| snippet:
|
Severe Weather Detection Product for Hurricane Milton October 2024. |
| summary:
|
Severe Weather Detection Product for Hurricane Milton October 2024. |
| extent:
|
[[-92.0770852654455,18.8146421807991],[-73.7770864874842,36.7146432501091]] |
| accessInformation:
|
NASA; John Cooney, Kris Bedka |
| thumbnail:
|
thumbnail/thumbnail.png |
| maxScale:
|
1.7976931348623157E308 |
| typeKeywords:
|
["Data","Service","Map Service","ArcGIS Server"] |
| description:
|
<div style='text-align:Left;'><div><div><p><span><b>Date of Images:</b></span></p><p><span>Post-Event: 10/10/2024</span></p><p><span><b>Summary:</b></span></p><p><span>Severe weather clouds exhibit distinct patterns in satellite imagery. These patterns, commonly referred to as overshooting cloud tops (OTs) and above anvil cirrus plumes (AACPs), are routinely observed by the human-eye in satellite imagery atop the most intense storms on Earth. With the latest geostationary high spatio-temporal resolution satellite imagers such as, GOES-16 and GOES-17, these severe weather patterns are being observed better than ever before. Advances in deep learning methods allow us to not only detect these signatures but also automate the detection process for real-time applications. Our software code was developed in Python and contains up to 3 model types, U-Net, MultiResU-Net, and AttentionU-Net with varying geostationary satellite channels used as model inputs. The MultiresU-Net model has been found to identify the severe storm signatures better than the other 2 model types and is the default provided in the software release. The models' were trained using a large database of OT and AACP patterns identified from human analysts based on 1-minute GOES-16 ABI imagery. In testing, the models have been shown to successfully detect OTs and AACPs with high precision. The software has also successfully been run for real-time and archived applications. The goal of this software is for users to accurately detect these severe storm signatures for a variety of forecast and research applications.</span></p><p><span>Version 2 released on 6 March 2024. LARGE model performance enhancement, particularly with AACP detection. Added A LOT of new model input combinations as possibilities, including TROPDIFF (10.3 μm - GFS Tropopause Temperature), DIRTYIRDIFF (12.3 μm - 10.3 μm), SNOWICE (1.6 μm), and CIRRUS (1.37 μm). Added new and improved checkpoint files associated with those model runs. Optimal models and thresholds have been updated since the Version 1 release. Changed how we interpolate the GFS data onto the GOES grid. We now follow the methods outlined in Khlopenkov et al. (2021). New version changes how we decide between daytime and nighttime model runs. Previously we used the maximum solar zenith angle within the domain and checked if that exceeded 85°, however, there was an issue of only nighttime models being used for CONUS domains due to how the satellite views the Earth and the data are gridded. Thus, now the software checks to see if more than 5% of the pixels in the domain have solar zenith angles that exceed 85°. If they do, then the software acts as if the domain is nighttime and if not the software acts as if it is daytime within the domain. Fixed major issue with looping over post-processing dates and the month or year changed during a real-time or archived model run. New version speeds up post-processing for OTs. Set variables to None after using them in order to clear cached memory. Catch and hide RunTime warnings that the User does not need to worry about. Software Users prior to 5 March 2024 will need to pull the latest python files from GitHub. This should be very fast. Users MUST also download the new checkpoint files from 'https://science-data.larc.nasa.gov/LaRC-SD-Publications/2023-05-05-001-JWC/data/ML_data.zip'. Once downloaded and unzipped, replace the old model_checkpoints subdirectory with the latest one that was just downloaded. This directory includes all of the improved model detection files as well as checkpoint files for the new input combination. If you have issues downloading and unzipping the file from the browser, run the run_download_model_chkpoint_files.py script included with the software. Users have experienced issues downloading the checkpoint files from the browser and this script has been successful.