Adding the previous crop at another location increased one-year linear forecast error in six MASTREE+ studies

We tested whether the previous crop at a second location improves a one-year crop forecast. We used 964 matched sets of two species at two locations from MASTREE+ V2. The sample has 141 records and 16 tree species in six source studies. Each set has 20 to 69 common consecutive years. A local linear model used the target record’s previous crop. The added predictor was the previous crop of the same species at the other location. Fits used earlier years only. Mean squared error increased in all six study means. The geometric mean increase was 11.59%, with a nominal 95% study t interval of 1.86% to 22.25%. The added predictor from a different species gave an 11.28% increase. Its difference from the same-species model was uncertain. A fixed ridge penalty reduced the same-species increase to 5.65%. A smaller sample scored after 20 initial years gave an uncertain 0.56% increase. These results apply to the tested linear models and selected site and region records. They do not rule out useful spatial forecast methods. The package has source data, all primary forecasts, source records, code, checks, and figures.

masting · seed production · forecast evaluation · spatial synchrony · open data · MASTREE+ · negative result