Although the troposphere is often thought of as the layer of the atmosphere closest to Earth's surface, the lowest layer of the troposphere, the planetary boundary layer (PBL), is actually the part that has the greatest influence on near-surface weather. . The 2018 Planetary Science Decadal Review identified PBL as an important scientific problem with the potential to enhance storm predictions and improve climate predictions.
“The PBL is where the Earth's surface interacts with the atmosphere, including the exchange of moisture and heat that causes severe weather and climate change,” said Adam Milstein, a technical staff member in Lincoln Laboratory's Applied Space Systems Group. “The PBL is also where humans live, and the turbulent movement of aerosols across the PBL is important for air quality that impacts human health.”
Although essential for studying weather and climate, important features of PBL, such as height, are difficult to resolve with current technology. For the past four years, Lincoln Laboratory staff have been studying PBL by focusing on two different tasks. One is to use machine learning to create his 3D scan profiles of the atmosphere, and the other is to more clearly resolve the vertical structure of the atmosphere to better predict droughts. .
This PBL-focused research effort builds on more than a decade of related research on fast, operational neural network algorithms developed by Lincoln Laboratory for NASA missions. These missions include the Time-Resolved Observation of Precipitation Structure and Storm Intensity (TROPICS) mission using small satellite constellations, and the Aqua satellite, which collects data on the Earth's water cycle and observes variables such as ocean temperature, precipitation, and water vapor. included. In the atmosphere. These algorithms obtain temperature and humidity from satellite instrument data and have been shown to significantly improve observation accuracy and usable global coverage compared to previous approaches. In the case of TROPICS, this algorithm helps obtain data used to characterize the rapidly evolving structure of storms in near real time, and in the case of Aqua, it helps improve predictive models, drought monitoring, and fire forecasting. Masu.
These operating algorithms for TROPICS and Aqua are based on classic “shallow” neural networks to maximize speed and simplicity, and are based on a one-dimensional Create a vertical profile. While this approach has improved observations of the entire atmosphere all the way to the surface, including the PBL, lab staff believe that to improve the details of the PBL, they need a newer “deep layer” approach that treats the atmosphere in the region of interest as a three-dimensional image. ” We decided that learning techniques were necessary. Even further away.
“We hypothesized that deep learning and artificial intelligence (AI) techniques could improve current approaches by incorporating better statistical representations of atmospheric 3D temperature and humidity images into the solution. ,” says Milstein. “But it took us time to figure out how to create optimal datasets that combine real and simulated data. We needed to be prepared to train these techniques.”
In a recent NASA-funded effort, the research team collaborated with Joseph Santanello of NASA Goddard Space Flight Center and William Blackwell, also of the Applied Space Systems Group, to demonstrate that these search algorithms We showed that the details of PBL can be improved, including a more accurate determination of . than previous state-of-the-art technology.
Improving our knowledge of PBL can broadly help improve our understanding of climate and weather, but one important application is drought prediction. According to the Global Drought Snapshot report released last year, drought is a pressing global problem that needs to be addressed by the international community. Lack of moisture near the surface, especially at his PBL level, is the main indicator of drought. Previous studies using remote sensing techniques have looked at soil moisture to determine drought risk, but studying the atmosphere can help predict when droughts will occur.
In an effort funded by Lincoln Laboratory's Climate Change Initiative, Milstein, along with lab staff member Michael Pieper, collaborated with scientists at NASA's Jet Propulsion Laboratory (JPL) to develop neural networks. Using technology to improve drought forecasts in the continental United States. This research builds on existing operational work done by JPL and (partially) incorporates the lab's operational “shallow” neural network approach for Aqua, but the team We believe that further improvements can be made by combining deep learning research work focused on PBL. Accuracy of drought prediction.
“Lincoln Laboratory has worked with NASA for more than a decade on neural network algorithms to estimate atmospheric temperature and humidity from spaceborne infrared and microwave instruments, including those aboard the Aqua spacecraft. “We've done that,” says Milstein. “To date, we have learned a lot about this issue by working with the scientific community, including learning what scientific challenges remain. Our years of experience working with this type of remote sensing and our experience using neural network technology gave us a unique perspective.”
Milstein said the next step for the project is to integrate deep learning results from the National Oceanic and Atmospheric Administration, NASA, and the Department of Energy, which were collected directly at the PBL using radiosondes, a type of weather instrument. to compare with the data set. balloon. “These direct measurements can be thought of as a kind of 'ground truth' for quantifying the accuracy of the techniques we have developed,” says Milstein.
This improved neural network approach demonstrates drought prediction beyond the capabilities of existing metrics and could become a tool scientists can rely on for decades to come, Milstein said.
