Although the troposphere is often thought of as the layer of the atmosphere closest to the Earth’s surface, the planetary boundary layer (PBL) – the lowest layer of the troposphere – is actually the part that most influences weather near the surface. In 2018 planetary science decade studyPBL was raised as important scientific problem which could improve storm forecasting and improve climate forecasts.
“PBL is where the surface interacts with the atmosphere, including the exchange of moisture and heat, which leads to severe weather and a changing climate,” says Adam Milstein, a technical fellow in Lincoln Laboratory’s Applied Space Systems Group. “The PBL is also a place where people live, and the turbulent movement of aerosols in the PBL is important for air quality, which affects human health.”
Although they are imperative for studying weather and climate, critical features of PBL, such as its altitude, are arduous to determine using current technology. Over the past four years, Lincoln Laboratory staff have been researching PBL, focusing on two different tasks: using machine learning to create 3D-scanned atmospheric profiles and more clearly determining the vertical structure of the atmosphere to better predict droughts.
This PBL-focused research effort builds on more than a decade of related work on quick, operational neural network algorithms developed by Lincoln Laboratory for NASA missions. These missions include Time-Resolved Observations of Precipitation structure and storm Intensity with a Constellation of Smallsat (TROPICS) and Aqua, a satellite that collects data on Earth’s water cycle and observes variables such as ocean temperature, precipitation and water vapor in the atmosphere. These algorithms retrieve temperature and humidity from satellite instrument data and have been shown to significantly improve the accuracy and useful global coverage of observations compared to previous approaches. For TROPICS, algorithms aid produce data used to characterize rapidly evolving storm structures in near real time, and for Aqua they helped improve forecast models, drought monitoring and wildfire prediction.
These operational algorithms for TROPICS and Aqua rely on classical “shallow” neural networks to maximize speed and simplicity, creating a one-dimensional vertical profile for each spectral measurement collected by the instrument at each location. While this approach improved overall observations of the atmosphere down to the surface, including PBL, lab staff determined that newer “deep” learning techniques that treat the atmosphere above an area as a three-dimensional image further down are needed to improve the details of PBL.
“We hypothesized that deep learning and artificial intelligence (AI) techniques could improve current approaches by incorporating better statistical representation of three-dimensional images of atmospheric temperature and humidity into the solutions,” says Milstein. “But it took some time to figure out how to create the best data set – a combination of real and simulated data; we had to prepare to train these techniques.”
The team worked with Joseph Santanello of NASA’s Goddard Space Flight Center and William Blackwell, also of the Applied Space Systems Group, on a recent project called A NASA-funded effort showing that these search algorithms can improve the detail of the PBL, including more precise determination of the height of the PBL than the prior art.
While improved knowledge of PBL is broadly useful for better understanding climate and weather, one key application is drought prediction. According to Global Drought Report published last year, droughts are a pressing planetary problem that the global community must address. Lack of moisture near the surface, especially at the PBL level, is a major indicator of drought. While previous studies using remote sensing techniques have already demonstrated this examined the soil moisture To determine drought risk, studying the atmosphere can aid predict when drought will occur.
As part of an effort funded by Lincoln Laboratory Climate Change InitiativeMilstein and lab staff member Michael Pieper are working with scientists at NASA’s Jet Propulsion Laboratory (JPL) to exploit neural network techniques to improve drought forecasting in the continental United States. While the work builds on existing JPL operational work, which included (in part) the lab’s operational approach to a “shallow” neural network for Aqua, the team believes this work and PBL-focused deep learning research can be combined to further improve the accuracy of drought forecasting.
“Lincoln Laboratory has been working with NASA for more than a decade on neural network algorithms to estimate atmospheric temperature and humidity from space-based infrared and microwave instruments, including those on the Aqua spacecraft,” Milstein says. “During that time, we have learned a lot about this problem by working with the scientific community, including learning what scientific challenges remain. Our long experience working on this type of remote sensing with NASA scientists, as well as our experience using neural network techniques, has given us a unique perspective.
According to Milstein, the next step for this project will be to compare the deep learning results with data sets from the National Oceanic and Atmospheric Administration, NASA and the Department of Energy collected directly at PBL using radiosondes, a type of instrument flown on a weather ship. balloon. “These direct measurements can be considered a kind of ‘ground truth’ for quantifying the accuracy of the techniques we have developed,” Milstein says.
This is an improved approach to neural networks promises to show drought forecast that could exceed the capabilities of existing metrics, Milstein says, and that will be a tool that scientists will be able to rely on for decades to come.