Comparative Analysis of the Land Use and Land Cover Changes in Different Governorates of Oman using Spatiotemporal Multi-spectral Satellite Data

Muhammad Shafi; Syed Mohsin Bokhari

REVIEW 4 major objections 6 minor 23 references

Comparative Analysis of the Land Use and Land Cover Changes in Different Governorates of Oman using Spatiotemporal Multi-spectral Satellite Data

T0 review · 4 major / 6 minor · reviewed 2026-08-07 · deepseek-v4-flash

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

Pith's one-line read Satellite classification finds built-up area grew in every Omani governorate from 2017 to 2021.

desk verdict A useful comparative idea, but the circular accuracy assessment and missing figures sink the current claims. read the letter →

arxiv 2505.23285 v1 pith:RMS3UPI6 submitted 2025-05-29 cs.LG

Muhammad Shafi , Syed Mohsin Bokhari This is my paper · ORCID

classification cs.LG

keywords landuseandcoverchangeSentinel-2supervisedclassificationOmanurbanexpansioncroplanddeclinegovernoratecomparisonsatelliteremotesensing

verification ladder T0 review T1 audit T2 compute T3 formal

The pith

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The reading

Using annual Sentinel-2 multispectral imagery and a supervised classifier, this paper aims to establish national and governorate-level land-cover trends for Oman. It reports that built-up area increased in all 11 governorates, led in absolute terms by North Batinah and in percentage terms by Wusta, while Muscat showed little change. It also reports that crop area decreased in most governorates, with the largest losses in Dhofar and North Batinah. The authors present these results as evidence for continued urbanization and a warning that agricultural land is not keeping pace with population and built-up growth.

What carries the argument

The central machinery is a supervised machine-learning classification pipeline applied to Sentinel-2 L2A 10-meter multispectral imagery, with manually labelled training points for six land-cover classes: Water, Trees, Crops, Built Area, Bare Ground, and Rangeland. The classifier converts raw reflectance into consistent land-cover maps, and the study then sums class areas by governorate and compares the maps across annual time steps. This pipeline is what produces every reported built-up and crop-area change, so the validity of the trends rests entirely on the quality of those classifications.

What would settle it

Have a different analyst label a fresh random sample of points from very high-resolution imagery for, say, 2021, without seeing the classifier's output, and compare those labels to the map; if the independent user's accuracies for built-up and crop classes are materially below the reported 94%, the governorate-level area changes are not established.

Watch

Extended reading notes

Core claim

On the paper's own terms, the central discovery is a spatiotemporal pattern: from 2017 to 2021, urban land cover expanded in every Omani governorate. The largest absolute increase was in North Batinah; Wusta had the largest percentage increase because its built-up base was small; Muscat did not change substantially, which the authors attribute to saturation and vertical construction. In the same period, crop area fell in most governorates, with Dhofar and North Batinah losing the most, although a few governorates showed marginal gains. The paper reads this combination—urban growth everywhere, cropland decline in most places, and crop area growing more slowly than population—as a signal that Oman needs deliberate policy to protect and expand agricultural land.

Load-bearing premise

The load-bearing premise is that the reported 94% overall accuracy is a genuine measure of map quality; the paper's own description does not rule out that the same 600 manually labelled points were used both to train and to test the classifier, in which case real accuracy could be substantially lower and every derived area change would be in doubt.

Editorial extensions

If this is right

Urban land in Oman is not static: every governorate added built-up area during the study period, so national planning must treat expansion as widespread rather than Muscat-centric.
North Batinah's absolute growth and Wusta's relative growth point to different pressures—large-scale coastal urbanization versus a small baseline—so governorate-level policies may need different targets.
The reported decline of crop area in most governorates, especially Dhofar and North Batinah, would imply that agricultural land is being lost while population and built-up area grow.
If these trends continue, urban land demand and food production will increasingly compete in the same governorates, making crop-area monitoring a policy-relevant indicator.

Reading between the lines

Editorial extensions of the paper, not claims the author makes directly.

Not in the paper: the same maps could be tested against independent validation points drawn from very high-resolution imagery, which would settle whether the reported accuracy holds outside the training sample.
A natural extension is to regress built-up growth against governorate population and area; Wusta's large percentage increase may be mostly a small-baseline effect rather than rapid urbanization.
A longer time series before 2017 would reveal whether the 2017-2021 pattern is part of a sustained urbanization wave or a short-term shift.
The cropland decline could be cross-checked against national agricultural statistics and groundwater salinity records, which would strengthen or weaken the satellite-only result.

