Future directions and project planning
SPARKLE Session 3 on climate data in infectious disease modelling
Presenter: Michael & Roslyn
Session 3 is a group discussion about what participants could and should be doing with climate data in their own work (no code). Start by sanity-checking whether a question even needs climate data (the section below), then run the breakout planning prompts. For anyone wanting a hands-on mechanistic model, the optional SEIRS practical now lives at the end of Session 2 (behind an expandable box), which keeps this session focused on discussion.
Purpose
The first two sessions built a workflow: from raw climate and malaria data to a fitted, climate-driven model with future projections. This session steps back and asks: what should you do with this in your own context? As in Session 2, we start from the research question, and the first thing to decide is whether climate data belongs in the answer at all. The rest of the session is spent in small groups planning concrete next steps.
Before we get started with the materials, please fill out the survey available at this link: https://redcap.unimelb.edu.au/surveys/?s=3HWR8NNF93TNKMAL
Does the question even need climate data?
This is the deliberate counterweight to the rest of the course: climate data is a tool, not a default, and part of using it well is recognising when not to. Walk through the spectrum below, from questions that obviously do not need climate data to ones that plausibly do. Land three distinctions: (1) is the outcome climate-sensitive at all? (2) if it responds to the environment, is it a short-term weather question (forecasting) or a long-term climate one (projection)? (3) even when climate-sensitive, will other drivers change faster than climate over the horizon of interest, making a forward projection misleading? This directly sets up the planning breakout.
Starting from the research question (as in Session 2) means the first thing to ask is whether it needs climate data at all. Some questions clearly do not; some need weather, not climate; and some are climate-sensitive but should not be projected decades forward. A rough spectrum, from clearly-not to yes-and-project:
| Example research question | Climate data? | Why |
|---|---|---|
| Efficacy of a new TB drug regimen in a trial | No | A clinical outcome; transmission and environment are outside the question |
| Effect of a hospital hand-hygiene programme on MRSA | No | Driven by behaviour and infection control, not weather |
| Vaccination coverage needed to eliminate measles in a city | Rarely | Any seasonality is mostly school-term contact behaviour; the lever is coverage, not climate |
| Will next week’s heatwave raise heat-stroke presentations? | Weather, not climate | A short-term operational forecast; long-range climate projections are the wrong timescale |
| Did the recent flooding trigger a cholera outbreak? | Weather (event), not projection | Near-term attribution of one event; projecting to 2100 does not answer it |
| How did the last El Niño affect dengue seasons? | Climate variability, but be careful projecting | Uses observed climate variability; a forward projection may be swamped by urbanisation, interventions, or immunity |
| How might dengue/malaria risk shift across regions by 2050 under emissions scenarios? | Yes, and project forward | Vector-borne, climate-sensitive, long horizon, the case-study type this course is built around |
Two distinctions are worth making explicit:
- Weather vs climate. Weather is the short-term state (today’s rain, next week’s heatwave); climate is the long-run statistics (what July rainfall typically looks like over decades). Operational questions (“what happens next week?”) need weather and forecasting; only questions about the shifting long-run baseline need climate data and projections.
- When not to project. Even a genuinely climate-sensitive outcome can be a poor candidate for forward projection if faster-moving drivers (interventions, drug resistance, urbanisation, demography, health-system change) will dominate the climate signal over the horizon of interest. The honest output there is a sensitivity analysis, not a headline projection.
Even a perfect climate-to-disease relationship captures one component of risk. Vector control, land use, human movement, immunity and health-system change all matter and are usually held fixed in a climate projection, so such a projection answers “how does the climate signal propagate”, not “how many cases will there be”.
Breakout groups: planning your own work
Suggested structure for ~60–75 minutes: 5 min framing, 30–40 min in groups of 3–4 working through the prompts, 20–25 min whole-group share-back. Ask each group to nominate a scribe and to leave with one concrete next step each. The prompts below are deliberately broad; encourage groups to anchor them to a real disease and a real decision in their setting.
In groups of three or four, work through the following. Nominate someone to jot down your group’s answers to share back.
Prompt A: The question (and does it need climate data?). Write down one decision or policy question relevant to your work. Then sanity-check it against the spectrum above: does answering it actually need climate data, or is it really a weather/forecasting question, or one where projecting decades forward would not make sense? If climate does belong, what would a useful answer look like?
Prompt B: The data. Which of the four data types from Session 2 (health surveillance, administrative boundaries, climate reanalysis, climate projections) can you realistically obtain? Where are the gaps, and how might you fill or work around them?
Prompt C: Ideal vs achievable. Recall the three tiers from Session 2: vector-explicit (climate → mosquito abundance → transmission → incidence), climate-forced mechanistic (SEIRS), and direct statistical. What is the most mechanistic model your data could actually support? For malaria we wanted the vector-explicit path but had no mosquito abundance data, so we dropped to the lower two tiers. Where does your problem sit, and what one dataset would let you move up a tier?
Prompt D: The caveats. List the two or three biggest caveats you would attach to any projection you produced. Consider in particular how you would convey model agreement (where do the climate models agree on the direction of change, and where do they not?) and the fact that climate is only one driver among many. Who needs to hear those caveats, and how would you communicate them?
Prompt E: The next step. Each person: write down one concrete thing you will do in the next month (obtain a dataset, reproduce a part of this workflow, talk to a data holder, etc.).
Contributors
- Dr Vassili Kitsios
- Prof. Roslyn Hickson
- Dr Michael Lydeamore
- Claude (Anthropic), assisted with drafting these session materials; all checked by the course team