HackMDMeeting Notes with Emilie V
- Emilie is a climate scientist (PhD in weather/climate modelling)
- Focus now on applied research (translating MO work into user applications)
- Worked in air pollution, wind energy
- MO group not sector-specific (e.g. transport, wild fires)
- Joined Joint Centre last year (Exeter + Met Office)
- James Salter
- RB: using synthetic populations, projecting to future, with climate/heat data to help local authorities plan adaptation stratgies
- EV: Agent based modelling you're working with - similar approach to Matt Thomas (DIMEX)? SPENSER?
- RB: UKCP 2.2 database? EV: good!
- RB 15 runs but 12 are selected? Why those 12?
- EV: I don't remember about the other 3. Models run at the scale of the whole globe because the resolution is coarser. When downscaling to a specific resolution. (output of the global one is input to the downscaling)
- EV: UKCP18 you only have 8.5 RCP scenario, so other climate projections may allow you to consider other scenarios
- RB 15 runs but 12 are selected? Why those 12?
- EV: the right term is "recalibration" but people say "bias correction"
- EV: climate models are a numerical representation of processes in the atmosphere (e.g. equations/physics-laws based)
- Issue is that when you put this into an equation to solve, you discretize everything so you're only accurate at that resolution and everything smaller you need to parameterize
- Introducing errors in the solutions of the equation (some may be systematic and be corrected for, others are stochastic)
- Bias correction is a field of research in an of itself; when looking at an application you need to look at domain-specific (e.g. for heat, looking at >30) and other variables it's less about the absolute value for trend
- Multivariate cases can be hard to correct for because of different considerations for variable types (lose consistency) vs. the case where you're only using temperature where you might not have this issue
- The year date is the year to use
- RB: Between the different runs, there is non-independence of climate variables (EV: yes)
- Because of the random stochastic nature of atmosphere, the global runs have slightly different conditions to start with
- RB: observational dataset - HADS Grid
- EV: yes this is the standard one/good; alternative one is (ECR5?)
- CHESS-SCAPE & CHESS-MET (EV: I've never heard of it)
- Recalibration & Bias Correction
- RB: is there no generally accepted approach?
- EV: yes, different approaches are better for different variables. Scaled Distribution Mapping is popular also Quantile Mapping. One of the reports on UKCP18 talks about bias correction. Falls short of telling you what to do
- EV: haven't heard about the R package that allows you to do multivariate (be careful); we primarily use Python
- EV: Santander package (Spanish) is a good one and used for the latest IPCC
- RB: is there no generally accepted approach?
- RB: our collaborators have developed a tool to share with local authorities; our role is to help advise them on data
- RB: We are hoping to use our adjusted data for all variables. They need some data for October, so we suggested CHESS-SCAPE. Want data on a daily time scale, but Gavin has suggested to not use the data like "this is the temperature in this year"
- EV: How much greenhouse gas in the atmosphere depends on what we do between now and then
- Taking the data as this is what's happening that year is wrong
- RB: there are 4 runs available
- EV: the more runs you have, the more indication of uncertainty, especially when we're thinking further in time
- RB: instead of RCP, convert to a global warming level?
- EV: Global warming levels calculated at the scale of the whole globe. At the regional level you might be at the same degrees, but you have an indication of where things are reached on average
- EV: We tend to use global warming levels with users
- In near future, slower moving factors don't matter as much. Usually users are focusing on a shorter horizon (e.g. 10 years time) decisions
- RB: We are hoping to use our adjusted data for all variables. They need some data for October, so we suggested CHESS-SCAPE. Want data on a daily time scale, but Gavin has suggested to not use the data like "this is the temperature in this year"
- CLIM-RECAL: mini project to compare different methods of BA for UKCP and document in an open, accesible way?
- EV: I think it would be good to find people in the MO who do these things (e.g. UKCP Projections team)
- Jason Lowe + Lizzie Kendon + Fai Fung
- Hope that when there is another release, there's a dataset that goes along with it
- EV: downside is that when we do recalibration there is a level of expert judgment to get a sense of what's good enough/not good enough
- EV: how to pick the dataset that is usedful for your questions at hand and how to choose the climate variables
- Question to answer is slightly different so that's why there is no standardized way
- EV: Love your idea of a big table
- EV: Table in UKCP18 report has a start on that
- Question to answer is slightly different so that's why there is no standardized way
- EV: I think it would be good to find people in the MO who do these things (e.g. UKCP Projections team)
- Who else to speak to (climate scientist focused on health)
- Christophe Sarran (Public Health England), Rosa Barciela
- Who are your users? Heat packs?
- EV: users change on project; no steady stream of projects at JCEEI
- EV: Bristol City Council dashboard on risk of mortality
- EV: Leeds or Manchester project on air quality (DIMEX); user is cities
- DfT, DEFRA (advisory)
- Worked with Heat Pack team for Bristol City
- EV team is focused on advancing DS techniques; not attached to a single sector or application
- Victoria's team is focused on urban climate services (UK applied science)
- Same skills and people move around
