Oceanographic-Advanced-QC

Neighbor Test (NT) – MATLAB Example

This repository provides a compact MATLAB example of the Neighbor Test (NT) for quality control of multiplatform oceanographic observations.

The example accompanies an associated study on advanced quality control and demonstrates the main workflow without reproducing the complete platform-specific processing and analysis.

Example

The example compares near-surface salinity observations from:

• Platform A: FerryBox (target)
• Platform B: E1-M3A buoy (neighbor)

The example dataset covers October 2017–October 2018 in the Cretan Sea, Eastern Mediterranean. A geographically reduced subset of the observations is provided for demonstration purposes.

Repository structure

Neighbor_Test/
│
├── README.md
├── run_NT_example.m
├── functions/
│   ├── nt_find_matchups.m
│   ├── nt_calculate_metrics.m
│   ├── nt_apply_threshold.m
│   └── nt_plot_sensitivity.m
├── example_data/
│   └── NT_example_data.mat
└── output/

run_NT_example.m is the main script.

Method

The example workflow:

1. loads and checks the two datasets;
2. evaluates different spatial (R) and temporal (TW) collocation criteria;
3. identifies point-to-point matchups;
4. calculates RMSD, bias, MAE, correlation, and number of matchups;
5. applies the selected collocation criteria;
6. calculates FerryBox–E1-M3A salinity differences;
7. applies the Neighbor Test threshold; and
8. produces diagnostic figures and saves the results.

For this compact example, the tested criteria are:

radiusDeg = [0.2 0.3 0.4 0.5 0.6];

timeWindowHours = [3 6 12 24 36 48 60 72 84];

The selected configuration for this comparison is:

R_selected_deg = 0.20;
TW_selected    = 60;

TW represents the full temporal-window width; therefore, TW = 60 h corresponds to ±30 h.

Spatial distances are calculated using the Haversine formula. For each target observation, neighboring observations within the spatial radius are identified and the observation with the smallest absolute temporal separation is selected. No temporal or spatial averaging is performed.

Neighbor Test threshold

Differences are calculated as:

ΔS = FerryBox salinity − E1-M3A salinity

For salinity, the reference variability is 0.20 psu and the NT threshold is:

NT_threshold = 2 * 0.20;   % ±0.40 psu

Matchups with |ΔS| > 0.40 psu are flagged as suspect. Threshold exceedance does not necessarily indicate an erroneous observation, as differences may also reflect genuine environmental variability or sampling differences.

Running the example

Set MATLAB to the repository directory and run:

run_NT_example

The supplied data file contains the generic variables:

timeA, lonA, latA, varA
timeB, lonB, latB, varB

Results are saved to:

output/NT_example_results.mat

Adapting the code

The functions use generic Platform A and Platform B inputs and can be adapted to other pairs of point-observation datasets by changing the input data, collocation criteria, and NT threshold.

Requirements

• MATLAB

Citation

Citation

If you use this code in scientific or scholarly work, please cite the associated publication:

Stamataki, N., Frangoulis, C., Tsiaras, K., De Mey-Frémaux, P., Petihakis, G., and Sofianos, S. (2026). Multiplatform and multivariate quality control of physical and biogeochemical ocean observations: From quality control to environmental event identification. Frontiers in Marine Science, Ocean Observation (manuscript submitted for publication).

The full publication citation and DOI will be added upon publication.

Scope

This repository provides a reproducible example of the point-to-point Neighbor Test workflow. It does not include the complete platform-specific processing chain, original full datasets, or all analyses performed in the associated study.