</span></p><p><span><b>Suggested Use:</b></span></p><p><i>Minimum Tropopause-Relative IR Temperature: </i>Minimum difference between the GFS Tropopause Temperature and the Cloud Top Infrared Brightness Temperature for each Grid Box During the Study Period (Kelvin). Calculated by GFS Tropopause Temperature - Cloud Top Infrared Brightness Temperature and then finding the minimum throughout the study period at each grid box. Infrared brightness temperatures nearer to or colder than the tropopause can indicate locations of more severe convection.</p><p><i>Cold Duration:</i> Number of Minutes in which the Cloud Top Infrared Brightness Temperature is Colder than the GFS Tropopause Temperature During the Study Period. Longer time periods can indicate locations of prolonged severe convective activity.</p><p><i>Overshooting Cloud Top Tropopause-Relative IR Temperature</i>: Minimum difference between GFS Tropopause Temperature and the IR Brightness Temperature for Overshooting Top Detections During the Study Period. Locations where algorithm detected the Overshooting Top severe storm signature and its associated GFS Tropopause Temperature - IR Brightness Temperature (Kelvin). Infrared brightness temperatures nearer to or colder than the tropopause can be another indication of a location experiencing severe convection.</p><p><i>Overshooting Cloud Top Storm Updraft Detections:</i> Earliest Overshooting Top Detection Date and Time During the Study Period. Locations where algorithm detected Overshooting Top severe storm signature and the earliest time the model detected it. This variable can indicate approximately when and where severe weather occurred. </p><p><span></span></p></div></div></div> |
| licenseInfo:
|
<div style='text-align:Left;'><p><span style='font-weight:bold;'>Data Availability Statement</span></p><p><span>The master set of OT and AACP labels used in this study are available at </span><a href='https://science-data.larc.nasa.gov/LaRC-SD-Publications/2023-05-05-001-JWC/data/master_labels.zip' rel='nofollow ugc' target='_blank'>https://science-data.larc.nasa.gov/LaRC-SD-Publications/2023-05-05-001-JWC/data/master_labels.zip</a><span>. These labels serve as our “truth”, and “masks” to train, validate and test the ML models. The labelers consisted of 2 experienced analysts and 3 lesser-experienced student analysts trained to identify these features. The various labels were combined to arrive at this optimal “master” label database.</span></p><p><span>There are 4 classes, ‘Overshoot’ (OT), ‘Plume’, ‘Confident Plume’, and ‘Null’. Confident and non-confident AACPs are differentiated to allow training of the model on the features most apparent to an analyst, which should also be most easily detectable by an ML model. AACPs can be either cold or warm, depending on the level of the tropopause relative to the storm top and their distance downstream from the parent OT (Murillo & Homeyer, 2022). Warm anomaly regions of AACPs, directly adjacent to an OT, are considered confident detections because they are easiest to see in the sandwich imagery. In addition, AACPs can span hundreds of kilometers and the peripheries may not be near to the parent OT. Thus, models may confuse similarly textured anvil regions with AACPs. Null labels were strategically positioned within GOES imagery without OT/AACP to attempt to teach the model the difference between ordinary convection and our desired features.</span></p><p><span style='font-weight:bold;'>Publication and Citation</span></p><p><span>If you use this software in your project, please contact our team prior to publication and cite our contribution. When using this software, credit the technical background and validation for this software as described here:</span></p><p><span>Cooney, J. W., Bedka, K. M., Liles, C. A., and Homeyer, C. R. (in review). Automated Detection of Overshooting Tops and Above Anvil Cirrus Plumes Within Geostationary Imagery Using Deep Learning. Artificial Intelligence for the Earth Systems.</span></p><p><span>More information can be found at:</span></p><p><a href='https://github.com/nasa/svrstormsig' rel='nofollow ugc' target='_blank'>https://github.com/nasa/svrstormsig</a><br /></p></div> |
| catalogPath:
|
|
| title:
|
Severe Weather Detection Product for Hurricane Milton October 2024 |
| type:
|
Map Service |
| url:
|
|
| tags:
|
["Severe weather","troposphere","NASA","NASA Disasters","Hurricane","Hurricane Milton","storm","updraft"] |
| culture:
|
en-US |
| name:
|
svrstormsig |
| guid:
|
C499FB70-3813-48CE-B6E3-726174546F2E |
| minScale:
|
0 |
| spatialReference:
|
GCS_WGS_1984 |