AI in Weather and Climate Science PDF

AI in Weather and Climate Science Week 1: Fundamental principles in Earth System Science Lecture 5: Modern weather and climate models UK Met Office Playlist on weather and climate https://www.youtube.com/playlist?list=PLDpNAhwWNIinkC dT27ETKMCV7Mu4Xpr_Z What is global circulation?...

AI in Weather and Climate Science Week 1: Fundamental principles in Earth System Science Lecture 5: Modern weather and climate models UK Met Office Playlist on weather and climate https://www.youtube.com/playlist?list=PLDpNAhwWNIinkC dT27ETKMCV7Mu4Xpr_Z What is global circulation? | Part One | Differential heating: https://www.youtube.com/watch?v=7fd03fBRsuU What is global circulation? | Part Two | The Three Cells https://youtu.be/xqM83_og1Fc?si=hzQxWPLv3Y96ahZT What is global circulation? | Part Three | Coriolos https://youtu.be/PDEcAxfSYaI?si=9cCbC0l6-JMa1RMJ 2/24 Spatiotemporal continuum of processes Processes that need modelling vary across large range of space (spatio) and time (temporal) scales. Specific models can be developed to focus on specific spatiotemporal scales. Global weather and climate models somehow have to attempt to cover all ranges!!?? https://www.youtube.com/watch?v=tBSlQfgpXfw 3/25 Climate and climate change Boundary forcing: Solar energy (approx. constant) Atmospheric Composition Greenhouse Gases increasing Aerosol emmissions fluctuating Boundary forcing changes the probabilities of the weather Forcing of monsoons related to aerosols example 4/25 Climate and climate change Boundary forcing changes the probabilities of the weather Forcing of monsoons related to aerosols example https://doi.org/10.1175/JCLI-D-21- 0412.1 Sahel in West Africa saw large decline in rainfall through 70s Modern research has shown this was due to European and N. American aerosol pollution Once Clean Air Acts were passed, rainfall recovered. 5/25  Weather Prediction What is the actual likelihood of a weather event This will fall within the “envelope” for weather that is set by the climate. Climate modelling is about accurate representation of this envelope Weather prediction is about precisely determining what will happen on a given hour or day or week. 6/25 Timescales of Predictability 7 Analogy for model to predict weather No because non-linear dynamical systems can have an “infinite sensitivity to initial conditions” Mathematics of this system here: https://arxiv.org/pdf/1910.12610 8 So how to construct a model? 9/25 Primitive Equations in derivatives Navier-Stokes equations Thermodynamic Energy Equation Continuity Equation Ideal Gas Law Moisture conservation u, v, ω, T, α, Φ, and q Newton: equations of motion Cellular numerical model of the weather: Richardson’s concept Navier – Stokes: Newton for fluids: quantifies how and why fluids (winds and oceans) move Lewis Fry Richardson: first conceived of models: How to turn Navier-Stokes into spatial calculations Climate models Earth system models  Coupled models Atmospheric General Circulation Model (AGCM) Ocean General Circulation Model (OGCM) Land Topography (m) in UK Met Office HadCM3 model There is a price to be paid for dividing the world into boxes: the loss of detail Height section along 47 N (through the Alps) HADLEY CENTRE CLIMATE MODELS (degrees) (degrees) Progression in Horizontal Resolution AR5 (2010-14): 70km max. horizontal resolution; ~90/ 60 layers in the atmosphere/ McGuffie ocean and Henderson- Sellers,200 AR6 even greater: 5. Climate Modelling data output = Primer. Wiley. 20-40 petabytes! (see Eyring et al, 2016) 17 Spatiotemporal continuum of processes Primitive equations are always truncated, especially in space. What to do…? 