Clara Arbizu-Barrena, David Pozo-Vázquez, José A. Ruiz-Arias, Joaquín Tovar-Pescador
SOLAR RADIATION AND
ATMOSPHERE MODELLING GROUP (MATRAS)
DEPARTMENT OF PHYSICS UNIVERSITY OF JAEN
SPAIN University of Jaén
Spain
16TH WRF USER WORKSHOP, BOULDER, JUNE 2015
ASSESSMENT OF THE CAPABILITY OF WRF MODEL TO ESTIMATE CLOUDS AT DIFFERENT TEMPORAL AND SPATIAL SCALES
Mo#va#ons of this work
Ø
Improvement of solar radiation forecasting reliability: a
key issue for solar energy grid integration
Ø
Reliability of the cloud forecasts is the most important
factor that limits this accuracy
Ø
Scarce works evaluating the reliability of the WRF cloud
estimates at high temporal and spatial resolution (i.e. site
locations), needed in solar energy applications
Ø
Evaluation studies are important to improve cloud
representation in the WRF
Motivations of this work
Evaluation of the cloud representation reliability in the WRF model:
a complex task
Ø
Clouds are characterized by microscopic and macroscopic
parameters
Ø
Different types of clouds an cloud-related processes
Ø
Clouds parameters difficult to measure
Ø
Discrete nature of clouds: double penalty effect
Ø
The evaluation of the cloud representation in WRF involves
the analysis of the role of:
•
Microphysics, cumulus and PBL parameterizations
•
Cloud fraction models
•
Cloud overlapping approaches
Aims of this work
In this work we aim to evaluate the role of the:
Ø
Microphysics parameterizations
Ø
Cloud fraction models
Ø
Cloud overlapping approaches
Ø
Spatial and temporal scales
..in the reliability of the WRF model cloud macroscopic
characteristics representation, i.e.:
Ø
Cloud occurrence
Ø
CBH and CTH
Ø
Cloud fraction
Evalua#on loca#on and data
UNIV. JAEN METEO STATION:
Ceilometer (Jenoptik 15k-Nimbus)
Ø CBH and CTH estimates based on LIDAR technique Ø Up to 5 cloud layers simultaneously
Ø Accuracy of ±5 m, range 5 m to 15 km
Ø Measurement every 15 seconds: 5 minutes average
TSI-880 Sky camera
Ø Hemispheric cloud cover
measurements every 30 seconds Study period: 21 days along 2013 Ø Different types of sky conditions
WRF set up
1. GFS initial and boundary conditions
2. 50 vertical levels
3. 24 hours spin-up
4. outputs saved every 5 minutes
5. 4 nested domains 34,
12, 4 and 1.3 km (evaluated)
Acronym Microphysics Scheme Reference
WSM6 WRF Single-‐Moment 6-‐class scheme [Hong and Lim, 2006]
THOM New Thompson et al. scheme [Thompson et al., 2008]
MILB Milbrandt-‐Yau Double-‐Moment 7-‐class
scheme
[Milbrandt and Yau, 2005]
MORR Morrison double-‐moment scheme [Morrison et al., 2009]
SBLI Stony Brook University (Y. Lin) scheme [Lin and Colle, 2011]
NSSL NSSL 2-‐moment scheme [Mansell et al., 2010]
Other physics prescribed for all the simulations:
• YSU PBL (Hong et al., 2006), RRTMG short- and long-wave radiation (Iacono et al., 2008), Noah land surface parameterization (Tewari et al., 2004).
• The parameterization for the cumulus scheme was disabled
Cloud-fraction (CF) is used to verify cloud structures simulated
by the WRF against ground observations (ceilometer and sky
camera).
Two CF parameterizations have been evaluated:
1: Binary CF (BCF), based on a threshold over the cloud
liquid water and ice mixing ratios. Only values 0 and 1 are
allowed
2: Xu and Randall [1996] CF (XCF), continuous CF value
between 0 and 1 are allowed.
WRF modeled cloud fraction, cloud occurrence, CBH, CTH
and cloud cover
Cloud occurrence in the model is here considered whenever the
modeled CF >0
The WRF-modeled CBH (CTH) estimates are derived from the
height of the lowest (highest) model layer with CF>0
Modeled cloud cover is derived from the CF values using a
cloud overlapping scheme. Here, we have evaluated 3:
1. maximum overlap,
2. random overlap,
3. maximum-random overlap.
WRF modeled cloud fraction, cloud occurrence, CBH, CTH
and cloud cover
10
ceilometer Sky camera image
12km 4km 1.3 km
12 km
4 km CBH
Evaluation procedure
Cloud occurrence:
contingency table
Frequency bias:
𝐹𝐵
=
𝐴
+
𝐵/𝐴
+
𝐶
Cloud occurrences predicted by WRF divided by the total number of cloud occurrences reported by the ceilometer. Perfect model FB=1.
