CRTF Telecon - 04-29-2015

Would a centralized database for reanalysis observations that all centers can write to and read from be desirable?

If so, what software should it based on? ECMWF ODB?

Notes from NCRTF Telecon.

Paul Poli (ECMWF) presented results and plans from ERA-20C. ERA-20C is the first 20th Century reanalysis by ECMWF, adopting an approach very similar to NOAA 20CR. The system is composed of a T159 atmospheric model with 91 vertical layers (same as ERA-40), computes increments at T95, but uses a 2013 version of the model. The reanalysis includes time varying forcing such as SST, greenhouse gases, solar cycle, and aerosols, following CMPI5 forcing when available. System assimilated surface pressure and mean sea level pressure as well as some winds from ICOADS. There are two productions run thus far, including one that utilizes a 10-member ensemble whereas the other is a deterministic re-run. The system utilizes 4DVAR with a longer assimilation window of 24 hours and includes inline variational bias correction of surface pressure observations for all platforms.

** Note: at this time there were technical difficulties and we actually moved on to Jerry Potter’s presentation. We were able to fix the glitch, and Paul was able to complete his presentation after Jerry. However, I am including the notes from Paul’s talk here for consistency with the above.

Products and data have been available since October 1st: http://apps.ecmwf.int/datasets. More data will be available next soon including observation feedback with interactive feature where you can click on variable/observation, source, and usage.

In evaluating the performance, ERA-20C did the best with for the October 1987 European when using the dynamic ensemble due to (flow dependent, evolving) structure functions, with the deterministic system performing slightly worse. For long term evaluations, ERA-20C is found to have a systematic dry bias (of 2mm or so), though is still able to nicely capture the water cycle. The dry bias can likely be attributed to the lack of assimilation of moisture observations. Despite this dry bias, ERA-20C compares favorably to RSS and HOAPS, even more so than ERA-Interim and JRA-55. ERA-20C is also shown to have the ability to reproduce realistic interannual variability. An interesting finding that is somewhat inexplicable is the fact that the reduction in ensemble spread for 2 meter temperature coincides with an increase in variability, perhaps an unexpected consequence of the ensemble method used.

When ERA-20C is compared to independent observations such as radiances from 1970 NIMBUS-4 IRIS using RTTOV, the standard deviation is reasonable. Comparisons to sensors such as SSMI (sensitive to water vapor) show larger standard deviations (up to 7K) whereas ERA-Interim was smaller (2-3K), which is not surprising given that ERA-Interim assimilated such data. ERA-20C explains about 90% of the variances in SSMI channel 3 compared to 98% explained in ERA-Interim. ERA-20C also appears to be missing tropical instability waves, likely due to the use of monthly SST forcing.

There were some issues found in ERA-20C including the treatment of moored and drifting buoy winds in the same way. Some of the winds appear to have 180 degree (sign) errors. While NOAA 20CR effectively assimilated many tropical cyclone bogus observations, ERA-20C ensemble and deterministic integrations rejected most of the data due to background checks and/or time-series checks (observations being constant over time). There is also a problem noted at southern high latitudes, with a drop in MSLP over time.

Plans for the future include an extension backward in time and to incorporate ocean (coupling). ERA-Interm is being replaced by ERA-5. Work will need to be done in terms of addressing model bias through the 20th century with a significantly changing observing system.

During Q/A, it was asked how the ship winds were handled. It was mentioned that the heights of the anemometers are actually unknown, and that it is likely that many were assimilated with the wrong heights. Despite this, the assimilation of the ship winds yielded better forecasts. There was another question about the Southern Hemisphere issue, and trying to address what is real, what is reanalysis, and what are the observations showing. The sea-ice between NOAA 20CR ERA-20C is quite similar. However, NOAA (new 2c) has recently redone with significantly different sea-ice and there is some hint that it has to be the ships (which are biased high). However, ECMWF hasn’t found any particular collection of data that is worse than another. Innovation feedback will be helpful, but time and further research is needed.

In between Paul’s presentation, Jerry Potter presented on CREATE-IP: collaborative reanalysis technology environment. The idea is based on providing a means for distributing climate model output (with a focus on CMIP5 for now), and establishing standards for reanalysis.

There is a transition occurring to expand the scope of the project in order to provide data for a reanalysis intercomparison…., i.e. create-ip (IP: Intercomparison). This will serve as a launchpad such intercomparisons, adding higher fidelity in addition to monthly data for CMIP5, includeing innovation statistics to investigate biases, gridding of observations used in assimilation, and potentially by providing model forecasts initialized from reanalyses. In order to make access easier, vetted users are provided internal access instead of moving enormous amounts of data around. There is a CREATE-IP website with advanced data search capability, which includes data descriptions, levels, variables, and in general a one-stop-shop for reanalysis with the data all being provided in the same format and accessibility. Thus far, the data includes MERRA, ERA-Interim, CFSR, JRA25, JRA55, 20CR, and there is permission to include ERA-20C and NOAA 20CR v2.

To make things easier and to establish a standard, the data is converted to monthly averaged CMIP5 format. There has been some effort to move toward higher temporal resolution, starting with hourly precipitable water and potentially extending to 3d winds, temperature, humidity, and surface pressure. They are possibly looking into assisting in generation of restart data for climate models.

Using the example of reanalysis differences for Russian heat wave, CFSR got the latent/sensible heat diurnal cycle correct. In looking at JJA 2010 precipitation anomaly (same heat wave), 20CR reproduced GPCP anomaly quite well. As another example, they have taken a look at impact of assimilation of ASMU in MERRA on precipitation anomaly. There was an error/bias found in JRA55 ozone, whereas the other reanalyses looked similar. In looking at JJA 2010 prec

There are new visualization tools being developed, and an example movie was shown for CFSR precipitable water for an atmospheric river event resulting in California precipitation in December 2014.

The standardization of data has made huge contribution for IPCC (CMIP3, CMIP5, etc.), and there is a desire to do the same for reanalysis products.

During the Q/A, it was mentioned that in using ESGF to download CMIP5 results, the format provided is good for people that do not understand the products of the data. The response was general agreement, and this will allow for more people to easily access the data. However, this type of effort (standardization of data?) is proving to be more difficult than anticipated. There was another question about common time period for the reanalysis data provided. The response was that everything that is being produced is being updated by year and 2014 is already completed. They are contemplating doing monthly updates but there has been some resistance. Lastly, there was a question about the idea to provide more variables to assist with CFMIP: Cloud Feedback Modeling Inter-Comparision project to use reanalysis to start models and then investigate systematic errors. There was a follow on inquiry regarding the enormity of the data. The response was that they choose a subset of case studies or periods, such as the Year of Tropical Convection, MJO initiation events, etc. For example of the efficacy of this idea, they started the NCAR climate model from ERA40 and MERRA and were able to glean a lot about the model bias.

Hello, Just to reiterate to Jerry my interest in using CREATE-IP for feedback comparisons. ERA-20C observation feedback should be available online very soon now, could that be a starting point for the ERA-20C feedback? We should have ERA-Interim on the same platform at some point as well. Thanks again for this great initiative, it could turn out into a huge service to the whole community. Paul

In stratosphere session, what types of observations are needed to ensure accurate strat analyzes going forward?

Dear Task Force, The new 20CR version 2c (1851-2011) is now online from NOAA, NCAR, and NERSC (every member). Data links are at go.usa.gov/XTd. best wishes, gil