Share X Bluesky LinkedIn Reddit HN

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, and a circularity audit.

Desk Editor's Note

read the letter

Quick take: this is a standard Sentinel-2 LULC application to Oman's governorates, with a useful comparative framing but a load-bearing validation flaw. The governorate-level five-year comparison is genuinely new for Oman—prior work covers single regions or one time step. The writing is clear and the domain motivation is solid; the references to Omani LULC studies are relevant.

The paper does not, however, establish its reported trends. Section 2.4 says they picked 100 random points per class, labeled them, trained the classifier, then picked another 100 points per class from the classified data for visual assessment. The confusion matrix in Table 3 shows exactly 100 reference points per class. The authors never say the two sets are disjoint, so the most natural reading is that the accuracy is computed on the training labels. That is circular, and it means the 94% overall accuracy does not measure generalization.

Even taken at face value, the numbers do not add up. The diagonal of Table 3 sums to 579 out of 600, which is 96.5%, not 94%. That is a small arithmetic discrepancy, but it doesn't inspire confidence. More importantly, the actual results—the built-up and crop area changes by governorate—are only in Figures 3-6, which are absent from the manuscript. There are no per-governorate area tables or annual values anywhere, so the central claims cannot be checked.

The method section is also under-specified: no classifier name, no hyperparameters, no preprocessing details. That alone would block reproducibility.

To be fair, the paper is honest about being an applied comparative study, and the idea of an annual governorate-level LULC dataset for Oman is a reasonable contribution if done properly. The authors should redo the accuracy assessment with a held-out set, report the algorithm and parameters, and include the area tables and figures.

My recommendation: as submitted, this does not deserve peer review. The flaws are fundamental and the key evidence is missing. I would desk reject with an invitation to resubmit after a proper validation and full reporting. That is more constructive than sending it out in the current state.

Referee Report

4 major / 6 minor

Summary. The manuscript reports a supervised classification of Sentinel-2 L2A imagery (10 m resolution) for the years 2017-2021 across 11 governorates of Oman, with the goal of quantifying land use/land cover change. Six land-cover classes are used (Water, Trees, Crops, Built-up Area, Bare Ground, and, in the accuracy tables, Rangeland). The authors state that 600 manually labeled points were used to train a supervised classifier and that an overall accuracy of 94% was obtained. Based on the resulting classified maps, the paper claims that built-up area increased in all governorates—largest absolute increase in North Batinah, largest relative increase in Wusta—and that crop area decreased in most governorates, particularly in Dhofar and North Batinah. The paper concludes with policy recommendations for urban planning and agricultural sustainability.

Significance. If the reported trends were reliable, the paper would provide useful sub-national evidence on urbanization and agricultural land change in Oman, an under-studied region for LULC analysis. The use of freely available Sentinel-2 data and the comparative governorate-level design are appropriate and potentially valuable for local planning. That said, the central quantitative basis is not currently verifiable: the accuracy assessment appears to be circular, the stated overall accuracy contradicts the paper's own confusion matrix, and the figures and tables that would support the change claims are missing. As submitted, the findings therefore cannot be evaluated, and the paper does not yet constitute a citable scientific contribution.

major comments (4)

[2.4, Tables 3-4] The validation protocol does not establish generalization. The text says that 100 random points per class were labeled for supervised learning (600 total), and then that 'Again 100 points were picked randomly from the classified data' for visual assessment. It never states that the second sample is spatially disjoint from the training points or withheld from training, and Table 3's reference totals sum to exactly 600, the same as the training sample. If the assessment set is the training set, the confusion matrix reports only reclassification accuracy, and the 94% figure cannot support the land-cover maps or any change statistic derived from them. A proper independent validation design—with a separate, spatially stratified, and temporally explicit sample—is required.
[2.4, Table 3] The reported overall accuracy is not reproducible from the paper's own numbers. The diagonal of Table 3 sums to 98+96+97+96+96+96=579 out of 600, i.e., 96.5%, not the 94% stated in the text. The authors should recompute the confusion matrix and reconcile the text with the table, or provide the raw classification/reference pairs if the table contains a different aggregation.
[3, Figures 3-6] The core empirical claims are unsupported by any quantitative data in the manuscript. Figures 3-6, which are supposed to show built-up area, percentage increase in built-up area, crop area, and percentage change in crop area by governorate, are not included, and no per-governorate area values or annual change tables are given elsewhere. As a result, the claims of universal built-up increase, the North Batinah maximum, the Wusta maximum percentage increase, and the Dhofar/North Batinah crop decrease cannot be checked or reproduced. These figures and the underlying area statistics must be provided for the Results section to be meaningful.
[2.3-2.4] The classification methodology is underspecified to the point of non-reproducibility. The text refers to 'supervised machine learning algorithms' without naming the algorithm, its hyperparameters, features, or software; nor does it explain whether one classifier is applied across all years or separate classifiers are trained per year. In addition, the class nomenclature is inconsistent: Section 2.3 defines five classes, while Tables 3-4 include a sixth class, Rangeland, which is never defined. These details are essential for any interpretation of the classified outputs.

