In midlatitude regions, year-round coastal water level forecast systems serve the public with reliable and consistent forecasts ranging from high-tide floods to extreme events. The Stevens Flood Advisory System (SFAS) is a probabilistic coastal water level forecast system forced by a superensemble of global atmospheric forecast products (Canadian, United States, and European atmospheric forecast systems, termed subensembles). We assessed 2 years of forecasts for the Mid-Atlantic/New York Bight region to demonstrate the advantages and challenges of these systems and determine how they can be improved. The accuracy of the SFAS ensemble-based central forecast and spread is compared to predictions using each subensemble and the National Oceanic and Atmospheric Administration's (NOAA) Probabilistic Extratropical Storm Surge (P-ETSS) forecasts. Averaged across 14 stations, the current SFAS central forecast achieves a 1%-20% lower root-mean-square error (RMSE) on semidiurnal water level peaks than its subensembles; P-ETSS has nearly 20%-50% higher RMSE than the current SFAS central forecast. Assessing the probabilistic forecast spread, the coverage of observation by forecast area of uncertainty (COU) of the SFAS 5th-95th percentile is close to the 90% optimal value, better than the spreads of P-ETSS and the subensembles. Results show that model simulations and ensemble forecasts created with the highest resolution forcing result in the lowest error and bias. Analysis of the wind forcings for significant storm surge events reveals both underestimation and a counterclockwise wind direction bias over the inner continental shelf, both signatures of "coastal blurring" of land-atmosphere boundary layer winds in the model forcing. This leads to underestimation of along-coast winds and Ekman transport-driven storm surge, important for midlatitude regions.
With the intensification of global warming, the Sichuan-Chongqing region of China has experienced a growing incidence of heat waves and associated disasters. However, due to complex topographic conditions, model-based early warning capabilities remain inadequate, particularly on the sub-seasonal to seasonal (S2S) scale. In this study, we analyze the prediction skills of five S2S models (CMA, ECCC, ECMWF, NCEP and UKMO). Additionally, focusing on models that demonstrate superior skills in forecasting summer daily maximum air temperature (Tmax), we develop a set of bias correction approaches and conduct a comprehensive evaluation on them. Results indicate that the CMA performs best overall, particularly in predicting heat events, while the ECMWF shows better capability to capture the temporal variability of daily Tmax among the five models. Forecast errors of both models exhibit notable sensitivity to terrain features and are correlated with El Niño-Southern Oscillation (ENSO) variations. The quantile mapping bias correction method that integrates complex topography, ENSO phases, and model forecast attributes reliably improves the forecast skill of both CMA and ECMWF in the SCR, whether in hindcast or independent validation. It can help ECMWF to successfully capture heat events that were widely missed in the original forecasts. Furthermore, it delivers effective improvement in forecast accuracy of the two models for extreme heat waves in August 2006 and 2022.
Abstract The extrascientific value of a weather forecast depends on how it is used in decisions. To explore a wide variety of possible use cases, we propose a notional formula for aggregate economic value based on the canonical cost:loss model. Using this framework, we identify characteristics of high-value forecasts and use cases and show how the distribution of these characteristics may be shaped by end users’ metadecisions—earlier decisions that affect use-case parameters that, in turn, determine forecasts’ economic value. Because aggregate economic value depends on which use cases are most prevalent, we analyze how common verification metrics implicitly weight different forecasts and use cases. This analysis can help forecasting organizations identify high-value use cases and end users and inform the development and selection of verification metrics for communication with end users and in forecast model and product development. Significance Statement The fundamental purpose of weather forecasts is to inform end users so they can better adapt to weather. The value of weather forecasts depends on both forecast accuracy and how well forecasts align with end users’ use cases. This paper examines how end-user decisions, including longer-term decisions made before receipt of a specific forecast, affect forecast value and identifies characteristics of use cases that are likely to be most prevalent. In addition, this paper shows how common verification metrics implicitly weight different forecast use cases and analyzes how this aligns with their contribution to aggregate economic value. Such metrics are gaining even more importance due to their role in developing data-driven weather models. These findings suggest the types of use cases the weather community should target for outreach and criteria for developing and selecting verification metrics.
Abstract Estimating tropical cyclone (TC) intensity from satellite observations remains challenging, and microwave (MW) imagery has been shown to improve skill relative to infrared-only approaches. This study introduces D-MINT183v2, an update to the D-MINT183 convolutional neural network that estimates current TC intensity from 183-GHz MW imagery. Unlike the original D-MINT183, which was restricted to the inner 50% of cross-track swaths to avoid limb effects, D-MINT183v2 is trained using limb-corrected full-swath MW imagery and includes scan angle as an additional predictor. A baseline model is developed for cross-track sounders (AMSU-B/MHS/ATMS) and then adapted to SmallSat sensors (TROPICS and Tomorrow.io TMS) using transfer learning. Across sensors, D-MINT183v2 maintains or improves accuracy relative to the original D-MINT183 within the inner swath and provides consistent skill across the swath, thereby substantially increasing the number of usable MW-based intensity estimates. Relative to the IR-only D-PRINT baseline, D-MINT183v2 produces more skillful intensity estimates at MW overpass times. SHAP-based attribution is used to quantify the roles of viewing geometry and channel-dependent brightness-temperature information in the retrieved intensities.
Abstract Satellite spectral bands have been used to identify patterns associated with thunderstorms such as strong updrafts, deep vertical development, cloud-top cooling, and cloud particle phase. Because of this, satellite data are incorporated into forecasting applications for identification of initiating thunderstorms including the Day Cloud Phase Distinction (DCPD) and Nighttime Microphysics (NtMicro) false color red-green-blue (RGB) composite imagery and machine learning models for short-term forecasting (nowcasting) of thunderstorms and their hazards. However, initiating lightning-producing and non-lightning-producing convective storms can appear similar in satellite imagery, making it difficult to determine which developing storms will become mature thunderstorms. As a consequence, nowcasting models relying solely on satellite imagery, such as LightningCast and ThunderCast, can be imprecise when used to nowcast at local scales. To further verify this and examine the conditions under which non-lightning-producing convective storms appear similar to developing thunderstorms in satellite imagery, four case studies are presented with LightningCast’s and ThunderCast’s predictions. Each case study contained at least one non-lightning-producing convective storm with RGB signatures consistent with cloud-top glaciation. Overall, the case studies demonstrate non-lightning-producing convective storms, appearing similar to developing thunderstorms in satellite imagery, are not limited by latitude and can form in an assortment of meteorological environments including near a sea breeze, in high humidity, and in an environment with key meteorological ingredients for severe weather. While geostationary meteorological satellites provide critical qualitative and quantitative insights for convective weather analysis and nowcasting, these results illustrate their limitations when distinguishing locally between non-lightning-producing and lightning-producing storms.
