This document discusses nowcasting and forecasting of solar irradiance using meteorological data. Nowcasting uses observations from the past 6 hours to predict clouds and irradiance up to 2 hours ahead for a specific site. Forecasting uses numerical weather prediction models to predict clouds and irradiance out to days or weeks ahead on regional to global scales. The document outlines various nowcasting techniques including the use of sky cameras, satellites, and neural networks. It also describes several forecast models run operationally at ECMWF and AEMET including HIRLAM, HARMONIE, and the ECMWF model. Prognostic aerosols are also modeled to improve irradiance forecasts.

1. Future Guidelines on solar
forecasting: the meteorological view
Isabel Martínez Marco
imartinezm@aemet.es
AEMET
In collaboration with Emilio Cuevas, Pilar Fernández,
Enric Terradellas and Javier Calvo

2. Outline
• Introduction
• Nowcasting:
• Cloud and irradiance nowcasting from Total-Sky cameras
• SAF of Nowcasting (NWC SAF)
• Forecasting:
• HIRLAM and HARMONIE Models
• ECMWF Model
• Quick overview of the MACC/ECMWF aerosol analysis and
forecasting system
• WMO SDS-WAS program
• Dust forecast

3. 3
Nowcasting is a technique for very short-range forecasting (normally
within 6h ahead) covering only a very specific geographic region.
In cloud nowcasting we map the current cloudiness and, using an
estimate of its speed and direction of movement, we forecast the
cloudiness a short period ahead (1-2h) for a specific site (1 km2) —
assuming the weather will move without significant changes.
10 20 30 40 50 60 70 80 90 100 120 130 140
150 minutes
Nowcasting
Forecasting
In-situ observations
Satellite
Neural network models
Satellite information
Post-processed Numerical Weather Prediction Model Da
Nowcasting concept

4. 4
 Total sky imagery can be used to make forecasts in quasi-real time (with a delay
of only 15-30 minutes) by applying image processing and cloud tracking techniques
to digitized sky photographs. Hazy skies can make difficult to properly identify
clouds.
 Under cloudless skies, when most Concentrated Solar Power (CSP) and
Concentrated Photovoltaics (CPV) plants operate, aerosol optical depth (AOD)
becomes the driver factor. Large portion of uncertainty can be attributed to the lack
of accurate aerosol data used to model DNI.
 Satellite imagery applies total sky imagery methods to cloud scenes (e.g. the
SEVIRI cloud-motion winds derived from successive satellite images can be used
to predict the DNI at ground level with sufficient accuracy.
 DNI and GHI attenuation by different types of clouds and aerosols must be
parameterized by sensitivity studies using Radiative Transfer Models (RTF).
So, a combined approach using in-situ observations (total-sky cameras and
radiometers), satellite observations (SEVIRI), and RTF models appears to be able
to provide the most accurate results for cloud-DNI-GHI nowcasting at CSPs and
CPVs
Main considerations

5. 5
- CCD sensor. 640x480 pixels, 8 bit, color response from 400 to
700nm and monochrome response from 400 to 1000nm.
- Very durable aluminium housing.
- Borosilitate dome.
- Rotating shadow band.
- Cooling/heating system (-10º to +50ºC).
- Fast frame rates (up to 70 fps).
- Adjustable JPEG compressed still-images or live MJPEG
streaming video.
- Transfer of images via FTP, RTP or HTTP.
- Camera control via HTTP, XML-RPC, Telnet
Component#1: SONA
Cloud Observation Automatic System

6. 6
Cloud detection: Neural network
Cloud flow determination:
To cluster the motion field we have
based on a Density-Based Algorithm
for Discovering Clusters in Large
Spatial Databases with Noise
(DBSCAN)

7. 7
Component #2: In-situ column aerosol content determination
Preliminary (excellent)
results of total column
aerosol content obtained
with a new inexpensive CI
radiometer
compared with AOD from
AERONET

8. 8
Cloud Mask Cloud Type
Cloud Top Height Cloud Top Pressure
Component#3: NWC SAF cloud products

9. 9
Component#4: Cloud and dust
sensitivity analysis with neural
network models and LibRadtran
DNI attenuation by altocumulus
GHI
DNI
DHI

10. 10
Summarizing the nowcasting state of the art
Neural network
modeling
DNI/GHI
nowcasting
Cloud
observation
DNI, GHI
observation
Aerosol/dust
observation
Optical flow
Cloud
height/type
DNI/GHI
Cloud
attenuation
Aerosol, Dust
and water vapor
sensitivity
analysis
SAF/NWC
Development
of low-cost
instruments

11. SAF de Nowcasting
(NWC SAF)
The Nowcasting Satellite Application Facility (SAF) was
established in 1996 between EUMETSAT and former
Instituto Nacional de Meteorología (Spanish National
Weather Service, AEMET (Agencia Estatal de Meteorología)
since 2008).
Under the leadership of the Spanish Meteorological
Agency (AEMET), the NWC SAF is developed by a
Project Team involving France (Météo-France), Sweden
(SMHI) and Austria (ZAMG) Meteorological Services.