18/25  Need for Parametrizations Clouds process of formation Temperature, humidity dust, ice crystals How much cloud, how much rain? collective effect of clouds Albedo (relectivity) Transmission of different spectral bands Cloud water content Aerosol scavenging Vegetation albedo Precipitation Convection, Diffusion ‘Buoyancy wave’ effects INCLUSION OF EFFECTS RATHER THAN PROCESSES  Land and ocean in Cloud in white black and grey Discrete algorithmic approaches to ‘continuous’ processes: Cnceptual representation of convective triggering in the Met Office model New Model: Convection permitting Africa (CP4-Africa) simulations N512 global 25x39km (768 x 1024 x 85) CP4Africa 4.5km (2000 x 2100 x 80) 10 year simulations: present-day (1997/03-2007/02) future RCP8.5 (2097/03-2107/02) Segolene Berthou, Rachel Stratton, Simon Tucker, Sonja Folwell First Rains Project Simulations 2.2km Met Office Unified Model Testing of configurations for use in 3-month long prediction experiments (seasonal prediction) This animation shows 3 days of outgoing longwave radiation (a proxy for deep convective cloud) 23 Summary Weather and climate models were developed to focus on quite different problems: weather prediction an initial condition problem, climate simulation a sensitivity to changing boundary conditions (atmospheric composition firstly). However, these models share the same fundamental primitive equation sets which include Navier-Stokes on a rotation sphere and thermodynamic equations. Both weather and climate models have progressed with the frontier of increase in computing power. However truncation of the equation set, especially in space, means need for parametrisations is always present. Recent breakthroughs in model resolution and configurations have enabled explicit rather than parametrised deep convection to be used in weather and climate simulations. Critically important advance for simulating tropical weather. 24/25  AI in Weather and Climate Science Week 1: Fundamental principles in Earth System Science Lecture 4: The physics of weather Navier-Stokes Equation on a rotating sphere  Navier-Stokes Equation is tricky One of 6 unsolved millennium prize problems https://www.quantamagazine.org/what-makes-the-hardest-equations-in-physics-so-difficult-20180116/ 2/17  Navier-Stokes Equation is tricky One of 6 unsolved millennium prize problems Since understanding the Navier–Stokes equations is considered to be the first step to understanding the elusive phenomenon of turbulence, the Clay Mathematics Institute in May 2000 made this problem one of its seven Millennium Prize problems in mathematics. It offered a US$1,000,000 prize to the first person providing a solution for a specific statement of the problem: Prove or give a counter-example of the following statement: In three space dimensions and time, given an initial velocity field, there exists a vector velocity and a scalar pressure field, which are both smooth and globally defined, that solve the Navier–Stokes equations. 3/17 And it is beautiful…. 4/17  So lets unpick what this says: Flow velocity change: acceleration and advection 5/17  So lets unpick what this says: Pressure gradient force 6/17  So lets unpick what this says: Friction 7/17 Flow in a pipe? 8/17 So what of flow in the atmosphere or ocean? Earth is rotating, therefore all mass on Earth has angular momentum This angular momentum must be accounted for in geophysical fluids. So we end up with Navier-Stokes in a rotating frame of reference 9/17 The Coriolos Effect (a pseudo force) For full derivation see geophysical fluid dynamics textbook such as Vallis’s 🡪 Or see Geophysical Fluid Dynamics course taught by Grae Worster (if it is offered this year?) 10/17 Scale Analysis of the equation 11/17 Geostrophic Approximation 12/17  Geostrophic Approximation In the ocean this balance between pressure gradient and Coriolos effect holds even more strongly due to greater density of water https://www.pbslearningmedia.org/resource/buac17-912-sci-ess-perpocean/perpetual-ocean/ 13/17 Quasi-geostrophic approximation This is the starting point for much of the linearization required to pursue analytical solutions to the equation 14/17  Barotropic Rossby Wave Equation First numerical weather forecast was attempted using a simulation of this equation Run on ENIAC computer If interested to read more on this history: Platzman’s description is great: https://www.jstor.org/stable/26218664 https://www.sciencedirect.com/science/article/ abs/pii/S0065268708601703 http://camp.cos.gmu.edu/CLIM-715/NWP-Hist ory.pdf 15/17 Primitive Equations in derivatives Navier-Stokes equations Thermodynamic Energy Equation