the Creative Commons Attribution 4.0 License.
the Creative Commons Attribution 4.0 License.
Spatial influence of agriculture residue burning and aerosols on land surface temperature
Abstract. The biophysical effect of agriculture-residue based fire through excessive release of energy and carbonaceous aerosols essentially unaccounted globally. Elucidating climate feedback from residue-based fire however, remain pertinent as energy released from fire pose potential to modify land surface temperature (LST) thereby, regional climate. Here, an observation-driven assessment of spatial change in LST due to concurrent release of energy and aerosols has been explored over northwest India using multiple satellite and reanalysis-based datasets. Initially, year-specific fire pixel density was computed to identify intensive fire zone encompassing only medium to large fire. Spatial analysis revealed positive correlation among FRP (fire radiative power), LST and AOD (aerosol optical depth) across the intensive fire zone. Residue-based fire accounted an increase in LST by 0.48 °C and AOD by 0.19 yearly during peak fire season over intensive fire zone. A Random Forest non-linear model was used to regress potential influence of FRP and AOD on LST. Two pre-constructed scenarios were evaluated to ascertain FRP-AOD-LST nexus. Interestingly, both scenarios recognized FRP as a top predictor to influence LST followed by solar radiation and AOD. A significant enhancement in relative feature importance of FRP was also noted during days having high fire intensity and positive association against LST. Geographically Weighted Regression further explained spatial heterogeneity in LST modulation by FRP. Our analysis therefore, provides first evidence on crop residue-based fire on modifying regional climate by altering land surface temperature. It also underlines that extent of such perturbation is subject to year-specific fire intensity and govern by meteorology.
- Preprint
(1845 KB) - Metadata XML
-
Supplement
(949 KB) - BibTeX
- EndNote
Status: final response (author comments only)
- RC1: 'Comment on egusphere-2025-3163', Anonymous Referee #1, 04 Sep 2025
-
RC2: 'Comment on egusphere-2025-3163', Anonymous Referee #2, 09 Sep 2025
Pandey et al.’s study "Spatial influence of agriculture residue burning and aerosols on land surface temperature" presents an observation-driven study of how crop-residue fires in northwest India influence land surface temperature (LST) and aerosol loading. They identify year-specific intensive fire zones using VIIRS FRP, retrieve VIIRS AOD and MODIS Aqua daytime LST, and use AgERA5 meteorology to control for meteorological context. They apply a space-for-time grid comparison to estimate ΔLST and ΔAOD associated with fire, compute Hurst exponents for persistence, and develop two Random-Forest (RF) regression scenarios (broad fire season and high-correlation windows) to quantify relative feature importance. Finally, they run a Geographically Weighted Regression (GWR) of FRP and LST to map spatial heterogeneity. The paper reports an average fire-induced ΔLST ≈ +0.48°C (range −0.55 to 1.69°C) and ΔAOD ≈ +0.19 yr-1 during the peak season, and finds FRP is the top RF predictor of LST in both scenarios (with much higher RF performance in the "scenario 2" windows). Crop-residue burning in NW India and other parts of South Asia has major air-quality and climate implications, this study’s focus on crop-burning and LST is important for this region.The use of VIIRS FRP, VIIRS AOD, MODIS LST, MODIS LC data and AgERA5 meteorology enables a multi-angle observational assessment. The space-for-time comparison, Hurst analysis, random forest for non-linear attribution, and GWR for spatial heterogeneity form a coherent methodological ensemble. However, there are some major concern and queries that needs to be properly addressed at this stage:
- LST is strongly influenced by near-surface air temperature, PBL height, soil moisture, recent precipitation, cloud cover, surface albedo and vegetation state (NDVI/LAI). Although AgERA5 meteorology (At, Sr, Pr, RH) is included as one of the predictors, the manuscript does not convincingly demonstrate that the estimated ΔLST (and RF / GWR results) are not driven by meteorological covariates or systematic land-cover differences between “fire” and “no-fire” grids. Without stronger control for these confounders, the causal attribution “fire to AOD and LST” remains tentative. In the space-for-time comparison, conducting matched comparisons, for instance for each fire grid choose one or more no-fire grids matched by NDVI, elevation, distance to major urban areas, and climatological mean LST. This reduces bias from non-random spatial placement of fires. Propensity-score matching or simple stratified matching would help. Additional proxies including but not limited to PBL height, surface soil moisture, and in-situ atmospheric radiative impacts induced by the fire-emitted aerosols themselves used in the predictor set may help clarify this relationship and strengthen the findings. However, I welcome the authors to instead post a rationale on why not including these variables and this suggested approach may still suffice in relationship quantification.
- Provide details on RF hyperparameter tuning (max_depth, max_features, min_samples_leaf). The manuscript uses n_estimators=100 with a fixed seed — please show whether you tuned parameters (grid search / CV) or at least show sensitivity to n_trees and max_features. To further imrpve RF model valiadtion, spatial and temporal block cross-validation (e.g., leave-one-year-out, or K-fold blocking by contiguous spatial clusters) and report cross-validated R2, RMSE, MAE. This may provide more robust predictive skill.
- The GWR model for scenario 2 is using only FRP, SR and AOD as predictor for LST, I do not understand the rationale of leaving out other local factors, included but limited to those mentioned in point 1 above. Are the authors testing the concept of using these specific variables exclusively in relationship to LST? However, I am confused if other meteorological variables and aerosol types (their optical varialbility in terms of scattering and absorption, and how these may influence atmospheric heating/radiative forcing and near-surface based cooling/radiative forcing (Freychet et al 2019; Tiwari et al. 2023) and surface albedo (Hou et al. 2025) when running GWR could bias the local coeffcients. Local coefficients maybe absorb the effect of omitted spatially-varying covariates. I am confused why scenario 2 is missing out important variables. Adiitionally, please also include bandwidth and kernel details of the AICc minimization you mention.
