A new federal report describes a hidden economy of science and engineering jobs in the U.S. — and reveals persistent inequities within it.

The big picture: There's been heavy investment — in the U.S. and other countries — in developing a STEM (science, technology, engineering and mathematics) workforce to fuel economies, improve quality of life and protect national security.

• For workers, it's a potential opportunity: The STEM workforce is projected to grow faster than all occupations in the next decade.
• But access to the typically higher-paying STEM jobs — and the agenda-setting power they can wield in boardrooms — has been limited for Black, Hispanic and other workers of color and for people with disabilities.

What's new: The National Science Foundation's biennial "Diversity and STEM" report released this week expands the definition of a STEM job.

• Previous analyses limited it to "white lab coat" jobs — engineers and scientists in the fields of math, computer science, biology, physics and social scientists — that typically require at least a bachelor's degree and earn an average of $90,000 per year.
• But the new report includes science- and engineering-related (S&E) jobs that draw on STEM skills and expertise. These are health care workers, technologists, technicians, K-12 science teachers and others whose typical salary is $67,000 per year.
• It also includes middle-skill STEM occupations — construction and maintenance jobs in areas like solar infrastructure and scientific instrumentation that require scientific or technical knowledge but not typically a bachelor's degree. The average salary for these jobs is $50,000.

By the numbers: When the authors re-analyzed 2011 data using the expanded definition, STEM workers as a percentage of the total workforce increased from 21% in that year to 24% in 2021.

• The report found an increase in the number of STEM workers with a disability but the "proportion of these workers in the STEM workforce was unchanged from 2011 to 2021." Just 3% of the workforce reported having at least one disability.
• In 2011, Black workers made up 7% of the STEM workforce and Hispanic or Latino workers made up 15%. That increased to 9% and 15%, respectively, in 2021.

Zooming in: They found 44% of Black STEM workers were employed in S&E-related jobs.

• And, 63% of Hispanic or Latino STEM workers and 52% of American Indian or Alaska Native workers held middle-skill jobs.
• 46% of STEM workers with at least one disability worked middle-skill jobs compared to 38% without a disability.

What they're saying: "You can be a part of the STEM workforce on a practical level," says Amy Burke, head of the analysis team at the National Center for Science and Engineering Statistics at NSF that produced the report.

• "It doesn't necessarily come at the cost of a four-year degree, which has been very prohibitive" for a lot of people, she adds.
• Other occupations can be an entry point and can help to build a STEM base and skill set that can potentially be "a launching pad" for more, she says.

The big question: How?

• People can grow and learn in some professions, and over time earn more money. But others require a degree or accreditation to advance. Without it, "you'll never be quite eligible for some of the roles that we need Black and brown people in, like chief information officers and chief technology officers," says Janeen Uzzell, chief executive officer of the National Society of Black Engineers (NSBE).
• Higher education institutions and industry need to "create better pathways, and that's what still continues to be missing in these conversations," says Rochelle Williams, chief program officer at NSBE.
• "How do we really move the needle more than just 2% every couple of years? And we're not until this country gets real about racism," she says.

Read the entire story.

A new study relying on machine learning methods finds the climate thresholds enshrined in the Paris agreement may be coming up faster than previously anticipated, Axios' Andrew Freedman reports.

Why it matters: The world is already suffering the impacts of 1.1°C (1.98°F) to 1.2°C (2.16°F) of warming to date, and passing 1.5°C or 2°C above preindustrial levels could dramatically increase the risks to society and ecosystems.

The big picture: The study, published Monday in the Proceedings of the National Academy of Sciences, uses neural networks trained on climate model simulations to predict the time remaining until the Paris targets will be reached.

• The data-driven approach involves identifying patterns in historical temperature observations that provide clues to the time remaining until a warming level is exceeded.
• The researchers from Stanford University and Colorado State University found the world has about a decade until the 1.5-degree target is reached and then exceeded, which is consistent with previous findings.
• Notably, it finds that even the lowest emissions scenario, featuring steep cuts to fossil fuel use in the next few decades, has a significant chance of breaching the 2°C target. The timeline for hitting 2°C ranges from 2043 to 2058, with a central estimate of 2050, the study finds.

Between the lines: The study has some significant caveats, including the fact that the machine learning techniques may be biased by the computer models they were trained on.

• Study lead author Noah Diffenbaugh, a climate scientist at Stanford, tells Axios the new paper took multiple steps to try to minimize the likelihood of such biases.
• The AI-enabled study is part of a trend in climate science toward focusing on developing new data analysis techniques. Some scientists are concerned this may come at the expense of potential new discoveries about how the climate system functions.

What they're saying: "It's always a bit tricky to know how much faith to put in machine-learning methods like this given the absence of physical modeling of the systems involved," Zeke Hausfather, climate research lead at payments company Stripe, who was not involved in the new study, tells Axios via email.

• He said it offers "another reason for caution" that an emissions scenario assumed to hold warming to under 2°C may in fact "have a real chance of resulting in higher warming than we expect." 

Machine learning could aid in the search for signs of intelligent alien life in the universe, according to a study published this week, Axios' Miriam Kramer reports.

Why it matters: Searching for radio signals emitted by technologically advanced extraterrestrial societies requires scanning large portions of the sky with powerful radio telescopes, producing a lot of data scientists need to parse through.

• Machine learning tools could make that job easier.

Driving the news: A new study published in Nature Astronomy this week details the use of an algorithm that helped researchers sift through data previously analyzed in 2017.

• The machine learning tool was trained to differentiate radio signals that might originate in distant star systems from human-created sources like cell phones and satellites.
• The tool turned up eight new radio signals of interest — possible signs of intelligent extraterrestrial life — within the dataset. Those signals came from five different stars that are 30 to 90 light-years away.

But, but, but: Those signals don't mean they found aliens.

• Brief follow-up observations didn't turn up a repeat signal, and it's still possible these eight targets of interest were created by chance, not emitted by alien civilizations.
• More observations and analyses are now being conducted, according to a press release from the University of Toronto.

The big picture: Scientists may need machine learning for these types of searches in the future.

• "With our new technique, combined with the next generation of telescopes, we hope that machine learning can take us from searching hundreds of stars, to searching millions," one of the authors of the study, Peter Ma, of the University of Toronto, said in the press release.

Water is weird. A new type of ice could help us understand why (Emily Conover — Science News)

The book of leaves (Brett Foxwell, via Aeon)

The existential wonder of space (Marina Koren — The Atlantic)

A rare green comet made its closest pass by Earth this week in 50 millennia.

What's happening: Comet C/2022 E3 came within 26 million miles of Earth last night, after already passing its closest point to the Sun along an orbit that stretches to the outer edges of the solar system.

• It's possible it won't return. As the comet travels back past other planets, their gravity could change its orbit, and the comet could then escape the Sun's pull, according to the Adler Planetarium.

Details: Astronomers Bryce Bolin and Frank Masci discovered the comet last March as it passed Jupiter using the Zwicky Transient Facility telescope at the Palomar Observatory in California.

• The comet head, or coma, appears green due to its complex carbon-containing molecules that are broken down by light as the comet approaches the Sun.
• The reaction forms diatomic carbon — an unstable gas of two carbon molecules bonded together that only exists at extremely high temperatures.

What to watch: The comet — it's not too late. It should be visible in the night sky for another week.

• You can, of course, track its journey.