Continuity Equation Ideal Gas Law Moisture conservation u, v, ω, T, α, Φ, and q The Primitive Equations are discretized Treatment of this grid really matters for forecast accuracy: https://www.ecmwf.int/sites/default/files/elibrary/2016/17262-new-grid-ifs.pdf 17 The Primitive Equations are discretized… also in time Mengaldo, G., Wyszogrodzki, A., Diamantakis, M. et al. Current and Emerging Time-Integration Strategies in Global Numerical Weather and Climate Prediction. Arch Computat Methods Eng 26, 663–684 (2019). https://doi.org/10.1007/s11831-018-9261-8 18 So bigger computer, finer grids, smaller timesteps = better forecasts? No because non-linear dynamical systems can have an “infinite sensitivity to initial conditions” And as you learnt in lecture 2 we already have observational uncertainty in what the atmosphere looks like at a given time 19 A non-linear (dynamical) system Initial condition ensemble forecasts are necessary https://charts.ecmwf.int/products/cyclone?base_time=202501090000 &product=tc_strike_probability&unique_id=05S_DIKELEDI_2025 21 The Quiet revolution in NWP (numerical weather prediction) Homework: Bauer, P., Thorpe, A. & Brunet, G. The quiet revolution of numerical weather prediction. Nature 525, 47–55 (2015). https://doi.org/10.1038/nature14956 Read by Monday. Additional – read by end of Sunday 26 Jan. Highly valuable in advance of Shruti’s lecture on Monday 27 Jan. Bauer, P. (2024). What if? Numerical weather prediction at the crossroads. Journal of the European Meteorological Society, 1, 100002. https://doi.org/10.1016/j.jemets.2024.100002 22 Timescales of Predictability 23 Summary Fundamental physical equations describing the spatial and temporal evolution of the atmosphere and ocean are known. Weather forecasts encode discretized (space and time) versions of these equations. Forecast starts with best estimate of current state of the atmosphere Steps forward in time until forecast horizon is reached (10-15 days) Operational weather prediction means speed of simulation is critically important hours of simulation = weeks of forecast Ensemble prediction is cornerstone of managing mathematical uncertainty inherent in dynamical systems. 24  AI in Weather and Climate Science Week 1: Fundamental principles in Earth System Science Lecture 3: Climate and climate change What controls Earth’s energy budget? Core learning goals Week 1 1. Overview of fundamental weather and climate concepts i. Observations available ii. Basics of Earth’s Radiation budget i. Climate as a boundary value problem iii. Key formulation of the Navier-Stokes Equation i. Weather forecasting as an initial value problem iv. Formulation of ocean-atmosphere models i. Basic conceptual structure ii. Approximations required iii. Typical format of model output 2 Today’s lecture Learn about energy source and distribution which drives the climate Why this is important for understanding climate change Considering this as a boundary value problem with uncertainty and unobserved outcome Highlight areas where ML may improve climate simulations (to revisit in week 3). With thanks to Prof Richard Washington for some of the slides 3 What drives our climate? It all starts with the sun Earth’s habitability is due to It’s distance from the sun Size, which controls its atmospheric composition This makes Earth a “goldilocks” planet Not to hot; not to cold Critical: Water can exist in all three phases (Solid, liquid, gas) 4/30 EM radiation emission and reflection by Earth For wavelengths < 5 μm solar radiation is dominant For wavelengths > 5 μm radiation of earth is dominant Solar radiation Earth radiation Watt/ m2 and micron Visible Visible Infrared and 5 Infrared  6000K short wave visible  6000K short wave visible 288 K long wave infrared Simplest mathematical model for climate Sigma: Plank’s constant 5.67 x 10-8 S, solar constant: 1370W/m^2 What temperature does this give us? 8/30 We need to include the atmosphere S(1-α) Fa τlwFg Atmosphere Ta Fa τswS(1-α) Fg Ground Tg This model works a little better. Surface temperatures are 13oC. That is an improvement! 