- I am also confused with the descritption of scenario 2, specifically if the reported relative feature importance (RFI) is normalized in the right way? As you mention this is a normzalized metric. But for scenario 2 FRP was 0.503 SR was 0.143 and Aerosol loading was 0.68. For these three predictors the normalized RFI sum more than 1. Is this a typographical error, a misunderstanding on my part, or is there some calculation mistake?
- ΔLST is reported as +0.48°C (mean) with range, but it’s unclear whether this difference is statistically significant after accounting for temporal variability and dependence, and how many grid cells underpin the estimates. Provide confidence intervals (e.g., bootstrapped CIs) for ΔLST and ΔAOD. Additionally, consider how comparison of pre-post events within the same grid for fire vs. similar non-fire grids) could help strengthen causual inference.
- Justify selection of FRP density threshold (>5 MW grid⁻¹), the 1500 MW threshold and the 50% growth/decline rule for scenario 1, and the r >=5 threshold for scenario 2. Add rationale and sensitivity checks (e.g., try thresholds (+20%, -20%).
- The Hurst exponent computed and interpreted as persistence (> 0.5), is relevant when there is large number of data points which are specifically not impacted by seasonal trends, however, in this case, with only 5-year dataset and strong seasonality, Rescaled Range (R/S) analysis for Hurst estimation can be sensitive to trend and seasonality. This is an important featured previously determined by various observational studies in this part of the world where both inter- and intra-annual variability is common (Lin et al. 2020; Liu et al. 2024 etc.). Did the authors conduct detrended fluctuation analysis (DFA) or remove seasonal cycle before computing Hurst. Furthermore, the author’s interpret values of H > 0.5 as indicating persistence and suggest that anomalies may “remain stable in the near future.” While H > 0.5 indeed indicates statistical persistence or long-term data analysis, this interpretation could overstate the predictive implications of the Hurst exponent, especially given the relatively short five-year data record and the presence of strong seasonal cycles (such as monsoon and agricultural seasonality) inherent in the dataset. I recommend the authors temper the predictive language by replacing claims that anomalies “will” persist with the more cautious and appropriate statement that H > 0.5 indicates statistical persistence. Additionally, the authors are encouraged to clarify whether seasonal cycles were accounted for or removed prior to computing the Hurst exponent, as this can significantly affect estimates derived from R/S analysis.
- There are several small typos/grammatical slips (e.g., “Dring” typo of “During” (Page 19), “reginal” typo of “regional” (Page 19 3.4), please go through the manuscript carefully and correct these and similar mistakes.
References:
Freychet, N., Tett, S. F. B., Bollasina, M., Wang, K. C., & Hegerl, G. (2019). The local aerosol emission effect on surface shortwave radiation and temperatures. Journal of Advances in Modeling Earth Systems, 11, 806–817. https://doi.org/10.1029/2018MS001530
Hou, Z., Zhang, L., Peng, J. et al. Radiative forcing reduced by early twenty-first century increase in land albedo. Nature 641, 1162–1171 (2025). https://doi.org/10.1038/s41586-025-08987-z
Liu, J., Cohen, J.B., He, Q. et al. Accounting for NOx emissions from biomass burning and urbanization doubles existing inventories over South, Southeast and East Asia. Commun Earth Environ 5, 255 (2024). https://doi.org/10.1038/s43247-024-01424-5
Lin, C., Cohen, J. B., Wang, S., Lan, R., & Deng, W. (2020). A new perspective on the spatial, temporal, and vertical distribution of biomass burning: quantifying a significant increase in CO emissions. Environmental Research Letters, 15(10), 104091. https://doi.org/10.1088/1748-9326/abaa7a
Tiwari, P., Cohen, J.B., Wang, X. et al. Radiative forcing bias calculation based on COSMO (Core-Shell Mie model Optimization) and AERONET data. npj Clim Atmos Sci 6, 193 (2023). https://doi.org/10.1038/s41612-023-00520-1
Citation: https://doi.org/10.5194/egusphere-2025-3163-RC2
Viewed
HTML | XML | Total | Supplement | BibTeX | EndNote | |
---|---|---|---|---|---|---|
1,706 | 23 | 9 | 1,738 | 19 | 25 | 17 |
- HTML: 1,706
- PDF: 23
- XML: 9
- Total: 1,738
- Supplement: 19
- BibTeX: 25
- EndNote: 17
Viewed (geographical distribution)
Country | # | Views | % |
---|
Total: | 0 |
HTML: | 0 |
PDF: | 0 |
XML: | 0 |
- 1
This manuscript try to address the relationship between fire radiative power (FRP), aerosol optical depth (AOD), and land surface temperature (LST) in northwestern India using multi-source remote sensing data combined with machine learning (random forest) and spatial regression (GWR). The topic is timely and relevant, particularly in the context of agricultural residue burning and its climatic impacts. The integration of multiple data sets and methods is commendable.
However, the current version has several shortcomings: the grammars and sentences are so poor, the transparency of data and methodology is limited, the interpretation of results is sometimes superficial and overly focused on correlations, and the discussion of mechanisms and uncertainties is insufficient. The conclusions also need to highlight the novelty and practical implications more clearly. With revisions to strengthen the grammars, methodological rigor, deepen interpretation, and improve clarity of presentation, this paper could make a valuable contribution. I recommend a major revision before it could be accepted.
Major Comments
Minor Comments