analysis of two missed summer severe rainfall forecasts zuohao cao 1 and da-lin zhang 2 1...
TRANSCRIPT
Analysis of two missed summer severe rainfall forecasts
Zuohao Cao1 and Da-Lin Zhang2
1Environment Canada, Toronto, Ontario, Canada
2Department of Atmospheric and Oceanic Science, University of Maryland, College Park, USA
Outline
• Motivation and objective
• Data and methodology
• Results and discussions
• Conclusions and recommendations
Motivation
• Summer severe rainfall (SSR) has high impacts on our society and economy especially in highly populated areas such as the Great Lakes regions, but it has low predictability.
• Despite considerable progress in mesoscale numerical weather prediction (NWP), the ability to predict SSR in terms of amount, location, and timing remains very limited due to its association with convective or mesoscale phenomena.
Summer severe rainfall - one of the least understood issues
• Complicated dynamic and thermodynamic processes– Upper-level forcing,
– Boundary-layer forcing,
– Secondary generation
• Integrated elements (e.g., Doswell et al. 1996)– Moisture availability
– Instability
– Lifting
Current status of operational forecasting techniques for SSR
• Conceptual models (partially derived from pattern recognition)
• Persistent nowcasts using observational precipitation amount and phase from approaching upstream weather systems
• Numerical weather prediction (NWP) model based precipitation forecasts
• NWP models often show errors in precipitation amount, location, and timing, mainly due to
– lack of sufficient observations in both space and time, and limited use of these observations,
– missing or inadequate inclusion of physical processes in NWP models,
– our limited understanding of basic atmospheric physical processes.
Why SSR or missed SSR
• Analyses of observations during the last two decades over Ontario, Canada indicate a substantial increase in the frequency of SSR events (e.g., Cao and Ma 2009; Cao 2008).
• However, most of them are difficult to predict accurately, and some of them are completely missed.
• In this study, we examine two representative missed SSR events.– in Ottawa on 24 July 2009, associated with three cyclones
– in Hamilton on 26 July 2009, a quasi-stationary low-pressure system
– identified with inputs from a senior meteorologist (Robinson 2014, personal communication) of OSPC of EC.
Objective
• To assess the model ability in forecasting SSR events
• To provide evidence for endorsements on continuously improving NWP model initial conditions and simulating SSR events using cloud-permitting NWP models
• These efforts are relatively new since we are examining SSR events that were missed operationally whereas most of published case studies (if not all) look at “successful” events.
Definition of SSR in Ontario
As defined by the Ontario storm prediction center, a SSR event is considered when
•its rainfall rate exceeds 50 mm d-1 or
•75 mm in 48 h in Ontario, Canada.
Data
• Hourly operational GEM regional model forecast data with a 15-km horizontal resolution and 58 vertical levels (Cote et al. 1998)
• 3 hourly NARR data with a horizontal resolution of 32-km and 29 constant pressure levels (Mesinger et al. 2006)
• Daily rain gauge data plus a small number of the hourly rainfall data (the national/Ontario climate center 2005)
• Radar estimated rainfall accumulation and rainfall rates at 10-min intervals (the Canadian national radar network 2005)
• Observed soundings (the University of Wyoming 2014)
Methodology
• The GEM regional model predicted rainfall is verified against rain gauge observations, extracted from the Canada Climate Center’s archived data using spreadsheet software, and radar observations, obtained from the Canadian radar network data using the URP (Unified Radar Processor) software.
• Synoptic environments associated with the SSR are analyzed by comparing the GEM regional model forecast and the NARR fields such as MSLP and geopotential height.
• The GEM regional model-predicted cloud coverage and precipitable water (a threshold of 35 kg m-2 used for operationally forecasting severe rainfall; see Johnson and Moser 1992; Mainville 2004) are assessed with the NARR.
• The model-predicted CAPE and soundings are evaluated against the observed CAPE and soundings obtained from the University of Wyoming as well as the NARR.
Mean sea level pressure at 03 UTC 24 July 2009
(a) NARR
(b) GEM regional model forecast
Comparison of 24-h rainfall accumulation in Ottawa city (45.42oN, 75.70oW) and its vicinity between observation available on July 24, 2009 and GEM regional model forecast
* It is calculated based on the distance between the surface station and the Parliament Hill of Ottawa city.
SourceLocation or
station name (ID)
Distance from the
measurement point to
Ottawa (km)*
24-h rainfall accumulation
(mm)
RadarFranktown, Ontario - 75-100
Rain gaugeOttawa CDA
RCS(ID: 6105978)
4.71 67.2
GEM regional model forecast
- - 5
GEM regional model predicted rainfall accumulation (00 to 04 UTC)
(a) Grid-resolvable scheme
(b) Parameterized convection scheme
Tracks of three surface cyclones (A, B, and C)
(a) NARR
(b) GEM regional model forecast
A
B
Ottawa
C
0
0
0
3
3
3
6
6
6
9
9
9
18 21 24
12(15)
1215
1821
24
(a)
40N
35N
45N
50N
55N
A
B
C
Ottawa
0
0
0
3
3
3
6
6
6
9
9
9
12(15)
1215
182124
18
21(24)
(b)
70W80W90W100W
40N
35N
45N
50N
55N
700-hPa omega ω fields at 03 UTC 24 July 2009
(a) NARR
(b) GEM regional model forecast
Geopotential height (in decameter) at 03 UTC 24 July 2009
Left: NARR Right: GEM
500 hPa
700 hPa
1000 hPa
Comparison of cloud coverage between the NARR and the GEM regional model forecast for 3 cases in Ottawa (45.42oN, 75.70oW) and Hamilton (43.25oN, 79.87oW)
Cloud coverage (%)
03 UTC 24 July 2009,
Ottawa
15 UTC 24 July 2009,
Ottawa
15 UTC 26 July 2009, Hamilton
LowLevel
NARR60-80 60-80 60
GEM regional model
> 85 100 90
MiddleLevel
NARR60-80 > 60 50
GEM regional model
28 27 30
High level
NARR> 80 80-100 80
GEM regional model44 0 0
Cloud coverage (%) at 03 UTC 24 July 2009 (NARR: colors; GEM: solid lines)
(a) High level
(b) Low level
Condensation heating in Ottawa at 03 UTC 24 July 2009
Comparison of CAPE (J kg-1) between the NARR and the GEM regional model forecast for 3 cases in Ottawa (45.42oN, 75.70oW) and Hamilton (43.25oN, 79.87oW)
Data source 03 UTC 24 July 2009, Ottawa
15 UTC 26 July 2009, Hamilton
NARR 690.5 672.0
GEM regional model 148.8 0.0
Atmospheric soundings (T: red; Td: green) at
03 UTC 24 July 2009
(a) NARR
(b) GEM regional model forecast
Over-mixing in the model
• McTaggart-Cowan and Zadra (2015) found that excessive vertical mixing in the planetary boundary layer (PBL) scheme is responsible for the model prediction error in temperature about 10oC for a freezing rain event occurred on 22 March 2007 over the Ontario and Quebec regions, contributing to late issuing the freezing rain warnings.
• By introducing Richardson (Ri) number hysteresis in the PBL scheme, they improved the model-predicted temperature for this case through suppressing the generation of turbulent kinetic energy (TKE) and mixing thereby.
• Their scheme has a positive impact on this stable winter case, but it shows relatively little impact on summer cases due to the prevalence of unstable PBL (McTaggart-Cowan and Zadra 2015).
Mean sea level pressure of NARR on 26 July 2009
(a) 00 UTC
(b) 24 UTC
Highway flash flooding caused by SSR occurred in Hamilton on July 26, 2009
Parking lot and basement flash flooding caused by SSR occurred in Hamilton on July 26, 2009
Comparison of 24-h rainfall accumulation around Hamilton city (43.25oN, 79.87oW) between observation available on July 26, 2009 and GEM regional model forecast
Source Location24-h rainfall accumulation
(mm)
Radar King City, Ontario
50-75
GEM regional model forecast
- 5-10
The rainfall rate (mm hr-1) on 26 July 2009
Left: Radar Right: GEM
15 UTC
16 UTC
17 UTC
700-hPa omega ω fields at 15 UTC 26 July 2009
(a) NARR
(b) GEM regional model forecast
Comparison of cloud coverage between the NARR and the GEM regional model forecast for 3 cases in Ottawa (45.42oN, 75.70oW) and Hamilton (43.25oN, 79.87oW)
Cloud coverage (%)
03 UTC 24 July 2009,
Ottawa
15 UTC 24 July 2009,
Ottawa
15 UTC 26 July 2009, Hamilton
LowLevel
NARR60-80 60-80 60
GEM regional model
> 85 100 90
MiddleLevel
NARR60-80 > 60 50
GEM regional model
28 27 30
High level
NARR> 80 80-100 80
GEM regional model44 0 0
Cloud coverage (%) at 15 UTC 26 July 2009 (NARR: colors; GEM: solid lines)
(a) High level
(b) Low level
Comparison of CAPE (J kg-1) between the NARR and the GEM regional model forecast for 3 cases in Ottawa (45.42oN, 75.70oW) and Hamilton (43.25oN, 79.87oW)
Data source 03 UTC 24 July 2009, Ottawa
15 UTC 26 July 2009, Hamilton
NARR 690.5 672.0
GEM regional model 148.8 0.0
Conclusions
• Results reveal the following limitations of the GEM regional model in predicting SSR events: – the model predicted rainfall is phase-shifted to an undesired
location that is likely caused by the model initial condition errors;– the model is unable to resolve the echo training process due to the
weakness of the parameterized convection and/or coarse resolutions.
