Ocean & Atmosphere Analysis
Technical & Scientific Documentation

1. Scope & Proposed Toolbox

The module is organized by geophysical phenomenon, not by satellite name. MODIS, Sentinel-3 OLCI, Sentinel-3 SLSTR, numerical weather models, current products and later altimetry are data sources feeding the same analysis concepts.

Ocean & Atmosphere
│
├─ Ocean Dynamics
│  ├─ Sea Surface Temperature (SST)
│  ├─ SST Anomaly
│  ├─ Thermal Front Detection
│  ├─ Ocean Front Detection
│  ├─ Upwelling Detection
│  ├─ Current Speed & Direction
│  └─ Ocean Forecast
│
├─ Marine Water Properties
│  ├─ Chlorophyll-a
│  ├─ Turbidity
│  ├─ Total Suspended Solids / Matter (TSS / TSM)
│  ├─ CDOM
│  ├─ NDCI
│  ├─ Floating Algae Index (FAI)
│  ├─ Maximum Chlorophyll Index (MCI)
│  └─ Algal Bloom Detection
│
├─ Atmosphere
│  ├─ Cloud Mask
│  ├─ Cloud Fraction
│  ├─ Aerosol Optical Depth (AOD)
│  └─ Water Vapor
│
├─ Weather Forecast
│  ├─ Temperature Forecast
│  ├─ Rainfall Forecast
│  ├─ Wind Forecast
│  ├─ Humidity Forecast
│  ├─ Pressure Forecast
│  └─ Weather Nowcasting
│
└─ Fisheries Intelligence
   └─ Potential Fishing Zone (PFZ)

2. Sensor & Model Sources

3. Scientific Workflow

3.1 Observation versus forecast

SLSTR → SST
OLCI → Chlorophyll
MODIS → FAI
Cloud bands → Cloud Mask

NWP / Ocean Model
initial state + physics
      ↓
+3h / +6h / +24h / +72h

4. Recommended Data Contract

4.1 Generic raster metadata

• sensor / product family;
• acquisition or valid time;
• variable and physical unit;
• band center wavelength for spectral algorithms;
• scale and offset;
• NoData and quality flags;
• cloud/land/ice mask where relevant;
• projection and geolocation;
• product level and processing lineage.

4.2 Remote-sensing reflectance

Ocean-colour algorithms often use remote-sensing reflectance R_rs, conventionally expressed per steradian. The exact input variable must be declared rather than treating any RGB-like reflectance as interchangeable.

4.3 Forecast metadata

Forecast rasters require both:

• initialization/reference time — when the model cycle starts;
• valid time / lead time — when the forecast applies.

model_run = 2026-08-17 00:00 UTC
lead_time = +24 h
valid_time = 2026-08-18 00:00 UTC

5. Sea Surface Temperature (SST)

SST represents the temperature of the ocean surface or near-surface skin/subskin according to the product definition. For the platform, the preferred route is to consume a validated Level-2/Level-3 SST product when available. A raw split-window retrieval should be an explicit advanced mode, because coefficients depend on sensor, calibration and atmospheric conditions.

5.1 Generic split-window formulation

5.2 Kelvin to Celsius

5.3 Output

Input : SLSTR / MODIS validated SST or thermal product
Output: 1-band Float32 SST raster
Unit  : °C or K, explicitly recorded
Masks : cloud / land / invalid retrieval

6. SST Anomaly

SST anomaly quantifies departure from a reference mean/climatology for the comparable calendar period.

6.1 Standardized anomaly

6.2 Output

SST anomaly raster:
negative → cooler than baseline
zero     → close to baseline
positive → warmer than baseline

Optional:
standardized anomaly (z-score)
hot/cold anomaly class raster

7. Thermal Front Detection

A thermal front is a zone of strong horizontal SST gradient. The core product is a gradient magnitude; a front mask is derived by thresholding or edge-detection logic.

Direction of increasing temperature can be expressed as:

7.1 Recommended modes

• Sobel / finite-difference gradient;
• multi-scale gradient;
• adaptive threshold based on local statistics;
• optional morphological thinning to produce front lines.

SST
 ↓
quality mask
 ↓
optional spatial smoothing
 ↓
gradient magnitude
 ↓
threshold
 ↓
Front Strength Raster + Front Mask

8. Ocean Front Detection

Ocean fronts need not be purely thermal. Ocean colour gradients can reveal boundaries that are weak in temperature. A multi-variable front detector may combine normalized SST and chlorophyll gradients.

8.1 Output

• front-strength continuous raster;
• binary/categorical front mask;
• optional vectorized front line.