9/30 We need to include the atmosphere 10/30 Atmospheric Composition Water vapour and carbon dioxide absorb strongly in longwave. Ozone absorbs strongly in shortwave. Atmosphere is largely transparent to 11/30 shortwave. KeySome Implication: history… CO2 sensitivity Eunice Foote 1856 Glass cylinder containing carbon dioxide heats up more than other constituent gasses in air John Tyndall 1861 Quantified the radiative absorption of CO2, Water Vapour (H2O), and Methane (CH4). Showed these gasses were very effective at absorbing and re-emitting radiation. Svante Arrhenius 1896 Calculated extent to which the greenhouse effect might increase Earth surface temperatures Guy Callendar 1938 Quantified estimate of the effect of anthropogenic carbon emissions on global mean temperature Key Some history….CO2 sensitivity Implication: Hawkins and Jones (2013) On increasing global temperatures: 75 years after Callendar. Q. J. R. Met. Soc. Imbalance of energy input and output? notice how - the latitudinal variation of Sn is far larger than that of Ln and dominates that of R - the zonal asymmetry of R (land-sea contrast) is rather small - the desert areas over land are radiatively deficient (anomalously low R for their latitude, on account of the large Ln loss) Slide: Richard Washington Annual cycle of surface air temperature note that the amplitude of the annual temperature range is higher at: - higher latitudes - over land rather than over water [this does NOT occur in terms of net radiation R] - over large land masses, especially their eastern side Slide: Richard Washington x2 CO2 GFDL: The first stable coupled climate model x4 CO2 warming = Cretaceous type conditions transient CO2 increase experiment (+1% per year compounded). [Source: adapted from Syukuro Manabe and Ronald Stouffer, Nature, 15 July 1993.] Tett et al 1996 “Human Influence on the Atmospheric Vertical Temperature Structure: Detection and Observations” Science, 247, 1170-1173 Superceded Santer et al 1996 Santer et al 1) forced models with Ozone, CO2, sulphates in separate runs 2) then added the results linearly Tett et al force model with Greenhouse Gas (G), G+Sulphates (G+S), G+S+Ozone (G+S+O) Undoes linearity assumption of Santer i.e. adding the results of the experiment together separately once experiments are complete  For each experiment…. 1986-1995 minus 1961-1980 90N 90S 90N 90S = Blue = colder 1986-1995 Red = warmer 1986-1995 90N 90S Red = warmer last ten years Blue = colder last ten years  So problem is solved yes? No Structure of the atmosphere is defined by the temperature profile (dT/dz) Troposphere (about the lowest 10km) is our concern with weather Why does temperature decrease with height in the troposphere? The closer to the sun – the colder it gets! 21/30 Trying to simulate this… Kiehl, J. T., and K. E. Trenberth (1997), Earth's Annual Global Mean Energy Budget, Bull. Am. Met. Soc., 78(2), 197-208. Redrafted by IPCC, 2001. Earth System Models Evolution Jones, C. D. (2020). So what is in an Earth system model? Journal of Advances in Modeling Earth Systems, 12, e2019MS001967 Earth System Models Evolution Global mean surface air temperature anomalies (relative to 1986-2005) from CMIP5 concentration-driven experiments. Shading is the 5-95% range across the distribution for individual models. See Collins and Knutti 2013, IPCC AR5 Chapter 12 Machine learning to save the day?!? Cloud climate sensitivity very poorly constrained…improve with ML derived parametrisations Atmosphere/ocean – biosphere interactions poorly constrained..ML to improve Even basic turbulence and therefore fluxes, poorly simulated so ML to improve? NeuralGCM guest lecture will discusses an approach to completely do away with individual parametrisations as described above…but at present it does not have an ocean couple to the atmosphere. 26/30  Summary https://interactive-atlas.ipcc.ch/ Basics of climate is a thermodynamics problem Includes: Observations, CMIP5, and CMIP6 Well-understood However many interacting factors create uncertainty in climate response to atmospheric composition changes. 