minor comments (6)

[Abstract] The abstract states the study covers 2016 to 2021, while Section 2.2 lists 2017-2021. Please align the time period.
[Keywords] The keywords line has a doubled colon (': : Classification, ...').
[Table 2] The governorate name 'Al Batinah North' in Table 2 is rendered as 'North Batinah' in the text and Section 3.1; use one convention consistently.
[Figures 1-2] Figures 1 and 2 (study-area map and methodology flowchart) are also only placeholder captions; both should be included.
[Section 3.1] The statement that 'the Muscat governorate has not seen any significant increase' is vague and potentially contradicted by the claim that built-up area increased in all governorates; quantify 'significant' or rephrase.
[References] Several references lack volume/page/DOI details (e.g., [2], [19], [20]), which will need completion for publication.

Circularity Check

0 steps flagged · score 0.0 of 10

No demonstrated circularity: the validation sample is described as a second random draw, so the accuracy claim is not by construction a training-set refit; the unresolved numerical and visual inconsistencies are correctness and completeness issues, not circularity.

full rationale

The paper's derivation chain is an empirical supervised classification workflow: 100 labeled points per class are selected for training, supervised learning is applied, and then "Again 100 points were picked randomly from the classified data and that was visually assessed if the label had been correctly assigned or not" (Section 2.4). On its face this is the standard two-sample design — a training sample followed by an independent accuracy-assessment sample — so the reported overall accuracy is not demonstrably a reclassification of the training labels. The text does not explicitly state that the validation points are spatially disjoint from the training points, but neither does it state or imply that they are the same points; interpreting the second draw as the training set would be speculation, which the review rules prohibit. Several serious non-circular problems do exist: the diagonal of Table 3 sums to 579/600 = 96.5%, not the reported 94%; Table 3 includes a "Rangeland" class that is not defined in Section 2.3; and Figures 3-6, which are the only quantitative support for the governorate-level change claims, are absent from the manuscript. These undermine reproducibility and confidence, but they are not cases where a prediction reduces to an input by construction. There are no self-citations, no imported uniqueness theorems, no fitted parameters renamed as predictions, and no equations that make a result definitionally equivalent to its inputs. The area-change conclusions inherit the uncertainty of an insufficiently documented accuracy assessment, but unvalidated is not the same as circular. Therefore the appropriate circularity score is 0.

Assumptions & free parameters 2 free parameters · 3 assumptions · 0 invented entities

The paper introduces no new entities. Its empirical results rest entirely on the classifier's labels and on the chosen training/reference points, so the ledger is dominated by domain assumptions and the unvalidated accuracy claim.

free parameters (2)

Training sample size per class = 100 points
The number of manually labeled points per class is chosen by the authors (Section 2.4); this choice affects the classifier's fit and the accuracy table, and no analysis of its sensitivity is given.
Classification algorithm and hyperparameters = not reported
The supervised machine learning algorithm is never named (Section 2.4), so its internal parameters are unstated, making the classification a black box with unconstrained flexibility.

assumptions (3)

domain assumption Sentinel-2 L2A surface reflectance at 10 m resolution is sufficient to distinguish water, trees, crops, built-up, bare ground, and rangeland in Oman
Stated in Sections 2.2-2.3; no spectral separability analysis is provided.
domain assumption Manual visual labeling of 100 random points per class in QGIS produces reliable ground truth
Section 2.4 uses these labels as reference data; no independent field verification or expert cross-check is described.
ad hoc to paper The classifier's accuracy on the training points generalizes to the entire governorate
Section 2.4 reports 94% overall accuracy, but the evaluation appears to use the same points as training, so generalization is assumed rather than demonstrated.