• These limitations are reflected by the ensuing model-predicted features: – vertical motion in the areas of SSR occurrence is unfavorable for
triggering parameterized convection and grid-scale condensation; – convective available potential energy is lacking for initial model
spin up and later for elevating latent heating to higher levels through parameterized convection, giving rise to less precipitation;
– the conversion of water vapor into cloud water at the high and middle levels is underpredicted.
Recommendations
• The Ottawa-24 July SSR event primarily involved with prediction errors in location, which are likely associated with errors in the model initial conditions. – Continuously improving the model initial conditions is very helpful
in accurate prediction of SSR events through taking more observational data into consideration in data assimilation processes, such as more radar data and the data from non-meteorological organizations (e.g., conservation authority).
– To trigger the KF scheme at right locations and times, we need right sign and magnitude of Wt, the threshold grid-scale vertical velocity. More accurate initial conditions may be helpful to avoid the model predicted Wt having an incorrect sign and magnitude.
– It is suggested that Wt needs to be tuned based on not only the model horizontal but also the model vertical resolution (in the current GEM regional model Wt is tuned based on the model horizontal resolution only), because Wt is dependent upon model resolution, and there is a consistence requirement between the model horizontal and vertical resolutions.
Recommendations
– To initiate and maintain convection, CAPE is needed for a parcel to continue rising vertically after its initial displacement. However, over-mixing in PBL leads to low CAPE in the model. ▪ McTaggart-Cowan and Zadra (2015)’s Richardson (Ri) number
hysteresis in the PBL scheme is mainly effective for some winter cases due to the stable PBL assumption. ▪ So far there is no successful scheme for SSR in operational
NWP models (including the GEM model). ▪ One of possibility is to use a variational method (e.g., Xu and
Qiu 1997; Cao and Ma 2005; Cao et al. 2006; Cao and Ma 2009) to retrieve temperature and moisture profiles in the PBL when sensible and latent heat flux observations are available. The principle of this variational approach is to minimize the differences between the computed and the observed fluxes so that it can adjust the computed profiles toward the “true” value.
Recommendations
• The Hamilton-26 July SSR event was associated with the quasi-stationary low-pressure system, with the most intense rainfall being generated through the echo training process in a modest CAPE environment. –The echo training process may require the use of high-resolution cloud-permitting NWP models to simulate the periodic initiation and subsequent propagation of convective cells along the same path, which is difficult for most of operational NWP models. –For example, Lavers and Villarini (2013) recently examined the ability of the world’s most advanced weather forecasting models to predict the Sept. 9-16, 2013 extreme rainfall that caused severe flooding in Boulder, Colorado. They found that these models tended to underestimate rainfall amounts and placed the rainfall in the wrong area, even though they provided an indication that a period of heavy rainfall was going to affect parts of Colorado.
Recommendations
– On the other hand, information on the NWP errors in precipitation forecasts may be used to minimize NWP model bias.
– The proper physical packages (especially dealing with condensation and precipitation processes) matched with high resolution NWP models are much more important than increasing model resolutions only.
– Since precipitation is an end product of multi-scale interactions, there will be always uncertainties in deterministic forecasts. As an alternative, ensemble-based NWP predictions might provide some uncertainties for SSR forecasts, e.g.,
▪ National Center for Atmospheric Research 3 km ensemble forecasting system (http://www.image.ucar.edu/wrfdart/ensemble/index.php) and
▪ Storm Prediction Center storm-scale ensemble of opportunity (http://www.spc.noaa.gov/exper/sseo/).
Acknowledgements
• Glenn Robinson, a senior forecaster of OSPC of EC, for his inputs in identifying missed SSR events over Ontario of Canada,
• Dave Patrick, for his practical and useful help to solve various problems associated with the scripts and codes for running the radar URP programs,
• Mark Alliksaar and Helen Yang for their efforts to run the URP programs to generate the radar observed precipitation accumulation,
• Victor Chung and Rob Kuhn for their carefully reading part of the early version of this work and providing constructive suggestions, and
• William Burrows and Norman Donaldson for their constructive comments on the part of early version of this work.
• We appreciate Lin Zhu’s assistance for plotting some figures used in
this work.