Impacts of Dropsonde Observations on Forecasts of Atmospheric Rivers and Associated Precipitation in the NCEP GFS and ECMWF IFS models

October 31, 2023

A paper titled “Impacts of Dropsonde Observations on Forecasts of Atmospheric Rivers and Associated Precipitation in the NCEP GFS and ECMWF IFS models” by Laurel DeHaan (CW3E), Anna Wilson (CW3E), Brian Kawzenuk (CW3E), Minghua Zheng (CW3E), Luca Delle Monache (CW3E), Xingren Wu (NOAA/NCEP and Axiom Consultants), David A. Lavers (ECMWF), Bruce Ingleby (ECMWF), Vijay Tallapragada (NOAA/NCEP, AR Recon Co-PI), Florian Pappenberger (ECMWF), and F. Martin Ralph (CW3E Director and AR Recon PI) was recently accepted in Weather and Forecasting. This paper investigates the differences in skill between forecasts that assimilated dropsonde data from Atmospheric River Reconnaissance (AR Recon) and forecasts that did not assimilate the dropsonde data, and illustrates improvements in the forecasts that do include the additional dropsonde data. This work supports the Atmospheric Rivers (AR) Research and Applications Priority Areas in CW3E’s 2019-2024 Strategic Plan, and represents an important international, interagency collaboration, in the framework of a Research And Operations Partnership, to diagnose the impact of AR Recon data. Forecast comparison is made in terms of both precipitation and integrated vapor transport (IVT) at multiple thresholds (13, 25, and 50 mm for precipitation, and 250 and 500 kg m-1 s-1 for IVT) for two global numerical prediction models: the Integrated Forecast System (IFS) from the European Centre for Medium-Range Weather Forecasts (ECMWF) and the Global Forecast System (GFS) from the National Centers for Environmental Prediction (NCEP). The comparison was made for 22 different Intensive Observation Periods (IOPs) in 2019 and 2020 at lead times from one to five days.

Differences between the control and denial forecasts were measured in terms of mean absolute error (MAE) and spatial correlation for both IVT and precipitation skill. In addition, the difference in precipitation was also measured using Fractions Skill Score (FSS) and a watershed intensity metric. Figure 1 shows an example of the comparison between the control and denial forecasts in terms of MAE and spatial correlation averaged over multiple thresholds. In this example, both models show generally modest improvement with the inclusion of dropsondes in IVT MAE and show larger improvements in IVT using the correlation metric. The NCEP model shows significant improvement at all three lead times for IVT correlation (Fig. 1f). For precipitation, the ECMWF model has significant improvements in MAE with the control forecasts at three lead times (Fig. 1c) and the NCEP model has significant improvements at two lead times (Fig. 1g). The correlation of precipitation has mixed results, with the only significant improvements with the control forecasts occurring at 48 and 72-hour lead times for the NCEP model (Figs. 1 d, h).

Combining this example with the other comparisons in the publication, this work illustrates that, more often than not, forecasts were improved when dropsonde data were assimilated. Both the ECMWF IFS and the NCEP GFS models show many improvements in forecast skill with the added information from the dropsondes. In particular, significant improvements in the control forecast IVT generally occur in both models, especially at a higher threshold. Significant improvements in the control forecast precipitation also generally occur in both models, but the two models are not consistent in the lead times and metrics that demonstrate the improvements.

Figure 1: (Figure 6 from DeHaan et al 2023): Averages of differences (control – denial) in error or correlation across all thresholds. Boxes are the interquartile range; the middle line is the median and the asterisk shows the mean. Blue colors indicate the control has less MAE or higher correlation in the mean; red colors indicate the denial has less MAE or higher correlation in the mean. Darker shades indicate significant differences in the mean based on a 90% confidence interval computed with bootstrapping.

DeHaan, L. L., and Coauthors, 2023: Impacts of Dropsonde Observations on Forecasts of Atmospheric Rivers and Associated Precipitation in the NCEP GFS and ECMWF IFS models. Wea. Forecasting, https://doi.org/10.1175/WAF-D-23-0025.1, in press.

Corresponding author: Laurel Dehaan