AI Technology

AI Local Rainfall Nowcasting Using Weather Radar Maps

To provide accurate precipitation nowcasting, Beijing startup ColorfulClouds Tech is applying machine learning (ML) techniques using observed radar echo maps to generate high resolution, minute-by-minute rainfall forecasts on its ColorfulClouds Weather mobile app.

Whether it’s walking the dog or the morning commute — an up-to-date and accurate weather forecast can mean the difference between a pleasant stroll under an umbrella or getting soaked through to the bone.

Current weather forecasts are based on traditional weather prediction (NWP) models which use complicated mathematical algorithms representing physical atmospheric principles to predict how weather changes over time. A variety of real-time meteorological observations are generated from ground sensors, weather radars and satellites monitoring land masses, oceans and the upper atmosphere. Variables such as humidity, temperature, wind direction and speed, etc. are fed into the models to produce daily and long range forecasts for as far as two weeks ahead on regional, national and global scales.

Such traditional methods however can struggle when it comes to short-term, small-scale precipitation forecasting, due to incomplete observation data, imperfect modeling or insufficient atmospheric understanding.

To provide accurate precipitation nowcasting, Beijing startup ColorfulClouds Tech is applying machine learning (ML) techniques using observed radar echo maps to generate high resolution, minute-by-minute rainfall forecasts on its ColorfulClouds Weather mobile app.

weather forecast.jpg
Sample minute-by-minute rainfall forecast in ColorfulClouds Weather app

ColorfulClouds has built up an end-to-end model which intakes authoritative radar map data from the China NMIC (National Meteorological Information Center) and the US NOAA (National Oceanic and Atmospheric Administration) to complete forecasts through machine learning powered technical processes of segmentation and prediction.

Segmentation filters and removes false echoes from non-precipitation objects such as buildings, hills, birds and aircraft that may appear in the radar reflectivity images. Model training is based on U-Net and SegNet. Segmented images are then processed with deep neural networks to create time-specific fuzzy prediction and output future radar maps.

false echos.jpg
Echo signals from Beijing buildings and mountains reflected on radar image
yuntu.jpg
Before (left) vs After (right) radar map image segmentation

The company uses a self-developed tool for data labeling. In China, this data covers maps from over 200 radar sites for different seasons, latitudes and terrains to ensure a high quality dataset for model training.

ColorfulClouds says its latest ML models’ weather forecast accuracy for the next six hours is comparable with that of traditional weather prediction models.

The ColorfulClouds Weather app is designed to supplement traditional weather forecast methods, allowing users to stay informed on local weather conditions such as temperature, pressure and wind; while also providing additional public awareness information (e.g. smog). Air pollutant concentrations and air quality index (AQI) data is currently available in Asia and North America.

aqi demo.jpg
Sample air quality data from ColorfulClouds Weather

Along with precipitation nowcasting, ColorfulClouds is also considering applying ML to other aspects of weather forecasting, such as expanding accurate short-term precipitation predictions into long-term precipitation forecasts.

Machine Learning is not new to the field of weather forecasting. Meteorologists and researchers are continuously developing and improving data assimilation and forecasting algorithms and updating related parameters. Common ML applications in contemporary weather forecasting include tropical cyclone intensity prediction and severe convective weather forecasts.

ColorfulClouds believes ML is also applicable in NWP models where fuzzy estimation is required, for data interpolation based on observation, coefficients estimation, data assimilation, etc. There remains however a talent shortage, as very few professionals in the field have comprehensive understanding of meteorological models.

Inspired by the NeurIPS 2018 best paper Neural Ordinary Differential Equations, a ColorfulClouds long-term aim is to develop deep neural networks that allow backpropagation for the set of equations used in weather forecasting. Discovering an ML approach that surpasses tried-and-tested, 40-year-old NWP models will be both challenging and rewarding.

This article is based on a Synced interview with ColorfulClouds Algorithm Team Director Pengqiu Xu. Founded in 2014, ColorfulClouds provides weather forecasting and translation services.

Source: Synced China


Localization: Tingting Cao | Editor: Michael Sarazen

1 comment on “AI Local Rainfall Nowcasting Using Weather Radar Maps

  1. In normal poker game, you’re playing whoever’s sitting round that table.
    Once in the online market place there are hundreds of casinos it is possible to
    choose by way of. But there’s no denying you’re right ’bout one consideration. http://Www.Directoryoftheturf.com/externalsite.cfm?id=14582&webaddress=//kslot.app

Leave a Reply

Your email address will not be published.

%d bloggers like this: