3 Data delivery for Climate TRACE and OceanMind
3.1 AIS-based emissions model estimates
3.1.1 Monthly asset-level data
Using the Climate TRACE and OceanMind asset-level (i.e., vessel level) data schema, we currently produce monthly datasets of emissions for both domestic and international activity including voyages and port visits. Each dataset contains rows for voyage-level emissions and port-stay emissions for voyages and port-stays that ended during that month. We currently produce monthly datasets from ‘2015-01-01’ through ‘2026-06-30’.
We save monthly CSVs of domestic activity in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/domestic-shipping-ship/YYYYMM/GFW/asset-climate-trace_domestic-shipping-ship_GFW_MMDDYY_*.csv, where the folder name YYYYMM represents the delivery month of the data, and the MMDDYY in the file name represents the month of the data included in each CSV. Each month of data is split into multiple files that each have a corresponding number which replaces the * wildcard operator. This approach is used since there are hard limits that restrict how big files can be when they are exported from Google Big Query to Google Cloud storage; large files must be split into smaller files. From Google’s website, “The wildcard operator is replaced with a number (starting at 0), left-padded to 12 digits. For example, a URI with a wildcard at the end of the filename would create files with000000000000appended to the first file, 000000000001 appended to the second file, and so on.”
We save monthly CSVs of international activity in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/international-shipping-ship/YYYYMM/GFW/asset-climate-trace_international-shipping-ship_GFW_MMDDYY_*.csv, where the folder name YYYYMM represents the delivery month of the data, and the MMDDYY in the file name represents the month of the data included in each CSV. Each month of data is split into multiple files that each have a corresponding number which replaces the * wildcard operator.
3.1.2 Asset-level dataset schema
Each monthly asset-level CSV (domestic and international alike) follows the Climate TRACE / OceanMind asset-level schema. There is one row per vessel activity — a single voyage or a single port stay, distinguished by the other8 column. Emissions are in metric tonnes, and each emissions factor is expressed per unit of activity (emissions ÷ activity). This schema is produced from the AIS model result tables described in Section 1.4.
| Column | Description |
|---|---|
ssvid |
Vessel identifier (MMSI). |
date_formatted |
Date of the activity’s end (voyage arrival / port-visit end). |
start_date / end_date |
Activity start and end timestamps. |
asset_identifier |
gfw-imo-<imo> if an IMO is known, otherwise gfw-mmsi-<ssvid>. |
model_number |
Model version (e.g. _v20260714). |
asset_name |
Lower-cased ship name. |
iso3_country |
Vessel flag state (ISO3). |
location |
WKT LINESTRING for the voyage / port-visit route. |
type |
Vessel class. |
capacity |
Gross tonnage (GT). |
capacity_factor |
Activity ÷ capacity. |
activity |
Distance travelled (nautical miles); for port stays, floored at 1 nm. |
{CO2,CH4,N2O,SOX,NOX,NMVOC,PM2_5,PM10,CO}_emissions_factor |
Per-unit-activity emissions factor for each pollutant (emissions ÷ activity). |
{CO2,CH4,N2O,SOX,NOX,NMVOC,PM2_5,PM10,CO}_emissions |
Total emissions of each pollutant for the activity (mt). |
total_CO2e_100yrGWP / total_CO2e_20yrGWP |
CO₂-equivalent totals using 100-yr / 20-yr GWPs (IPCC AR6). |
other1 |
Activity duration, formatted d.hh:mm:ss. |
domestic_international |
domestic or international (from comparing origin/destination country). |
other8 |
Trip indicator: TRUE = voyage, FALSE = port stay. |
other2–other7, other9–other12 |
Reserved/placeholder columns in the Climate TRACE schema (currently NULL). |
3.1.3 Confidence and uncertainty
For all data entering into Climate TRACE platform, it is important to document both the confidence and uncertainty with each individual data attribute. According to the Climate TRACE definitions, confidence is a qualitative measure of how well the data are understood and how trustworthy they are (i.e., very high, high, medium, low, very low). To assign these qualitative categories, we follow the general recommendations proposed by Climate TRACE, which take into account multiple considerations including whether the data are directly observed or generated by a model, whether or not the data are self-reported, whether self-reported data are incentive compatible (i.e., the asset-owner has little reason to provide false information), whether or not the data can be validated, whether or not the data can be corroborated across multiple sources, etc. Confidence is assigned differently for two sub-groups of vessels: 1) those vessels that have known vessel characteristics from official vessel registries (which must include all of known vessel type, main engine power, gross tonnage, and length); and 2) those vessels that do not have all of these known vessel characteristics from official vessels registries, and for which we use the GFW vessel characteristics machine learning algorithm to generate any missing characteristics. We will call these sub-groups “high-information vessels” and “low-information vessels.” For “high-information vessels” where vessel characteristics are obtained from a registry, we assign a confidence value of “very high” to vessel characteristics (vessel type, capacity, and capacity factor). For “low-information” vessels where vessel characteristics are obtained through an algorithm, we assign a value of “low” to vessel characteristic attributes (vessel type, capacity, and capacity factor). Since emissions estimates depend on certain vessel characteristic metadata, vessels with higher confidence vessel characteristics will naturally have higher confidence emissions estimates.
