This website uses cookies. By continuing to use this website you are agreeing to our use of cookies. 



ACRUISE: deep-learning inferred shiptrack clouds from AQUA MODIS daylight satellite data for 2002-2021

Update Frequency: Not Planned
Latest Data Update: 2022-09-16
Status: Completed
Online Status: ONLINE
Publication State: Citable
Publication Date: 2022-09-09
DOI Publication Date: 2022-09-16
Download Stats: last 12 months
Dataset Size: 1.13M Files | 9TB


Large dataset of emission induced "shiptrack" clouds, detected using deep-learning, from satellite based remote sensing data with global coverage, from 2002 to 2021 for the Atmospheric Composition and Radiative forcing changes due to UN International Ship Emissions regulations (ACRUISE) project. Shiptracks were inferred from every daylight granule captured by the MODerate Imaging Spectroradiometer (MODIS) instrument, onboard the NOAA-AQUA satellite from 2002-2021 inclusive and stored in a compressed netcdf file. In addition, polygons corresponding to contours of level 0.5 and 0.8 from the inference images are provided as a light-weight alternative. These are stored in annual geopackages in the geographic projection.

The model is a standard neural-network based segmentation model with a UNet architecture, a resnet-152 backbone and sigmoid activation on the final layer that was pre-trained on the 2012 ImageNet Large Scale Visual Recognition Challenge 2012 (ILSVRC2012) ImageNet dataset. This model was trained to segment clouds formed by ship exhausts, known as shiptracks, from MODIS level 1b, day microphysics composite granules enhanced through histogram stretching.

The purpose of these data is to measure the effect that shipping fuel regulation has on climate change and to reduce the uncertainty in the relationship between aerosols and cloud formation and properties. This allows the determination of where tracks are more likely to form and the sensitivity of clouds to such perturbations.The data indicate a sharp reduction in tracks due to the more stringent ship emission regulations since 2020.

A small minority of granules (<0.5%) are missing due to a combination of missing or corrupt files and/or unexpected computational processing failures. These remained unresolved as they were judged insignificant compared to model uncertainties and and of negligible additional benefit to warrant the overheads to resolve each missing granule.

Citable as:  Watson-Parris, D.; Christensen, M.; Laurenson, A.; Clewley, D.; Gryspeerdt, E.; Stier, P. (2022): ACRUISE: deep-learning inferred shiptrack clouds from AQUA MODIS daylight satellite data for 2002-2021. NERC EDS Centre for Environmental Data Analysis, 16 September 2022. doi:10.5285/0d88dc06fd514e8199cdd653f00a7be0.
Abbreviation: Not defined
Keywords: Ship tracks, MODIS


Previous Info:
No news update for this record
Previously used record identifiers:
No related previous identifiers.
Access rules:
Public data: access to these data is available to both registered and non-registered users.
Use of these data is covered by the following licence: When using these data you must cite them correctly using the citation given on the CEDA Data Catalogue record.
Data lineage:

NASA supplied AQUA MODIS level 1a data, archived at Plymouth Marine Laboratory, was processed to level 1b using NASA SeaDAS script. Satpy was used to process level 1b data into ‘day microphysics’ composites from channels 1, 20 and 32 (corresponding to wavelengths of 645nm, 3.75µm and 12.5µm respectively). Histogram equalisation was then used to enhance the image prior to training and inference by the model.

Model output was written to a variable, "shiptracks", within a netcdf file that inherits the coordinates and metadata from the original day micro-physics granule. Contours at 0.5 and 0.8 were computed and stored in annual .gpkg files in the geographic projection.

For archiving, lossy compression was applied to inference granules: Variable "shiptracks" from float32 to unint16, with appropriate scaling factor. Latitude and Longitude coordinates from float64 to float32. Contours files were not compressed.

Duncan Watson-Paris (DWP), Matthew Christensen (MC) and Philip Stier (PS) designed the research, DWP carried it out. Angus Laurenson and Daniel Clewley supported analysis. MC and Ed Gryspeerdt provided label data. Data were prepared by the project team and uploaded to CEDA for archival

Data Quality:
No quality checks have been made by CEDA
File Format:
Data are NetCDF (.nc) formatted with additional geopackages files (.gpkg)

Related Documents

 ImageNet Website

Citations: 2

The following citations have been automatically harvested from external sources associated with this resource where DOI tracking is possible. As such some citations may be missing from this list whilst others may not be accurate. Please contact the helpdesk to raise any issues to help refine these citation trackings.

Watson-Parris, D., Laurenson, A., Clewley, D. (2022). Ship tracks detected using machine learning algorithm (Version 1.0) [Data set]. Zenodo.
Watson-Parris, D., Laurenson, A., Clewley, D. (2022). Ship tracks detected using machine learning algorithm (Version 1.0) [Data set]. Zenodo.

Process overview

This dataset was generated by a combination of instruments deployed on platforms and computations as detailed below.

Mobile platform operations

Mobile Platform Operation 1 Aqua Satellite orbit details

Computation Element: 1

Title shiptrack_semantic_segmentation_v1
Abstract A convolutional neural network with a Unet architecture, with a RESNET-152 backbone, trained to segment shiptrack clouds from enhanced day_microphysics imagery from AQUA MODIS Input Description AQUA MODIS level 1B day microphysics composite granules, enhanced with histogram stretch. Output Description Netcdf files with a single variable 'shiptracks' that contains shiptrack inference values and shares the coordinates of the original AQUA MODIS granule from which they are derived. Post-processing is required to extract contours and filter them by brightness temperature to obtain final results used in publication. Software Reference
Input Description None
Output Description None
Software Reference None
Output Description


  • long_name: inferred shiptrack clouds
  • var_id: shiptracks
  • names: inferred shiptrack clouds

Co-ordinate Variables

  • units: degrees_north
  • standard_name: latitude
  • var_id: latitude
  • names: latitude
  • units: degrees_east
  • standard_name: longitude
  • var_id: longitude
  • names: longitude
Temporal Range
Start time:
End time:
Geographic Extent