Jatan Buch
@JBuch7
Followers
568
Following
16K
Media
249
Statuses
3K
Researcher studying clouds, weather, and wildfires with machine learning and physics
new york city
Joined March 2013
We are actively reviewing applications for a Performance Engineer role at Aeolus https://t.co/CyxMLM5IV9 I'll be at NeurIPS later in the week, so DM me if you're interested in chatting more about the role or our approach to modern differentiable modeling more broadly!
expensive-legume-bf0.notion.site
Location: San Francisco (on-site only)
0
0
0
Clearing out my old office and finding transparencies from my talks from 25 years ago...
14
31
429
🚀 I am hiring a Founding Engineer for Inkboard Inc. 🚀👀 I’m confident there is no other role on the market like this - and for the right person, I believe this could be a career defining opportunity. General Requirements: - Good vibes. 😁 - You're a prolific generalist
5
9
39
Recently published! Simulating the air quality impact of prescribed fires using graph neural network-based PM2.5 forecasts 👉 https://t.co/a92xQhKWk4 By Kyleen Liao, @JBuch7, @karadlamb and @PierreGentine
#airquality #wildfires #forecasting
0
2
3
recht’s post might qualify for the “not even wrong” distinction, the more I think about it
0
0
1
Usually a good read, Ben Recht absolutely wrote a sophomore-level philosophy analysis of the Higgs discovery, and I’m glad Kyle (who was on the CMS team) corrected the record here!
My first draft of a reaction to @beenwrekt's blog post on the Higgs discovery. TL;DR, yes it did. https://t.co/EMLVB2Dby4
1
0
5
I've thought of this as one of the stronger consistency tests for AI NWP models -- glad there's a detailed study out now!
Can #AI-based weather forecasting models (trained on present-day data) provide skillful forecasts also in different colder and warmer states of the climate system? Our new preprint @arxiv explores this question: https://t.co/zIYPxk97SF Here is what we found so far (🧵1/6)
0
2
7
Are there any examples of differentiable modeling being applied to ecological systems, or discrete-state systems, more broadly?
0
0
1
Horrifying but not that surprising if you have spent any time in New Delhi.
Inspired by @ed_hawkins, @AtmosScience have developed a tool to generate air quality stripes. Here is the plot for Delhi, India. It shows clearly how we have made bad policy decisions when it comes to air quality in India https://t.co/gmZGVmt9ho
0
1
5
Update your Zoteros/bibliography managers ppl: there’s a new ERA5 reference in town!
0
0
8
I've tried my best to tune this out, but the response of sea ice to anomalous sea surface and atmospheric temperature perturbations over the past two years is a scary phenomenon.
Due to anomalous conditions ongoing in the #Arctic and #Antarctic, the total extent of sea ice globally is a record low for the date and nearly *4 million square kilometers* below the 1981-2010 average... More graphics: https://t.co/ecHYax1cql
0
3
9
I should add that the spatial and temporal scales for these forecasts unfortunately do not quite capture the dynamics of rapidly growing wildfires such as the #ParkFire : https://t.co/hfbA34Ttdw
The #ParkFire has grown to a mind-boggling 348,370 acres in less than 72 hours, becoming one of the fastest-growing wildfires in California history. The fire has been spreading at an average pace of over 60 football fields per minute since it began.
0
0
0
4. Last, these forecasts are experimental in nature and should not be used in any decision-making.
1
0
0
area appears to be at the upper end of ensemble spread for the mean size-high frequency (gpd-high-2sig) case and is best described by large size-high frequency (gpd-ext-high-2sig) scenario. The red shaded area is the variance of the ensemble mean whereas gray lines are means of
1
0
0
3. Focusing on the Middle Rockies, the ecoregion that includes the Blue Mountains which is the site of 4 large Oregon wildfires that have burned ~3000 km^2, we see that the models forecast fire frequency to be significantly lower than the climatological mean, while the burned...
1
0
0
2. The model forecasts fire probability in 12km x 12km grid cells, which can then be used to calculate the fire frequency for aggregate ecoregions. A separate model takes the stochastic ignitions, generating sizes of individual fires and, in turn, the burned area.
1
0
0
SEAS5 ensemble runs initialized in June. The ensemble mean predicted fire prob is shown on the left panel, whereas the ensemble uncertainty, derived using the standard deviation of member forecasts, is shown on the right. Both are normalized by the mean fire prob. from 2001-2020.
1
0
0
A few notes: 1. The model forecasts have been generated using the SMLFire1.0 model (see https://t.co/MSGhzcMcOt for more details) run in forecast mode, i.e with observations and reanalysis data for various fire-related predictors until May and climate model outputs of monthly
As we head into Fall, I wanted to share results from my paper (w/ stellar colleagues @peedublya, @caro_in_space, Winslow Hansen, and @PierreGentine) https://t.co/LNnB9A1UoU introducing a stochastic machine learning model for wildfires in the western US (WUS). A mini-🧵:
1
0
1
August 2024 appears to follow a similar pattern, with elevated fire danger in coastal Central/Southern California -- not a good sign for a region that saw several fires in July, notably the Lake Fire in Los Padres National Forest https://t.co/8LoVO36Gfw .
1
0
1
predicted an above baseline fire probability (warmer colors > 1) for large areas in the Middle Rockies, Blue Mts., Great Basin, as well as sections of the Sierra Nevada ( https://t.co/CnAgR8jn2h) with ~68% confidence across the 51 member ECMWF SEAS5 ensemble initialized in June.
app.watchduty.org
Real-time information about wildfire and firefighting efforts nearby
1
0
1