System and Method for Detecting Misidentified Hydrometeors in Weather Radar Data
A system for detection of misidentified hydrometeors includes a raw radar null event identifier and a null event time sequence creator. The raw radar null event identifier is configured to receive weather radar data and null event information and to identify a null event in the weather radar data. The null event time sequence creator is configured to receive the null event from the raw radar null event identifier and to form a null event time sequence based on the null event.
1 . A system for detection of misidentified hydrometeors comprising:
a raw radar null event identifier configured to receive weather radar data and null event information and identify a null event in the weather radar data; and
a null event time sequence creator configured to receive the null event from the raw radar null event identifier and form a null event time sequence based on the null event.
2 . The system of claim 1 , wherein the weather radar data comprises raw weather data and/or filtered radar data.
3 . The system of claim 1 , wherein the weather radar data comprises Maximum Expected Size of Hail (MESH) radar-derived product data.
4 . The system of claim 1 , wherein the null event information comprises one of the group consisting of machine learning null event information and human identified null event information.
5 . The system of claim 1 , further comprising a machine learning null event information module comprising an artificial neural network configured to produce the null event information.
6 . The system of claim 5 , wherein the neural network further comprises a long short-term memory (LSTM) architecture.
7 . The system of claim 6 , wherein the LSTM architecture comprises a plurality of gates configured to determine an output activation of a received input sequence and classify the sequence as either a normal behavior sequence or a null sequence.
8 . A computer based method for detection of misidentified hydrometeors comprising the steps of:
receiving weather radar data;
receiving null event information; and
identifying a null event in the weather radar data based upon the null event information.
9 . The method of claim 8 , wherein the weather radar data comprises raw weather radar data and/or filtered weather radar data.
10 . The method of claim 8 , further comprising the step of forming a null event time sequence based on the null event.
11 . The method of claim 10 , further comprising the step of storing the null event time sequence in a null temporal sequence data archive.
12 . The method of claim 8 , wherein identifying a null event in the weather data further comprises:
identifying expected hail size variations over time,
comparing changes in reported hail sizes over time according to the weather radar data with the expected hail size variations over time; and
if the changes in reported hail sizes over time are inconsistent with the expected hail size variations over time, declaring the reported hail sizes over time as anomalous.
13 . The method of claim 12 , wherein the expected hail size variations over time exhibit smooth transitions rather than step-function variation.
14 . The method of claim 8 , wherein the null event information further comprises machine learning null event information and/or human identified null event information.
15 . The method of claim 8 , further comprising the step of populating an archive of null temporal sequence data.
16 . The method of claim 8 , wherein identifying a null event in the weather radar data further comprises the steps of:
determining an output activation of a received input sequence; and
classifying the sequence as either a normal behavior sequence or a null sequence.