Yannis Kalfas
@kalfasyan.bsky.social
Deep Learning | Remote Sensing
PhD in Bioscience Engineering
☁️ cloud segmentation 🌍 land surface classification
@ VITO, Belgium
🇪🇺 LCFM project
🇬🇷 -> 🇸🇪 -> 🇧🇪
PhD in Bioscience Engineering
☁️ cloud segmentation 🌍 land surface classification
@ VITO, Belgium
🇪🇺 LCFM project
🇬🇷 -> 🇸🇪 -> 🇧🇪
Reposted by Yannis Kalfas
[1/2]
Colorful landscape at the border of Chile and Bolivia.
In the center is massive Cerro Chao lava dome.
Multispectral image taken by Sentinel-2 on Dec. 28, 2024.
Colorful landscape at the border of Chile and Bolivia.
In the center is massive Cerro Chao lava dome.
Multispectral image taken by Sentinel-2 on Dec. 28, 2024.
January 24, 2025 at 6:51 PM
[1/2]
Colorful landscape at the border of Chile and Bolivia.
In the center is massive Cerro Chao lava dome.
Multispectral image taken by Sentinel-2 on Dec. 28, 2024.
Colorful landscape at the border of Chile and Bolivia.
In the center is massive Cerro Chao lava dome.
Multispectral image taken by Sentinel-2 on Dec. 28, 2024.
"...new Perch model helps conservationists analyze audio faster to protect endangered species, from Hawaiian honeycreepers to coral reefs."
deepmind.google/discover/blo...
deepmind.google/discover/blo...
How AI is helping advance the science of bioacoustics to save endangered species
Our new Perch model helps conservationists analyze audio faster to protect endangered species, from Hawaiian honeycreepers to coral reefs.
deepmind.google
August 17, 2025 at 3:53 AM
"...new Perch model helps conservationists analyze audio faster to protect endangered species, from Hawaiian honeycreepers to coral reefs."
deepmind.google/discover/blo...
deepmind.google/discover/blo...
Thanks to @pycoders.com for mentioning my package `desto`!
github.com/kalfasyan/de...
pycoders.com/issues/692
github.com/kalfasyan/de...
pycoders.com/issues/692
PyCoder’s Weekly | Issue #692
PyPI, pedalboard, Django URL Patterns, and More
pycoders.com
August 5, 2025 at 12:53 PM
Thanks to @pycoders.com for mentioning my package `desto`!
github.com/kalfasyan/de...
pycoders.com/issues/692
github.com/kalfasyan/de...
pycoders.com/issues/692
Accelerating AI/ML Workflows in Earth Sciences with GPU-Native Xarray, Zarr, DAL and nvcomp
xarray.dev/blog/gpu-pip...
xarray.dev/blog/gpu-pip...
Accelerating AI/ML Workflows in Earth Sciences with GPU-Native Xarray, Zarr, DALI, and nvcomp (and more!)
How to accelerate AI/ML workflows in Earth Sciences with GPU-native Xarray and Zarr.
xarray.dev
July 17, 2025 at 1:02 PM
Accelerating AI/ML Workflows in Earth Sciences with GPU-Native Xarray, Zarr, DAL and nvcomp
xarray.dev/blog/gpu-pip...
xarray.dev/blog/gpu-pip...
Happy to share this open-source project I've been working on lately: 🙂
desto helps you run and manage your bash and Python scripts in the background, using tmux sessions. I built it to simplify my own workflow, and thought maybe it could be helpful for others too 👀
github.com/kalfasyan/de...
desto helps you run and manage your bash and Python scripts in the background, using tmux sessions. I built it to simplify my own workflow, and thought maybe it could be helpful for others too 👀
github.com/kalfasyan/de...
GitHub - kalfasyan/desto: tmux session manager made with nicegui
tmux session manager made with nicegui. Contribute to kalfasyan/desto development by creating an account on GitHub.
github.com
July 5, 2025 at 4:28 PM
Happy to share this open-source project I've been working on lately: 🙂
desto helps you run and manage your bash and Python scripts in the background, using tmux sessions. I built it to simplify my own workflow, and thought maybe it could be helpful for others too 👀
github.com/kalfasyan/de...
desto helps you run and manage your bash and Python scripts in the background, using tmux sessions. I built it to simplify my own workflow, and thought maybe it could be helpful for others too 👀
github.com/kalfasyan/de...
