Introduction to downscaling Jess Fernndez Jesus.Fernandez@unican.es - - PowerPoint PPT Presentation

▶
introduction to downscaling
SMART_READER_LITE
LIVE PREVIEW

Introduction to downscaling Jess Fernndez Jesus.Fernandez@unican.es - - PowerPoint PPT Presentation

Sep 26, 2022 625 likes •1.14k views

http://www.meteo.unican.es A multidisciplinary approach to weather & climate Santander Meteorology Group A multidisciplinary approach for weather & climate Introduction to downscaling Jess Fernndez Jesus.Fernandez@unican.es Grupo

slide-1
SLIDE 1

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Introduction to downscaling

SPECS Workshop on seasonal forecasting: Data access, BC & DS 8-12 Sep 2014, Santander

Jesús Fernández

Jesus.Fernandez@unican.es

Grupo de Meteorología de Santander Universidad de Cantabria

  • Dept. Matemática Aplicada y CC Comp.
slide-2
SLIDE 2

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Outline

Introduction to downscaling

  • Downscaling: why?
  • Approaches: dynamical and statistical DS
  • Pros and cons
  • Seasonal perspective
  • Link to bias correction
slide-3
SLIDE 3

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Outline

Introduction to downscaling

  • Downscaling: why?
  • Approaches: dynamical and statistical DS
  • Pros and cons
  • Seasonal perspective
  • Link to bias correction

FOCUS Lessons learnt from long-term RCM simulations

slide-4
SLIDE 4

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Downscaling: why?

Global model Resolution: 3.75° x 3.75°

(T30)

slide-5
SLIDE 5

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

The World as seen by a GCM

T30

slide-6
SLIDE 6

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

The World as seen by a GCM

Global model Resolution: 2.5° x 2.5°

(T62)

slide-7
SLIDE 7

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

The World as seen by a GCM

T62

slide-8
SLIDE 8

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

The World as seen by a GCM

T62

slide-9
SLIDE 9

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es 75 km (~ERA-Interim / System4)

slide-10
SLIDE 10

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es 75 km (~ERA-Interim / System4)

slide-11
SLIDE 11

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es 75 km (~ERA-Interim / System4)

slide-12
SLIDE 12

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Downscaling

  • Downscaling techniques try to adapt

the coarse global model output to the local features of a given region.

slide-13
SLIDE 13

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Downscaling

  • Downscaling techniques try to adapt

the coarse global model output to the local features of a given region.

  • This can be achieved through

dynamical techniques, which solve numerically the governing equations

  • f the atmosphere on a finer grid,

The most common are nested RCMs

slide-14
SLIDE 14

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

The World as seen by an RCM

45 km

slide-15
SLIDE 15

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Not seen by a GCM / RCM

slide-16
SLIDE 16

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Physical parameterizacions

A parameterization is a statistical representation of the net effect of processes occurring on spatial scales smaller than the grid spacing of a dynamical model (GCM, RCM, CRM, LES, …) over mean variables at each grid cell. Parameterizations are based on the physics of the processes, plus simplifying (closure) assumptions to relate unknown variables to prognostic (mean) model variables. Parameters are obtained from:

  • Theory (e.g. known physical constants)
  • Field campaigns
  • Higher resolution models (CRM, LES)
slide-17
SLIDE 17

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Downscaling

  • Downscaling techniques try to adapt

the coarse global model output to the local features of a given region.

  • This can be achieved through

dynamical techniques, which solve numerically the governing equations

  • f the atmosphere on a finer grid,
  • or by statistical techniques, which

seek empirical relationships between local and large-scale variables.

slide-18
SLIDE 18

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

General classes of downscaling Local climate = f (larger scale predictors) + locally forced variance Dynamical Two approaches Empirical-statistical Three main classes Perturbed observed RCM Hi-res GCM Weather Generators Transfer Functions

Trained on long term time series and atmospheric re-analysis data Conditioned by GCM parameters to capture low frequency variance Trained on time series that spans range of variability, and atmospheric re-analysis data Residual local scale variance added stochastically

Index / analogues

Requires long term data sets and uses weather typing or historical analogues

Source: Bruce Hewitson (CSAG)

DS techniques

slide-19
SLIDE 19

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Statistical downscaling approaches

Perfect Prognosis (PP) Calibrated in the training phase using observational data for both the predictands and predictors (reanalysis). Since different GCMs are used in the training and downscaling phases, large-scale circulation variables well-resolved by the models are typically chosen as predictors in this approach. Variables directly influenced by model parameterizations and

  • rography are not suitable predictors in this approach.

PP Training Downscaling Large scale Obs. GCM Local scale Obs. →

slide-20
SLIDE 20

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

1960 1970 1980 1990 2010 2020 2030 2070 2080 2090

…

2000

Present Climate Future Observations

Spain02, 20km

… ………………… … ………………. GCM scen.

