Pre-Submission Review for Remote Sensing Papers

Remote sensing reviewers often ask:

• why were these sensors, bands, spatial resolution, and dates chosen?
• are preprocessing, atmospheric correction, georeferencing, cloud masking, and compositing described enough?
• is ground truth reliable and independent?
• is there leakage across tiles, scenes, sites, seasons, or time?
• do validation results show geographic and temporal transfer?
• are baselines, ablations, uncertainty, and error maps strong enough?
• are data, labels, code, and trained models available or constrained?
• does the paper fit IEEE TGRS, Remote Sensing of Environment, Remote Sensing, IJAEOG, an application journal, or a computer-vision venue?

The manuscript has to prove the result is remote sensing, not just image modeling.

In our pre-submission review work, remote sensing papers most often fail when the model result is reported without enough geospatial validation discipline.

Leakage risk: training and test data are split by patches or pixels even though nearby areas share scene conditions, labels, or acquisition artifacts.

Ground-truth opacity: labels come from maps, manual annotation, field surveys, or public datasets, but accuracy and independence are not clear.

Transfer gap: a model performs well in one region or season but the manuscript claims broad applicability.

Sensor-method mismatch: the paper uses SAR, LiDAR, hyperspectral, multispectral, or UAV data without explaining what the sensor contributes.

Application overclaim: the manuscript writes policy, ecology, disaster, or agriculture conclusions before validation supports that decision use.

A useful review should identify the first remote-sensing-specific objection a reviewer would raise.

IEEE TGRS describes its scope around sensing the land, oceans, atmosphere, and space, plus processing, interpretation, and dissemination of that information. Its author information emphasizes novel methodological advancement and experimental completeness. Remote Sensing instructions ask authors to consider ethics, data, figures, references, and full datasets where possible. IJAEOG author guidance sits at the applied earth-observation boundary, where application value and geoinformation quality matter.

Remote sensing is not the same as ordinary image classification. Spatial autocorrelation, sensor physics, ground truth, atmospheric effects, geolocation, and transfer all shape whether reviewers trust the result.

These expectations make validation design a first-order submission issue.

Send the manuscript, target journal, sensor and acquisition table, preprocessing workflow, ground-truth description, spatial and temporal split plan, validation maps, baseline comparisons, ablations, uncertainty analysis, data and label access plan, code repository plan, figures, supplement, and prior reviewer comments.

For ML remote-sensing papers, include split logic, leakage safeguards, model cards if available, training details, data provenance, and external validation. For application papers, include how the remote-sensing output connects to the environmental, agricultural, urban, ocean, or disaster decision.

A useful remote sensing pre-submission review should include:

• remote-sensing contribution verdict
• sensor and preprocessing critique
• ground-truth and validation review
• leakage and transfer check
• baseline, ablation, and uncertainty review
• data, label, and code readiness note
• journal-lane recommendation
• submit, revise, retarget, or diagnose deeper call

The review should not only say "add validation." It should say which validation design would make reviewers trust the result.

Before submission, authors often need to:

• explain sensor choice and acquisition timing
• add preprocessing and quality-control details
• redesign train-test splits by site, scene, time, or geography
• add external validation or cross-region tests
• document ground-truth source and uncertainty
• add baselines, ablations, and error maps
• limit application claims to validated conditions
• retarget from TGRS or Remote Sensing of Environment to Remote Sensing, IJAEOG, a domain journal, or a computer-vision venue

These fixes make the earth-observation claim easier to trust.

A SAR paper needs speckle, polarization, incidence-angle, and interpretation discipline. A hyperspectral paper needs spectral preprocessing, feature logic, and ground truth. A UAV paper needs flight, sensor, ground-control, and scale detail. A land-cover paper needs label quality and geographic transfer. A disaster-mapping paper needs time sensitivity and independent validation. A geospatial AI paper needs leakage control, baselines, and external testing. A sensor-fusion paper needs to explain what each source contributes.

The AI manuscript review can flag whether the blocking risk is sensor logic, leakage, ground truth, transfer, or journal fit.

Use this page when the manuscript's submission risk depends on earth-observation data, sensor physics, geospatial validation, ground truth, leakage across space or time, remote-sensing application value, or remote-sensing journal fit. Use computer vision review when the paper is mainly about a general model or benchmark. Use oceanography review when ocean process interpretation dominates and remote sensing is only one data source.

That distinction keeps the page focused on the remote-sensing buyer's actual problem.

Do not submit a remote sensing paper if validation is only a random pixel or patch split. Spatial autocorrelation can make a model look strong while failing on new sites, seasons, sensors, or acquisition conditions.

Also pause if ground truth is not independent. If labels come from products derived from the same imagery or closely related sources, reviewers need to know that limitation.

For method papers, pause if baselines are weak or undertuned. Remote sensing reviewers are used to seeing new architectures beat inadequate comparisons.

For application papers, pause if the decision claim is stronger than the validation. A flood, crop, forest, urban, or ocean product may be useful, but the manuscript should state where it works and where it is not yet proven.