how reviews work

0 comments

Cite this review

Pith. "Pith review of Comparative Analysis of the Land Use and Land Cover Changes in Different Governorates of Oman using Spatiotemporal Multi-spectral Satellite Data." pith.science (2026). https://pith.science/paper/RMS3UPI6

@misc{pith2026250523285,
  author       = {Pith},
  title        = {Pith review of: Comparative Analysis of the Land Use and Land Cover Changes in Different Governorates of Oman using Spatiotemporal Multi-spectral Satellite Data},
  year         = {2026},
  howpublished = {\url{https://pith.science/paper/RMS3UPI6}},
  note         = {Machine review of arXiv:2505.23285}
}

read the original abstract

Land cover and land use (LULC) changes are key applications of satellite imagery, and they have critical roles in resource management, urbanization, protection of soils and the environment, and enhancing sustainable development. The literature has heavily utilized multispectral spatiotemporal satellite data alongside advanced machine learning algorithms to monitor and predict LULC changes. This study analyzes and compares LULC changes across various governorates (provinces) of the Sultanate of Oman from 2016 to 2021 using annual time steps. For the chosen region, multispectral spatiotemporal data were acquired from the open-source Sentinel-2 satellite dataset. Supervised machine learning algorithms were used to train and classify different land covers, such as water bodies, crops, urban, etc. The constructed model was subsequently applied within the study region, allowing for an effective comparative evaluation of LULC changes within the given timeframe.

Discussion (0). Continue with ORCID to comment.

Reference graph

Works this paper leans on

23 extracted references · 23 canonical work pages

[1]

Mapping Land Use and Land Cover Change in the Pearl River Delta Region, China,

X. Yang et al., "Mapping Land Use and Land Cover Change in the Pearl River Delta Region, China," Remote Sensing, 2020

work page 2020
[2]

Object-Based Image Analysis for Land Use Classification,

T. Blaschke et al., "Object-Based Image Analysis for Land Use Classification," International Journal of Remote Sensing, 2024

work page 2024
[3]

Object-Based Classification for LULCC in Coastal Kerala, India,

R. Vijay et al., "Object-Based Classification for LULCC in Coastal Kerala, India," Journal of Environmental Management, 2024

work page 2024
[4]

Post-Classification Comparison for LULCC in the Three Gorges Reservoir, China,

X. Hao et al., "Post-Classification Comparison for LULCC in the Three Gorges Reservoir, China," GIScience & Remote Sensing, 2019

work page 2019
[5]

Image Differencing for Land Cover Change Detection,

D. Panuju et al., "Image Differencing for Land Cover Change Detection," Photogrammetric Engineering & Remote Sensing, 2020

work page 2020
[6]

Monitoring Wetland Transformation Using Image Differencing in Pearl River Delta, China,

W. Ji et al., "Monitoring Wetland Transformation Using Image Differencing in Pearl River Delta, China," Ecological Indicators, 2019

work page 2019
[7]

J. R. Jensen, Remote Sensing of the Environment: An Earth Resource Perspective, Prentice Hall, 2009

work page 2009
[8]

NDVI-Based Land Cover Change Analysis in the Yangtze River Delta, China,

J. Wan et al., "NDVI-Based Land Cover Change Analysis in the Yangtze River Delta, China," Remote Sensing Applications: Society and Environment, 2020

work page 2020

Show all 23 references

[9]

Urban Expansion and Land Use Transitions: A Global Perspective,

K. C. Seto et al., "Urban Expansion and Land Use Transitions: A Global Perspective," PNAS, 2012

2012
[10]

Urbanization and Land Use Dynamics in Tianjin Binhai, China,

H. Long et al., "Urbanization and Land Use Dynamics in Tianjin Binhai, China," Urban Studies, 2014

2014
[11]

High-Resolution Global Maps of 21st-Century Forest Cover Change,

M. C. Hansen et al., "High-Resolution Global Maps of 21st-Century Forest Cover Change," Science, 2013

2013
[12]

Quantifying Forest Cover Loss in the Brazilian Amazon Using Satellite Data,

A. Tyukavina et al., "Quantifying Forest Cover Loss in the Brazilian Amazon Using Satellite Data," Global Change Biology, 2017. 9 of 9

2017
[13]

