That is: the transactions did not (directly) cause that cost, but they are the "benefits" of incurring that cost. It makes sense to think about how many "benefits" you got from a given amount of costs, or inversely, how much cost there was for each unit of "benefit". To put it another way, if the costs are identical, it's better to have 1000000 transactions than just 100 transactions; "environmental impact per transaction" (or its inverse, "transactions per unit of environmental impact") captures that perfectly.
I don't think it's possible to decouple # of transactions and the price of Bitcoin though. Every additional Bitcoin transaction proves it is a feasible means of exchange, driving the price higher.
I think the reason people (media) talk about transactions is that blocks and mining rewards are hard to understand.
It really doesn't have anything to do with transactions(the fees are only a fraction of the profits). The energy impact would be precisely correlated to the fiat price since the miners will be paying for the energy and equipment in fiat.
Transactions or not, the higher the price, the dirtier the proof of work blockchain is.
Just look at the most recent block: https://www.blockchain.com/btc/block/00000000000000000002865...
reward: 6.25BTC fees for transactions: 0.5BTC
total revenue: 6.75BTC, that is ~ $350K at the moment.
Each block is mined every 10min on average, so it is 2.1M per hour. So the total energy consumption is limited to $2.1M per hour or $1.5B per month at this price. It would be $150M if the price was $5k instead and $30B if the price goes to $1M per BTC. When that happens, 1 transaction will amount to the energy consumption of an American family for 1 year or more. At $1M per BTC, $360,000,000,000 worth of energy would be wasted per year on calculating a hash of a block that is functionally exactly the same with the same amount of transaction as the one calculated for $360, $3.6 or $0.03.
I'm sure I'm missing something like investors betting on price increases, infrastructure elasticity etc. but the correlation must be roughly like that.