12. Objectives
 Development of Nowcasting products derived from both
Geostationary (MSG) and Polar Platform (PPS) satellite
systems
 To be delivered to users as SW Packages
 Products are generated locally at user premises
Responsible for
 Development and maintenance of the NWC products
 Development and maintenance of the SW Packages
 User's support and training tasks
89 users at the date
http://www.nwcsaf.org

13. Clear Air, Precipitation & Wind MSG products
Clouds & Convection MSG products
Cloud & Precipitation PPS products
Meteorological Systems MSG products
Products responsibility

14. Cloud Products MSG
Cloud Top Temperature & Height (CTTH)
Detailed cloud analysis with information on
the major cloud classes for all the pixels
identified as cloudy.
Information on the cloud top temperature,
pressure and height for all pixels identified
as cloudy.
Cloud Type (CT)
Cloud-free pixels delineation in a satellite
scene with a high confidence.
Also: snow/sea ice, dust clouds and
volcanic plumes.
Cloud Mask (CMa)

15. Cloud Products PPS
Cloud Type (CT)Cloud Mask (CMa) Cloud Top Temperature &
Height (CTTH)

16. Precipitation & Convection Products (MSG&PPS)
Precipitating Clouds (PC)
Probability of precipitation intensities in pre-
defined intensity intervals.
Convective Rainfall Rate (CRR)
Precipitation estimated rate associated to
convective clouds. Instantaneous rain rate
and hourly accumulations.
Rapid Development Thunderstorm (RDT)
Identification, monitoring and tracking of
intense convective systems, and detection of
rapidly developing convective cells
PPSMSG

17. Clear Air Products MSG
SAFNWC Physical Retrieval (SPhR)
Optimal estimation algorithm to obtain Stability Parameters:
Total and Layered Precipitable Water and Instability Indexes
TPWLPW-HLLPW-MLLPW-BL
LI KI SHW

18. HRW v2011
HRW v2012
 Up to 7 SEVIRI Channels
 Improved Height assignment
Wind Products MSG
Detailed and frequently updated
sets of Automatic Motion Winds
including wind pressure level
information and quality control
flags.

19. 19
10 20 30 40 50 60 70 80 90 100 120 130 140
150 minutes
Nowcasting
Forecasting
In-situ observations
Satellite
Neural network models
Satellite information
Post-processed Numerical Weather Prediction Model Da
Forecasting

20. Numerical weather prediction
• The behaviour of the atmosphere is governed by a set of physical laws
which express how the air moves, the process of heating and cooling, the
role of moisture, and so on.
• Equations cannot be solved analytically, numerical methods are needed.
• Given a description of the current state of the atmosphere, numerical
models can be used to propagate this information forwards to produce a
forecast for future weather.
• Additionally, knowledge of initial conditions of system is necessary.
• Incomplete picture from observations can be completed by data
assimilation.
• Resolution of the model is determined by available computing resources.
It does not correspond to any natural scale separation.

21. Numerical Weather Prediction
• Processes not resolved by the model must be ‘parametrized’.
• Effective resolution is not same as model grid spacing.
• Numerical algorithms are compromise between accuracy and speed; care
needed to ensure numerical stability.
• Interactions between atmosphere and land/ocean important

22. Forecast ranges
• Short-range weather forecast (0-2 days ahead)
• Detailed prediction - regional forecasting system
• Produce forecast few hours after observations are made
• Medium-range weather forecast (2 days - 2 weeks ahead)
• Less detailed prediction - global forecasting system
• Produce forecast up to several hours after
observations are made
• Long-range weather forecast (more than 2 weeks ahead)
• Predict statistics of weather for coming month or season
• Climate prediction
• Predicts the climate evolution on the basis of pre-defined
scenarios (CO2, O3, …)

23. HIRLAM (High Resolution Limited Area Model)
• Model Formulation:
• Horizontal resolution: 0.16º latxlon (ONR) and 0.05º latxlon
(HNR)
• Boundary Conditions:
• ONR: from ECMWF with 0.25º
• HNR and CNN: from ONR with 0.16º (nesting models)
• Analysis: 3-dimensional variational method (3D-VAR)
• The Resolution in space
• Vertical Resolution: 40 hybrid levels
• Horizontal grid: regular rotated longitude/latitude

