Update slides, fix broken image fetching

Added new references, added slides on differentiable models and
foundation models. Websites now seem to be blocking downloads of
images, possibly due to the rise in AI bot scraping therefore set
quarto to dynamically load images.

Author: isaacaka

slides/applications.qmd

@@ -3,7 +3,7 @@ title: "Climate Machine Learning Applications"
 subtitle: "NCAS Summer school 2025"
 format:
   revealjs:
-    embed-resources: true
+    embed-resources: false
     slide-number: true
     chalkboard: false
     preview-links: auto
@@ -31,12 +31,7 @@ bibliography: references.bib
 
 ## Teaching Material Recap {.smaller}
 
-Over the ML sessions at the summer school we have learnt about:
-
-- Classification - categorising items based on information
-- Regression - using information to predict another value
-
-using:
+We have learnt the theory behind using:
 
 - ANNs - using input _features_ to make predictions
 - CNNs - using _image-like_ data as an input
@@ -46,18 +41,18 @@ using:
 
 ## _Image-like_ data
 
-![](https://web.archive.org/web/20250812154555if_/http://colah.github.io/posts/2014-10-Visualizing-MNIST/img/mnist_pca/MNIST-p1815-1.png){.absolute width=30% top=20% left=35%}
+![](http://colah.github.io/posts/2014-10-Visualizing-MNIST/img/mnist_pca/MNIST-p1815-1.png){.absolute width=30% top=20% left=35%}
 
 ::: {.fragment}
-![](https://web.archive.org/web/20250812154555if_/http://colah.github.io/posts/2014-10-Visualizing-MNIST/img/mnist_pca/MNIST-PCA1.png){.absolute width=30% top=20% left=35%}
+![](http://colah.github.io/posts/2014-10-Visualizing-MNIST/img/mnist_pca/MNIST-PCA1.png){.absolute width=30% top=20% left=35%}
 :::
 
 ::: {.fragment}
-![](https://web.archive.org/web/20250812154621if_/http://colah.github.io/posts/2014-10-Visualizing-MNIST/img/mnist_pca/MNIST-PCA1-4.png){.absolute width=30% top=20% left=35%}
+![](http://colah.github.io/posts/2014-10-Visualizing-MNIST/img/mnist_pca/MNIST-PCA1-4.png){.absolute width=30% top=20% left=35%}
 :::
 
 ::: {.fragment}
-![](https://web.archive.org/web/20171016213111if_/http://mdpi.com/atmosphere/atmosphere-08-00024/article_deploy/html/images/atmosphere-08-00024-g005.png){.absolute width=70% top=15% left=15%}
+![](http://mdpi.com/atmosphere/atmosphere-08-00024/article_deploy/html/images/atmosphere-08-00024-g005.png){.absolute width=70% top=15% left=15%}
 :::
 
 ::: {.attribution}
@@ -74,7 +69,7 @@ NB: [colah provides an excellent article](https://colah.github.io/posts/2014-10-
 
 ## Applications in geosciences: {.smaller}
 
-See review of @kashinath2021physics
+See review of @kashinath2021physics and @gmd-16-6433-2023
 
 :::: {.columns}
 ::: {.column width="50%"}
@@ -88,17 +83,21 @@ See review of @kashinath2021physics
 - Downscaling/Upsampling  
   [@harris2022generative]
   <br>
+- Climate Emulators  
+  [@watt2025ace2; @chapman2025camulator; @Dheeshjith_2025]  
+  <br>
 :::
 
 ::: {.column width="50%"}
 - Time series forecasting  
-  [@shao2021deep]  
+  [@shao2021deep; @nguyen2023climax; @bodnar2025foundation]  
   <br>
 - Equation discovery  
   [@zanna2020data; @ma2021data]  
   <br>  
 - Complete forecasting  
-  [@rasp2020weatherbench; @pathak2022fourcastnet; @bi2022pangu]  
+  [@pathak2022fourcastnet; @bi2022pangu; @rasp2024weatherbench; @Kochkov_2024; @nathaniel2024chaosbench]  
+  <br>
   <br>
 
 :::
@@ -188,7 +187,7 @@ Additional challenges:
 - Train to predict _'image'_ from coarsened version.
   - Topography?
 
