Winter precipitation gave no clear cherry bloom forecast gain at Liestal and Washington in 1988-2025

We tested whether January-February precipitation improved March 1 cherry bloom forecasts beyond January and February temperature and a time trend. We used fixed bloom records and daily station weather at Liestal and Washington. Each target year used the prior 30 calendar years for training. Four site and period tests in 1988-2006 and 2007-2025 used 74 held-out dates. Mean absolute error rose by 0.09 to 0.23 days with precipitation. All four adjusted block intervals included zero. All planned sensitivity intervals also included zero. These data gave no clear gain from the tested precipitation term. They do not show that water has no role in flowering. Fixed sites and paired errors reduce coverage bias. Missing visit records, different bloom rules, nearby weather stations, selected complete years and short tests limit inference. This is a retrospective prediction test, not a causal climate or pollution test. The package gives code, source hashes, aggregate results, a figure and a method to regenerate external evidence.

phenology · cherry bloom · precipitation · temperature · forecast validation · observation bias · null result