24. Integration Domains

25. HIRLAM (High Resolution Limited Area Model)
• In development: HARMONIE
Hirlam Aladin Regional/Meso-scale Operational NWP In Europe
• A new model formulation:
• Horizontal resolution: 2.5km
• Vertical resolution: 65 hybrid levels
• Analysis: 4-dimensional variational method (4D-VAR)
• Horizontal grid: Spectral representation
• Vertical grid: finite differences
• Non-hydrostatic dynamical kernel from ALADIN Model

26. The ECMWF Numerical Weather
Prediction (NWP) Model
• High-resolution model
• T1279 spectral
resolution
• 16 km global grid
• 91 hybrid levels from
the surface to a height
of 80km
• Variables at each grid
point
• Wind
• Temperature
• Humidity
• Cloud water, ice, cloud
fraction
• Ozone
• Pressure at surface

27. • A number of radiation schemes are in use at ECMWF. As of January 2011 are
active
• McRad including RRTM_LW and RRTM_SW is used in the forward model for
operational 10-day forecasts at TL1279 L91, EPS 15-day forecasts at TL639
L62, and seasonal forecasts at TL159 L62.
• The tangent linear and adjoint of the “old” SW radiation scheme in a 2-
spectral interval version is used for Data Assimilation.
• The tangent linear and adjoint of the “old” LW radiation scheme with 6
spectral intervals, replacing a neural network version of the same “old” LW
radiation scheme (Morcrette, 1991; Janiskova and Morcrette, 2005), is used
for DA.
• … and all the dedicated RT scheme used to simulate radiances (RTTOV-
based) in the analysis of satellite data.
The ECMWF radiation schemes

28. The ECMWF radiation schemes
Adiabatic processes
Winds Temperature Humidity
Cloud Fraction
Cloud Water
Diffusion Radiation Cumulus
convection
Stratiform
precipitation
Friction Sensible
heat flux
Evaporation
Ground
roughness
Ground
temperature
Snow Ground
humidity
Snow
melt

29. The ECMWF radiation schemes
• Differences with other physical processes
• There exists a well known theory (from Quantum Mechanics to
Spectroscopy to Radiation Transfer).
• Radiation is exchanged with the outside space: radiative balance
determines the climate.
• The sun providing the energy input, radiation undergoes regular
forcings: seasonal, diurnal.
• Radiation at ToA has been globally measured since the 60’s (by
operational satellites), with real flux measurements from ERB (1978),
ERBE (1985), ScaRaB (1993), CERES (1998).
• Surface radiation has been (roughly) measured at points over almost 40
years. Present programs like ARM, BSRN, SURFRAD measure it with high
accuracy. Also satellite-derived SW (and LW) radiation is becoming
available.
• Therefore, there exist some relatively extended possibilities of
validation/verification (radiation in the SW visible and near-IR, in the
LW, … in the mW).

30. The ECMWF radiation schemes
• What is required to build a radiation transfer scheme for
a GCM?
• 5 elements, the last, in principle in any order:
• a formal solution of the radiation transfer equation
• an integration over the vertical, taking into account the
variations of the radiative parameters with the vertical
coordinate
• an integration over the angle, to go from a radiance to a flux
• an integration over the spectrum, to go from monochromatic to
the considered spectral domain
• a differentiation of the total flux w.r.t. the vertical coordinate
to get a profile of heating rate

31. The ECMWF radiation schemes
• In the ECMWF model, the 3-D distributions of T, H2O, cloud fraction
(CF), cloud liquid water (CLW), cloud ice (CIW) are given for every
time-step by the prognostic equations.
• Other parameters, i.e., O3, CO2 and other uniformly mixed gases of
radiative importance (O2, CH4, N2O, CFC-11, CFC-12 and aerosols)
have to be specified (prognostic O3 soon interactive with rad?).
• Prognostic aerosols (as part of GEMS/MACC project)
Radiation black
box
Efficient radiation
transfer
algorithms
Profiles of T,
q, CF, CLW,
CIW,
O3
Climatological
data:
other trace
gases, aerosols
OUTPUT
updated
from
time to time
to be used in
the
thermodynamic
equation
DFLW, DFSW to be
used in the surface
(soil) energy
balance equation
Rad
t
T



32. MACC Daily Service Provision
Air
quality
Global
Pollution
Aerosol UV index
Biomass
burning
http://www.gmes-atmosphere.eu

33. Radiation Transfer in NWP: Lecture 4
Forward modelling of aerosols
• As part of the GEMS/MACC/MACC II projects, the IFS has been modified to include
prognostic aerosols (sea-salt SS, dust DU, organic OM and black carbon BC, sulphate
SO4).
• Sources for SS and DU are linked to some of the model surface parameters (U10,
soil moisture, UVis albedo, stdev orography, snow mask).
• Sources for OM, BC and SO4 are taken from climatologies and/or inventories
(GFEDm, GFED8d, SPEW, EDGAR databases). For NRT FCs, OC, BC and SO4 linked to
fire emissions are linked to an analysis of the MODIS and Geostationary Satellites
“fire hot spots”
• Aerosols are transported by advection, vertical diffusion and convection, and
undergo their specific processes, i.e., sedimentation, dry deposition, wet
deposition by large-scale and convective precipitation, and for OM and BC
hygroscopic effects. Transfer between SO2 and SO4 is handled with a time-scale
simply dependent on latitude.
• The TL159 (GEMS) and the TL255 (MACC) L60 models have been simulating aerosols
for the 2003-2008 (GEMS) and 2003-2010 (MACC)-AER reference period. Since
September 2008, an experimental pre-operational near-real time aerosol analysis
followed by a 5-day FC is produced every day.
• Comparisons with MODIS and AERONET data.
http://www.gems-atmosphere.eu/d/services/gac/nrt/nrt_opticaldepth/

34. Prognostic AERosols in the ECMWF IFS

35. Prognostic AERosols in the ECMWF IFS
• Aerosol model to represent the main characteristics of the 4D distribution of
aerosols, while keeping the computational burden within the parameters of a
future operational configuration.
• Aerosol model formulation originally taken from the LOA/LMD-Z model (Reddy et
al., 2005, JGR), and adapted to the IFS
• Adapted to the ECMWF IFS model dynamics and physics:
• with original developments to include N (=12) new prognostic variables for
the aerosols
• and original developments/upgrades to the sedimentation, wet deposition
and radiative diagnostics.
• Extensive validation against MODIS t550 (aerosol optical depth at 550 nm),
AERONET t500, t865, CALIPSO aerosol/cloud mask
• ECMWF IFS model including prognostic aerosols can be run in two
configurations:
• In aerosol free-wheeling mode: aerosol advection and “full” (but simplified)
aerosol physics using temperature, humidity, winds etc. from the
analyses/forecasts every 12 hours
• In analysis mode with subsequent forecasts

36. Quick overview of the MACC/ECMWF
aerosol analysis and forecasting
system
•
•
•
•
Forward model Analysis

37. Evaluation with MODIS/SEVIRI and AERONET
Saharan dust outbreak: 6 March 2004
Model simulation Assimilation MODIS
SEVIRI
Cape Verde Dakar
AERONET
Assimilation
Simulation
Aerosol optical depth at 550nm (upper)
and 670/675nm (lower)

38. Comparison of GEMS simulated and
analysed aerosol optical depth with
MODIS and MISR for July 2003

39. WMO SDS-WAS programme
Regional Center for
Northern Africa, Middle East
and Europe
http://sds-was.aemet.es
sdswas@aemet.es
• WMO SDS-WAS program
• Dust forecast

40. WMO SDS-WAS program
Mission:
Improve the capacity of countries to produce and deliver to end
users timely and precise atmospheric dust forecasts
Structure:
•Regional Center for Northern Africa, Middle East and Europe.
Barcelona, Spain
•Regional Center for Asia, Beijing, China
•Regional Center for Pan America, Orange, Ca, USA

41. The Regional Center is managed by the Spanish Met. Agency
(AEMET) AND THE Barcelona Supercomputing Center (BSC-CNS)
Nexus II building
Catalonia Tech. University MareNostrum supercomputer
The Regional Center NA-ME-E

42. http://sds-was.aemet.es
sdswas@aemet.es eterradellasj@aemet.es

43. Forecast products

44. MODEL INSTITUTION RUN
TIME
DOMAIN DATA
ASSIMILATION
BSC-
DREAM8b
BSC-CNS 12 Regional No
CHIMERE LMD 00 Regional No
LMDzT-INCA LSCE 00 Global No
MACC ECMWF 00 Global MODIS AOD
DREAM-
NMME-MACC
SEEVCCC 12 Regional MACC analysis
NMMB/BSC-
Dust
BSC-CNS 12 Regional No
MetUM U. K. Met
Office
00 Global MODIS AOD
GEOS-5 NASA 00 Global MODIS
reflectances
NGAC NCEP 00 Global No
Dust models

45. Dust optical depth at 550 nm
RUN: 15 Apr 2013
VALID: 15 Apr 2013 12:00 – 18 Apr 2013 00:00

46. Surface concentration
RUN: 15 Apr 2013
VALID: 15 Apr 2013 12:00 – 18 Apr 2013 00:00

47. Numerical products
netCDF format
SFC
concentration
Dust AOD 550 µm

48. Evaluation with AERONET data
SANTA_CRUZ_TENERIFE

49. Evaluation with satellite products
24 Apr 2013