180 likes | 364 Views
How do models work?. METR 2021: Spring 2009 Lab 10. How do we forecast the weather?. Use Numerical Weather Prediction (NWP) models Fancy name for we let the computers crunch the numbers and we interpret the results
E N D
How do models work? METR 2021: Spring 2009 Lab 10
How do we forecast the weather? • Use Numerical Weather Prediction (NWP) models • Fancy name for we let the computers crunch the numbers and we interpret the results • Models are run at least 4 times per day and forecast for anywhere between 12 to 384 hours into the future • But how does the model work?
History of NWP • Lewis Fry Richardson first to attempt NWP • His attempt failed miserably in 1922 • What were some limitations he faced? • Jule Charney 1950 first successful try • Equivalent barotropic model • Only 500-hPa heights forecasted • Today • Run models on desktop computers, forecast to 384 hours!
NWP basics • Each Model has a certain Domain or area over which it predicts the weather • Two types of models • Gridded models • Spectral models • Each model has some representation of the topography of Earth, some are better than others • Each grid point model has a certain resolution or spacing to grid points within the model • Spectral models increase resolution by keeping more parts of the wave series
Model Resolution Vertical Resolution Horizontal Resolution
Forecasting: The Musical Act One: • Sfc Observations (input: worldwide) • Radiosondes, Wind Profilers • Satellite, radar, aircraft, ship obs Act Two: • Data Analysis and Assimilation • Error Checking (QC) • Gridding of Data (interpolation) • Smoothing (initialization) • Input into models (assimilation)
Forecasting: The Musical cont. Act Three: • Numerical Modeling • Computer works to find solutions to the equations of the atmosphere Act Four: • Forecast Dissemination (output: worldwide) • This is what meteorologists look at!!
Limitations to NWP Models • The Equations are inexact!!! • Must parameterize things • Errors in Initial State • The model is limited by quality of data going in • Inadequate Resolution • Not enough computer power to do it all • Most operational models cannot resolve T-Storms
Parameterizations • Are we able to resolve boundary layer processes? • Can we know of all radiation scattering? • Is a typical convective cell 20 km wide? • Need to parameterize (make the model produce) phenomenon that are sub-grid scale
Parameterizations cont. • Many different parameterizations have been developed over the years • Each parameterization behaves differently • Convective Schemes • Kain–Fritsch • Betts–Miller • Boundary Layer • Which closure to use? (zero, 1, 1.5, or 2)
Inadequate Resolution • We keep getting smaller and smaller grid spacings…which helps, somewhat • What can be resolved for a given grid spacing? • 20 km (2x = 40 km) • 4 km (2x = 8 km) • Are there a lot of different phenomenon occurring that are smaller than 8 km?
Errors in Initial State • Error in data collection • Instrumentation errors • Gridding up observations • Smoothing of data • Too few observations for analysis • Features not resolved in observations data
Ensemble Forecasting • To account for errors in initial state, run model a number of times with slightly varying initial conditions (Lorenz – Chaos) • Used to asses confidence in models forecast ability • Also called spaghetti plots • Many different model runs for the same time • If it looks like spaghetti, low confidence • Less Variability = high confidence in forecast
Research Models • NWP are sometimes developed for research purposes • Usually higher resolution • Take a couple of days to a couple of weeks to run one model simulation that lasts 24 hours • Models to predict… • T-Storms • Tornadoes • Hurricanes