Reconstruction GWMF using ML

.remark-slide-number .progress-bar-container {
  position: absolute;
  bottom: 0;
  height: 4px;
  display: block;
  left: 0;
  right: 0;
}

.remark-slide-number .progress-bar {
  height: 100%;
  background-color: rgba(2, 70, 79, 0.874);
}

.remark-slide-number .numeric {
  position: absolute;
  bottom: 4%;
  height: 4px;
  display: block;
  right: 2.5%;
  font-weight: bold;
}

.remark-slide-number .numeric-out{
  color: rgba(2, 70, 79, 0.874);
}

</style>

---
count: false

## Estimating Balloon-observed Gravity Wave Momentum Fluxes using Machine Learning and Input from Reanalysis

.pull-left-60[

.stress[Sothea Has1, Riwal Plougonven2, Aurélie Fischer1, ...]<hbr>
.center[
.ref[1 CNRS, Laboratoire de Probabilités, Statistique et Modélisation]<h0br>

.ref[2 Laboratoire de Meteorologie Dynamique, Ecole Polytechnique]
]
]
.pull-right-40[
<img src="img/mainz/scale.png" align="left" width="350" height="75">
]

---
count:false
### What is gravity wave momentum flux (GWMFs)?

.center[
<img src="./img/mainz/gravitational_wave.png" align="center" width="750px" height="400px"/>
]

---
count:false

### What is gravity wave momentum flux (GWMFs)?

.center[
<img src="./img/mainz/gravitational_wave2.png" align="center" width="750px" height="400px"/>
]

---
count:false

### What is gravity wave momentum flux (GWMFs)?

.center[
<img src="./img/mainz/gravitational_wave3.png" align="center" width="750px" height="400px"/>
]

---
count:false

### What is gravity wave momentum flux (GWMFs)?<hbr>

- .stress[Gravity waves]: oscillations in atmosphere due to the restoring force of gravity.

- .stress[Momentum flux]: waves `$\leftrightarrow$` mean flow.

- Sources: topographic features, convective processes, flow over mountains, ...

---
count:false

### What is gravity wave momentum flux (GWMFs)?<hbr>

- .stress[Gravity waves]: oscillations in atmosphere due to the restoring force of gravity.

- .stress[Momentum flux]: waves `$\leftrightarrow$` mean flow.

- Sources: topographic features, convective processes, flow over mountains, ...

### Why does it matter?<hbr>

- Knowledge of its impact is crucial:
    - weather forecasting 
    - climate modeling
    - studies of atmospheric dynamics ...
]

---
template: inter-slide
class: left, middle

## <svg aria-hidden="true" role="img" viewBox="0 0 576 512" style="height:1em;width:1.12em;vertical-align:-0.125em;margin-left:auto;margin-right:auto;font-size:inherit;fill:rgb(16, 111, 171);;overflow:visible;position:relative;"><path d="M565.6 36.2C572.1 40.7 576 48.1 576 56V392c0 10-6.2 18.9-15.5 22.4l-168 64c-5.2 2-10.9 2.1-16.1 .3L192.5 417.5l-160 61c-7.4 2.8-15.7 1.8-22.2-2.7S0 463.9 0 456V120c0-10 6.1-18.9 15.5-22.4l168-64c5.2-2 10.9-2.1 16.1-.3L383.5 94.5l160-61c7.4-2.8 15.7-1.8 22.2 2.7zM48 136.5V421.2l120-45.7V90.8L48 136.5zM360 422.7V137.3l-144-48V374.7l144 48zm48-1.5l120-45.7V90.8L408 136.5V421.2z"/></svg> .bold-blue[Outline]

---
template: inter-slide
class: left, middle
count: false

---
### I. Strateole 2 campaign <img src="./img/mainz/logo_strateole2_4x4cm_300dpi.png" align="right" width="100px"/><hbr> .subtitle[Hasse et al. (2018)] <hbr>

- First campaign (Nov 2019 to Feb 2020):

- 8 superpressure balloons drifting along the tropics
  - altitude `$\sim 18$`km - `$20$`km.
  
- Data collected: 30s observations of in-situ **wind, pressure** and **temperature**.

- GWMFs can be derived using wavelet analysis [.stress[Hertzog et al. (2012)]].
 
.center[
<img src="./img/mainz/vd_2020_01_1_25_bal1.gif" align="center" width="370px"/>
<img src="./img/mainz/vd_2020_01_1_25_bal3.gif" align="center" width="370px"/>
]

---
count: false

### I. Strateole 2 campaign <img src="./img/mainz/logo_strateole2_4x4cm_300dpi.png" align="right" width="100px"/><hbr> .subtitle[Hasse et al. (2018)] <hbr>

- First campaign (Nov 2019 to Feb 2020): 
  - 8 superpressure balloons drifting along the tropics
  - altitude `$\sim 18$`km - `$20$`km.