9. Upwelling Detection

Coastal upwelling often brings relatively cool, nutrient-rich water toward the surface. Satellite indicators can therefore combine negative SST anomaly, chlorophyll enhancement and frontal structure. However, the relationship varies by region and season, so this should be implemented as a calibrated upwelling indicator/probability, not a universal physical retrieval.

9.1 Normalized indicator components

9.2 Composite upwelling score

9.3 Advanced wind-driven upwelling

When surface wind is available, Ekman transport can provide a physically grounded indicator:

10. Ocean Current Speed & Direction

MODIS, OLCI and SLSTR do not directly provide horizontal current vectors. For true current speed/direction, the platform should accept U/V current components from an ocean model/reanalysis, or later derive geostrophic current from suitable sea-level products.

10.1 From eastward and northward components

For a compass-style direction convention, conversion must explicitly define whether the direction means toward or from and how angles are normalized to 0–360°.

10.2 Geostrophic current from sea-surface height

For sufficiently large-scale, away-from-equator geostrophic flow:

10.3 Output

current_speed.tif       m/s
current_direction.tif   degree
optional current_vector layer
optional U/V source bands retained

11. Ocean Forecast

A future ocean-current or ocean-temperature forecast should primarily ingest an operational ocean-model product. Satellite time series can be used for bias correction, validation or data-driven forecasting, but a single image is not sufficient.

11.1 Forecast state notation

11.2 Potential forecast variables

• SST / subsurface temperature where model provides it;
• eastward current U;
• northward current V;
• current speed/direction;
• sea surface height/anomaly;
• mixed-layer properties where available.

11.3 Platform output

Variable: current_speed
Run time: T0

+03 h → raster
+06 h → raster
+12 h → raster
+24 h → raster
+48 h → raster
+72 h → raster

12. Chlorophyll-a

Chlorophyll-a is a widely used proxy for phytoplankton biomass. For open ocean, operational ocean-colour processors often use blue/green band-ratio families such as OCx. For turbid/coastal waters, red/red-edge algorithms such as NDCI or regional algorithms can be more appropriate. The platform should distinguish validated product ingestion from algorithmic retrieval from reflectance.

12.1 Generic OCx polynomial structure

12.2 Output

1-band Float32 raster
Variable: chlorophyll_a
Unit: typically mg m^-3
Quality: algorithm/product flags must be retained

13. Turbidity

Turbidity describes optical cloudiness caused by suspended particles and other constituents. Single-band red/NIR semi-analytical algorithms are widely used, but coefficients depend on wavelength and calibration.

13.1 Generic Nechad-style form

13.2 Output

turbidity.tif
Unit: e.g. FNU/NTU depending on calibrated algorithm
Metadata:
- algorithm name/version
- wavelength
- coefficients
- valid range

14. Total Suspended Solids / Total Suspended Matter

TSS/TSM/SPM retrievals use optical reflectance as a proxy for suspended particulate concentration. A generic semi-analytical family can be expressed similarly to turbidity:

Exact coefficients, units and valid ranges require calibration/validation against in-situ concentration measurements. Therefore the platform should support:

• official/sensor-specific coefficient presets;
• custom coefficients;
• regional calibration mode;
• metadata recording of calibration source and concentration units.

15. Colored Dissolved Organic Matter (CDOM)

CDOM strongly absorbs blue light and can affect chlorophyll retrieval. Satellite products may estimate a CDOM absorption coefficient, often at a reference wavelength, through empirical ratios or semi-analytical inversion. There is no single universal band-ratio equation suitable for every water type.

15.1 Generic empirical model

15.2 Spectral slope model

16. Normalized Difference Chlorophyll Index (NDCI)

NDCI was developed for chlorophyll-sensitive analysis in turbid productive waters using a red and red-edge band.

For MERIS-style wavelengths, the original concept uses bands near approximately 708 nm and 665 nm:

17. Floating Algae Index (FAI)

FAI detects enhanced NIR reflectance from floating algae relative to a linear spectral baseline connecting red and SWIR bands.

17.1 Baseline

17.2 FAI

The original MODIS-oriented formulation uses suitable red, NIR and SWIR bands. For another sensor, wavelengths must be adapted carefully; it is not enough to pick bands solely by ordinal number.

17.3 Output

• continuous FAI raster;
• optional thresholded floating-algae mask;
• optional polygonized extent.

18. Maximum Chlorophyll Index (MCI)

MCI measures the height of a red-edge spectral peak relative to a linear baseline. It is useful for intense surface plankton blooms and related high-biomass features when the sensor provides appropriate bands.