First LCFM products are released! (building on shoulders of Worldcover giants!) 🗺️
www.linkedin.com/posts/copern...
www.linkedin.com/posts/copern...
🗺️ First #Copernicus Land Cover and Forest Monitoring (LCFM) products are now available in the #CDSE Copernicus Browser. | Copernicus Data Space Ecosystem
🗺️ First #Copernicus Land Cover and Forest Monitoring (LCFM) products are now available in the #CDSE Copernicus Browser. Copernicus #LCFM provides 10 m annual land cover maps allowing users to detect ...
www.linkedin.com
June 30, 2025 at 8:56 AM
First LCFM products are released! (building on shoulders of Worldcover giants!) 🗺️
www.linkedin.com/posts/copern...
www.linkedin.com/posts/copern...
Reposted by Yannis Kalfas
A close-up view of this week's destination: #Vienna🇦🇹
From 23-27 June, the city will welcome the @esaearth.esa.int Living Planet Symposium #LPS25, bringing together scientists, researchers and users of Earth observation data from all over the world to share the latest findings on Earth science.
From 23-27 June, the city will welcome the @esaearth.esa.int Living Planet Symposium #LPS25, bringing together scientists, researchers and users of Earth observation data from all over the world to share the latest findings on Earth science.
Earth from Space: Vienna’s Danube divide
A close-up view of Vienna, Austria’s capital city, is featured in this image from April 2025.
www.esa.int
June 23, 2025 at 8:13 AM
A close-up view of this week's destination: #Vienna🇦🇹
From 23-27 June, the city will welcome the @esaearth.esa.int Living Planet Symposium #LPS25, bringing together scientists, researchers and users of Earth observation data from all over the world to share the latest findings on Earth science.
From 23-27 June, the city will welcome the @esaearth.esa.int Living Planet Symposium #LPS25, bringing together scientists, researchers and users of Earth observation data from all over the world to share the latest findings on Earth science.
Reposted by Yannis Kalfas
Amazing differences within the EU.
Via prof. Marco te Brömmelstroet on LinkedIn.
NB: 2019 data.
Via prof. Marco te Brömmelstroet on LinkedIn.
NB: 2019 data.
May 22, 2025 at 7:12 PM
Amazing differences within the EU.
Via prof. Marco te Brömmelstroet on LinkedIn.
NB: 2019 data.
Via prof. Marco te Brömmelstroet on LinkedIn.
NB: 2019 data.
Reposted by Yannis Kalfas
Strange clouds over the North Sea, jets leaving clear contrails — and Sentinel-2 captures it all. A clever use of band offsets even lets us calculate the planes’ speed and altitude! A great example of what satellite images can show. #BCEarthGallery #EarthObservation #RemoteSensing
👉 bit.ly/4k7htTG
👉 bit.ly/4k7htTG
May 7, 2025 at 9:36 AM
Strange clouds over the North Sea, jets leaving clear contrails — and Sentinel-2 captures it all. A clever use of band offsets even lets us calculate the planes’ speed and altitude! A great example of what satellite images can show. #BCEarthGallery #EarthObservation #RemoteSensing
👉 bit.ly/4k7htTG
👉 bit.ly/4k7htTG
Reposted by Yannis Kalfas
Almost half of global average temperature rise and a third of sea-level rise can be attributed to the world’s 122 largest fossil fuel and cement producers, a recent paper shows.
By 2300, greenhouse gases from the companies’ past production could result in 35 centimeters of sea-level rise.
By 2300, greenhouse gases from the companies’ past production could result in 35 centimeters of sea-level rise.
122 companies responsible for a third of present day sea-level rise: Study
What’s new: Almost half of global average temperature rise and a third of sea-level rise can be attributed to the “carbon majors,” the world’s 122 largest fossil fuel and cement producers, a recent…
news.mongabay.com
April 10, 2025 at 1:17 AM
Almost half of global average temperature rise and a third of sea-level rise can be attributed to the world’s 122 largest fossil fuel and cement producers, a recent paper shows.
