General comment

This manuscript tackles an important and timely question: whether multi-year satellite data on phenology (based on NDVI/EVI/NDWI processed with TIMESAT) can classify soil degradation states in smallholder grasslands and meaningfully relate to on-the-ground soil condition. The study spans two Kenyan landscapes and couples a remote-sensing classification (2013–2018) with field sampling at 90 sites (Oct–Nov 2019) for a suite of physical, chemical, and microbial variables (0–10 cm). The overall conclusion—that only microbial biomass C (and to a lesser extent bulk density) consistently aligns with the remote sensing classes—has practical implications for monitoring and restoration. However, several aspects of the methodology need clarification or strengthening before the evidence can fully support the claims.

Major concerns.

The paper mixes sensors (Landsat TM/OLI and Sentinel-2) and resamples to 10 m, but the harmonization/preprocessing steps are not fully described.

The specific land cover ESA product used for masking is not named or discussed in terms of accuracy/limitations for these mosaics. 

Terminology should be standardized (e.g., “Normalized Difference Vegetation Index,” and clarify that your NDWI formulation uses NIR–SWIR, i.e., Gao-type, to avoid confusion with the original NDWI.)

Degradation states are defined from average distributions of TIMESAT metrics and then selected by visual consistency with Google Earth, without an independent accuracy assessment. At minimum, the manuscript should report a quantitative agreement/uncertainty analysis for the remote sensing maps.

With only ~35 scenes over 2013–2018 (≈ 6 per year) and no explicit treatment of cloud cover impacts on phenology fits, TIMESAT-derived timing metrics are likely uncertain. Moreover, remote sensing labels summarize 2013–2018 whereas field sampling is in 2019 a gap that can be consequential in smallholder systems. These choices plausibly weaken soil–RS correspondence.  

Several sections are overly detailed (lab methods) while key methodological choices (RS preprocessing, TIMESAT parameters) are terse. 

Please also state whether a research permit/ethical clearance was obtained. 

L78-80: Reporting a single stocking rate (1–2 cattle ha⁻¹) without nuance is misleading; please contextualize it by describing the different production systems.

L82-83; The discussion of soil degradation is overly simplified. Even if not the central objective, the manuscript should briefly address the complexity of degradation processes and site-specific drivers at the study locations

l172: Use ‘difference’ rather than ‘differential’ here.”

L310: This is an isolated citation. 

Summary

This study investigates the use of satellite-derived vegetation indices to identify soil degradation in Eastern Kenyan grasslands. The authors classify fields into three categories (equilibrium, transition, and degraded) and validate these classifications against a large set of laboratory soil analyses of topsoil samples collected from a subset of the sites. Due to limited initial findings, a PCA and k-means clustering approach was applied to all the soil data, identifying two primary clusters which represent degraded and non-degraded soils.



While the development of rapid, remote-sensing-based degradation assessments is of great practical importance, there are significant concerns that need to be addressed. The major points are outlined below under "General comments" with more details provided in the "Detailed comments".

General comment

While the authors’ objectives and overall workflow are clear, the manuscript suffers from a lack of general clarity and technical detail.

There is an inconsistency in the terminology used throughout the paper, particularly regarding the degradation and soil property classifications (see detailed comments). Furthermore, while the authors provide an extensive list of laboratory-measured soil properties, reasons for their selection and the analysis remain unclear and confusing for the reader.

The description of the remote sensing methodology is incomplete. Vital information is missing regarding the selection (e.g. specific sensor details, image selection criteria, total number of images used, reasons for the temporal coverage) and the pre-processing (e.g. cloud masking is likely crucial in the humid tropics) of satellite imagery. Statistical analyses also require further explanations and details: It is unclear on what data the tests were performed (e.g. was there data splitting by site? What were explaining and response variables?) and whether model assumptions were verified before performing ANOVA. Additionally, the authors utilized pairwise t-tests to compare three levels of a factor (degraded, transitional, and equilibrium); a post-hoc test such as Tukey’s HSD could be more statistically robust and appropriate?

The results section generally lacks structure. Several results mentioned in the text are not supported by data in the manuscript or the supplementary material (see detailed comments below). Some visualizations would help the reader to better understand the results (e.g. boxplot and score plot; see below). The discussion section requires major rewriting. It is currently a mix of introduction and a repetition of the results, also often drifting into unrelated topics. Furthermore, the discussion does not cite enough literature (in my opinion, 10 references in the entire discussion section is not enough to discuss the results obtained (there are numerous high-quality articles published in the field)).