- Data collected: 30s observations of in-situ **wind, pressure** and **temperature**.

- GWMFs can be derived using wavelet analysis [.stress[Hertzog et al. (2012)]].

.center[
<img src="./img/mainz/balloon_trajectory.png" align="center" height="220px" width="770px"/>
.caption[The balloons' trajectories during the first campaign.]
]

---
### I. Strateole 2 campaign <img src="./img/mainz/logo_strateole2_4x4cm_300dpi.png" align="right" width="100px"/><hbr> .subtitle[Related works]<hbr>

.right[.stress[Corcos et al. (2021)] &nbsp; 👉 &nbsp; <img src="./img/mainz/gwmf_corcos.png" align="center" width="450px"/>]

.left[<img src="./img/mainz/cloud.png" align="center" width="400px"/> &nbsp; 👈 &nbsp; .stress[Bramberger et al. (2022)]]

.right[.stress[Wilson et al. (2023)] &nbsp; 👉 &nbsp; <img src="./img/mainz/turbulence.png" align="center" width="400px"/>]

.left[<img src="./img/mainz/parametrization.png" align="center" width="420px"/> &nbsp; 👈 &nbsp; .stress[Lott et al. (2023)]]

---
count: false

### I. Strateole 2 campaign <img src="./img/mainz/logo_strateole2_4x4cm_300dpi.png" align="right" width="100px"/><hbr> .subtitle[Related works]<hbr>

.right[.stress[Corcos et al. (2021)] &nbsp; 👉 &nbsp; <img src="./img/mainz/gwmf_corcos.png" align="center" width="450px"/>]

.left[<img src="./img/mainz/cloud.png" align="center" width="400px"/> &nbsp; 👈 &nbsp; .stress[Bramberger et al. (2022)]]

.right[.stress[Wilson et al. (2023)] &nbsp; 👉 &nbsp; <img src="./img/mainz/turbulence.png" align="center" width="400px"/>]

.left[<img src="./img/mainz/parametrization1.png" align="center" width="420px"/> &nbsp; 👈 &nbsp; .stress[Lott et al. (2023)]]

---
template: inter-slide
class: left, middle
count: false

---
### II. Reanalysis (ERA5) & Machine Learning (ML)<hbr> .subtitle[5th Generation of the European Reanalysis (ERA5)]<hbr>

- Reanalysis: historical observations `$+$` numerical models `$\rightarrow$` datasets.

- ERA5: state-of-the-art global atmospheric dataset ( 1 to 137 vertical levels ).

- Extracted variables: .stress[precipitation], .stress[pressure], .stress[wind] and .stress[temperature] profile ( `$67$` vertical levels ) at `$5\times 5$` horizontal grid points of `$1^\circ\times1^\circ$` ( `$\sim 100$`km ) resolution.

.center[
<img src="./img/mainz/bal_grid.png" align="center" width="800px"/> 
.caption[Including ERA5 grid points.]
]

---
### II. Reanalysis (ERA5) & Machine Learning (ML)<hbr> .subtitle[Machine Learning Methods (ML)]<hbr>

- ML methods: tree-based ensemble machine learning, <h0br>

- random forest (RF)
 - extremely randomized trees (Extra-trees)
 - adaptive boosting (Adaboost).
 - <del>.stress[DNN], .stress[CNN], .stress[LSTM], ...</del> `$\leftarrow$` lack of observations (16k)!

.pull-left-60[
- Inputs:
 - Temperature: **temp** (*K*)
 - Zonal and meridional wind: **u** and **v** (*m/s*)
 - Log surface pressure: **lnsp**
 - Solar zenith angle: **sza** (*°*)
 - Precipitation: **tp**, **tp_mean** & **tp_sd** (*m*).
 
]
.pull-right-40[
<img src="./img/mainz/ERA52.png" align="center" width="220px"/>
]

- Targets: .stress[absolute], .stress[eastward] & .stress[westward] GWMFs:<h0br>

- High frequency waves (**HF**): period 15mn `$\rightarrow$` 1h.
  - Wide frequency waves (**WF**): period 15mn `$\rightarrow$` 1day.

---
### II. Reanalysis (ERA5) & Machine Learning (ML)<hbr> .subtitle[Machine learning method: Decision trees]<hbr>