A strong daily level, with hourly detail.
Each complete day is a row of 24 hourly readings. PCA (principal component analysis) finds the combinations of hours along which days differ most, ordered by how much of the variation each one captures. The first, PC1, mainly reflects a day’s overall level; later components capture contrasts between hours.
PC1 holds 61–78% of the day-to-day variation in every sensor, and the largest share, 77.6%, in S4.
Scree plots and loadingsEach bar is one sensor’s own PCA, fitted to its 357 complete days × 24 hourly readings. The coloured segments are the shares of PC1, PC2 and PC3, grey is the rest, and the number at the end of each bar is PC1’s share.
Compare weekdays and weekends.
For S1 and S3, a day’s PC2 score is strongly correlated with whether it falls on a weekend (Pearson r ≈ −0.64, on a scale from −1 to +1).
Weekdays are circles and solid lines; weekends are squares and dashed lines. A score is a day’s position along a PCA component. For S1 and S3, weekend days are shifted along PC2, the vertical axis, relative to weekdays. The curves average the raw hourly readings of all weekdays and of all weekends, each day weighted equally. Use the S1–S5 buttons to switch sensor.
S4’s dominant pattern follows weather.
S4’s PC1 score is closely correlated with the day’s absolute humidity and temperature, and with its position in the date sequence (retained-day order).
Its correlations with the NO₂ and NOx readings of the reference analysers are much weaker.
Detailed conclusionsBars to the right of 0 are positive correlations, bars to the left negative. Weather and day order use all 357 complete days; NO₂ and NOx use the 323 days with reference readings. Daily averages use the hours with readings. Retained-day order numbers the complete days 0–356 by date.