Flood Mapping Anomaly Detection

Owned by Madeline Lambert
Last updated: Oct 27, 2020
3 min read

Data Pipeline

The pipeline in the figure above stems from the portion of the larger product pipeline outlined with a light blue dashed line in the Standard Product Pipeline figure.

The user can select an SLC track to push through the ISCE TopsStack PGE to create a coregistered SLC stack, by searching facets on Tosca. This stack is then input for the PredRNN PGE, which utilizes Machine Learning to predict future flood maps from the stack of data. PredRNN then publishes a predicted flood map using items 0 through (N-1) of the stack. Then, another PGE will compare item N to the prediction, and will publish a wet/dry anomaly map.

Finally, the anomaly map will be fed through a TBD “Production Rules for Detected Anomalies” PGE, in which some TBD value of the map will be compared to a threshold, and if there is a detected anomaly, the PGE will (something; email/notify the person who submitted the job?)

TopsStack Processor (​TopsStack Processor (PGE))

Description:

Trigger dataset/action: On-demand Action “topsStack Processor”

Inputs: SLC (faceted on Tosca), master date, and bbox (min_lat, max_lat, min_lon, max_lon) (bbox is required in multiprocessing and optional in GNU Parallel)

Outputs: directory structure containing coregistered SLC stack, merged geometric data, baselines, etc

PredRNN++ PGE

Description:

Trigger dataset/action:

Inputs:

Outputs: Predicted flood map (naming convention, format?)

Flood Map Comparison PGE** Uncertain as of right now if this comparison will be a separate PGE, or encompassed in the PredRNN++ PGE

Description:

Trigger dataset:

Inputs:

Outputs:

Use Cases

Use Case 1: On-demand request for flood anomaly map for an AOI.

  • Customer externally requests flood anomaly map for an AOI.
    Note to developer: At this time, we will have only developers add AOIs and run data through this pipeline.
    Science story: Requests will come from agencies/companies interested in monitoring flooding during an event in which flooding is expected. Customers will want predicted flood anomaly maps to determine extent of flooding, not to predict flooding itself.

  • User (developer) inputs flood mapping AOI information, including number of before-flood scenes/steps to include.
    Note to developer: General information needed will be 1. AOI/bbox, 2. Number of pre-flood scenes to include in stack

  • System creates coregistered SLC stack.

  • System automatically runs coregistered SLC stack through PredRNN++, and publishes flood anomaly map to S3

  • User logs on to system to browse results.

Use Case 2: Forward keep-up processing of PredRNN+ for rapid response flood mapping

  • Customer externally requests flood anomaly map for an AOI

  • User (developer) inputs flood mapping AOI information, including number of before-flood scenes/steps to include.
    Note to developer: General information needed will be 1. AOI/bbox, 2. Number of pre-flood scenes to include in stack

  • System creates coregistered SLC stack.

  • System updates coregistered SLC stack with each new data acquisition in the AOI.

  • System automatically runs coregistered SLC stack through PredRNN++ and publishes flood anomaly map to S3

  • User logs on to system to browse results.

Questions:

  • How do you input the amount of time/number of steps you want to process? - will need to be an input for the PredRNN PGE

  • Identify production rule(s) - how do we want to trigger this?