Predicting origins of coherent air mass trajectories using a neural network—the case of dry intrusions
Abstract Mid‐latitude cyclones are complex weather systems that are tightly related to surface weather impacts. Coherent air streams are known to be associated with such systems, in particular dry intrusions (DIs) in which dry air masses descend slantwise from the vicinity of the tropopause equatorw...
| Main Authors: | , , |
|---|---|
| Format: | Article |
| Language: | English |
| Published: |
Wiley
2021-03-01
|
| Series: | Meteorological Applications |
| Subjects: | |
| Online Access: | https://doi.org/10.1002/met.1986 |
| _version_ | 1818083053133627392 |
|---|---|
| author | Vered Silverman Stav Nahum Shira Raveh‐Rubin |
| author_facet | Vered Silverman Stav Nahum Shira Raveh‐Rubin |
| author_sort | Vered Silverman |
| collection | DOAJ |
| description | Abstract Mid‐latitude cyclones are complex weather systems that are tightly related to surface weather impacts. Coherent air streams are known to be associated with such systems, in particular dry intrusions (DIs) in which dry air masses descend slantwise from the vicinity of the tropopause equatorward towards the surface. Often, DIs are associated with severe surface winds, heavy precipitation and frontogenesis. Currently, DIs can only be identified in hindsight by costly Lagrangian calculations using high resolution wind field data. Here, we use a novel method aiming to simplify the detection procedure of DI origins to allow their future identification in climate datasets, previously inaccessible for such diagnostic studies. A novel adaptation of a segmentation‐oriented neural network model is hereby presented as a successful tool to identify DI origins based solely on three ERA‐Interim reanalysis geopotential height fields, representing the state of the atmosphere. The model prediction skill is tested by calculating both the grid‐point and DI object based Matthews correlation coefficient. We find the model highly skilful in both reconstructing accurately the climatological distribution and predicting the vast majority of the individual DI origin objects. The skill decreases for relatively small objects and for objects occurring at locations where such cases are relatively less frequent. This indicates that geopotential height variability is related to the dynamic mechanisms involved in DI initiations. The results serve as a proof of concept for predicting DIs and other coherent air mass trajectories even when high resolution wind field data are not available, such as for model output for future climate projections. |
| first_indexed | 2024-12-10T19:31:52Z |
| format | Article |
| id | doaj.art-a46c706a877e4bb5b1c142a3cb63a9d2 |
| institution | Directory Open Access Journal |
| issn | 1350-4827 1469-8080 |
| language | English |
| last_indexed | 2024-12-10T19:31:52Z |
| publishDate | 2021-03-01 |
| publisher | Wiley |
| record_format | Article |
| series | Meteorological Applications |
| spelling | doaj.art-a46c706a877e4bb5b1c142a3cb63a9d22022-12-22T01:36:14ZengWileyMeteorological Applications1350-48271469-80802021-03-01282n/an/a10.1002/met.1986Predicting origins of coherent air mass trajectories using a neural network—the case of dry intrusionsVered Silverman0Stav Nahum1Shira Raveh‐Rubin2Department of Earth and Planetary Sciences Weizmann Institute of Science Rehovot IsraelDepartment of Earth and Planetary Sciences Weizmann Institute of Science Rehovot IsraelDepartment of Earth and Planetary Sciences Weizmann Institute of Science Rehovot IsraelAbstract Mid‐latitude cyclones are complex weather systems that are tightly related to surface weather impacts. Coherent air streams are known to be associated with such systems, in particular dry intrusions (DIs) in which dry air masses descend slantwise from the vicinity of the tropopause equatorward towards the surface. Often, DIs are associated with severe surface winds, heavy precipitation and frontogenesis. Currently, DIs can only be identified in hindsight by costly Lagrangian calculations using high resolution wind field data. Here, we use a novel method aiming to simplify the detection procedure of DI origins to allow their future identification in climate datasets, previously inaccessible for such diagnostic studies. A novel adaptation of a segmentation‐oriented neural network model is hereby presented as a successful tool to identify DI origins based solely on three ERA‐Interim reanalysis geopotential height fields, representing the state of the atmosphere. The model prediction skill is tested by calculating both the grid‐point and DI object based Matthews correlation coefficient. We find the model highly skilful in both reconstructing accurately the climatological distribution and predicting the vast majority of the individual DI origin objects. The skill decreases for relatively small objects and for objects occurring at locations where such cases are relatively less frequent. This indicates that geopotential height variability is related to the dynamic mechanisms involved in DI initiations. The results serve as a proof of concept for predicting DIs and other coherent air mass trajectories even when high resolution wind field data are not available, such as for model output for future climate projections.https://doi.org/10.1002/met.1986air massair streamsdeep learningdry intrusionsmachine learningneural networks |
| spellingShingle | Vered Silverman Stav Nahum Shira Raveh‐Rubin Predicting origins of coherent air mass trajectories using a neural network—the case of dry intrusions Meteorological Applications air mass air streams deep learning dry intrusions machine learning neural networks |
| title | Predicting origins of coherent air mass trajectories using a neural network—the case of dry intrusions |
| title_full | Predicting origins of coherent air mass trajectories using a neural network—the case of dry intrusions |
| title_fullStr | Predicting origins of coherent air mass trajectories using a neural network—the case of dry intrusions |
| title_full_unstemmed | Predicting origins of coherent air mass trajectories using a neural network—the case of dry intrusions |
| title_short | Predicting origins of coherent air mass trajectories using a neural network—the case of dry intrusions |
| title_sort | predicting origins of coherent air mass trajectories using a neural network the case of dry intrusions |
| topic | air mass air streams deep learning dry intrusions machine learning neural networks |
| url | https://doi.org/10.1002/met.1986 |
| work_keys_str_mv | AT veredsilverman predictingoriginsofcoherentairmasstrajectoriesusinganeuralnetworkthecaseofdryintrusions AT stavnahum predictingoriginsofcoherentairmasstrajectoriesusinganeuralnetworkthecaseofdryintrusions AT shiraravehrubin predictingoriginsofcoherentairmasstrajectoriesusinganeuralnetworkthecaseofdryintrusions |