![\"n](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5838132%2F7bb604cc0c6ff4e9ed85070baf1ca9a4%2FE3SM-cloud-resolving-model.jpg?generation=1713557912839814&alt=media\")
To understand more about the context of the competition, lets take a look at the E3SM-MMF model, which has generated the data for this competition.
\nFrom the offical project website, the E3SM-MMF (Energy Exascale Earth System Model - Multiscale Modeling Framework) is a project developed by the Exascale Computing project, supported by the U.S. Department of Energy. It's goal is to find a faster approach to cloud resolving modeling based on an MMF approach, employing superparameterization. Specifically, it can accurately incorporate cloud physics while also obtaining the throughput necessary for multidecade, coupled high-resolution climate simulations.
\nThere's lots of confusing terms here, lets quickly deconfuse them.
\nA cloud resolving model, or CRM, is a model that allows performing numerical simulations of different types of convective clouds (i.e., clouds formed by the process of convection: warmer air raises and gets concentrated since it is less dense than the surrounding colder air). They employ realistic representations of cloud microphyisical processes and can predict (or, resolve) the evolution of cloud systems in time, structure and life cycle. They also can predict interactions between clouds, different types of outgoing longwave and incoming solar radiation, as well as ocean and land surface processes. This should be clear, as all of these are names of features or targets in the competition. This paper provides a nice review of CRMs.
\n
A Multiscale Modeling Framework, or MMF, is a modelling framework typically employed for physical processes and phenomena that can be modeled at varying degrees of complexity and at different scales. A good starting point is the book Principles of Multiscale Modeling by Weinan E.
\n![\"The](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5838132%2Fa051bb2f96bc53303137378a1d66df9c%2Fe3_image2.png?generation=1713557941366886&alt=media\")
And finally, superparameterization refers to a specific technique employed with MMFs. The issue of MMFs is that they require information that appears at both larger and smaller scales, some of which may be missing. For ths purpose, this missing information is condensed into model parameters. While simple parameterization refers to estimating parameters without simulating the process directly, superparameterization replaces these parameterizations with a seconf model that does similate the precess, thus yielding more accurate parameter values. A good primer is this blog by Walter Hannah.
\n![](\"https://www.googleapis.com/download/storage/v1/b/kaggle-forum-message-attachments/o/inbox%2F5838132%2F22717b2a71320c2d375fd6cd96716ab9%2Fjame20937-fig-0004-m.jpg?generation=1713557975073287&alt=media\")
I find climate related machine learning to be a very interesting field. Let's use this as an opportunity to learn more about it!
", "messages": [ { "id": 2761419, "postDate": "2024-04-19T20:17:47.070Z", "content": "To understand more about the context of the competition, lets take a look at the E3SM-MMF model, which has generated the data for this competition.
\nFrom the offical project website, the E3SM-MMF (Energy Exascale Earth System Model - Multiscale Modeling Framework) is a project developed by the Exascale Computing project, supported by the U.S. Department of Energy. It's goal is to find a faster approach to cloud resolving modeling based on an MMF approach, employing superparameterization. Specifically, it can accurately incorporate cloud physics while also obtaining the throughput necessary for multidecade, coupled high-resolution climate simulations.
\nThere's lots of confusing terms here, lets quickly deconfuse them.
\nA cloud resolving model, or CRM, is a model that allows performing numerical simulations of different types of convective clouds (i.e., clouds formed by the process of convection: warmer air raises and gets concentrated since it is less dense than the surrounding colder air). They employ realistic representations of cloud microphyisical processes and can predict (or, resolve) the evolution of cloud systems in time, structure and life cycle. They also can predict interactions between clouds, different types of outgoing longwave and incoming solar radiation, as well as ocean and land surface processes. This should be clear, as all of these are names of features or targets in the competition. This paper provides a nice review of CRMs.
\n
A Multiscale Modeling Framework, or MMF, is a modelling framework typically employed for physical processes and phenomena that can be modeled at varying degrees of complexity and at different scales. A good starting point is the book Principles of Multiscale Modeling by Weinan E.
\n
And finally, superparameterization refers to a specific technique employed with MMFs. The issue of MMFs is that they require information that appears at both larger and smaller scales, some of which may be missing. For ths purpose, this missing information is condensed into model parameters. While simple parameterization refers to estimating parameters without simulating the process directly, superparameterization replaces these parameterizations with a seconf model that does similate the precess, thus yielding more accurate parameter values. A good primer is this blog by Walter Hannah.
\n
I find climate related machine learning to be a very interesting field. Let's use this as an opportunity to learn more about it!
