Lily-belle Sweet
@lilybellesweet.bsky.social
PhD student at UFZ - interested in explainable machine learning, agriculture and food security, compound climate events 🌾
This quote in particular!
May 26, 2025 at 12:24 PM
This quote in particular!
📢 Tomorrow (27th May) at 16:00 CET, join our next online seminar on knowledge-guided machine learning for crop growth modelling! We will be hearing from Jingye Han, a researcher at Wageningen University. Open to all - get in touch for the Zoom link or join the AgML mailing list.
May 26, 2025 at 11:11 AM
📢 Tomorrow (27th May) at 16:00 CET, join our next online seminar on knowledge-guided machine learning for crop growth modelling! We will be hearing from Jingye Han, a researcher at Wageningen University. Open to all - get in touch for the Zoom link or join the AgML mailing list.
This is a great bit of writing from a paper by Meyer & Pebesma which discusses this in more detail www.nature.com/articles/s41...
May 20, 2025 at 7:15 PM
This is a great bit of writing from a paper by Meyer & Pebesma which discusses this in more detail www.nature.com/articles/s41...
Predicting peaty soil presence, I got excited here because they mentioned spatial CV and cited Roberts' excellent paper! But as far as I can tell the evaluation metrics reported actually use a randomly-sampled 20% test set. So, it looks like they know it's wrong, but they're doing it anyway??
May 19, 2025 at 8:15 PM
Predicting peaty soil presence, I got excited here because they mentioned spatial CV and cited Roberts' excellent paper! But as far as I can tell the evaluation metrics reported actually use a randomly-sampled 20% test set. So, it looks like they know it's wrong, but they're doing it anyway??
And for modelling surface drainage and erosion features, they do something kind of weird where they split based on the third numerical value of the spatial index of each datapoint, 'to ensure even spatial coverage in each of these sets'. Even coverage is not what you want for a good test set, though
May 19, 2025 at 8:12 PM
And for modelling surface drainage and erosion features, they do something kind of weird where they split based on the third numerical value of the spatial index of each datapoint, 'to ensure even spatial coverage in each of these sets'. Even coverage is not what you want for a good test set, though
For upland bare peat modelling using a CNN, they use stratified random sampling to select a validation set (again, bad)
May 19, 2025 at 8:09 PM
For upland bare peat modelling using a CNN, they use stratified random sampling to select a validation set (again, bad)
For vegetation modelling they use XGBoost, and it's hard to parse the exact method but it sounds like they use either random or stratified random sampling to define their test set (either of which would be incorrect for this application)...
May 19, 2025 at 8:06 PM
For vegetation modelling they use XGBoost, and it's hard to parse the exact method but it sounds like they use either random or stratified random sampling to define their test set (either of which would be incorrect for this application)...
Another nice office for the morning - the State Library of Western Australia in Perth
December 10, 2024 at 2:09 AM
Another nice office for the morning - the State Library of Western Australia in Perth
Another day, another office...the Edge at State Library Queensland. Great view, lovely coffee!
November 27, 2024 at 8:21 AM
Another day, another office...the Edge at State Library Queensland. Great view, lovely coffee!
Saying goodbye to Gold Coast & its beautiful orb-weaver spiders - I'll be in Brisbane until Saturday (visiting CSIRO on Friday & hopefully giving a short talk). Get in touch if you're in the area!
November 26, 2024 at 8:09 AM
Saying goodbye to Gold Coast & its beautiful orb-weaver spiders - I'll be in Brisbane until Saturday (visiting CSIRO on Friday & hopefully giving a short talk). Get in touch if you're in the area!
Welcome to my office (for today)...
November 19, 2024 at 3:40 AM
Welcome to my office (for today)...
Pretty happy with my office today! this is the State Library in Melbourne, where I did a couple of hours' work before visiting a local colony of fairy penguins. I'm in Australia for a few weeks, reach out if you're also down under :)
November 17, 2024 at 2:33 PM
Pretty happy with my office today! this is the State Library in Melbourne, where I did a couple of hours' work before visiting a local colony of fairy penguins. I'm in Australia for a few weeks, reach out if you're also down under :)
We're halfway through our Kaggle competition runtime, with 148 submissions so far!
Despite using the same training data (1980-2020), submitted data-driven yield projections vary widely (median and interquartile range plotted).
Try it for yourself:
www.kaggle.com/competitions...
Despite using the same training data (1980-2020), submitted data-driven yield projections vary widely (median and interquartile range plotted).
Try it for yourself:
www.kaggle.com/competitions...
July 25, 2024 at 2:14 PM
We're halfway through our Kaggle competition runtime, with 148 submissions so far!
Despite using the same training data (1980-2020), submitted data-driven yield projections vary widely (median and interquartile range plotted).
Try it for yourself:
www.kaggle.com/competitions...
Despite using the same training data (1980-2020), submitted data-driven yield projections vary widely (median and interquartile range plotted).
Try it for yourself:
www.kaggle.com/competitions...
For wheat, the median estimate is zero productivity change, but some models predict large positive increases and a couple expect enormous drops. Luckily, as this experiment is based on simulated data, we have ground truth to compare to at the end of the competition. Join the competition now!
July 2, 2024 at 9:54 AM
For wheat, the median estimate is zero productivity change, but some models predict large positive increases and a couple expect enormous drops. Luckily, as this experiment is based on simulated data, we have ground truth to compare to at the end of the competition. Join the competition now!
Can you learn from the past to predict future climate impacts? Our Kaggle competition has been running for 3 weeks, with 45 submissions so far. Despite using the same training data, models vary in their projected changes in global maize productivity... www.kaggle.com/competitions...
July 2, 2024 at 9:52 AM
Can you learn from the past to predict future climate impacts? Our Kaggle competition has been running for 3 weeks, with 45 submissions so far. Despite using the same training data, models vary in their projected changes in global maize productivity... www.kaggle.com/competitions...
Sounds easy? In a practice run, using a different crop model, here's how six different ML models predicted the global annual maize prediction (black line shows the true value). Which models do you think each line corresponds to? We were surprised by the results!
June 17, 2024 at 9:54 AM
Sounds easy? In a practice run, using a different crop model, here's how six different ML models predicted the global annual maize prediction (black line shows the true value). Which models do you think each line corresponds to? We were surprised by the results!
Can we learn from the recent past to predict future climate change impacts using machine learning? We have created a new benchmark dataset designed to help answer this question, and you can take part in the challenge: www.kaggle.com/competitions...
June 17, 2024 at 9:53 AM
Can we learn from the recent past to predict future climate change impacts using machine learning? We have created a new benchmark dataset designed to help answer this question, and you can take part in the challenge: www.kaggle.com/competitions...
Ioannis Athanasiadis kicking off the first AgMIP Machine Learning (AgML) workshop in Wageningen! So excited to meet everyone in person and do some cool work together 🌾🤖
January 22, 2024 at 9:20 AM
Ioannis Athanasiadis kicking off the first AgMIP Machine Learning (AgML) workshop in Wageningen! So excited to meet everyone in person and do some cool work together 🌾🤖
Join us in Wageningen in January for the first AgMIP Machine Learning (AgML) workshop! www.wur.nl/en/research-...
December 8, 2023 at 10:00 AM
Join us in Wageningen in January for the first AgMIP Machine Learning (AgML) workshop! www.wur.nl/en/research-...
INRAE knows how to put on a workshop lunch - top marks!
November 21, 2023 at 4:47 PM
INRAE knows how to put on a workshop lunch - top marks!