The scicast prediction market is shutdown atm (since early July?) so those numbers aren't live. But...
Network hash rate
3,255.17 PH/s (same block size)
5,032.64 PH/s (block size increase)
4,969.68 PH/s (no replace-by-fee)
3,132.09 PH/s (replace-by-fee)
Those numbers seem completely implausible: that's ~2.9-3.6 doublings of the current hashrate (< 400PH/s) in 17 months, when it's taken 12 months for the last doubling, and there's a block reward reduction due in that period too. (That might've been a reasonable prediction sometime in the past year, when doublings were slowing from once every ~45 days to once a year; it just doesn't seem a supportable prediction now)
That the PH/s rate is higher with bigger blocks is surprising, but given that site also predicts USD/BTC will be $280 with no change but $555 with bigger blocks, so I assume that difference is mostly due to price. Also, 12.5btc at $555 each is about 23 btc at $300 each, so if that price increase is realistic, it would compensate for almost all of the block reward reduction.
Daily transaction volume
168,438.22 tx/day (same block size)
193,773.08 tx/day (block size increase)
192,603.80 tx/day (no replace-by-fee)
168,406.73 tx/day (replace-by-fee)
That's only a 15% increase in transaction volume due to the block size increase; I would have expected more? 168k-194k tx/day is also only a 30%-50% increase in transaction volume from 130k tx/day currently. If that's really the case, then a 1.5MB-2MB max block size would probably be enough for the next two years...
(Predicting that the node count will drop from ~5000 to ~1200 due to increasing block sizes seems quite an indictment as far as centralisation risks go; but given I'm not that convinced by the other predictions, I'm not sure I want to give that much weight to that one either)