Early crop weights gave no clear later gain under a squared departure score at 33 MASTREE+ site groups

Can early seed crop records identify a mix of species with more stable output in later years? We tested three fixed weight rules in 33 local site groups from 13 MASTREE+ V2 source studies. Each group has 20 to 41 shared years and two to seven species. We fitted weights on the first half and tested them on the later half. Each species was divided by its early mean. The main score is the mean squared departure of the mixed index from one, so it includes both variation and a shift in mean output. A nonnegative minimum variance rule used early covariance with a fixed 50% reduction of off-diagonal terms. Its early score was 0.765 times the equal-weight score. Its later score ratio was 1.101, with a 95% source-study bootstrap interval of 0.838 to 1.494. The later ratio against inverse-variance weights was 1.057, with an interval of 0.994 to 1.137. Thus neither comparison established a later gain. Later centered variance and squared coefficients of variation also gave no clear gain. Resampling 11 spatial study clusters gave the same conclusion. The result is a retrospective benchmark for selected relative crop indices, mostly from the United States. It does not establish equal performance, measured food supply, or a response to planting. The package has unchanged source data, provenance, fixed rules, all results, figures, code and checks. An independent calculation matched all 2447 selected source values and all 107 fitted weight vectors. A fresh-directory replay matched every result and figure.

seed production · species mixtures · temporal stability · MASTREE+ · covariance · chronological validation · reproducible research