Evaluation of GFSv16 for Near-Real-Time Data Impact Studies During the Atmospheric River Reconnaissance Program 2022

March 16, 2026

A new paper entitled “Evaluation of GFSv16 for Near-Real-Time Data Impact Studies During the Atmospheric River Reconnaissance Program 2022” was recently published in the Journal of Geophysical Research: Atmospheres (JGR-A). The study was led by Vijay Tallapragada (NOAA/NCEP), with co-authors Xingren Wu (NOAA/NCEP), Minghua Zheng (CW3E), Luca Della Monache (CW3E), F. Martin Ralph (CW3E), Keqin Wu (NOAA/NCEP), Xiujun Wang (Axiom Consultants), M. M. Nageswararao (NOAA/UCAR), Jia Wang (CW3E), Anna Wilson (CW3E), Jason Cordeira (CW3E), Julie Kalansky (CW3E), Daniel Steinhoff (CW3E), and Patrick Mulrooney (CW3E).

The study examined how data collected from dropsondes through the Atmospheric River Reconnaissance (AR Recon) program improved precipitation forecasts associated with ARs in winter 2022. Using the Global Forecast System version 16 (GFSv16), the authors looked at simulations with and without assimilated dropsonde observations to evaluate the impact they had on forecasting geopotential height, horizontal wind and water vapor transport, and precipitation.

Results show that assimilating dropsonde data improved forecasts during moderate to strong AR events, particularly when dropsondes were deployed in the core of the AR. Dropsonde assimilation reduced IVT forecast errors near landfall by approximately 20 kg m^-1 s^-1 and led to substantial improvement in precipitation forecasts along the U.S. West Coast (Figure 1). For high-impact AR events affecting Washington and Oregon, precipitation forecast errors were reduced by between 20-40%, with improvements persisting through 48-hour forecasts (Figure 2).

Forecast improvements varied across cases, reflecting differences in AR intensity, structure, propagation direction, and interactions with complex coastal terrain. Overall, the findings demonstrate the value of targeted AR Recon dropsonde observations for improving forecasts of landfalling ARs and associated precipitation, and they support continued investment in advanced data assimilation systems and strategic observing approaches to reduce forecast uncertainty for high-impact West Coast events. This work supports the goals set out in CW3E’s 2025-2029 Strategic Plan by improving operational forecasts of AR and extreme precipitation using novel observations.

Figure 1. Precipitation forecast improvements. The percentage differences in precipitation between the Ctrl and Deny runs for (a, d) equitable threat score (ETS), (b, e) Spearman’s rank correlation coefficient (SCC), and (c, f) root mean square error (RMSE) over the PNW (upper panel) and WEST (lower panel) at forecast hours 24–120 over the period from 11 January to 25 March 2022. The difference is Ctrl–Deny for ETS and SCC, but Den–Ctrl for RMSE. Thus, red and blue indicate improvement and degradation from dropsondes, respectively. Values with frames are those passing the significance test, using the 90% confidence interval. The X-axis denotes thresholds for 24-hr accumulated precipitation. Figure 5 from Tallapragada et al. 2026.

Figure 2. Precipitation (mm d⁻¹) from (a, g) Stage-IV, (b, h) Ctrl, and (c, i) Deny runs on March 1 (top panel) and 2 (the third panel), 2022, and errors in (d, j) Ctrl and (f, l) Deny runs, and (f, l) the differences in absolute error (DAE) between the Ctrl and Deny at forecast hour 24 (the second panel) and 48 (bottom panel). The green contours in the second and bottom panels indicate regions with precipitation exceeding 50 mm d⁻¹ in Ctrl, Deny, and Stage-IV. The Ctrl and Deny runs were initialized from 0000 UTC on 28 February 2022 (IOP15), with dropsondes excluded in the Deny run. A negative DAE indicates an improvement due to the use of dropsonde data. Figure 10 from Tallapragada et al. 2026.

Citation:

Tallapragada, V., Wu, X., Zheng, M., Delle Monache, L., Ralph, F. M., Wu, K., Wang, X., Nageswararao, M. M., Wang, J., Wilson, A. M. Cordeira, J., Kalansky, J., Steinhoff, D., & Mulrooney, P. (2026). Evaluation of GFSv16 for Near-Real-Time Data Impact Studies During the Atmospheric River Reconnaissance Program 2022. Journal of Geophysical Research: Atmospheres, 131(4), e2025JD044818. https://doi.org/10.1029/2025JD044818