The hardest part about running PMF is to decide what is a good solution. There are many ways and approaches. The list here is far from complete and should only be concerned as a guideline to get started.
- Know your sampling site and your instrument. What sources can/cannot be resolved with this type of data? Do you expect some sources to be resolved (e.g. traffic if you are next to a main traffic road)?
- Inspect your key variables. Is it possible that this factor is present (all the time/sometimes)? How do they change over time? Do they correlate with externals? Do you have different slopes in the scatter point that can give you more information?
- Run unconstrained PMF over a wide range of factors. Do you see familiar profiles or time series patterns? Also check higher number of factors, some factors might only be resolved after mathematical splitting of factors has already started.
- How does your Q/Qexp change? Is it stabilizing at some point? Are the values reasonable or do you have over-/underestimated your error matrix?
- How clean are your profiles? Is it necessary to constrain? Do you constrain by time series or profile? Which reference profile suits your environment best?
- Do your factor time series correlate with external data?
- How does your solution change with different seed values? If you constrain, how does the tightness influence your solution?