18.1 Generic line-height equation

λ₂ is the target red-edge peak band, while λ₁ and λ₃ define the baseline. For MERIS/OLCI-like use, the relevant wavelengths are around the red/red-edge region; exact band mapping must be sensor-aware.

19. Algal Bloom Detection

Algal bloom detection should be treated as a multi-indicator decision tool rather than one universal threshold. Depending on water type and sensor, it may use chlorophyll-a, NDCI, MCI, FAI, fluorescence products, water temperature and contextual masks.

19.1 Generic score

19.2 Suggested output

0 = No / invalid
1 = Low indication
2 = Moderate
3 = High
4 = Very High

Optional:
continuous bloom score
polygonized bloom extent

20. Cloud Mask & Cloud Fraction

Cloud masking is a prerequisite for many ocean/atmosphere analyses. Whenever an official quality/cloud flag exists, using it is generally preferable to inventing a generic threshold-based mask. A custom multispectral mask can still be offered as an advanced fallback.

20.1 Cloud fraction

20.2 Suggested mask classes

0 = Clear
1 = Cloud
2 = Cirrus
3 = Cloud Shadow
4 = Snow/Ice
255 = NoData

For ocean products, a separate land mask should normally be retained instead of merging land with cloud.

21. Aerosol Optical Depth (AOD)

AOD represents column-integrated aerosol extinction at a specified wavelength. A basic radiative-transfer relationship can be illustrated using Beer–Lambert attenuation:

21.1 Platform recommendation

• prefer ingestion of validated satellite AOD products;
• record reference wavelength, e.g. 550 nm if defined by the product;
• retain retrieval quality flags;
• do not present a simple Beer–Lambert inversion as an operational satellite aerosol algorithm.

22. Atmospheric Water Vapor

Column water vapor can be estimated from spectral absorption features, but operational retrieval coefficients are sensor dependent. A simplified absorption-ratio concept is:

For production systems, validated product ingestion is recommended where available.

23. Weather Forecast

Numerical Weather Prediction (NWP) models assimilate observations and solve discretized atmospheric dynamics and physics to forecast the future state. A weather forecast tool should therefore ingest model fields rather than infer a 1–7 day forecast from a single satellite scene.

23.1 Core variables

23.2 Wind speed

23.3 Relative humidity concept

23.4 Rain accumulation

24. Weather Nowcasting

Nowcasting targets very short lead times, often minutes to a few hours. Satellite sequences can contribute by estimating cloud motion and growth/decay. This is conceptually different from medium-range NWP forecasting.

24.1 Advection model

24.2 Extrapolation

24.3 Recommended output leads

+30 min
+60 min
+120 min
+180 min

A later AI nowcasting engine could combine multi-temporal satellite imagery with radar and NWP variables when available.

25. Potential Fishing Zone (PFZ)

PFZ is best designed as a habitat/suitability indicator, not as guaranteed fish-catch prediction. Without historical catch/CPUE, species, effort and fleet information, the platform should not claim to predict actual catch.

25.1 Candidate environmental predictors

• SST and SST anomaly;
• chlorophyll-a;
• thermal/ocean fronts;
• upwelling score;
• current speed/direction when available;
• bathymetry;
• species-specific preferred ranges, if defined.

25.2 Suitability functions

A simple triangular or Gaussian preference curve can convert a variable to 0–1 suitability.

25.3 Composite PFZ score

This is a platform suitability model whose weights/preferences must be species/region calibrated. A generic configuration can be provided for exploration, but scientific users should be able to modify the curves and weights.

25.4 Suggested classes

0–25    Low
25–50   Moderate
50–75   High
75–100  Very High

25.5 Fish Catch Prediction is a different future tool

Historical catch / CPUE
+ species
+ fishing effort
+ location/time
+ SST / chlorophyll / currents / fronts
        ↓
Statistical / ML / GeoAI model
        ↓
Expected catch / CPUE prediction

26. Relationships Between Tools

26.1 Important parallel products

27. Recommended Platform Output Contract

27.1 Sidecar metadata

Every raster output should have a sibling statistics/provenance sidecar:

product.tif
product.statistics.json

For forecast products, the sidecar should include:

{
  "model": "GFS / other",
  "model_run": "...",
  "lead_hours": 24,
  "valid_time": "...",
  "variable": "temperature_2m",
  "unit": "K",
  "source_resolution": "...",
  "processing": ["subset", "reproject", "COG"]
}

28. Quality Control & Validation

28.1 Ocean-colour QC

• atmospheric correction must be appropriate for water;
• land, cloud, glint, ice and invalid pixels must be masked;
• algorithm valid range must be enforced or warned;
• reflectance convention and units must match formula requirements;
• sensor wavelength mapping must use physical wavelengths, not generic band numbers.

