A Southeast US annual community fit reduced later error with HadISST but increased it with ERSST in 2006-2017

We tested early temperature fits on later annual changes in thermal community indices in four US trawl regions. Fits used pairs ending before 2005. Tests used adjacent pairs within 2005-2019, or to 2017 in Southeast US. With HadISST, mean squared error fell by 25% in Northeast US fall and 54% in Southeast US fall, but rose by 26% in Eastern Bering Sea and 93% in Gulf of Mexico. Southeast error rose by 30% with ERSST. Its HadISST gain was uncertain when early fitting uncertainty was included. All four joint intervals with a four-comparison allowance crossed zero. Fixed survey strata, area weights and record checks reduce some observation bias. Full-record index construction, short records, catchability and surface temperature limit inference. This is conditional transfer with observed temperature, not a prospective forecast or a causal climate effect. Code, aggregate results, source hashes and verified input retrieval are included.

community temperature index · time transfer · sea surface temperature · observation bias · trawl survey · temperature product · uncertainty