HKUST's 4,116-city model finds more than a tenth of sustainability gains beyond city limits

Start with the elasticity, because that is where the mechanism lives. A 10% improvement in other cities' particulate (SDG 11.6), gaseous-pollutant (12.4) or greenhouse-gas (13.2) indicators moves the focal city's matching score by an average of 0.2% to 0.45% ^[8]. Divide through and the spillover coefficient sits between 0.02 and 0.045 ^[16]. On those terms a single neighbour's cleanup is nearly invisible in your own annual figures, so the headline "more than 10%" ^[1] has to describe a whole connected system improving at once, and the account released with the paper does not state the counterfactual behind it ^[2]. The two numbers answer different questions, and only the smaller one arrives with a range.

The identification problem is the part worth sitting with. Pollutant and greenhouse-gas fields came from satellite products ^[2], which record what is in the air over a city rather than what that city emitted. A spatial model fitted to concentrations must distinguish a plume crossing a boundary from a factory crossing one, and the summary does not say how this one does it. The authors' own examples split across those channels and are hedged accordingly: Hong Kong's 2020 gain on CO2 is described as possibly attributable to industrial upgrading among Greater Bay Area neighbours ^[10], Copenhagen's carbon performance as positively correlated with its connected cities and potentially reflecting shared climate governance, transport and energy systems ^[11], and Beijing's PM2.5 and PM10 as negatively correlated with its neighbours, potentially through relocation of pollution-intensive activity ^[12]. Correlation with a plausible story attached is the right way to report a spatial regression. It is not the same as showing that coordination caused the gain.

Some arithmetic worth having. 93.51% of 4,116 cities is about 3,849 cities whose emission-related scores demonstrably move with someone else's ^[17]. If synergy outnumbers trade-off two to one over the 11 years ^[13], and if those two categories cover the affected set, roughly 1,283 of them spent the decade being pulled the wrong way ^[18]. The paper reports the ratio rather than the split, so that figure rests on an assumption, and I flag it as mine.

What this does not tell you is whether any coordination mechanism delivers the modelled gain. There is no intervention here and no control region: 11 years of observed indicators across cities producing more than 90% of global GDP ^[6], fitted to a spatial structure ^[2]. Lu's framing, that conventional environmental governance treats urban emission reduction as a purely local issue ^[15], holds up comfortably under that caveat. The 10% is a property of the fitted structure, and the same structure shows that one route to a better local number is moving the smokestack across the line ^[12].