I'm not sure what to make of those dips and spikes in June/July in the area data.
Upthread I had smoothed the sea ice area data somewhat, by using 5-day running averages. But even then, the fluctuations are fairly large.
The NSIDC sea ice area is quite sensitive to melt ponds. So it's more variable than extent. If weather conditions change, then the melt pond fraction could change relatively quickly.
It's probably the main reason why area is a better predictor than extent: melt ponding in June and July preconditions the ice for extent losses later in the melt season.
Another factor is that only 11 years of data were used in the regression analysis (2007-2017). It's a small sample size. Doing the same analysis for 1979-2017, then the curve decreases more smoothly, and area is still a better predictor than extent during early summer, although by a smaller margin.