AR4 ~250km

Control scenario: 20c3m B1, A1B, A2 … ………………. GCM reanal.

ERA40, 250km

… ……………….

Precip

Projections

Spain02, 20km

… ………………… SDM Precip = 0.8 MSLP + 1.2 Q850 ………………… … SDM Precip = 0.8 MSLP + 1.2 Q850

MSLP, Q850, etc.

Precip = 0.8 MSLP + 1.2 Q850 … …………………

Statistical model

SDM

  • Assumption 2: Choosing consistent predictors:
  • Assumption 1: Reanalysis choice
  • Assumption 3: Stationarity/robustness: SDM SDM

Statistical downscaling (Perfect Prog)

slide-21
SLIDE 21

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Typical predictors

slide-22
SLIDE 22

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es Grids of atmospheric patterns

for a given day n

Predictands: precip., etc.

for a given day n

Yn

(T(1ooo mb),..., T(500 mb);

Z(1ooo mb),..., Z(500 mb); .......; H(1ooo mb),..., H(500 mb)) = Xn

Linear regression:

Yn= a Xn+ b Logistic regression Probabilistic prediction Yn= F(a Xn+ b)

1. Transfer Functions

  • 1. Transfer functions
slide-23
SLIDE 23

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Variable Nivel Hora Geopotencial 500 0 UTC Geopotencial 1000 0 UTC Temperatura 500 0 UTC Temperatura 850 0 UTC Humedad Relativa 850 0 UTC

140 parameters (5 variables, 28 gridboxes), n=16434

Redundancy (correlation): Principal Components (domain selection) Nearest grid-boxes Both (e.g. 15 CPs + values in 1 gridbox) Predictor selection

slide-24
SLIDE 24

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Statistical downscaling approaches

Model Output Statistics (MOS) Predictors are taken from the global (or regional) model for both training and downscaling phases. They require the model

  • utput to have day-to-day correspondence with observations.

These methods can work with the variable of interest as

  • predictor. For instance, local precipitation can be derived from

the direct model precipitation forecasts.

MOS Training Downscaling Large scale GCM GCM Local scale Obs. →

slide-25
SLIDE 25

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

SDS: Classification

Source: SPECS D52.1

slide-26
SLIDE 26

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

S2D statistical DS

Source: SPECS D52.1

slide-27
SLIDE 27

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Pros and cons

Statistical downscaling  Variables available as long as there are obs.  Different variables probably do not keep physical or even spatial consistency  Variables keep the representativity of obs. ☺ Computationally cheap ☺ Biases are low ☺ Non-meteorological variables (e.g. impact indices) could be directly produced.

slide-28
SLIDE 28

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Pros and cons

Dynamical downscaling ☺ Plenty of variables, including 3D ☺ Subdaily data available ☺ Variables are physically consistent  Variables represent areal averages  Computationally expensive  Biases are commonplace

slide-29
SLIDE 29

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Seasonal scale considerations

Statistical downscaling ☺ Full hindcasts with many members easily DS'd ☺ No stationarity problem  Need to handle seasonal forecast drift  Weak large-to-local scale relationships and poor

training data where seasonal forecast skill is high.

Dynamical downscaling  Full hindcasts are costly  Biases still present ☺ Physical process analysis of seasonal anomalies.

Role of regional feedbacks.

slide-30
SLIDE 30

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Added value

Source: Nikulin et al. (2012)

  • J. Clim. 25:6057-6078
slide-31
SLIDE 31

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Potential to narrow uncertainty

Source: SMHI

RCM GCMs

RCA4 nested into 8 ESMs

slide-32
SLIDE 32

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Potential to narrow uncertainty

RCM GCMs

Not always the case. This is a seasonal forecast example, using ECMWF System3 (left) and downscaled by RegCM3 (right). Source: Diro et al, 2012

JGR 117:D16103

slide-33
SLIDE 33

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

DD seasonal skill (pr T1 SON)

ECMWF System3 (1981-2003) ECMWF System3 + RCA5 Source: Díez et al, 2011

Tellus A 63:757

slide-34
SLIDE 34

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

DD seasonal skill

Source: Diro et al, 2012

JGR 117:D16103

slide-35
SLIDE 35

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

1960 1970 1980 1990 2010 2020 2030 2070 2080 2090

…

2000

Present Climate Future Observations

E-OBS, 25km

… ………………… … ………………. GCM long term

AR4 ~250km

… ………………. Control simulations (20C3M) Scenarios (B1,A1B,A2)

Dynamical downscaling (Climate change)

GCM reanal.