27/30  AI in Weather and Climate Science Week 1: Fundamental principles in Earth System Science Lecture 2: Observations  Course Structure Week 1 (Neil): Introduction to Atmospheric (& Earth System) Science Aim: Equip you with sufficient domain knowledge to begin applying your ML skills Week 2 (Neil and Shruti): ML to explore emergent phenomena in weather and climate Aim: Gain appreciation of ways ML tools have helped us understand inner workings of our atmosphere Week 3 (Shruti): Prediction – how deep learning is (and is not) revolutionising this field of science Aim: as the title says. 2  Course Structure Practicals: Week 1 and into 2: Causal Inference Networks as a way to explore Earth Observation data and global teleconnections. Starts today Week 3: Deep Learning for flood prediction Assessments: Monday Week 2: Online Quiz to assess learning based on Week 1 (30% of grade) Friday Week 3: Individual oral exam discussion will be held for me to evaulate the advancement in your knowledge across the 3 weeks of the course against the core learning outcomes/goals. (70% of grade) 3  Guest Lectures Dates and timing are still to be finalized but provisionally… Stephan Hoyer (Deepmind) talking about development of NeuralGCM https://www.nature.com/articles/s41586-024-07744-y 18:45-19:45 Tues 28 January Ilan Price (Deepmind) talking about development of GenCAST https://www.nature.com/articles/s41586-024-08252-9 14:30-15:30 Thurs 30 January 4 Core learning goals Week 1 1. Overview of fundamental weather and climate concepts i. Observations of weather and climate available to us ii. Basics of Earth’s Radiation budget i. Climate as a boundary value problem iii. Key formulation of the Navier-Stokes Equation i. Weather forecasting as an initial value problem iv. Formulation of ocean-atmosphere models i. Basic conceptual structure ii. Approximations required iii. Typical format of model output 5 The earliest forms of observation Traditional knowledge/observations often tied to lunar cycles NAMES OF MONTHS IN LUNDA: 1.January CHINZEZHI-'nvula wanokang'a muzowa‘ (it rains continously) 2.February KUSOLU- 'wasola ntamba na yilung'u‘ (sweet potatoes and yams reach a peak in their growth, due to adequate rain water) ----------------------- 3.March MAYINZAMANENI (a month of heavy rains;storms) --------------------- 4.April KULUNDU- 'kasa kannong'a‘ (beginning of the dry season) ---------------------- 5.May KASHIKANA (the cold season sets in) ----------------------- 6.June CHISHIKAMUNENI- 'wocheli ndimbu‘ (so cold,it makes manioc 'cassava' leaves to wither) -------------------- 7.July KAPUPULU- 'kapupula tudya natububu‘ (flappy flappy;there is excitement in the air) ----------------------- 8.August KAKANZAKANZA- 'nkwang'a yalunjika mavunda‘ (strong winds from July continue shaking trees of the forest) ----------------------- 9.September NKANZAMUNENI- 'chumisha menzhi‘ (heat that dries water in brooks,streams, rivers and dams) ------------------------ 10.October KAVUMBI- 'kavumbi kawandimi‘ (planting season) ------;---------------- 11.November IVUMBIMUNENI (at the height of the planting season;cultivation all around) ----------------------- 12.December KWALI- 'kwa mpusa na wunja‘ (a season for mushrooms) 6 The earliest forms of observation Traditional knowledge/observations often tied to lunar cycles Requires careful stellar and lunar observation Driving force for the Calanais Standing Stones development of much early mathematics…prediction of key agricultural and associated religious dates. al-Battani First Measurement Instruments Temperature Pressure  First Measurement Instruments Maintenance of weather observatories requires ongoing human involvement Protection of stationarity of records Expanse of Shipping https://vimeo.com/200805506 created by Philip Brohan Philip is doing very interesting things with ML and weather and climate data: https://brohan.org/ & https://github.com/philip-brohan Expanse of Shipping buoys Different bucket observation protocols and Engine room instrumentation intake Challenge for creating long-term temperature record IPCC AR5 WG1 Growing observational networks Upper Surface Surface network 1023 stations Upper air network 171 stations HYDROLOGY & WATER RESOURCES Gilbert Walker, droughts, and El Nino El Nino Southern Oscillation Gilbert Walker, second