Generally speaking, we have very high confidence in the AIS spatiotemporal data itself, since it is in essence directly observed, so we assign a value of “very high” to the activity attribute. While the bottom-up engineering emissions estimation methodology uses these directly observed AIS data, and while the approach is based on the published ICCT and IMO methodology, it is still a model and does not represent directly observed emissions data. We therefore generally classify emissions estimates from high-information vessels as “medium”, and emissions estimates from low-information vessels as “very low”. We provide confidence estimates for each vessel by month.
For the uncertainty column, following the recommendations by CT and to be consistent with the OM methodology, for now we assign “Standard deviation” to all numeric attributes. Standard deviation for each of these attributes is calculated by vessel, across both port visits and voyages, for port visits and voyages that end in each year of the dataset. In cases where there was only one observation, the standard deviation was set to 0. However, we believe an interesting and important area of future work could be to think more about what the most appropriate uncertainty measures are for each attribute. While standard deviation provides information on the overall distribution of any given data attribute across the entire dataset, it does not provide information on the uncertainty of that attribute for any given vessel, voyage, or port-visit. Ultimately, the most appropriate measure of uncertainty, and the scale at which to measure it, will depend on how the emissions estimates will be used by various actors and for different interventions.
We save the domestic confidence dataset in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/domestic-shipping-ship/YYYYMM/confidence-climate-trace_international-shipping-ship_mmyy_GFW_MMDDYY.csv, where the folder YYYYMM represents the delivery date of the data, mmyy represents the time period corresponding to the data (each confidence CSV represents a single month of activities that ended in that month, so these numbers will the month and year of the data), and the suffix on the file name _MMDDYY.csv represents the delivery date of each CSV to GCS. We follow the Climate TRACE and OceanMind confidence data schema.
We save the international confidence dataset in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/international-shipping-ship/YYYYMM/confidence-climate-trace_international-shipping-ship_mmyy_GFW_MMDDYY.csv, where the folder YYYYMM represents the delivery date of the data, mmyy represents the time period corresponding to the data (each confidence CSV represents a single month of activities that ended in that month, so these numbers will the month and year of the data), and the suffix on the file name _MMDDYY.csv represents the delivery date of each CSV to GCS. We follow the Climate TRACE and OceanMind confidence data schema.
We save the domestic uncertainty dataset in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/domestic-shipping-ship/YYYYMM/uncertainty-climate-trace_international-shipping-ship_mmyy_GFW_MMDDYY.csv, where the folder YYYYMM represents the delivery date of the data, mmyy represents the time period corresponding to the data (each uncertainty CSV represents all activities that ended across an entire year, so these numbers will show 01/January as the month along with the appropriate year), and the suffix on the file name _MMDDYY.csv represents the delivery date of each CSV to GCS. We follow the Climate TRACE and OceanMind uncertainty data schema.