By 2300, greenhouse gases from the companies’ past production could result in 35 centimeters of sea-level rise.
By 2300, greenhouse gases from the companies’ past production could result in 35 centimeters of sea-level rise.
Reposted by Yannis Kalfas
Animation showing a north-south rupture in the ground west of the Irrawaddy River, Mandalay, Myanmar, after a 7.7-magnitude earthquake struck on 28 March.
In response to the disaster, #CopernicusEMS and the #DisastersCharter have been activated to provide satellite imagery to help relief efforts.
In response to the disaster, #CopernicusEMS and the #DisastersCharter have been activated to provide satellite imagery to help relief efforts.
April 1, 2025 at 12:22 PM
Animation showing a north-south rupture in the ground west of the Irrawaddy River, Mandalay, Myanmar, after a 7.7-magnitude earthquake struck on 28 March.
In response to the disaster, #CopernicusEMS and the #DisastersCharter have been activated to provide satellite imagery to help relief efforts.
In response to the disaster, #CopernicusEMS and the #DisastersCharter have been activated to provide satellite imagery to help relief efforts.
Reposted by Yannis Kalfas
January 2025 was the 18th month in a 19-month period in which the global-average temperature was more than 1.5°C above pre-industrial level. 🧵
February 11, 2025 at 8:26 AM
January 2025 was the 18th month in a 19-month period in which the global-average temperature was more than 1.5°C above pre-industrial level. 🧵
Reposted by Yannis Kalfas
Friends, for something to be open source, we need to see
1. The data it was trained & evaluated on
2. The code
3. Model architecture
4. Model weights.
DeepSeek only gives 3, 4. And I'll see the day that anyone gives us #1 without being forced to do so, because all of them are stealing data.
1. The data it was trained & evaluated on
2. The code
3. Model architecture
4. Model weights.
DeepSeek only gives 3, 4. And I'll see the day that anyone gives us #1 without being forced to do so, because all of them are stealing data.
January 29, 2025 at 5:01 AM
Friends, for something to be open source, we need to see
1. The data it was trained & evaluated on
2. The code
3. Model architecture
4. Model weights.
DeepSeek only gives 3, 4. And I'll see the day that anyone gives us #1 without being forced to do so, because all of them are stealing data.
1. The data it was trained & evaluated on
2. The code
3. Model architecture
4. Model weights.
DeepSeek only gives 3, 4. And I'll see the day that anyone gives us #1 without being forced to do so, because all of them are stealing data.
Reposted by Yannis Kalfas
Why does everyone get so excited about performance on some benchmarks when we don't know what the training data is? How do you know that they're not testing on the training set? How do you reject papers that don't tell you the splits but get so excited by these PR announcements?
January 29, 2025 at 5:14 AM
Why does everyone get so excited about performance on some benchmarks when we don't know what the training data is? How do you know that they're not testing on the training set? How do you reject papers that don't tell you the splits but get so excited by these PR announcements?
Reposted by Yannis Kalfas
We have updated our wildfire summary for 2024 with the latest #CopernicusAtmosphere data from December. Last year, North and South America saw the most wildfire activity.
Dive into the analysis, charts and data: https://bit.ly/4ai4ydO
Dive into the analysis, charts and data: https://bit.ly/4ai4ydO
January 16, 2025 at 10:30 AM
We have updated our wildfire summary for 2024 with the latest #CopernicusAtmosphere data from December. Last year, North and South America saw the most wildfire activity.
Dive into the analysis, charts and data: https://bit.ly/4ai4ydO
Dive into the analysis, charts and data: https://bit.ly/4ai4ydO
"We still don't know why training deep learning models with highly non-convex, 100+ billion parameter loss functions actually works. While we've found many effective tricks, we're only slowly starting to understand them..." ~ David Sauerwein
www.linkedin.com/posts/davids...
www.linkedin.com/posts/davids...