Director General of the Observatories in India from 1904-1924, developed the first objective forecast of Indian Monsoon in 1909 Surface 13/30 Early Weather forecasting Enabled by the telegraph Observations could be gathered centrally Surface pressure points used to create charts of isobars Storm structure (in terms of pressure) created Observing in 3-dimensions: Atmosphere Attach temperature, humidity and pressure sensor to large Hydrogen or Helium filled balloon https://weather.uwyo.edu/upper air/sounding.html 15/30 Height of human-recorded observations Unfortunately many observation stations have stopped reporting Very sparse observations in some parts of Africa. James et al 2018, Bull. Am. Meteorol. Soc. Global Telecommunications System (GTS) Automatic weather stations (electronic measurement techniques) Weather balloons Ship observations Ocean buoys Airplanes send in flight data which can be converted into temperature and windspeed observations. Maahuis 2018: 10.13140/RG.2.2.21395.53288 Global Telecommunications System (GTS) But not all stations reporting….in real-time for weather forecasting https://wdqms.wmo.int/nwp/land_surface/six_hour/availability/pressure/all/2025-01-12/18 Global Telecommunications System (GTS) Even fewer for climate records https://wdqms.wmo.int/gcos/land_surface/availability/2024-10 Global Telecommunications System (GTS) Even fewer balloons for vertical structure of atmosphere https://wdqms.wmo.int/gbon/land_upper-air/daily/availability/all/2025-01-12 Observing in 3-dimensions: Ocean T Ocean surface obs also in GTS https://wdqms.wmo.int/nwp/marine_surface/six_hour/quality/temperature/all/2025-01-08/18 Summary for in-situ observations Extensive, but not comprehensive coverage of globe. Multiple types of measurement devices Inhomogenous in space… …and time Mix of digitized and paper-based observations. But what are we actually trying to observe? https://met3d.wavestoweather.de/met-3d.html 24 Remote Sensing Observing from afar (ex situ) as opposed in situ. RADAR Observations after world war two. Airplane observations Satellite observations. 25/30  Satellite-based observations An era that began ~1976 Possible with invention of digitial imaging technology Relies on electromagnetic radiation sensing Electromagnetic Radiation infrared ~3-14microns visible microwave ~0.4-0.7 microns 5-500mm EM radiation emission and reflection by Earth For wavelengths < 5 μm solar radiation is dominant For wavelengths > 5 μm radiation of earth is dominant Solar radiation Earth radiation Watt/ m2 and micron Visible Visible Infrared and Infrared Visible spectrum imaging T High reflectance High reflectance Sun Glint Very thick Snow clouds Desert Very thin clouds over land Bare Soil Forest Very thin Ocean, Sea clouds over ocean Low reflectance Low reflectance 29/30 Infrared spectrum imaging We cannot see this radiation. It is being emitted by the Earth and features near the Earth (e.g. clouds) It tells us about the temperature of the emitting body In infrared images white is cold and black is hot. White = longer wavelengths Black = shorter wavelengths 23rd February 2018, 12pm Infrared 9.8-11.8µm channel Meteosat Links to other sciences: Astronomy example Hubble's visible and infrared views of the Monkey Head Nebula. Credit: NASA and ESA Acknowledgment: the Hubble Heritage Team (STScI/AURA), and J. Hester What do these pairs of images have in common? Meteosat visible and infrared views of Earth: Credit: Eumetsat and Dundee Satellite Receiving Station Links to other sciences: Astronomy example Artist’s image of GOES-T the next geostationary satellite to be launched (March 2022). Credit: NOAA’s National Environmental Satellite, Data, and Information Service (NESDIS) What do these satellites have in common? Recently launched James Webb Space Telescope. Image courtesy NASA - https://web.archive.org/web/20100527230 418/http://www.jwst.nasa.gov/images_arti st13532.html (direct link), Public Domain, https://commons.wikimedia.org/w/index.p hp?curid=7732621  Active remote-sensing from space Two key active sensor types: RADAR or LIDAR RADAR typically used for: Sea Surface Height (SSH) Rainfall