Ultimately, the study demonstrates that while laboratory analysis successfully identifies soil degradation status, the proposed remote sensing method does not. Given that the efficacy of laboratory analysis to identify soil degradation is already well-established, the value of this paper lies in a deeper exploration of why the remote sensing approach underperformed. The authors should focus their discussion on comparing their remote sensing results with existing studies to identify specific challenges or limitations.

Detailed comments

L24: Not all soils across Sub-Saharan Africa are degraded! Please rephrase.

L30: What is equilibrium, transition, and degraded soil? The reader has absolutely no idea what is meant. Please introduce the terms.

L40: What are stable and transient soil variables? Please explain.

L125-158: It would be more useful to have a description of the sampled sites instead of the hydrological catchments / basins. Also, since topography is a major factor influencing soil degradation this would be crucial to add.  

L156-158: The degradation classification should be properly introduced, and the terms should be kept consistent throughout the manuscript. It is not clear to the reader what equilibrium, transition, and degraded soil really means (see above).

L198-204: It is not relevant which parameters were calculated if they were not used within this study.

L171-225 and Table 1: It is not clear to the reader what kind of models were trained. It is also unclear why models were compared or what their underlying parameters are. Moreover, I recommend moving any data or outcomes presented in this section to the results (if it is relevant).

L235-236: This visual inspection seems arbitrary.

L241-243 and L246: What is “the status of the locations” and “land use history check”? Please add more details on this. The long-term history of the grassland is crucial to understanding soil degradation.

L255–310: The soil sample analyses section is currently disorganized and lacks critical information. Sample processing (e.g. drying, sieving, grinding) and citations for laboratory methods are missing. To improve clarity, the authors should remove any variables that were not further investigated within this study. I recommend presenting the measured soil properties in a table, for example categorized by their subsequent classification (stable/transient).

L312: Which soil variables?

L314: Why and how were variables classified and what does transient and stable mean? The information is missing completely. Literature is needed to justify such a classification too.

L325: Why are there missing observations? Specify. Number of samples should be added to the overview table (see comment above).

L330-337 and Table 2: It is not clear to the reader how the models were trained (which data, which explaining and response variables). Also, have the model assumptions been met? The dataset is unbalanced, has this been considered? Why was a t-test applied to a factor with three levels (degradation status). A Tukey’s HSD test might be more appropriate in this case..?

L337: What is the “content of the mean differences”?

L343-344 and L349-360: Why was the Gaussian mixture model compared to the k-means clustering? What does an application of a model “for reference” mean? This is again very arbitrary, and the entire paragraph is irrelevant and should be removed. L355-360 sounds like a justification for why k-means clustering was selected in the end. In my opinion this is not relevant.

L384-386: It is not clear to the reader how the t-test was used to determine the separation of populations.

L388-391: It is not necessary to describe the used R functions. The R packages and their versions should be sufficient.

L 394-395: It would be helpful to the reader to visualize the results of the variables which differ significantly between the degradation classes in a graph (for example in a boxplot).

Please correct line numbers (they should be continuous but randomly restart here).

L10-30: Numerous soil properties which are reported in this section are not presented in the manuscript (for example inorganic N, organic N, pH, clay, bulk density). When you report results, always refer to the table or figure where the reader can find these results (this is completely missing with one exception of the density plots). Moreover, it would be nice to have a graph showing the results of the principal component analysis and the k-means clustering.

Table 3: It would be interesting to see which sites have been correctly classified and which didn’t (for example using a confusion matrix). This table is otherwise irrelevant.

L23-25, Figure 3, and Figure 4: Use the same terms for soil properties in figures and text. Otherwise, the reader cannot easily relate the text to the figures. In these two figures it is not clear what the p-value indicates. Also, the number of samples in each cluster needs to be added.

L26: What are degradation labels?

L26-28: This belongs to the discussion and a reference is missing.

L44-53: This is a repetition of the introduction.

L71-72: Completely unrelated sentence to the topic.

L75: This is the first time rainfall appears and seems very arbitrary.

L85-96: This is repetition of your results. This does not belong to the discussion.

L97: Salinisation? Again, very arbitrary.

L105: Since when were these ratios used? These properties have not been introduced and occur the first time.

L96-108: Some discussion, but no reference at all.

L108-109: Repetition

L110-117: This part of the discussion seems interesting. But again, literature is missing and it needs to be extended.