David Sauerwein on LinkedIn: #deeplearning #machinelearning #llm | 71 comments
We still don't know why training deep learning models with highly non-convex, 100+ billion parameter loss functions actually works. While we've found many… | 71 comments on LinkedIn
www.linkedin.com
December 30, 2024 at 1:40 PM
"We still don't know why training deep learning models with highly non-convex, 100+ billion parameter loss functions actually works. While we've found many effective tricks, we're only slowly starting to understand them..." ~ David Sauerwein
www.linkedin.com/posts/davids...
www.linkedin.com/posts/davids...
Reposted by Yannis Kalfas
Can we say something about the phase of water in these clouds? I’m thinking about this recent paper www.science.org/doi/10.1126/...
Glaciation of liquid clouds, snowfall, and reduced cloud cover at industrial aerosol hot spots
The ability of anthropogenic aerosols to freeze supercooled cloud droplets remains debated. In this work, we present observational evidence for the glaciation of supercooled liquid-water clouds at ind...
www.science.org
December 23, 2024 at 3:19 PM
Can we say something about the phase of water in these clouds? I’m thinking about this recent paper www.science.org/doi/10.1126/...
Reposted by Yannis Kalfas
🧪🛰️ A great example at display in this GOES nighttime image over NE North America.
Pollution change cloud properties is an example of an #aerosol-cloud interactions (a rather complicated atmospheric process). The controlling factors the determine a cloud evolution (macro like meteorology
1/2
Pollution change cloud properties is an example of an #aerosol-cloud interactions (a rather complicated atmospheric process). The controlling factors the determine a cloud evolution (macro like meteorology
1/2
December 23, 2024 at 2:48 PM
🧪🛰️ A great example at display in this GOES nighttime image over NE North America.
Pollution change cloud properties is an example of an #aerosol-cloud interactions (a rather complicated atmospheric process). The controlling factors the determine a cloud evolution (macro like meteorology
1/2
Pollution change cloud properties is an example of an #aerosol-cloud interactions (a rather complicated atmospheric process). The controlling factors the determine a cloud evolution (macro like meteorology
1/2
Reposted by Yannis Kalfas
🌍 My new map shows Germany's breathtaking land cover, powered by the highly detailed ESA WorldCover dataset!
Want to see how I create these maps? Watch this video:
🎥 youtu.be/EuLip_J0x3I
#dataviz #GIS #Cartography #DataViz #Germany
Want to see how I create these maps? Watch this video:
🎥 youtu.be/EuLip_J0x3I
#dataviz #GIS #Cartography #DataViz #Germany
December 11, 2024 at 4:36 PM
🌍 My new map shows Germany's breathtaking land cover, powered by the highly detailed ESA WorldCover dataset!
Want to see how I create these maps? Watch this video:
🎥 youtu.be/EuLip_J0x3I
#dataviz #GIS #Cartography #DataViz #Germany
Want to see how I create these maps? Watch this video:
🎥 youtu.be/EuLip_J0x3I
#dataviz #GIS #Cartography #DataViz #Germany
Reposted by Yannis Kalfas
ᴍᴀᴊᴏʀ ᴛᴏᴍ ꜰʟᴏᴀᴛɪɴɢ ɪɴ ᴛʜᴇ ʟᴀᴛᴇɴᴛ ꜱᴘᴀᴄᴇ 🧑🚀️
𝐆𝐥𝐨𝐛𝐚𝐥 𝐚𝐧𝐝 𝐃𝐞𝐧𝐬𝐞 𝐎𝐟𝐟𝐢𝐜𝐢𝐚𝐥 𝐄𝐦𝐛𝐞𝐝𝐝𝐢𝐧𝐠𝐬 𝐨𝐟 𝐌𝐚𝐣𝐨𝐫 𝐓𝐎𝐌 - 𝐎𝐩𝐞𝐧 𝐚𝐧𝐝 𝐅𝐫𝐞𝐞 𝐀𝐜𝐜𝐞𝐬𝐬 𝐨𝐧 @hf.co
Embedding Earth observation data with pre-trained AI models is starting to become a hot topic, for all the good reasons.