https://www.youtube.com/watch?v=SSKv4A_Cj5s LIDAR typically used for: Aersols with lidar (and winds most recently) Ocean surface winds with microwave scatterometers Most recently upper-atmosphere winds: AEOLUS Where to put the satellites? Orbits including Polar orbits Satellites can be categorised by their orbits including Polar orbits https://www.youtube.com/watch?v=17jymDn0W6U Trade off: Temporal vs Spatial coverage Timing of the Instrumental Record Summary of available time periods of data Proxy records Ocean heat content Satellite Rainfall CO2 Rainfall Upper air temperature Mix/max Temperature Sea Level Pressure Sea Surface Temperature Land Surface temperature Summary We need observations We have a vast network that is heterogenous in terms of Observation types Time sampling Spatial coverage Temporal coverage Uncertainty sources Creating robust and scientifically-sound mergers of data is very challenging, time-consuming, and important! ML to deal with raw data is possible, but needs very careful thought Data pipelining is possibly the most time consuming and critical parts of any ML application you may develop…Week 3 Lecture 2  To come… Make fuller use of these observations requires data assimilation This is the optimal fitting of a 4-dimensional (latitude, longitude, height, time) model to the observations. Needs to account for observational uncertainty, inhomogeneity in space and intermittency in time. This will be covered in more detail in lecture on ERA5 (ECMWF Reanalysis 5) First we need to appreciate some key points about weather climate and the differences in modelling each. These topics will be covered in lecture 3 (Wed) and lecture 4 (Thurs) Links to explore observations(and forecasts) yourself Useful websites: http://windy.com & http://ventusky.com https://www.ncdc.noaa.gov/gibbs/calendar/2023 https://climate.copernicus.eu/ http://climexp.knmi.nl/start.cgi https://earthengine.google.com/  AI in Weather and Climate Science Week 1: Fundamental principles in Earth System Science Lecture 1: Problem-driven research AI in Climate aka ML & AI in Weather and Climate Week 1: Introduction to Atmospheric (& Earth System) Science Aim: Equip you with sufficient domain knowledge to begin applying your ML skills Week 2: ML to explore emergent phenomena in weather and climate Aim: Gain appreciation of ways ML tools have helped us understand inner workings of our atmosphere Week 3: Prediction – how deep learning is (and is not) revolutionising this field of science Aim: as the title says. 2 Today’s lecture Think about what is means to work in a problem-driven research field Introduction to my research Discussion about your experience of weather and climate problems. Take notes: Specifically note down assumptions I am making in your knowledge, Things I clearly don’t explain in much detail A secondary aim of this lecture is to set up questions in your minds that we will answer in following lectures. 3 Two Lecturers Neil Hart…that’s me. [email protected] Shruti Nath: [email protected] 4 When will the rains start? Managing climate risks in a warming world.  NAMES OF MONTHS IN LUNDA: 1.January CHINZEZHI-'nvula wanokang'a muzowa' (it rains continously) ----------------------- 2.February KUSOLU- 'wasola ntamba na yilung'u' (sweet potatoes and yams reach a peak in their growth,due to adequate rain water) ----------------------- 3.March MAYINZAMANENI (a month of heavy rains;storms) --------------------- 4.April KULUNDU- 'kasa kannong'a' (beginning of the dry season) ---------------------- 5.May KASHIKANA (the cold season sets in) ----------------------- 6.June CHISHIKAMUNENI- 'wocheli ndimbu' (so cold,it makes manioc 'cassava' leaves to wither) -------------------- 7.July KAPUPULU- 'kapupula tudya natububu' (flappy flappy;there is excitement in the air) ----------------------- 8.August KAKANZAKANZA- 'nkwang'a yalunjika mavunda' (strong winds from July continue shaking trees of the forest) ----------------------- 9.September NKANZAMUNENI- 'chumisha menzhi' (heat that dries water in brooks,streams, rivers and dams) ------------------------ 10.October KAVUMBI- 'kavumbi kawandimi' (planting season) ------;---------------- 11.November IVUMBIMUNENI (at the height of the planting season;cultivation all around) ----------------------- 12.December KWALI- 'kwa mpusa na wunja' (a season for mushrooms) Diplodocus Lucy the “first” You are here hominid? Farming T-Rex & Triceratops What might this mean for change in the start of the southern African rains? Let’s think through this from basic climate processes.  