We save the international uncertainty dataset in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/international-shipping-ship/YYYYMM/uncertainty-climate-trace_international-shipping-ship_mmyy_GFW_MMDDYY.csv, where the folder YYYYMM represents the delivery date of the data, mmyy represents the time period corresponding to the data (each uncertainty CSV represents all activities that ended across an entire year, so these numbers will show 01/January as the month along with the appropriate year), and the suffix on the file name _MMDDYY.csv represents the delivery date of each CSV to GCS. We follow the Climate TRACE and OceanMind uncertainty data schema.
| Data attribute | Confidence | Uncertainty |
|---|---|---|
| type | Very high (high-information vessels) Low (low-information vessels) |
NULL |
| capacity | Very high (high-information vessels) Low (low-information vessels) |
Standard deviation |
| capacity_factor | Very high (high-information vessels) Low (low-information vessels) |
Standard deviation |
| activity | Very high | Standard deviation |
| CO2_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| CH4_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| N2O_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| SOX_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| NOX_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| VOCS_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| PM 2_5_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| PM10_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| CO_emissions_factor | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| CO2_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| CH4_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| N2O_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| SOX_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| NOX_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| VOCS_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| PM2_5_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| PM10_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| CO_emissions | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| total_CO2e_100yrGWP | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
| total_CO2e_20yrGWP | Medium (high-information vessels) Very low (low-information vessels) |
Standard deviation |
3.2 Dark fleet emissions model estimates
Using the Climate TRACE country-level data schema, we currently produce a single dark fleet dataset where each row is a month, and there are columns for emissions from the dark fleet for each of the pollutants. All units in this table are in metric tonnes. We currently produce data for 2017 through 2026. The dataset currently uses Sentinel-1 satellite data, as described in Section 2.1. These data provide insight into previously undisclosed and unmeasured emissions, and can therefore directly inform Climate Trace’s known gaps estimates for the shipping sector. Note that since these estimates are for the entire globe, and not differentiated by country, we put UNK in the iso3_country column — Climate TRACE’s convention for global-scale emissions that feed into a country-level data type.
We save this CSV in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/international-shipping-ship/YYYMM/GFW/global-climate-trace_dark-fleet-shipping-ship_GFW_MMDDYY.csv, where the folder name YYYYMM represents the delivery date of the data, and the suffix on the file name _MMDDYY.csv also represents the delivery date of the data.
3.2.1 Dark fleet dataset schema
The delivered dark fleet dataset sums the gridded model output (the s1_time_gridded_dark_fleet_model table, Section 2.3) over all cells into a single global emissions time series, formatted to the Climate TRACE schema. There is one row per month.
| Column | Description |
|---|---|
start_date / end_date |
Start and end of the month. |
iso3_country |
UNK (global-scale estimate; see above). |
{CO2,CH4,N2O,CO,NOX,SOX}_emissions_tonnes |
Total dark-fleet emissions of each pollutant (mt). |
total_CO2e_100yrGWP / total_CO2e_20yrGWP |
Dark-fleet CO₂-equivalent totals using 100-yr / 20-yr GWPs (IPCC AR6). |
3.2.2 Confidence and uncertainty
For the dark fleet emissions confidence estimates, we again follow the general recommendations proposed by Climate TRACE for assigning qualitative values for each numeric value (i.e., very high, high, medium, low, very low). We assign a value of very low to all numeric values given the technical challenges and difficulty of determining emissions for vessels that do not broadcast AIS. We save this CSV in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/international-shipping-ship/YYYMM/GFW/confidence-global-climate-trace_dark-fleet-shipping-ship_GFW_MMDDYY.csv, where the folder name YYYYMM represents the delivery date of the data, and the suffix on the file name _MMDDYY.csv also represents the delivery date of the data.
For the dark fleet emissions uncertainty estimates, there is not currently a standard Climate TRACE methodology for assigning these values since this is a brand new data format. Therefore, we adopt an analogous approach to what is currently done for the vessel-level emissions data. For each year and each pollutant, we calculate the standard deviation of emissions from across the twelve monthly estimates. This gives us an uncertainty estimate for each year of data and each pollutant. We save this CSV in Climate TRACE’s Google Cloud Storage bucket as climate_trace_internal/transportation/international-shipping-ship/YYYMM/GFW/uncertainty-global-climate-trace_dark-fleet-shipping-ship_GFW_MMDDYY.csv, where the folder name YYYYMM represents the delivery date of the data, and the suffix on the file name _MMDDYY.csv also represents the delivery date of the data.