More info in thread 🧵
𝐆𝐥𝐨𝐛𝐚𝐥 𝐚𝐧𝐝 𝐃𝐞𝐧𝐬𝐞 𝐎𝐟𝐟𝐢𝐜𝐢𝐚𝐥 𝐄𝐦𝐛𝐞𝐝𝐝𝐢𝐧𝐠𝐬 𝐨𝐟 𝐌𝐚𝐣𝐨𝐫 𝐓𝐎𝐌 - 𝐎𝐩𝐞𝐧 𝐚𝐧𝐝 𝐅𝐫𝐞𝐞 𝐀𝐜𝐜𝐞𝐬𝐬 𝐨𝐧 @hf.co
Embedding Earth observation data with pre-trained AI models is starting to become a hot topic, for all the good reasons.
More info in thread 🧵
December 10, 2024 at 7:47 AM
ᴍᴀᴊᴏʀ ᴛᴏᴍ ꜰʟᴏᴀᴛɪɴɢ ɪɴ ᴛʜᴇ ʟᴀᴛᴇɴᴛ ꜱᴘᴀᴄᴇ 🧑🚀️
𝐆𝐥𝐨𝐛𝐚𝐥 𝐚𝐧𝐝 𝐃𝐞𝐧𝐬𝐞 𝐎𝐟𝐟𝐢𝐜𝐢𝐚𝐥 𝐄𝐦𝐛𝐞𝐝𝐝𝐢𝐧𝐠𝐬 𝐨𝐟 𝐌𝐚𝐣𝐨𝐫 𝐓𝐎𝐌 - 𝐎𝐩𝐞𝐧 𝐚𝐧𝐝 𝐅𝐫𝐞𝐞 𝐀𝐜𝐜𝐞𝐬𝐬 𝐨𝐧 @hf.co
Embedding Earth observation data with pre-trained AI models is starting to become a hot topic, for all the good reasons.
More info in thread 🧵
𝐆𝐥𝐨𝐛𝐚𝐥 𝐚𝐧𝐝 𝐃𝐞𝐧𝐬𝐞 𝐎𝐟𝐟𝐢𝐜𝐢𝐚𝐥 𝐄𝐦𝐛𝐞𝐝𝐝𝐢𝐧𝐠𝐬 𝐨𝐟 𝐌𝐚𝐣𝐨𝐫 𝐓𝐎𝐌 - 𝐎𝐩𝐞𝐧 𝐚𝐧𝐝 𝐅𝐫𝐞𝐞 𝐀𝐜𝐜𝐞𝐬𝐬 𝐨𝐧 @hf.co
Embedding Earth observation data with pre-trained AI models is starting to become a hot topic, for all the good reasons.
More info in thread 🧵
Reposted by Yannis Kalfas
For folks that don't use conda, how do you debug problems that only occur on some particular Python or CUDA version?
With conda, I have a few envs set up with various python and CUDA versions, so I just activate one and get to work.
With something like uv, what do you do?
With conda, I have a few envs set up with various python and CUDA versions, so I just activate one and get to work.
With something like uv, what do you do?
December 4, 2024 at 2:45 AM
For folks that don't use conda, how do you debug problems that only occur on some particular Python or CUDA version?
With conda, I have a few envs set up with various python and CUDA versions, so I just activate one and get to work.
With something like uv, what do you do?
With conda, I have a few envs set up with various python and CUDA versions, so I just activate one and get to work.
With something like uv, what do you do?
This well-written blog post makes it very easy to explain what I'm working and why. We improve upon the Worldcover maps in the LCFM project at Vito, a Copernicus initiative developing high-resolution 10m global land cover maps for 2020-2026 🛰️🌎
www.landcarbonlab.org/insights/glo...
www.landcarbonlab.org/insights/glo...
Global Land Cover Maps’ Accuracy and Applications Explained
This article examines the accuracies for four global land cover data sets and explains their strengths and weaknesses and how best to apply them.
www.landcarbonlab.org
November 29, 2024 at 6:19 AM
This well-written blog post makes it very easy to explain what I'm working and why. We improve upon the Worldcover maps in the LCFM project at Vito, a Copernicus initiative developing high-resolution 10m global land cover maps for 2020-2026 🛰️🌎
www.landcarbonlab.org/insights/glo...
www.landcarbonlab.org/insights/glo...