All-sky assimilation impacts of the Tomorrow.io Microwave Sounder constellation on global weather forecasts

This is a preprint of a manuscript submitted for peer-review to the Quarterly Journal of the Royal Meteorological Society on 2025-07-24.

Abstract

The Tomorrow.io Microwave Sounder (TMS) builds on a heritage of passive microwave sounders (PMWSs) that have well-established value in weather forecasts. The TMS constellation of cubesats uses a combination of sun-synchronous and inclined orbits to fill gaps in current public PMWS data. This work establishes a data assimilation methodology for TMS, and demonstrates its real global weather forecast impacts with Observing System Experiments (OSEs). The OSE control uses techniques similar to the National Oceanic and Atmospheric Administration (NOAA), and additionally assimilates TMS-S02 and TMS-S04.

We introduce a full assimilation methodology for TMS encompassing all-sky error modeling and quality control, and novel variational bias correction (VarBC) predictors.  The all-sky error model is based on techniques in earlier literature, and a new mismatched cloud impact quadrant filter enables more effective assimilation of surface-sensitive TMS channels.  TMSs in inclined orbits are exposed to a transient thermal environment that motivates VarBC predictors responsive to unique night-day cycle of each satellite.  The new VarBC approach mitigates transient instrument biases that otherwise degrade upper-level forecasted temperature verification.

Assimilating two TMS instruments in January 2025 achieves 6-hour forecast accuracy improvements similar to two Advanced Technology Microwave Sounder (ATMS) instruments for water vapor, and 50\% of two ATMSs for tropospheric temperature.  Statistically significant forecast accuracy improvements from TMS persist for water vapor up to three days, and for temperature and winds up to two days. TMS provides complementary information content to public PMWSs, and fills in gaps as designed.
 
We use software tools from the Joint Effort for Data assimilation Integration (JEDI) and the NOAA atmosphere-only Unified Forecast System (UFS). Our TMS assimilation configurations are available for general application by the broad and growing JEDI user-base, and are ready for the next phase in the research to operations pipeline at weather forecast centers.

Data assets

Two additional datasets are provided by The Tomorrow Companies Inc.  They contains L1-TBR V01-00 Tomorrow.io Microwave Sounder (TMS) observations, formatted for consumption in Joint Effort for Data assimilation Integration (JEDI) applications using the Interface for Observation Data Access (IODA) version 3 format (e.g., HofX, Variational). Observations are supplied from the TMS-S02 and TMS-S04 instruments during January 2025. Each file contains data from a one hour window. The TMS file names correspond to the central date of a 6-h DA window, with a +/- 3 h bound on each side.  So "TMS02-merged-2025-01-25-06-subwindow001.nc" corresponds to raw granules that start on or after 2025-01-25T03Z and before 2025-01-25T04Z.  It might include some data after 025-01-25T04Z.

This is an addendum to the preprint (in review) publication.  This limited subset of an otherwise proprietary dataset provides all of the TMS data and metadata needed to replicate the experiments described therein.  As a commercial mission, full access to the TMS datasets is otherwise restricted.  Samples of the TMS level 1B-TBR products in their as-distributed format are available at https://www.tomorrow.io/satellite-data/data-catalog/l1b-tbr/.  At the time of submission, additional data from TMS are under evaluation by NOAA's CWDP, and are available to academic and international partner institutions.