CBH, CTH and cloud cover
Ceilometer Y N WRF Y A B N C D
Evaluation procedure
6 microphysics parameterization 2 cloud fraction models
3 cloud overlapping methods 3 spatial resolutions: 12, 4 and 1.3 km
1. Cloud occurrence
2. CBH
3. CTH
4. Cloud cover
Temporal resolution: aggregations starting at 5 minutes to 6 hours.RESULTS: WRF CLOUD OCCURRENCE PREDICTION SKILL
• WRF over-predict the number of observed cloud occurrence events, except for low level clouds at 4 and 1 km
• FB values for high-level clouds are considerably higher than for middle- and low-level clouds.
• All MPs performs similarly, except WSM6 , that shows the best FB.
• FB slightly better for the 4 and 1.3 km resolutions, caused by the low levels clouds
RESULTS: CBH AND CTH PREDICION SKILL
5 minutes samples. XCF
• Overall, scarce dependence of the results on MPs, except for the BIAS • Spa\al resolu\on only important for low level clouds BIAS
• Model tends to yield too low CBHs and too high CTHs, irrespec\vely of the cloud level considered. Thus, it tends to produce thicker clouds than the observed ones.
RESULTS: CBH and CTH prediction skill
5 minutes samples. XCF
High level clouds
• Model systematically underestimates the CBH of high-level
clouds by ≈1100 m, regardless MP and the domain spatial
resolution (model locates cloud bases below observed values)
• Contrarily, CTH of high-level clouds are overestimated (≈700 m )
• As a consequence, the WRF-modeled high-level clouds appear
RESULTS: CBH AND CTH PREDICTION SKILL
5 minutes samples. XCF
Low level clouds
• Low-level clouds shows lower BIAS and RMSE values for both
CBH and CTH compared to high and middle level clouds
(vertical resolution!!)
• Significant dependence of the BIAS on the spatial resolution.
• CBH RMSE lower for 4 and 1.3 km
RESULTS: WRF CLOUD COVER PREDICTION SKILL
• WRF tend to over-predict cloud fraction, positive bias
• In general, 4 and 1 km experiments, more reliable cloud cover estimates. • XCF: WSM6/NSSL MPs best/worst estimates
• BCF: lower RMSE values, MORR the best performing MPs • Differences in RMSE values are lower than 10%
• There is little dependence on the choice of CF overlapping scheme Evalua\on carried out in an area of 12 × 12 km centered at the sta\on loca\on
RESULTS: WRF CLOUD COVER PREDICTION SKILL
Time aggrega\on experiment
The modeled and observed cloud covers are averaged for aggrega\ng \me intervals in the range from 5 minutes to ~5 hours, by 5 minutes \me increments
• For shorter averaging time intervals, the experiments with finer spatial resolutions provide lower RMSE values.
• As the averaging time interval increases, RMSE decreases at a rate of ≈0.03 cloud cover unit per hour.
• For averaging time intervals longer than 4 hours, the RMSE decreasing rate slows down and RMSE does not appear to depend on the spatial resolution anymore • BCF lower RMSE values than XCF.
SUMMARY
Cloud occurrence
1. WRF over predicts cloud occurrence of high-level clouds while tends to under-predict the cloud occurrence of low-level clouds for the domains with 4 and 1 km cell spacing.
2. Better prediction skill of the 4 and 1.3 km experiments specially for low level clouds
3. Scarce role of the MPs and the cloud fraction parameterization
CBH and CTH
1. Model tends to yield too low CBHs and too high CTHs, irrespectively of the cloud level considered. Thus, it tends to produce thicker clouds than the observed ones.
2. The role of the domain spatial resolution has proven to be only important for low-level clouds, with decreasing CBH error for increasing spatial resolution.
3. The choice of MPs has little influence in the model performance, except for high-level clouds.
SUMMARY
Cloud cover
1. The model tends to over-predict cloud cover and produce estimates with RMSE values of ≈0.5 cloud cover unit.
2. 4 km and 1 km experiments higher reliability 3. Better performance of the WSM6 MPs
4. Scarce role of the cloud overlapping schemes
5. Temporal aggregation analysis has shown a nearly linear decrease of RMSE as the size of the averaging window increases.
6. Maximum WRF reliability has been observed for averaging time intervals longer than 4 hours. RMSE reduces from about 0.48 to 0.35.
Arbizu-‐Barrena et al.,
Clara Arbizu-‐Barrena, David Pozo-‐Vázquez, José A. Ruiz-‐Arias, Joaquín Tovar-‐Pescador
SOLAR RADIATION AND
ATMOSPHERE MODELLING GROUP (MATRAS)
DEPARTMENT OF PHYSICS UNIVERSITY OF JAEN
SPAIN University of Jaén Spain
16TH WRF USER WORKSHOP, BOULDER, JUNE 2015
ASSESSMENT OF THE CAPABILITY OF WRF MODEL TO ESTIMATE CLOUDS AT DIFFERENT TEMPORAL AND SPATIAL SCALES
Arbizu-‐Barrena et al., Under review J. Geophys. Res. Atmos.