", "rawMarkdown": "To understand more about the context of the competition, lets take a look at the E3SM-MMF model, which has **generated the data** for this competition.\n\nFrom the [offical project website](https://www.exascaleproject.org/research-project/e3sm-mmf/), the E3SM-MMF (Energy Exascale Earth System Model - Multiscale Modeling Framework) is a project developed by the Exascale Computing project, supported by the U.S. Department of Energy. It's goal is to find a faster approach to **cloud resolving modeling** based on an **MMF** approach, employing **superparameterization**. Specifically, it can accurately incorporate cloud physics while also obtaining the throughput necessary for multidecade, coupled high-resolution climate simulations. \n\nThere's lots of confusing terms here, lets quickly deconfuse them.\n\nA **cloud resolving model**, or CRM, is a model that allows performing numerical simulations of different types of convective clouds (i.e., clouds formed by the process of convection: warmer air raises and gets concentrated since it is less dense than the surrounding colder air). They employ realistic representations of cloud microphyisical processes and can predict (or, **resolve**) the evolution of cloud systems in time, structure and life cycle. They also can predict interactions between clouds, different types of outgoing longwave and incoming solar radiation, as well as ocean and land surface processes. This should be clear, as all of these are names of features or targets in the competition. [This paper](https://www.tandfonline.com/doi/full/10.1080/16000870.2017.1373578) provides a nice review of CRMs.\n\n\n\nA **Multiscale Modeling Framework**, or MMF, is a modelling framework typically employed for physical processes and phenomena that can be modeled at varying degrees of complexity and at different scales. A good starting point is the book [Principles of Multiscale Modeling](https://web.math.princeton.edu/~weinan/papers/weinan_book.pdf) by Weinan E.\n\n\n\nAnd finally, **superparameterization** refers to a specific technique employed with MMFs. The issue of MMFs is that they require information that appears at both larger and smaller scales, some of which may be missing. For ths purpose, this missing information is condensed into model parameters. While simple parameterization refers to estimating parameters without simulating the process directly, **superparameterization** replaces these parameterizations with a seconf model that does similate the precess, thus yielding more accurate parameter values. A good primer is [this blog](https://hannahlab.org/blog/what-is-super-parameterization/) by Walter Hannah.\n\n\n\nI find climate related machine learning to be a very interesting field. Let's use this as an opportunity to learn more about it!", "votes": 17 }, { "id": 2761651, "postDate": "2024-04-20T01:24:56.760Z", "content": "Indeed Sarvazyan, climate +ML it's a compelling subject. Stunning topic and thanks for the article (A short review of numerical cloud-resolving models) link.
", "rawMarkdown": "Indeed Sarvazyan, climate +ML it's a compelling subject. Stunning topic and thanks for the article (A short review of numerical cloud-resolving models) link." } ], "comments": [ { "id": 2761651, "author_name": "Marília Prata", "author_url": "", "post_date": "2024-04-20T01:24:56.760000", "content": "Indeed Sarvazyan, climate +ML it's a compelling subject. Stunning topic and thanks for the article (A short review of numerical cloud-resolving models) link.
", "votes": 0, "replies": [] } ], "raw_markdown_by_id": { "2761419": "To understand more about the context of the competition, lets take a look at the E3SM-MMF model, which has **generated the data** for this competition.\n\nFrom the [offical project website](https://www.exascaleproject.org/research-project/e3sm-mmf/), the E3SM-MMF (Energy Exascale Earth System Model - Multiscale Modeling Framework) is a project developed by the Exascale Computing project, supported by the U.S. Department of Energy. It's goal is to find a faster approach to **cloud resolving modeling** based on an **MMF** approach, employing **superparameterization**. Specifically, it can accurately incorporate cloud physics while also obtaining the throughput necessary for multidecade, coupled high-resolution climate simulations. \n\nThere's lots of confusing terms here, lets quickly deconfuse them.\n\nA **cloud resolving model**, or CRM, is a model that allows performing numerical simulations of different types of convective clouds (i.e., clouds formed by the process of convection: warmer air raises and gets concentrated since it is less dense than the surrounding colder air). They employ realistic representations of cloud microphyisical processes and can predict (or, **resolve**) the evolution of cloud systems in time, structure and life cycle. They also can predict interactions between clouds, different types of outgoing longwave and incoming solar radiation, as well as ocean and land surface processes. This should be clear, as all of these are names of features or targets in the competition. [This paper](https://www.tandfonline.com/doi/full/10.1080/16000870.2017.1373578) provides a nice review of CRMs.\n\n\n\nA **Multiscale Modeling Framework**, or MMF, is a modelling framework typically employed for physical processes and phenomena that can be modeled at varying degrees of complexity and at different scales. A good starting point is the book [Principles of Multiscale Modeling](https://web.math.princeton.edu/~weinan/papers/weinan_book.pdf) by Weinan E.\n\n\n\nAnd finally, **superparameterization** refers to a specific technique employed with MMFs. The issue of MMFs is that they require information that appears at both larger and smaller scales, some of which may be missing. For ths purpose, this missing information is condensed into model parameters. While simple parameterization refers to estimating parameters without simulating the process directly, **superparameterization** replaces these parameterizations with a seconf model that does similate the precess, thus yielding more accurate parameter values. A good primer is [this blog](https://hannahlab.org/blog/what-is-super-parameterization/) by Walter Hannah.\n\n\n\nI find climate related machine learning to be a very interesting field. Let's use this as an opportunity to learn more about it!", "2761651": "Indeed Sarvazyan, climate +ML it's a compelling subject. Stunning topic and thanks for the article (A short review of numerical cloud-resolving models) link." } }