Zambezi River: Lake Kariba lowest in 40 years BBC / Kennedy Gondwe Zambia electricity crisis: Drought hits hydro-powered Kariba Dam - BBC News https://www.worldweatherattribution.org/el-nino-key-driver-o https://www.bbc.co.uk/news/articles/cx2krr137x9o.amp f-drought-in-highly-vulnerable-southern-african-countries/ Zambezi River: Lake Kariba lowest in 40 years https://www.worldweatherattribution.org/el-nino-key-driver-o f-drought-in-highly-vulnerable-southern-african-countries/  Rio Negro EPA https://www.bbc.co.uk/news/i n-pictures-67087949.amp Espinoza, JC., Jimenez, J.C., Marengo, J.A. et al. The new record of drought and warmth in the Amazon in 2023 related to regional and global climatic features. Sci Rep 14, 8107 (2024). https://doi.org/10.1038/s41598-024-58782-5  Land and ocean in Cloud in white black and grey  Air rises in tropics, sinks in subtropics Red: Air rising in warm tropical thunderstorms Blue: Air descending into subtropics, inhibiting thunderstorms  Energy Conversions in the Hadley Cell Air cools by longwave radiation Descent, adiabatic Tropical Heating compression and drives warming uplift and evaporation Net radiation warms air and evaporates water Subtropical Equatorial High pressure Thunderstorms What of this year? 23 And what about southern Africa particularly? Contemporary Change in SICZ cloud bands Contemporary Change in SICZ cloud bands A risk forewarned may be a disaster averted… Can we predict the start of the rains weeks to months ahead? What the models need to represent. Details in Hart et al 2018 J. Climate Typical climate models rain too much!  The rain belt as a marionette The sun provides the same forcing each year, but the rain belt as a lot of freedom of movement…can we simulate its dance? What has been missing? What has been missing? Thunderstorms!  New Model: Convection permitting Africa (CP4-Africa) simulations N512 global 25x39km (768 x 1024 x 85) CP4Africa 4.5km (2000 x 2100 x 80) 10 year simulations: present-day (1997/03-2007/02) future RCP8.5 (2097/03-2107/02) Segolene Berthou, Rachel Stratton, Simon Tucker, Sonja Folwell Explicit thunderstorms increase strength of tropical ascent South North Typical climate model Height South North New model with thunderstorms Height Explicit thunderstorms increase strength of subtropical descent = less rainfall South North Typical climate model Height South North New model with thunderstorms Height Hart et al, under review Why does this matter? Application to understanding climate models Hart et al 2018, Geophysical Research Letters What the models need to represent. Details in Hart et al 2018 J. Climate Group conversations Did you grow up in a rural or urban setting? How has weather affected your live? Do you remember particularly bad seasons? Do your parents or grandparents talk about what the weather was like in years past? Are there changes that have affected the livelihoods of your family or community? 47/30 Problems/Experiences 1. 2014 Floods in Khartoum 2. 2023 Drought Zimbawe…grains for planting, late planting (December) 3. Dafur, Bedouin tribes…Drought in south, rains in north, but now switched….migration driven by change in rainfall….much conflict 4. Ghana 1981-1983 drought, bushfire 5. 2017-2018 South Africa drought (Cederburg)…2 years of drought, then flooding 48 AI in Climate aka ML & AI in Weather and Climate Week 1: Introduction to Atmospheric (& Earth System) Science Aim: Equip you with sufficient domain knowledge to begin applying your ML skills Week 2: ML to explore emergent phenomena in weather and climate Aim: Gain appreciation of ways ML tools have helped us understand inner workings of our atmosphere Week 3: Prediction – how deep learning is (and is not) revolutionising this field of science Aim: as the title says. 49