Intelligent Earth CDT Project
Harmonising tropical weather feature identification for evaluation of dynamical and AI-based models
In tropical and subtropical meteorology, current knowledge is largely based on the identification of different, yet interconnected, modes of variability that modulate synoptic-scale weather patterns. These features are often closely coupled with moist convection, which dominates the energetics of the tropical atmosphere and manifests as deep, towering cumulonimbus clouds which can organise into large thunderstorms. Understanding these features is crucial for forecasting impactful weather extremes, as well as predicting how their frequency and intensity might change under climate forcing.
Unlike in the midlatitudes, where quasi-geostrophic theory can mathematically describe most frequent weather regimes with reasonably good accuracy, there is no unifying theory that connects all identifiable tropical modes of variability. Techniques for identifying tropical and subtropical weather features are often designed to locate one specific type of feature: for example, equatorial waves (Kiladis et al., 2009; Yang et al., 2021), the intertropical convergence zone (Bain et al., 2011), mesoscale convective systems (Feng et al., 2023), and cloud bands (Hart et al., 2012). Even among these, there are discrepancies in outputs from different tracking methods and application to different datasets (e.g. for equatorial waves, see Knippertz et al., 2022; for mesoscale convective systems, see Feng et al., 2024). Often, they are designed with specific application to one region (e.g. African Easterly Waves over Africa; Bain et al., 2014; Lawton et al., 2022). Many of these weather features appear to be connected, but there is a gap in fundamental science to link them through an underlying framework.
Until now, our understanding of the features has been further limited by our inability to represent interactions between processes at different scales – from kilometre-scale convection up to mesoscale (100-500km) and synoptic-scale (~1000km) weather patterns. However, the very latest generation of models now produce global climate simulations with grid spacing on the order of kilometres, allowing for deep moist convection to be explicitly represented over extended simulation timescales. These “convective-scale” models allow for a thorough interrogation of the dynamics and moist thermodynamics of tropical features. Meanwhile, various organisations are developing AI-based weather and climate models which perform remarkably well on some metrics – but sometimes fail to capture fundamental properties of the atmosphere.
The aim of this project is to utilise a unique dataset of human-labelled tropical features (produced by expert tropical weather forecasters at the Met Office - example here) to generate a scale-aware algorithm using machine learning techniques which can identify a range of tropical features in various datasets, including reanalysis, convective-scale simulations, and AI weather predictions. Such an algorithm could have vast practical applications to provide a novel and valuable approach for forecast and model evaluation. Crucially, applying machine learning techniques to these datasets could also lead to fundamental new discoveries about the nature of tropical weather features.
There is opportunity in harmonising tropical weather identification across regions and scales, through fusing conventional assessment of dynamical features with established and novel machine learning approaches. Harmonised tropical weather features identification could:
Trained meteorologists are experts at such harmonised feature identification and have produced a rich archive of Operational Meteorological guidance: in particular, this project aims to utilise the daily Tropical Analysis reports created by the Expert Weather Hub in the UK Met Office. This archive will be leveraged through modern image recognition techniques, including rapidly maturing transformer-based deep-learning video analysis and creation approaches, to map existing human-labelled weather features to diagnostics in reanalysis data. This unique Met Office human-labelled archive is rich in information and is yet to be used in peer-review literature. It has potential to be a key differentiator in evaluating model behaviour and perhaps even training AI models.
The project is CASE-sponsored by the Met Office and benefits from expert Met Office supervisors, collaborators in tropical forecasting, and access to datasets from the Met Office including high-resolution climate model outputs from the K-SCALE project, and new AI-based model outputs from upcoming Met Office project AI4NWP.
The project will also align with multiple international collaborations, such as the DYAMOND3 initiative (Phase 1 and Extension) which aims to produce a new generation of global climate models at kilometre-scale resolution in conjunction with meteorological centres across the world. Another project that aligns well with this project is the new ‘21st Century Weather’ initiative of the ARC Centre of Excellence in Melbourne, Australia.