28.2 Thermal QC

• apply official cloud/quality flags;
• preserve SST quality level;
• check Kelvin/Celsius conversion;
• do not interpolate large cloud gaps silently.

28.3 Current QC

• verify U/V units and grid staggering if applicable;
• define direction convention;
• avoid geostrophic inversion near the equator without suitable treatment;
• record model run and valid time.

28.4 Forecast QC

• never mix forecast cycles without explicit rule;
• distinguish instantaneous vs accumulated precipitation;
• retain pressure level / vertical level;
• record forecast lead time;
• optionally compare with observations to produce bias/RMSE verification.

28.5 PFZ QC

• use species/region-specific preferences if available;
• do not label suitability as guaranteed catch;
• report missing predictors;
• report the score weights and input dates;
• avoid combining satellite fields from incompatible times without warning.

29. Analysis Server Implementation Notes

29.1 Sensor-aware adapter layer

ocean_atmosphere/
├─ adapters/
│  ├─ generic
│  ├─ sentinel3_olci
│  ├─ sentinel3_slstr
│  ├─ modis
│  ├─ weather_model
│  └─ ocean_model
│
├─ ocean_dynamics/
├─ water_properties/
├─ atmosphere/
├─ forecast/
└─ fisheries/

The analysis algorithms should consume normalized variable names such as SST, Rrs_665, u_current, v_current, while adapters map source-specific bands/products to those variables. This avoids creating a separate algorithm for every satellite.

29.2 Do not duplicate generic raster mathematics

NDCI, FAI, MCI, gradients and weighted models can reuse the common raster expression / block-processing engine where possible. The new tool defines the scientific preset, required wavelengths, masks, metadata and output semantics.

29.3 Block processing

29.4 Unit harmonization

Before calculation, the engine should explicitly normalize units:

K ↔ °C
Pa ↔ hPa
m/s
mm or kg m^-2 precipitation
nm ↔ μm wavelengths
mg m^-3 / mg L^-1 only when physically equivalent under the stated density convention

29.5 Provenance example

{
  "analysis": "ocean.upwelling",
  "inputs": {
    "sst": "layer-...",
    "chlorophyll": "layer-...",
    "wind": null
  },
  "methods": {
    "sst_normalization": "local_anomaly",
    "chlorophyll": "log_anomaly",
    "weights": {"sst": 0.55, "chlorophyll": 0.45}
  },
  "warning": "Satellite indicator; no wind/current forcing used"
}

30. Recommended Development Phases

Phase 1 — New features with low risk

Phase 2 — External model integration

• Current Speed & Direction from U/V products;
• Weather Forecast from NWP;
• Ocean Forecast from model/reanalysis;
• AOD and Water Vapor validated product ingestion;
• Weather Nowcasting from time-series imagery.

Phase 3 — Advanced oceanography

• Sea Surface Height / Sea Level Anomaly;
• geostrophic current;
• eddy detection;
• species-specific PFZ;
• Fish Catch / CPUE Prediction using historical fisheries data.

31. Scientific & Technical References

Primary papers and official technical resources relevant to the concepts in this document.

1. European Space Agency (ESA), Sentinel-3 Instruments — OLCI and SLSTR mission/instrument overview. ESA.
2. Copernicus, Sentinel Online — Sentinel-3. Official Sentinel portal.
3. NASA Ocean Biology Processing Group, MODIS Level-2 Ocean Color Format Specifications. NASA Ocean Color.
4. NASA Ocean Color, Ocean Color Chlorophyll Algorithm / OCx Reprocessing Documentation. NASA Ocean Color.
5. Mishra, S. & Mishra, D. R. (2012), Normalized Difference Chlorophyll Index: A novel model for remote estimation of chlorophyll-a concentration in turbid productive waters, Remote Sensing of Environment 117, 394–406. Paper.
6. Hu, C. (2009), A novel ocean color index to detect floating algae in the global oceans, Remote Sensing of Environment. Paper.
7. Gower, J., King, S. & Goncalves, P., Global monitoring of plankton blooms using MERIS MCI. Paper record.
8. Nechad, B., Ruddick, K. G. & Neukermans, G. / related multisensor calibration work, generic multisensor turbidity/TSM retrieval. Remote Sensing of Environment.
9. NOAA National Centers for Environmental Information, Global Forecast System (GFS). NOAA.
10. NOAA NCEI, Numerical Weather Prediction. NOAA.
11. NOAA NCEI, Global Data Assimilation System (GDAS). NOAA.
12. JAXA EORC, Algorithm Theoretical Basis Document — Floating Algae Index. JAXA ATBD.