-![](https://web.archive.org/web/20250907164310if_/https://earthdatascience.org/images/earth-analytics/climate-data/downscale-climate-data-met.jpg)
+![](https://earthdatascience.org/images/earth-analytics/climate-data/downscale-climate-data-met.jpg)
 
 ::: {.attribution}
 Image by [Earth Lab](https://www.earthdatascience.org/courses/use-data-open-source-python/hierarchical-data-formats-hdf/intro-to-MACAv2-cmip5-data/)
@@ -198,10 +197,10 @@ Image by [Earth Lab](https://www.earthdatascience.org/courses/use-data-open-sour
 ## Forecasting {.smaller}
 
 - Time-series
-  - popular use
   - Recurrent Neural Nets
 - Complete weather
-  - FourCastNet, Pangu-Weather, GraphCast, NeuralGCM
+  - FourCastNet, Pangu-Weather, 
+  - GraphCast, NeuralGCM
 
 ![](https://media.springernature.com/full/springer-static/image/art%3A10.1038%2Fs41586-023-06185-3/MediaObjects/41586_2023_6185_Fig2_HTML.png?as=webp){.absolute bottom=0% left=0% width=48%}
 
@@ -216,6 +215,28 @@ Global image from NVIDIA FourCastNet
 Speaker notes go here.
 :::
 
+## Differentiable Models
+
+- Online training
+  - End-to-end differentiable GCMs
+  - Greater stability
+
+![](https://storage.googleapis.com/gweb-research2023-media/images/NeuralGCM-img1.width-1250.png){.absolute bottom=10% left=10% width=100%}
+
+::: {.attribution}
+Images from Google NeuralGCM
+:::
+
+
+## Foundation Models
+- Pretrained on large amount of heterogeneous climate datasets
+- Aims to learn general purpose representations of dynamic
+- Fine tune for specific forecasting tasks
+- Examples:
+  - ClimaX, Microsoft Aurora 
+
+
+
 # Challenges
 
 ## Training data - considerations {.smaller}
@@ -287,7 +308,7 @@ Replacing physics-based components of larger models (emulation or data-driven) r
 - Language interoperation
 - Physical compatibility
 
-![]( https://web.archive.org/web/20250718183653if_/https://upload.wikimedia.org/wikipedia/commons/thumb/5/55/Mathematical_Bridge_tangents.jpg/250px-Mathematical_Bridge_tangents.jpg ){style="border-radius: 50%;" .absolute top=40% left=30% width=40%}
+![]( https://upload.wikimedia.org/wikipedia/commons/thumb/5/55/Mathematical_Bridge_tangents.jpg/250px-Mathematical_Bridge_tangents.jpg ){style="border-radius: 50%;" .absolute top=40% left=30% width=40%}
 
 ::: {.attribution}
 [Mathematical Bridge](https://upload.wikimedia.org/wikipedia/commons/thumb/5/55/Mathematical_Bridge_tangents.jpg/250px-Mathematical_Bridge_tangents.jpg)
@@ -303,7 +324,7 @@ used under [CC BY-SA 3.0](https://creativecommons.org/licenses/by-sa/3.0/deed.en
 
 ![](https://raw.githubusercontent.com/pytorch/pytorch/main/docs/source/_static/img/pytorch-logo-dark.png){style="background-image: radial-gradient(gray 40%, black);" .absolute top=65% left=5% width=20%}
 
-![]( https://web.archive.org/web/20250825230804if_/https://upload.wikimedia.org/wikipedia/commons/1/11/TensorFlowLogo.svg ){.absolute top=75% left=7.5% width=15%}
+![]( https://upload.wikimedia.org/wikipedia/commons/1/11/TensorFlowLogo.svg ){.absolute top=75% left=7.5% width=15%}
 
 
 
@@ -393,7 +414,7 @@ runtime
 
 ![](https://raw.githubusercontent.com/pytorch/pytorch/main/docs/source/_static/img/pytorch-logo-dark.png){style="background-image: radial-gradient(gray 40%, black);" .absolute bottom=12.5% right=22% height=10%}
 
-![](https://web.archive.org/web/20250825230804if_/https://upload.wikimedia.org/wikipedia/commons/1/11/TensorFlowLogo.svg){.absolute bottom=10% left=82% height=15%}
+![](https://upload.wikimedia.org/wikipedia/commons/1/11/TensorFlowLogo.svg){.absolute bottom=10% left=82% height=15%}