ERA40, 250km

… ……………….

day-to-day Correspondence

slide-36
SLIDE 36

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

1960 1970 1980 1990 2010 2020 2030 2070 2080 2090

…

2000

Present Climate Future Observations

E-OBS, 25km

… ………………… … ………………. GCM long term

AR4 ~250km

… ………………. RCMs

  • ENSEM. 25km

… ………………… Control simulations (20C3M) Scenarios (B1,A1B,A2) GCM reanal.

ERA40, 250km

… ……………….

day-to-day Correspondence

RCM VALIDATION OPTIMAL CONDITIONS

Dynamical downscaling

slide-37
SLIDE 37

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

1960 1970 1980 1990 2010 2020 2030 2070 2080 2090

…

2000

Present Climate Future Observations

E-OBS, 25km

… ………………… … ………………. GCM long term

AR4 ~250km

… ………………. RCMs

  • ENSEM. 25km

… ………………… RCMs

  • ENSEM. 25km

… ………………… Control simulations (20C3M) Scenarios (B1,A1B,A2) GCM reanal.

ERA40, 250km

… ……………….

day-to-day Correspondence

RCM VALIDATION OPTIMAL CONDITIONS

Dynamical downscaling

slide-38
SLIDE 38

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

1960 1970 1980 1990 2010 2020 2030 2070 2080 2090

…

2000

Present Climate Future Observations

E-OBS, 25km

… ………………… … ………………. GCM long term

AR4 ~250km

… ………………. RCMs

  • ENSEM. 25km

… ………………… RCMs

  • ENSEM. 25km

… ………………… Control simulations (20C3M) Scenarios (B1,A1B,A2) GCM reanal.

ERA40, 250km

… ……………….

day-to-day Correspondence

RCM VALIDATION OPTIMAL CONDITIONS

Dynamical downscaling

VALIDATION RCM-GCM

slide-39
SLIDE 39

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

1960 1970 1980 1990 2010 2020 2030 2070 2080 2090

…

2000

Present Climate Future Observations

E-OBS, 25km

… ………………… … ………………. GCM long term

AR4 ~250km

… ………………. RCMs

  • ENSEM. 25km

… ………………… … ………………… RCMs

  • ENSEM. 25km

… ………………… Control simulations (20C3M) Scenarios (B1,A1B,A2) GCM reanal.

ERA40, 250km

… ……………….

day-to-day Correspondence

RCM VALIDATION OPTIMAL CONDITIONS

Dynamical downscaling

VALIDATION RCM-GCM

slide-40
SLIDE 40

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es PROJECTION DELTA-METHOD

1960 1970 1980 1990 2010 2020 2030 2070 2080 2090

…

2000

Present Climate Future Observations

E-OBS, 25km

… ………………… … ………………. GCM long term

AR4 ~250km

… ………………. RCMs

  • ENSEM. 25km

… ………………… … ………………… RCMs

  • ENSEM. 25km

… ………………… Control simulations (20C3M) Scenarios (B1,A1B,A2) GCM reanal.

ERA40, 250km

… ……………….

day-to-day Correspondence

RCM VALIDATION OPTIMAL CONDITIONS

Dynamical downscaling

slide-41
SLIDE 41

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

GCM-RCM coupling

Turco et al. (2013) “Large biases and inconsistent CC signals in ENSEMBLES …” CC 120:859

slide-42
SLIDE 42

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

RCM biases

García-Díez et al., 2012

  • Biases depend on location and season.
  • They also depend on the time of the day

(maximum/minimum temperature)

slide-43
SLIDE 43

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Temperature biases

García-Díez et al., 2012

  • Biases depend on location and season.
  • They also depend on the time of the day

(maximum/minimum temperature)

  • Seasonal forecasting: dependence on forecast (lead) time
  • How does the drift propagate into the RCM?
  • RCM drift? (e.g. associated to soil initialization)
slide-44
SLIDE 44

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

DJF tasmin

García-Díez et al., 2012

Biases can, in general, affect differently mean and extreme values.

slide-45
SLIDE 45

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

JJA tasmax

García-Díez et al., 2012

This is an example of a very systematic bias along the whole temperature range.

slide-46
SLIDE 46

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

Common problem in RCMs

Qq-plots of ENSEMBLES RCMs in central Europe from Plavcova & Kysely (2011)

slide-47
SLIDE 47

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

RCM pr biases

Source: Ana Casanueva, UC

slide-48
SLIDE 48

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

RCMs vs SDMs

DJF mean precip pattern (unfair comparison)

Source: Ana Casanueva, UC

slide-49
SLIDE 49

A multidisciplinary approach to weather & climate

Santander Meteorology Group

A multidisciplinary approach for weather & climate

http://www.meteo.unican.es

RCMs vs SDMs

90pWET CDD

Source: Ana Casanueva, UC