Ant bait counts gave an uncertain gain in later colony estimates at eight Portal plots in 2000–2009

We tested whether counts of workers at seed baits improve estimates of annual ant colony counts in later years. We used PortalData 6.117.0, eight fixed plots, and 17 named taxa with comparable colony protocols. We fitted a separate mean for each plot and then added one common bait slope. Fits used eight survey years in 1988–1999. Tests used 80 plot-year records in 2000–2009. Adding bait counts reduced mean squared log error from 0.32728 to 0.26199, a reduction of 19.95%. A nominal 95% range from test-year blocks alone was 5.12% to 31.56%. When training years were also sampled and models were fitted again, the range was −31.98% to +41.67%. This range used 4,986 valid fits from 4,999 attempts. Fits restricted to plots without ant removal increased test error. All planned checks had ranges that included zero gain when training variation was included. The result leaves the added value of this bait model uncertain. It does not show that bait counts have no information. Counts at baits can depend on worker activity as well as colony number. The comparison assumes complete surveys within selected plot-years. It applies to the stated models and fixed plots at one site. The package preserves unchanged source tables, source hashes, a complete species grid, all test estimates, code, figures, and independent checks.

ants · biodiversity monitoring · colony counts · seed baits · temporal validation · Portal Project