Adames, Á.F. and Maloney, E.D, 2021. Moisture Mode Theory’s Contribution to Advances in our Understanding of the Madden-Julian Oscillation and Other Tropical Disturbances. Current Climate Change Reports 7, 72–85.DOI: 10.1007/s40641-021-00172-4
Bain, C.L., De Paz, J., Kramer, J., Magnusdottir, G., Smyth, P., Stern, H. and Wang, C.C., 2011. Detecting the ITCZ in instantaneous satellite data using spatiotemporal statistical modeling: ITCZ climatology in the east Pacific. Journal of Climate, 24(1), pp. 216-230.DOI: 10.1175/2010JCLI3716.
Bain, C.L., Williams, K.D., Milton, S.F. and Heming, J.T., 2014. Objective tracking of African easterly waves in Met Office models. Quarterly Journal of the Royal Meteorological Society, 140(678), pp. 47-57.DOI: 10.1002/qj.2210
Feng, Z., Leung, L.R., Hardin, J., Terai, C.R., Song, F. and Caldwell, P., 2023. Mesoscale convective systems in DYAMOND global convection‐permitting simulations. Geophysical Research Letters, 50(4), p.e2022GL102603.DOI: 10.1029/2022GL102603
Feng, Z., Prein, A.F., Kukulies, J., Fiolleau, T., Jones, W.K., Maybee, B., Moon, Z., Ocasio, K.M.N., Dong, W., Molina, M.J. Albright, M.G., Feng, R., Song, J., Song, F., Leung, L.R., Varble, A., Klein, C., and Roca, R., 2024. Mesoscale Convective Systems tracking Method Intercomparison (MCSMIP): Application to DYAMOND Global km-scale Simulations. Preprint.
Gill, A.E., 1980. Some simple solutions for heat-induced tropical circulation. Quarterly Journal of the Royal Meteorological Society, 106, pp. 447-462.DOI: 10.1002/qj.49710644905
Hart, N.C., Reason, C.J. and Fauchereau, N., 2012. Building a tropical–extratropical cloud band metbot. Monthly Weather Review, 140(12), pp.4005-4016.DOI: 10.1175/MWR-D-12-00127.1
Kiladis, G. N., Wheeler, M. C., Haertel, P. T., Straub, K. H., and Roundy, P. E., 2009. Convectively coupled equatorial waves, Reviews of Geophysics., 47.DOI: 10.1029/2008RG000266
Knippertz, P., Gehne, M., Kiladis, G.N., Kikuchi, K., Rasheeda Satheesh, A., Roundy, P.E., Yang, G.-Y., Žagar, N., Dias, J., Fink, A. H., Methven, J., Schlueter, A., Sielmann, F., and Wheeler, M.C., 2022. The intricacies of identifying equatorial waves. Quarterly Journal of the Royal Meteorological Society, 148(747), pp. 2814–2852.DOI: 10.1002/qj.4338
Lawton, Q.A., Majumdar, S.J., Dotterer, K., Thorncroft, C. and Schreck III, C.J., 2022. The influence of convectively coupled Kelvin waves on African easterly waves in a wave-following framework. Monthly weather review, 150(8), pp. 2055-2072.DOI: 10.1175/MWR-D-21-0321.1
Matsuno, T, 1966. Quasi-Geostrophic Motions in the Equatorial Area. Journal of the Meteorological Society of Japan. 44(1) pp. 25–43.DOI: 10.2151/jmsj1965.44.1_25
Yang, G.Y., Ferrett, S., Woolnough, S., Methven, J. and Holloway, C., 2021. Real-time identification of equatorial waves and evaluation of waves in global forecasts. Weather and Forecasting, 36(1), pp. 171-193.DOI: 10.1175/WAF-D-20-0144.1
Zhang, C., 2005. Madden-Julian Oscillation, Reviews of Geophysics, 43.DOI: 10.1029/2004RG000158
