 0.028 via soil water content. ln(clay/sand) directly affected ΔT (0.183) and θv (0.163), and exerted a direct effect of 0.265 on Fs. θv directly affected ΔT (−0.363), and had a direct effect of −0.304 on Fs.” (Line 222-224)
Comment 5: Line 253, "Variable importance in the random forest model (Fig. 5)", but the RF variable importance graph is Fig. 4, and Fig. 5 is a partial dependence plot. The citation here should be changed to Fig. 4.
Re: The figure citation error has been corrected. The citation has been updated from Fig. 5 to Fig. 4.
“Soil pH was identified as the predictor affecting Fs, with its variable importance in the random forest model (Fig. 4) and its independent contribution in HP analysis both reaching significant levels (Fig.6).” (Line 267-268)
Comment 6: Line 221, Please check the superscript on the entire text.
Re: The superscript and subscript formats throughout the manuscript have been thoroughly checked and standardized, covering chemical formulas, physical units, and statistical metrics.
“Across 563 observations from global deserts, the mean nocturnal soil CO2 influx (Fs) was −0.195 μmol m-2 s-1 (−36.9 g C m-2 a-1).” (Line 232-233)
Comment 7: The method mentions using parameters such as AIC to evaluate SEM, but no relevant results are provided in the outcome.
Re: Model fit statistics for SEM have been explicitly reported in the revised manuscript. The SEM fit results have been added in Section 3.4, including Fisher’s C statistic, P value, degrees of freedom, and the actual AIC value..
“Model fit statistics indicated adequate fit (Fisher’s C = 2.397, P = 0.302, df = 2). The Akaike Information Criterion (AIC=4422.8) was used for comparative model selection, consistent with the methods described above.” (Line 218-220)
Comment 8: Line 237-238, "The mean Fs values for the Central Asian Desert Area, Circum Mediterranean Desert Area, North American Desert Area, Southern African Desert Area, and Antarctic Desert Area were ranged from −0.138 to −0.224 μmol m-2 s-1 (Fig. 2)", grammatical error.
Re: We have corrected the grammar.
“The mean Fs values for the Central Asian Desert Area, Circum Mediterranean Desert Area, North American Desert Area, Southern African Desert Area, and Antarctic Desert Area ranged from −0.138 to −0.224 μmol m-2 s-1 (Fig. 2).” (Line 251-252)
Comment 9: Line 254, It states its independent contribution in VPA analysis, but in the method, hierarchical partitioning (HP) is used instead of VPA. Please unify it to HP, or explain that VPA is a different name for the same analysis.
Re: The terminology inconsistency has been corrected. The term has been unified as hierarchical partitioning (HP) throughout the manuscript, and all references to “VPA” have been removed.
“Soil pH was identified as the predictor affecting Fs, with its variable importance in the random forest model (Fig. 4) and its independent contribution in HP analysis both reaching significant levels (Fig.6).” (Line 266-267)
Comment 10: Line 289, "Previous studies in different desert regions had referred the various mechanisms underlying nocturnal soil CO2 influx.", grammatical error.
Re: We have corrected the verb error.
“Previous studies in different desert regions referred to the various mechanisms underlying nocturnal soil CO2 influx.” (Line 301)Citation: https://doi.org/10.5194/egusphere-2026-4692-AC3
-
AC3: 'Reply on RC1', Jia-Bin Liu, 05 Nov 2026
reply
Viewed
| HTML | XML | Total | Supplement | BibTeX | EndNote | |
|---|---|---|---|---|---|---|
| 293 | 172 | 96 | 561 | 121 | 69 | 74 |
- HTML: 293
- PDF: 172
- XML: 96
- Total: 561
- Supplement: 121
- BibTeX: 69
- EndNote: 74
Viewed (geographical distribution)
| Country | # | Views | % |
|---|
| Total: | 0 |
| HTML: | 0 |
| PDF: | 0 |
| XML: | 0 |
- 1
This paper focuses on the nocturnal CO2 absorption process in desert soil, based on global multi-regional observational data. It comprehensively employs methods such as LMM, RF, HP, and SEM to analyze the effects of environmental factors, and further proposes a physical-chemical coupling mechanism framework. The research is novel and has ecological significance, and the data integration and multi-method analysis are relatively comprehensive. Overall, the manuscript has the basis for publication, but it still needs further improvement in terms of statistical analysis consistency, and some mechanism explanations. It is recommended to be accepted after minor revisions. Specific comments are as follows: