Forecasting for the energy system

1 « Short » timescales

Riwal Plougonven,
Laboratoire de Météorologie Dynamique / IPSL
Ecole Polytechnique
Prelude

Forecasting,
what is familiar ?
Prelude

Forecasting,
what is familiar ?

Forecast for Wednesday 02/10/2024,

initiated Saturday 28/09/2024
Prelude

Forecasting,
what is familiar ?
Prelude

Forecasting,
what is familiar ?
Further examples of charts, for October 2, 2024

Mean Sea Level pressure and 850 hPa wind 2 m tempertature and 30 m wind

100 m wind Pseudo-satellite image of clouds

- what is represented on these maps ? Why pressure ?

- how are these forecasts obtained ? How do we model weather ?

- how reliable is this forecast ? How can we quantify this ?

- what are sources of uncertainty ? how far in time can we forecast ?

- how do we apply these forecasts to the energy sector ?

- what other sources of information for the energy sector ?

- what is represented on these maps ? Why pressure ?

- how are these forecasts obtained ? How do we model weather ?

- how reliable is this forecast ? How can we quantify this ?

- what are sources of uncertainty ? how far in time can we forecast ?

- how do we apply these forecasts to the energy sector ?

- what other sources of information for the energy sector ?

Outline
1. Atmospheric circulation

2. What is a Numerical Weather Prediction Model

3. Forecasting for energy – examples for short timescales

1. Atmospheric circulation

Motivations to briefly discuss some aspects of atmospheric circulation :

drivers of atmospheric circulation

main processes affecting renewable energies (focus on wind)

timescales involved

→ notion on what may be expected

Starting point : visualization of the atmospheric flow

https://earth.nullschool.net/
 1. Atmospheric circulation : examples for 28/09/2024, from earth.nullschool.net

Atlantic / Europe Atlantic / Europe

Surface 250 hPa (~10 km)

Low pressure Strong,
systems over meandering
the Atlantic eastward winds

500 hPa (~5 km) 70 hPa (~18 km)

 1. Atmospheric circulation : examples for 28/09/2024, from earth.nullschool.net

South Pole South Pole

Surface South Pole 250 hPa (~10 km) South Pole

Low pressure Strong,
systems over meandering
the Atlantic eastward winds :
‘jet stream’

500 hPa (~5 km) 70 hPa (~18 km)

What is the fluid we wish to describe ?
What is the fluid we wish to describe ?

Intuition / personal experience may suggest a three-dimensional fluid with complex

structures… true, on certain scales
What is the fluid we wish to describe ?

(left) Felix Baumgartner just before (right) The atmosphere seen from
his famous jump, 14 October 2012, the Internation Space Station.
from 38 960 m altitude above
New Mexico, U.S.A.

Thin layer ?
Aspect ratio
Radius of the Earth : ___ km ?

Earth circumference : ___ km ?

1 degree of latitude : ___ km ?
1 degree of longitude : ___ km ?

Weight of the atmosphere ? ___ kg

‘Thickness’ of the atmosphere ? ___ km
troposphere : ~ 10 km
(more precisely, height of the tropopause :
~ 8 km at polar latitudes
~ 10 km in midlatitutdes
~ 16 km in the Tropics )

Typical horizontal scale ___ km

Typical aspect ratio : L / H ~ ? ___

Illustration from oceanography in the next few slides.

Aspect ratio
Radius of the Earth : 6370 km

Earth circumference : 40 000 km

1 degree of latitude : 111 km
1 degree of longitude : 111 x cos(f) km , with f the latitude

Weight of the atmosphere ? ___ kg

‘Thickness’ of the atmosphere ? ~10 km ? (roughly 3/4 of the mass)
troposphere : ~ 10 km in extra-tropics
(more precisely, height of the tropopause :
~ 8 km at polar latitudes
~ 10 km in midlatitutdes
~ 16 km in the Tropics )

Typical horizontal scale : 1000 km

Typical aspect ratio : L / H ~ ? 100
Aspect ratio
Radius of the Earth : 6370 km

Earth circumference : 40 000 km

1 degree of latitude : 111 km
1 degree of longitude : 111 x cos(f) km , with f the latitude

Weight of the atmosphere ? ~ 5e18 kg (Earth’s surface (R~6370 km) x

mean sea level pressure (1e5 Pa))
Given the aspect ratio, and typical scales and orders of magnitude,
what terms dominate in the equations ?

EQUATIONS OF MOTION
Newton’s second law :

Acceleration Pressure Gravity Friction

gradient

and where
EQUATIONS OF MOTION
But : rotating frame of reference :

Motion relative to a galilean frame :

R

Earth Rate of rotation on Earth :

EQUATIONS OF MOTION
Newton’s second law in a rotating frame
(with a slight change of coordinates to get
rid of the centrifugal term) :

Coriolis
‘force’
EQUATIONS OF MOTION
In local coordinates : x in the West-East direction,
y in the South-North direction,
z in the local vertical

y
z
Projection of W on the local vertical
Coriolis parameter :

For example, at 45°N :

(corresponding period in hours ?)

z
VERTICAL MOMENTUM EQUATION

Scaling of the equation for the vertical component :

(estimating the order of magnitude of each term)

Equation :

Order of
magnitude :

With

Order of
magnitude :
→ HYDROSTATIC BALANCE
Scaling of the equation for the vertical component :
(estimating the order of magnitude of each term)

Equation :

To an excellent approximation,
the dominant balance is hydrostatic :
between gravity and the pressure force

Order of
magnitude :
HORIZONTAL MOMENTUM EQUATION
Scaling of the equation for the x component :
(estimating the order of magnitude of each term)
Equation :
HORIZONTAL MOMENTUM EQUATION
Scaling of the equation for the x component :
(estimating the order of magnitude of each term)
Equation :

Order of
magnitude :
With
GEOSTROPHIC BALANCE
Scaling of the equation for the x component :
(estimating the order of magnitude of each term)
Equation :

To a good approximation,
the dominant balance is geostrophic :
between the ‘Coriolis force’ and pressure
GEOSTROPHIC AND HYDROSTATIC BALANCE
A non-dimensional number to quantify the
relevance of geostrophic balance :

The Rossby number :

With the values used above :

 GEOSTROPHIC BALANCE
z
y

Low Wind

Pressure force High

x

Coriolis force

(Northern Hemisphere) :

The wind leaves the high pressure on its right.

Expression for geostrophic wind in Cartesian coordinates :

Very good approximation of the wind in the free troposphere, outside the Tropics

Geostrophic wind is
- parallel to isobars,
- stronger where the isobars are more tightly packed,
- rotates counter-clockwise around low pressure centers in the Northern Hemisphere

Pressure nearly plays the rôle of a streamfunction in the above expressions

Mean sea level pressure and 100 m wind
Northern Hemisphere Southern Hemisphere
Cyclonic rotation → counterclockwise Cyclonic rotation → clockwise
Extra-tropical storms and storm tracks
A short remark on clouds and pressure

Geostrophic wind and boundary layer

aloft : balance between pressure gradient and Coriolis force

Near the surface : three way balance between Coriolis, pressure and friction
→ convergence in low pressure systems

→ clouds near pressure lows

(not discussed because of time :
fronts, non-linear development,
turbulence in the boundary layer, ec..

→ clear skies in anti-cyclones

Forecast initiated
1/10/24, 12:OO UTC,

valid T+120 h
(Sunday 6, 12:00 UTC)
→ not the same day !
Brief Summary on Atmospheric Circulation
Aspect ratio
→ quasi-horizontal motions

Layers :
Troposphere (up to 10 / 15 km)
Stratosphere (from tropopause to ~50 km)

Dominant balances :
hydrostatic balance
geostrophic balance
→ key rôle of pressure (mass distribution of the atmosphere)

Atmospheric Dynamics :
Chaotic : sensitive to initial conditions
Predictability limit for atmospheric dynamics (10 days to 2 weeks)

Multi-physics :
Dynamics, radiative transfer, thermodynamics and microphysics

Multiscale :
clouds, boundary layer : unresolved
There are now many features of these maps that we now understand.

Mean Sea Level pressure and 850 hPa wind 2 m tempertature and 30 m wind

100 m wind Pseudo-satellite image of clouds

There are now many features of these maps that we understand.

Mean Sea Level pressure and 850 hPa wind 2 m tempertature and 30 m wind

100 m wind Pseudo-satellite image of clouds

But, how were they obtained ?

2. What is a Numerical Weather Prediction Model ?

Model based on physical laws (Navier-Stokes, physics of radiation) which calculates the evolution of the atmosphere
allowing to produce meteorological forecasts

Difficulties and sources of error/uncertainty :

- discretisation (fluid = continuum, numerical model = finite number of operations)

- unresolved processes (→ described by parameterizations, ie ad hoc sub-models)

- initialization (data assimilation)

- chaotic evolution of fluid motion (deterministic forecast vain beyond ~15 days)

Some vocabulary :
Data assimilation
Analysis, reanalysis

Deterministic vs probabilitic forecast ;

Ensemble forecast
Discretization
→ choice of grids

Laws of physics to advance in time

fluid motions, large-scale

Limited resolution
→ un-resolved processes
(subgrid-scale)
→ parameterizations for the
un-resolved processes :

clouds, precipitation
(boundary-layer) turbulence
surfaces and fluxes
radiative transfer
internal gravity waves
Discretization
→ choice of grids

Laws of physics to advance in time

fluid motions, large-scale

Limited resolution
→ un-resolved processes
(subgrid-scale)
→ parameterizations for the
un-resolved processes :

clouds, precipitation
(boundary-layer) turbulence
surfaces and fluxes
radiative transfer
internal gravity waves Bauer et al 2015
Focus on Data assimilation, analyses and reanalyses
Focus on Data assimilation, analyses and reanalyses

How was this figure obtained ?

Focus on Data assimilation, analyses and reanalyses

How was this figure obtained ?

Forecast for t0 + 3 days

Close-up :

Forecast three
days lead time

Initiated
25/03/2019, 00
valid
28/03/2019, 00
Close-up :

Verification :

What really
happened

28/03/2019, 00
 Focus on Data assimilation, analyses and reanalyses

How was this figure obtained ?

Initial condition Forecast for t0 + 3 days

NWP model
 Focus on Data assimilation, analyses and reanalyses

How was this figure obtained ?

Initial condition Forecast for t0 + 3 days

NWP model

Information for the initial condition :

- observations (meteorological stations, soundings, satellite obs…)
 Focus on Data assimilation, analyses and reanalyses

How was this figure obtained ?

Initial condition Forecast for t0 + 3 days

NWP model

Information for the initial condition :

- observations (meteorological stations, soundings, satellite obs…)

• Previous forecast (t0 + 12 h)

 Data assimilation :
Merging of all available information to obtain the best possible
estimate of the atmospheric state at t0

Analysis : best estimate of the atmospheric state, given observations,

a NWP model and data assimilation methodology
 Data assimilation :
Merging of all available information to obtain the best possible
estimate of the atmospheric state at t0

Analysis : best estimate of the atmospheric state, given observations,

a NWP model and data assimilation methodology
Now, say you wish to study the variability of the winds and temperatures over Asia for the
past four decades…

possible to use analyses ?

Now, say you wish to study the variability of the winds and temperatures over Asia for the
past four decades…

possible to use analyses ?

No : inhomogeneous dataset.
For example, the resolution changes with time as the grid has been refined.
 Data assimilation :
Merging of all available information to obtain the best possible
estimate of the atmospheric state at t0

Analysis : best estimate of the atmospheric state, given observations,

a NWP model and data assimilation methodology

Re-Analysis : analyses for re-forecasts carried out over a long period with a frozen
NWP model and data assimilation methodology
 Data assimilation :
Merging of all available information to obtain the best possible
estimate of the atmospheric state at t0

Analysis : best estimate of the atmospheric state, given observations,

a NWP model and data assimilation methodology

Re-Analysis : analyses for re-forecasts carried out over a long period with a frozen
NWP model and data assimilation methodology

Examples : ERA-Interim, ERA5, MERRA2, JRA...

3. Forecasting for energy – examples for short timescales

Motivation
Anticipating energy extraction and consumption
Planning maintenance
Example below from www.energy-charts.de, from Hong et al, 2019 :
June 2019, negative prices (8/6) and spike (25/6)

Hong et al 2019
3. Forecasting for energy – examples for short timescales

Timescales : minutes
Source of information

Solar Wind
sky camera nacelle anemometer
lidar (recent developments)

Methods : advection of observed clouds Statistical methods

A side remark on statistical approaches for forecasting

Exercise of forecasting based on data

for a specific location :

- interesting to carry out

- case specific

- difficult to generalize,
to compare,
to use…

Illustration of the variety of methods

investigated →

from a review article, Liu et al 2019

Liu et al, (2019) : Deterministic wind energy forecasting: A review of intelligent predictors and auxiliary methods,
Energy Conversion and Management, 195, 238-245
3. Forecasting for energy – examples for short timescales

Timescales : intra-hour to a few hours

Source of information

Solar Wind
Satellites NWP (Numerical Weather Prediction
NWP models)

Work from Sylvain Cros

(Réuniwatt in 2014, now LMD)
3. Forecasting for energy – examples for short timescales

Timescales : day-ahead to a few days

Source of information

Solar Wind
NWP NWP
From wind to power
Knowledge of the wind,
and manufacturer’s
power curve P(U)

→ power produced
From wind to power
Knowledge of the wind,
and manufacturer’s
power curve P(U)

→ power produced

In practice :
difference between modeled wind
and real wind at the turbine
spread around power curve
From wind to power
Knowledge of the wind,
and manufacturer’s
power curve P(U)

→ power produced

In practice :
difference between modeled wind
and real wind at the turbine
spread around power curve

availability of turbines
wakes
curtailment*

* imposed reduction of power produced, due

to noise or excess of renewables for ex
Using both local information and NWP forecasts
PhD work of Aurore Dupré, with Philippe Drobinski and Jordi Badosa (LMD)
Compare persistence (reference) with statistical models (ARMA, ANN) and with a linear
regression from the ECMWF forecast with or without using the initial observation

One example of a 3hr forecast Improvements over persistence

Dupré et al, 2020

Using both local information and NWP forecasts, and statistical
methods
Predictions out to 4 hours
for the locations of 5 wind
farms, using several
methods to combine
local measurement
and NWP

→ little sensitivity to the

method

Bouche et al, 2023

Using both local information and NWP forecasts, and statistical
methods
Predictions out to 4 hours
for the locations of 5 wind
farms, using several
methods to combine
local measurement
and NWP

→ little sensitivity to the

method

But quantifies which

variables are informative :
wind speed (100m)
wind speed ( 10m)
wind shear
not boundary layer info
Bouche et al, 2023
Similar but not identical across locations
Forecasting for energy has grown very large over past decade

Hong et al 2020

→ recommendations for reproducible research, and working together

Summary and Conclusion for short timescales (hours to a day)

Best to combine

local observation, and

NWP forecasts

Figure 7.1 from Troccoli et al, 2018

Chapter ‘Short-range forecasting for energy,
by Sue Ellen Haupt
Summary and Conclusion for short timescales (hours to a day)

Best to combine

local observation, and

NWP forecasts
How far ahead
can we forecast ?

Figure 7.1 from Troccoli et al, 2018

Chapter ‘Short-range forecasting for energy,
by Sue Ellen Haupt