Satellite Remote Sensing for Agricultural Land Use Monitoring and Food Security,

S. Fritz et al., "Satellite Remote Sensing for Agricultural Land Use Monitoring and Food Security," Land Use Policy, 2015

2015
[14]

Mapping Crop Types and Rotations in the Argentine Pampas Using Remote Sensing,

A. Baldassini et al., "Mapping Crop Types and Rotations in the Argentine Pampas Using Remote Sensing," Agricultural Systems, 2024

2024
[15]

Remote Sensing for Biodiversity Science and Conservation,

W. Turner et al., "Remote Sensing for Biodiversity Science and Conservation," Trends in Ecology & Evolution, 2016

2016
[16]

GIS-Based Land Surveys in Mountain Oases of Oman,

M. Al-Rawahi et al., "GIS-Based Land Surveys in Mountain Oases of Oman," Arabian Journal of Geosciences, 2014

2014
[17]

Supervised Classification for Land Use Change in Jabal Al Akhdar, Oman,

E. Luedeling and A. Buerkert, "Supervised Classification for Land Use Change in Jabal Al Akhdar, Oman," Remote Sensing of Environment, 2008

2008
[18]

Geospatial Analysis of Agricultural Shifts in Oman,

P. Deadman et al., "Geospatial Analysis of Agricultural Shifts in Oman," Journal of Arid Environments, 2016

2016
[19]

Land Use Changes in Oman’s Braka Province,

T. Caspari, D. Donato, and M. Jendryke, "Land Use Changes in Oman’s Braka Province," Land Degradation & Development, 2019

2019
[20]

Assessing Land Use in Wilayat Nizwa Using Remote Sensing,

E. Fadda, B. Al Shebli, and M. Al Kabi, "Assessing Land Use in Wilayat Nizwa Using Remote Sensing," Environmental Earth Sciences, 2019

2019
[21]

Spatiotemporal Analysis of Land Use in Musandam, Oman,

A. Megdiche-Kharrat et al., "Spatiotemporal Analysis of Land Use in Musandam, Oman," International Journal of Remote Sensing, 2019

2019
[22]

Remote Sensing of Land Use Changes in Dhofar, Oman,

E. Ramadan, T. Al-Awadhi, and Y. Charabi, "Remote Sensing of Land Use Changes in Dhofar, Oman," Applied Geography, 2021

2021
[23]

Impact of Tropical Cyclones Gonu and Phet on Land Use in Oman,

S. Al-Hatrushi and A. Al-Alawi, "Impact of Tropical Cyclones Gonu and Phet on Land Use in Oman," Natural Hazards, 2021

2021

Pith tools

Reviewed August 7, 2026 · model on record in the stance chip above.

[1] [1]

Mapping Land Use and Land Cover Change in the Pearl River Delta Region, China,

X. Yang et al., "Mapping Land Use and Land Cover Change in the Pearl River Delta Region, China," Remote Sensing, 2020

work page 2020

[2] [2]

Object-Based Image Analysis for Land Use Classification,

T. Blaschke et al., "Object-Based Image Analysis for Land Use Classification," International Journal of Remote Sensing, 2024

work page 2024

[3] [3]

Object-Based Classification for LULCC in Coastal Kerala, India,

R. Vijay et al., "Object-Based Classification for LULCC in Coastal Kerala, India," Journal of Environmental Management, 2024

work page 2024

[4] [4]

Post-Classification Comparison for LULCC in the Three Gorges Reservoir, China,

X. Hao et al., "Post-Classification Comparison for LULCC in the Three Gorges Reservoir, China," GIScience & Remote Sensing, 2019

work page 2019

[5] [5]

Image Differencing for Land Cover Change Detection,

D. Panuju et al., "Image Differencing for Land Cover Change Detection," Photogrammetric Engineering & Remote Sensing, 2020

work page 2020

[6] [6]

Monitoring Wetland Transformation Using Image Differencing in Pearl River Delta, China,

W. Ji et al., "Monitoring Wetland Transformation Using Image Differencing in Pearl River Delta, China," Ecological Indicators, 2019

work page 2019

[7] [7]

J. R. Jensen, Remote Sensing of the Environment: An Earth Resource Perspective, Prentice Hall, 2009

work page 2009

[8] [8]

NDVI-Based Land Cover Change Analysis in the Yangtze River Delta, China,

J. Wan et al., "NDVI-Based Land Cover Change Analysis in the Yangtze River Delta, China," Remote Sensing Applications: Society and Environment, 2020

work page 2020