Creating Synthetic Weather Radar Images Using Convolutional Neural NetworksChristopher J. Mattioli, Mark S. Veillette, Haig Iskenderian, Eric P. Hassey, Patrick M. Lamey: Massachusetts Institute of Technology, Lincoln Laboratory, Lexington, MA

Synthetic Weather Radar• The Synthetic Weather Radar (SWR) capability is a system

that creates radar-like mosaics of precipitation intensity and echo tops beyond the range of current weather radar

• At its core, this capability makes use of a Convolutional Neural Network (CNN) trained with truth sources over land

• Its architecture is that of a directed acyclic graph (DAG) where multiple output layers contribute to the loss

VIS

1 Channel at 1 km Resolution

4 Channels at 4 km Resolution

9 Densities at 2 km Resolution

10 Model Fieldsat 13 km Resolution

IR

Lightning

NumericalModel

CNN (Convolutional Neural Network)

Convolution RectifiedLinear Unit Pooling Dropout

Loss: MSEBackpropagation

Convolution RectifiedLinear Unit Pooling Dropout

Loss: MSEBackpropagation

Convolution RectifiedLinear Unit Dropout

Loss: MSEBackpropagation

Convolution RectifiedLinear Unit Pooling Dropout

Loss: MSEBackpropagation

Concatenation Convolution Convolution• • •

N{

Objective

Loss: MSE

Backpropagation

*MSE: Mean Squared Error

Current extent of weather radar cannot represent hazardous weather offshore

indicated by lightning flashes (white markers).

SWR applied to the same case is able to fill in the precipitation outside the

range of the radar.

Data Sources for Synthetic Weather Radar

Regions without weather radar have varying areas and times of data coverage. The challenge is to combine these data sources to create a radar-like depiction of weather.

Fixed Weather Radar

Surface Observations

Low-Earth Orbiting Satellite Lightning Strike Locations

Numerical Model

Geostationary Satellite

Global Expansion and Forecasting for DoD

MIT Lincoln Laboratory routinely performs quantitative evaluations of the synthetic weather radar against land-based NEXRAD radar and space-based radar from NASA’s Global Precipitation Measurement (GPM) mission Core Observatory satellite.

Quantitative Capability Evaluation

SWR Probability of Detection vs. GPM

NEXRAD Radar Synthetic Radar Comparison

Comparison

Probability of Detection (POD) = #Hits / (#Hits + #Misses)

05/21, 13Z

06/02, 19Z

06/03, 08Z

06/03, 18Z

06/06, 07Z

06/06, 17Z

07/13, 20Z

07/14, 06Z

07/20, 19Z

07/23, 18Z

07/25, 18Z

07/28, 17Z

08/09, 13Z

08/10, 12Z

08/18, 11Z

08/20, 11Z

08/20, 19Z

08/21, 08Z

08/25, 18Z

08/25, 19Z

08/26, 09Z

08/27, 18Z

08/28, 18Z

08/29, 17Z

08/30, 17Z0

0.2

0.4

0.6

0.8

1

Selected Overpasses from Summer 2015

Prob

abili

ty o

f Det

ectio

n

GPM Radar Synthetic Radar

GPM

Limited Weather Radar over Caribbean

The Caribbean region represents over a million square kilometers of offshore U.S. airspace and supports nearly 17% of all U.S. outbound passengers. Air traffic demand in the Caribbean is expected to grow rapidly by 5-6% per year over the next decade, and the lack of weather radar coverage in this airspace will continue to be a major hurdle for safe and effective air traffic control.

ZHU ZIX

ZMA

ZNY Non-Radar

ADS-B Challenges

International Coordination

Degraded NEXRAD

Destruction of Puerto Rico’s NEXRAD

Learned Features for Visible Satellite

“Peakyness” Detector “Bottom Edge” Detector

“Clear Sky” Detector “Cumulus” Detector

High

Low

Feat

ure

Inte

nsity

32 filters learned in first convolutional layer

* Expa

nded

Cap

abili

ties Global Proof of Concept Forecasting Prototype

GALWEM Numerical Model

Portable Doppler Radar

Advanced Satellites

New

Dat

a So

urce

s

DISTRIBUTION STATEMENT A. Approved for public release. Distribution is unlimited.This material is based upon work supported by the Federal Aviation Administration and Department of the Air Force under Air Force Contract No. FA8702-15-D-0001. Any opinions, findings, conclusions or recommendations expressed in this material are those of the author(s) and do not necessarily reflect the views of the Federal Aviation Administration and Department of the Air Force.© 2019 Massachusetts Institute of Technology.Delivered to the U.S. Government with Unlimited Rights, as defined in DFARS Part 252.227-7013 or 7014 (Feb 2014). Notwithstanding any copyright notice, U.S. Government rights in this work are defined by DFARS 252.227-7013 or DFARS 252.227-7014 as detailed above. Use of this work other than as specifically authorized by the U.S. Government may violate any copyrights that exist in this work.