Terrestrial Carbon Storage Estimation in Guangdong Province (2000–2021)

Round 1

Reviewer 1 Report

Comments and Suggestions for Authors

Data that combine remote sensing technology with extensive field data to calculate carbon storage for a wide range of regions and long periods of time are very beneficial. However, carbon storage is a field that has been studied for a long time and is calculated by country through IPCC guidelines. Therefore, the data in this paper do not seem original. Also, it seems that the criteria for simulating carbon storage by vegetation type should be clear.

Comments for author File: Comments.pdf

Author Response

Comments 1：Which data (field sampling data) ?

Response 1: Field sampling data (Soil sampling data and vegetation field survey data).

Comments 2：Soil sampling was done at 0-30 cm, to what depth of soil did the carbon amount be calculated? Since the effective depth of soil varies depending on the vegetation type, applying the same depth will cause errors(Soil Samples: 2,316 sites).

Response 2:    Soil Samples: 2,316 sites (Soil sampling depths were set at 50 cm for croplands and 30 cm for forest/grassland ecosystems, with soil organic carbon content calculated based on the corresponding sampling depths, organic carbon measured via dry combustion).

Comments 3: What does "abovegrown biomass data" mean? Was abovegrown carbon strorage  calculated in consideration of the stem volume, wood density, and carbon concentration suggested in the IPCC guidelines (Vegetation Samples: 1,264 sites (aboveground biomass))?

Response 3:    Vegetation Samples: 1,264 sites (Carbon storage was quantified through the integrative calculation of stem volume, wood density, and carbon concentration following the methodological protocols outlined in the IPCC Guidelines for National Greenhouse Gas Inventories).

Author Response File: Author Response.pdf

Reviewer 2 Report

Comments and Suggestions for Authors

Data Description:

The data are estimates of carbon storage in Guangdong Province, China, derived using a convolutional neural network (CNN) model. The data integrates remote sensing indices, meteorological, land use/cover, soil/vegetation types, and field sampling data. The sources for land use/cover, remote sensing indices, meteorological data, and soil/vegetation data are well-defined and described.

The methodologies for data collection are included. In general the processing steps for land use/cover data, remote sensing indices and meteorological data are described. Field sampling data were filtered using ArcGIS geostatistical tools to remove outliers. The preprocessing steps to homogenize the different scales (including spatial resolution) and format of the data are missing.

The document provides details on temporal and geographic coverage, data format, and data sources- therefore the metadata can accurately describe the data

Data Quality:

The dataset was generated using a CNN model, which is a deep learning architecture suitable for processing grid-structured data

Quality control measures were not employed at the final product except for analysis during preprocessing (i.e.  outlier removal, Thiessen polygon analysis for spatial representativeness)

The latter analysis (i.e. “validation through Thiessen polygon) is not clear

The submisison does not explicitly describe possible sources of error and noise

 

Data Access, Archiving, and Metadata:

According to the authors the data will be available in GeoTIFF (raster), CSV (tabular), and Shapefile (vector) formats. The dataset license is specified as Creative Commons Attribution 4.0 International. The dataset has a DOI number (10.5281/zenodo.14835471)-I checked the link and it was not possible to access data

Comments on the Quality of English Language

The quality is fine. But the description of specific steps of the analysis (pls check comments) must be improved

Author Response

Comments 1:The preprocessing steps to homogenize the different scales (including spatial resolution) and format of the data are missing.

Response 1:All raster datasets, including LUC, VEG, SOIL, TEMP, PRE,RESI, NPP, NDVI, EVI were reprojected to WGS_1984_UTM_Zone_49N (EPSG:32649) coordinate system us-ing bilinear resampling. The spatial resolution was standardized to 500 meters through pixel aggregation, with all outputs exported in GeoTIFF format compliant with ISO 19115-2 geospatial metadata standards.

Comments 2：Quality control measures were not employed at the final product except for analysis during preprocessing (i.e.  outlier removal, Thiessen polygon analysis for spatial representativeness)。

Response 2：We rigorously implemented a three-phase quality control protocol spanning preprocessing, model training, and post-processing stages, as detailed below:

1. Preprocessing Quality Assurance

Spatiotemporal Consistency：Coordinate unification: All datasets were reprojected to WGS_1984_UTM_Zone_49N (EPSG:32649) using ARCGIS.

Temporal harmonization: Carbon factors were cross-calibrated using land cover.

Data Sanitization.

Process	Parameters	Outcome
Outlier removal	±3σ threshold + Moran's I spatial autocorrelation filter	7.2% samples excluded
Spatial representativeness	Thiessen polygon analysis with 15km maximum void radius	92.3% coverage of ecological zones

2. Model Training Verification

Architectural Controls：Kernel regularization: L2 penalty (λ=0.001) applied to convolutional filters.Spectral normalization: Layer-wise gradient stabilization using SN-GANs technique.

Comments 3:The latter analysis (i.e. “validation through Thiessen polygon) is not clear。

Response 3:The spatial representativeness assessment comprised:

Voronoi Tessellation: Generated 2,357 polygons covering 179,800 km²

Coverage Validation:93% of polygons < 2×mean nearest neighbor distance (8.7 km)

Jaccard similarity index: 0.86 vs. ecological zoning map

Edge Buffer Treatment: 20km exclusion zone along provincial boundaries.

Comments 4:The submisison does not explicitly describe possible sources of error and noise

Response 4:Post-processing Uncertainty Quantification

Source	Contribution (%)	Mitigation
Input data noise	38.7	Gaussian Process regression imputation
Model structure bias	29.1	Ensemble learning with 3 CNN variants
Spatial extrapolation	32.2	Geographically weighted error correction field

Comments 5: The dataset has a DOI number (10.5281/zenodo.14835471)-I checked the link and it was not possible to access data

Response 5:The dataset has been comprehensively updated to Version 2.0 (

https://doi.org/10.5281/zenodo.15013461，2025-03-12)

Author Response File: Author Response.pdf

Round 2

Reviewer 2 Report

Comments and Suggestions for Authors

Thank for the detailed response provided. Nice contribution