In audit simulations a changed buck harvest rate flipped the deer trend

STEP 1 · The rebuild

Wisconsin and Pennsylvania grow herd figures from bucks killed

The Wisconsin Department of Natural Resources starts sex-age-kill at buck harvest and widens it through assumptions plus extra data. Its sheet names five inputs. Registered buck kill is what the agency already has measured. Buck harvest rate is drawn from radio-collared bucks or from the yearling portion of bucks killed with a recovery rate assumed on top. Adult doe-to-buck ratio compares the percentage of yearling bucks with the percentage of yearling does. Public and staff summer sightings provide the fawn-to-doe ratio. On that sheet the registered kill is what they enter as total harvest.

Antler restrictions sit inside the sex-age-kill model used by the Pennsylvania Game Commission. The January 2011 writeup says model output was never built to be a count of deer on the ground. Because of the assumptions the commission cannot say a management unit holds a specific number of deer. Those assumptions do not necessarily block tracking of the population trend. The antlered harvest rate is the divisor for antlered harvest. That quotient estimates antlered deer. Mature females equal the antlered estimate times the adult sex ratio. Juveniles equal mature females times the fawn-to-doe ratio measured in the harvest. Adding antlered deer, mature females, and juveniles gives the total. Figures from the most recent years should be treated as indices. The best estimate for a year arrives only when every cohort's records are finished.

Pennsylvania's sheet walks a worked example. Start from a 50 percent harvest on adult males. Killing 100 adult males at that rate produces an estimated 200. Move the true rate to 60 percent and the real adult male population is 167, so the estimate is too high. Keep 60 percent steady year after year and the trend in the estimate still follows the herd. Let the rate jump and that following stops.

STEP 2 · The break

Simulated herds reversed when the male harvest rate shifted

In a draft dated 29 November 2006, a panel chaired by Millspaugh examined sex-age-kill as Wisconsin applied it and tested simulated populations whose size was already known. The panel judged the model extremely sensitive when the male harvest rate changes abruptly. When that male harvest rate changed inside the simulations, estimates moved against the true population trend. Female harvest changes caused much less trouble. Bias in the population figure can come from a new buck rule. It can also come from hunters changing which bucks they choose. Earn-a-buck style rules were one case the panel named. Quality-deer hunting styles were the other.

The review had more comfort at statewide scale than inside one deer management unit. Across those 16 units, the model accounted for as much as 62 percent of year-to-year differences between the harvest it predicted and the harvest hunters actually took. Prediction was weak for some units. The panel said a printed deer-per-square-mile figure brings extra error. Someone must define the range deer can use. Inside that range the animals sit in patches.

STEP 3 · The check

Minnesota reads density beside kill and success

Minnesota's Department of Natural Resources builds its own reconstruction. Staff refuse to treat that model as the only evidence. In the 2024 trend report biologists read modeled results together with harvest metrics, hunter success, and goals the public helped set.

Published output is spring 2024 pre-fawning density from a run spanning 2019 through 2024. Starting values were settled with 500 Monte Carlo runs. The final model used 5,000 runs. Reported densities cover 105 of 129 permit areas. Those areas are drawn from 23 management units. Absolute density is to be read with caution. The trend carries more of the weight. Labels are above, at, or below goal. Staff apply a label only after they read modeled density with bucks killed and with hunter success.

Our read waits on agreement among the model, the buck kill, and hunter success before the direction counts as settled.

Where planes struggle, cameras are the check in the woods. Thirty cameras ran on public land in permit area 679 between 17 July and 12 September 2023. Picture total was 489,517. Deer detections inside them numbered 197. Only the 55 males, the 75 females, and the 28 fawns enter the ratios. Classification failed for the other 62 deer. Space-to-event work estimated 20,398 deer. That is about 20 deer per square mile. The 95 percent interval was 17 to 23 deer per square mile. Coefficient of variation was 7 percent. The run also gave 1 fawn for every 2.7 does. It gave 0.74 bucks per doe.

All of that rests on one permit area during one summer and on public land alone.

The report cites an earlier northeast camera effort at under 2 deer per square mile. The authors called that density rather low beside the population model.

Baited surveys from older work tried to identify individual bucks. They were dropped because of sightability bias, misclassification, time cost, and disease risk tied to bait.

The state deer plan for 2019 through 2028 says aerial surveys need snow deep enough and deer visible enough for observers. Reliability is poor for most permit areas on southwest farmland. It also fails in the northeast forest. The plan uses the model to follow relative abundance as it changes through time. Staff hold more confidence where several indices agree. That camera report reconsiders transition-zone aircraft counts over cost, flying danger, and uneven data when required snow is missing. Roadside distance sampling remains ongoing work for spring densities in farmland. No separate accuracy number for that method is given in the camera report.

STEP 4 · The kill

Days afield can shift bucks killed while the herd holds

Sitar and Roell in a June 2021 paper said reliable abundance estimates were unavailable for their Upper Peninsula analysis. They used buck harvest as the indicator for trend. They called it the best series they had on deer numbers in Michigan. Hunter numbers, regulation changes, poor firearm-season weather, and the opener's weekday can all move that kill.

They write that effort by hunters is less limited than the buck harvest series. Historical effort is missing, so long trends cannot use it.

Their hunter-days-per-buck figure divides days afield among all hunters by bucks killed. The Upper Peninsula graph shows days per buck rising as the buck kill falls.

Across northern Michigan they tied part of the buck-harvest decline over 35 years to tighter buck rules. Another part was nearly 40 percent fewer deer hunters over 20 years.

From the 2004 to 2005 seasons onward Pennsylvania has estimated harvest by a closed two-sample Lincoln-Petersen method. Aging teams mark deer during field checks. A marked animal enters the second sample when the hunter reports it.

Michigan's harvest page puts average mail-survey returns above 70 percent in the early 2000s and at 33 percent in 2021. Low response can bias the harvest estimate.

Our side-by-side look at Michigan's two harvest figures keeps registered totals separate from the survey estimate. The gap was large. The gap changed with the year. Folding them into one kill would make a third figure that no method produced.

STEP 5 · The read

Three series have to move with the model

Our rule treats the printed population as a signal of whether the agency wants the unit up, steady, or down. It is silent on how many deer use one farm.

Line the trend up with three series for matching years. Look at whether buck kill moved alongside it. Kill can fall while the herd holds if the opener is poor, if the opener lands on a weekday, or if hunter numbers drop.

Yearling share among killed bucks feeds both the sex ratio and the harvest rate inside sex-age-kill. Our jawbone and cementum age guide shows the supported result is the yearling-versus-adult split. An exact adult year is softer.

Place the fawn ratio beside antlerless permits. A reduced harvest may be fewer deer. It may be fewer doe tags instead. Our paper on the seven-state harvest keeps those causes apart. Iowa showed a drop that hit does harder. Wisconsin's headline drop changes if the counties at the start and the counties at the end are the same set.

Buck kill, yearling share, and fawn ratio may all point with the model. Count that trend as real inside the unit for those years only when that happens. A single series moving alone means the population figure is still only a hypothesis. With a steady buck harvest rate the reconstruction can follow the herd. With a rate the rules have just changed, it cannot.