Radar reflectivity data snapshot fine-grained atmospheric variations that cannot be represented well by numerical weather prediction models or satellites, which poses a limit for nowcasts based on model–data fusion techniques. Here, we reveal a multiscale representation (MSR) of the atmosphere by reconstructing the radar echoes from the Weather Research and Forecasting (WRF) model simulations and the Himawari-8 satellite products using U-Net deep networks. Our reconstructions generated the echoes well in terms of patterns, locations, and intensities with a root mean square error (RMSE) of 5.38 dBZ. We find stratified features in this MSR, with small-scale patterns such as echo intensities sensitive to the WRF-simulated dynamic and thermodynamic variables and with larger-scale information about shapes and locations mainly captured from satellite images. Such MSRs with physical interpretations may inspire innovative model–data fusion methods that could overcome the conventional limits of nowcasting.

The atmosphere is governed by laws of atmospheric physics and chemistry. For decades even centuries, these laws are represented by differential equations, usually solved numerically for the scale defined by model grid cells. However, this representation paradigm reaches its limits when the underlying physics or chemistry is too complex or even unknown, especially when considering the multiscale nature of the atmosphere. In our data era these laws are stored in immense datasets of various observations and numerical simulations, while artificial intelligence techniques can retrieve them from data albeit often with limited physical interpretations. Hybrid modeling [1] can thus balance the physical modeling and the data-driven modeling for a more comprehensive representation of the atmosphere. By representation learning, the multiscale features of the atmosphere can be learnt and encoded in the weights of connected neurons in the deep networks of multiple layers, as is fundamentally different from the traditional atmospheric representation using formulae and equations. Here we elaborate this new deep learning-based representation paradigm with two demonstrating cases. In the first case [2], we reveal a multiscale representation of the convective atmosphere by reconstructing the radar echoes from the Weather Research and Forecasting (WRF) model simulations and the Himawari-8 satellite products using U-Net deep networks. We then diagnose the physical interpretations of the learnt representation with a sensitivity analysis method. We find stratified features with small-scale patterns such as echo intensities sensitive to the WRF-simulated dynamic and thermodynamic variables and with larger-scale information about shapes and locations mainly captured from satellite images. Such explainable representation of the convective atmosphere would inspire innovative data assimilation methods and hybrid models that could overcome the conventional limits of nowcasting. In the second case [3], we employ deep convolutional neural networks (CNN) to represent the errors associated with fine particulate matter (PM2.5) forecasts of a chemistry-transport model (CTM), the Nested Air Quality Prediction Modeling System (NAQPMS), within 240-hour lead times across 180 monitoring sites in the Yangtze River Delta (YRD) region of China. The learnt multiscale error representation reduces the PM2.5 forecasts’ root mean square error (RMSE) by 16.3-34.2% on test cases in 2017-2018. We then probe the physical interpretation of the multiscale error representation using the deep learning important features (DeepLIFT) interpretability method. We quantify the significant contribution from sulfur dioxide (SO2, 31.3%) and ozone (29.4%), which are comparable to PM2.5 (31.1%) and about three times higher than nitrogen dioxide (8.2%). Such interpretations would suggest that improvements are needed in formulating the SO2-sensitive pollution in the ammonia-poor YRD region. We consider our representation studies as a step towards more comprehensive atmospheric hybrid models that take advantage of the mighty artificial intelligence technologies but are at the same time physically explainable.[1] Liao, Q., Zhu, M., Wu, L. et al. 2020. https://doi.org/10.1007/s40726-020-00159-z[2] Zhu, M., Liao, Q., Wu, L. et al. 2023. https://doi.org/10.3390/rs15143466[3] Zhu, M., Liao, Q., Wu, L. et al. 2024. In submission.

Medium-term regional forecasts of fine particulate matter (PM_2.5) are critical to mitigate its hazardous environmental effects. Deep learning (DL) has recently emerged as a promising technique to improve traditional forecasts based on chemistry-transport models (CTMs). However, challenges remain in (1) representing the complex physics of the PM_2.5 pollution processes including the emission, transport, and chemical conversion of the pollutants, and (2) probing the interpretability of DL models for trustworthy forecasts and mitigations. Here we learn U-Net deep networks from a reanalysis dataset—the data fusion information by blending available pollutant observations and CTM simulations in China—to represent the best available physics of PM_2.5 pollution in this region, and then assess this representation using interpretability methods. This hybridization between CTM and U-Net results in robust improvements by over 30% against the baseline CTM forecasts in root mean square error over 1391 stations of the China National Environmental Monitoring Centre network. Interpretability analysis demonstrates physical consistency, manifesting the roles of wind fields and precursor pollutants in modulating PM_2.5 concentration variations through dispersion and chemical processes. A decline in the emission importance from 2013 to 2020 suggests the potential of the deep networks to capture the evolving PM_2.5 pollution processes and growing emission uncertainties under China’s continuous significant emission reduction in this period. Such a combination of learning of physics and interpretability analysis is valuable for PM_2.5 forecasts that are accurate yet trustworthy for effective pollution mitigation.