Weather Tech Services LLC

Case 1 Model performance – self-learning function

The testing domain is located in California at USGS station 11447360, a residential area where soil infiltration and surface discharge differ significantly from rural areas. The regular WRF-Hydro model failed to accurately represent this case. However, our model, equipped with a self-learning function, adapts to land type changes. We tuned the parameters based on the first precipitation event and applied the same parameters for the following three months. The model results closely matched the observations over the three-month simulation. This self-learning function can be integrated into the daily operation of the real-time system to maintain optimal performance.

Model simulation from hydro-extension with horizontal grid resolution 100 meters.

Case 2 Reservoir /water management function

The model simulates reservoir discharge to align closely with observed data, measured in cubic meters per second (m³/s). This capability enables accurate simulation of downstream impacts, represented by downstream reservoir inflows and observations. In the results, simulated reservoir inflows are shown in light green, while observations are in blue.

This simulation illustrates reservoir input and discharge. The black line represents hourly reservoir discharge, the blue line shows hourly input from all rivers, and the red line depicts the total input (direct precipitation to the reservoir plus river input). This data can be utilized to manage reservoir discharge during flooding events.

Case 3 City flooding

The model includes parameters that capture urbanization features, making it suitable for forecasting city floods, estimating flooding risks, and planning urban construction. The figure shows a simulation of city flooding during the heavy precipitation event on July 20th in Zhengzhou. The filled color indicates the maximum surface water level during the simulation. Arrows mark the locations of tunnel entrances and exits, where significant water accumulation occurred due to large-scale hydrodynamic processes. This highlights the cause of tunnel flooding, which tragically resulted in substantial loss of life.

Case 4 Mountain flash flood warning

The advancements in flash flood forecasting are instigating a pivotal shift in how decision-makers perceive flood situations. Picture this transformation as a single comprehensive figure that encapsulates the entire flooding scenario. In this enlarged image, high-resolution models operate with the precision of street-level decision-making, pinpointing flood locations with remarkable accuracy.

In real-time applications, consider a vast 10,000 square kilometer domain where a 6-hour flood forecast can be generated in less than 5 minutes. This is not merely a glimpse of what’s achievable; it represents a significant stride towards highly accurate flash flood predictions. This progress is particularly pronounced when these models are enhanced with AI-driven precipitation data obtained from state-of-the-art satellite or radar-based nowcasting.

Presented here is the high-resolution hydrology model simulation depicting the flooding in Waverly (TN) on August 21, 2021. The initial visual showcases an animated representation of streamflow within the model’s geographical scope, while the subsequent illustration highlights the peak streamflow that appeared in Waverly. Our forecasting system is capable of real-time deployment in any location. In the real-time application, this information is integrated with radar-derived precipitation and forecasted products from the NOAA HRRR weather model. Additionally, the system automatically assimilates snow data when available.

Case 5 Atmospheric large scale circulation forecast

70 days forecast from modified WRF - 2 degree resolution. Left is the GFS analysis data, right is 70 days model forecast validated at 20160310.