1. Investigating AMV and NWP model errors using the NWP SAF monitoring
James Cotton
Met Office, FitzRoy Road, Exeter, EX1 3PB, United Kingdom
The NWP SAF (Satellite Application Facility for Numerical Weather Prediction) atmospheric motion vector (AMV) monitoring, aims to help better characterise AMV errors in order to aid improvements to the AMV derivation and their assimilation in NWP models. One of the ways this can be achieved is via long term monitoring of trends and patterns in observed minus background (O-B) statistics. The NWP SAF maintains an archive of O-B statistics against both the Met Office and ECMWF global model backgrounds, providing a framework in which we can attempt to separate error contributions. Differences between centres suggest a dependence on model error whereas similarities suggest either problems with the AMVs or problems shared by the models.
GOES winter negative speed bias over NE America
Background: A negative wind speed bias observed at low level over the Eastern US during boreal winter. The bias has previously been linked to observations over land with a ‘high’ height bias (cloud base height assignment only applied over sea) and is more pronounced versus the Met Office model.
Update: For winter 2012/13 the bias is much more extensive and spreads out over the Atlantic e.g. January 2013
High troposphere (above 180 hPa) speed bias in Tropical Easterly Jet (TEJ)
Background: A negative speed bias for Meteosat-7 and MSG winds in the high-troposphere of the tropics between June - Sept. This seasonal feature has been shown to coincide with the presence of the TEJ but it has not previously been investigated whether this is due to model or observation error.
Update: The plots below show an example from August 2013 for Meteosat-7 WV winds above 200 hPa height. The AMVs describe a much weaker easterly with O-B speed differences widely greater than 5 m/s (locally 8 m/s) versus the Met Office background. The ECMWF O-B’s show a similar pattern but the bias is smaller in magnitude. Could there be a contribution from model error?
GOES-13 IR operational
GOES-13 IR hourly
Bias much reduced over sea in ‘hourly’ GOES product
Lack of cloud-base height assignment over land
Mean observed vector/spd
Mean UKMO vector/spd
UKMO: vector/spd diff
ECMWF: vector/spd diff
When inversion detected over sea, AMVs now assigned to base of inversion layer
Over sea: model best-fit pressure suggest too high by ~50 hPa
Comparing mean Met Office and ECMWF analyses at 150 hPa reveals very large wind speed differences between the models. The Met Office analysed winds are up to 10 m/s faster in the same region as the AMV bias. At 200 hPa the sign reverses and ECMWF analyses are faster by up to 6 m/s.
Improved bias in hourly product allows us to focus on smaller signal e.g. Dec 2012
GOES-13 IR hourly
Mean OSTIA SST analysis
Distinctive curl in bias around 40°W which appears to match the position of the Gulf Stream SST front.
Is there a link?
Speed bias limited to narrow region along coast
150 hPa
200 hPa
Error in the vertical wind distribution?
Jet core - UKMO faster and extends further south
Some displacement in the vertical (e.g.10 m/s contour)
Case study: 30 December 2012, 12:00 UTC
Forecast shows a surface low tracking NE-wards along east coast, with cold air being advected south over the warm waters of the Atlantic consistent with a synoptic-scale cold air outbreak.
GOES-13 imagery shows clear skies immediately along the coast with cloud formation ~50 km offshore, forming into cloud streets.
IR AMVs extracted at this time show a band of negative speed bias for observations tracking the formation of cloud streets and holes within the cloud layer.
AMV assigned pressures are mostly below 850 hPa but model best-fit estimates are not well-constrained for the problem winds. Nowhere low enough in the model profile to fit the slow AMVs?
Met Office
ECMWF
Forecast error verified against own analysis
Plots (right) show drift in forecasts away from their respective analyses for different lead times.
UKMO forecasts weaken with lead time: 4-6 m/s weaker by day 5. This suggests an analysis problem rather than a forecast model error.
ECMWF forecasts tend to strengthen in the first 24 hours - may explain why, although the models are very different at T+0, the O-B’s tend to be more similar. Forecasts are slightly weaker at day 5.
T+24
T+72
GOES-13 IR hourly AMVs (QI2>80, P<700 hPa)
A few WV intercept heights hPa
O-B’s in excess of minus 3 m/s
Overall, strong evidence that the negative AMV speed bias is in large part due to an excessively strong representation of the TEJ in the Met Office analysis.
Improving the assimilation of AMVs in this area could help to reduce the analysis error.
T+120
Most cloudbase heights below 850 hPa.
Spuriously fast AMVs at low level (below 700 hPa)
Background: Both MTSAT and Meteosat-7 show regions with very fast wind vectors assigned at low levels, e.g. for MSTAT IR AMVs in the South China Sea.
Update: The O-B speed bias statistics for MTSAT-2 show a large difference between the IR and VIS.
The few AMVs assigned WV intercept heights are likely assigned too high, but these are not causing for the more widespread bias. For cloudbase heights the negative bias is worse below 900 hPa - suggests not a case of height assignment error since assigning higher would only increase the bias.
UKMO T+6 forecast valid at 12 UTC
UKMO (T+6) – EC (T+12)
VIS IR
Aug 2013 versus the ECMWF model bg.
Visible: good agreement with collocated model wind speeds
IR: significant biases against both Met Office and ECMWF forecasts.
How well do the models agree?
In the area of moderate winds circled, the Met Office forecast is stronger than ECWMF e.g. red shading around 68-70°W, 38-40°N. The difference is ~2 m/s and coincides well with the AMV speed bias of a similar magnitude shown above.
E.g. 18 UTC on 5 July (below) shows a very marked disagreement for a line of winds to the S of Japan.
Large differences in both speed/direction suggest a large error in height assignment. AMVs assigned heights are in the range hPa whilst best-fit estimates are constrained to hPa.
Imagery shows this gross height assignment error is from tracking cirrus cloud.
Overall there is some evidence to suggest that the Met Office model may be overestimating the strength of the low level winds.
However, its also clear that the AMVs with the largest negative speed bias tend to be located near to where the stratocumulus is breaking up. This suggests that the AMV tracking algorithm may not be handling this kind of situation well and/or that such clouds do not make good tracers.
The reason for the cloud deck breaking up in that location is likely related to the cold air encountering the warm SST’s of the Gulf Stream. The vertical transfer of heat and moisture from the ocean surface will promote mixing of the lower layers helping to break up the cloud. This would explain why the monthly mean speed bias pattern in appears to coincide with the SST gradients.
Another cold-air outbreak on 22 December 2013 resulted in a similar cloud pattern and bias.
Observations
Forecast
Pressure
Best-fit
© Crown copyright Met Office and the Met Office logo are registered trademarks
In the subsequent 0 UTC cycle the IR bias is similar but there are no visible AMVs tracking the cirrus in the upper level easterly flow. This explains why the visible AMVs appear better in the monitoring.
Met Office FitzRoy Road, Exeter, Devon, EX1 3PB United Kingdom Tel: Fax: