A comparison of summer, fall and winter estimates of monarch population size before and after milkweed eradication from crop fields in North America

Measures of the area occupied by overwintering monarchs in México since the mid‐1990s show a decline. Summer surveys of monarchs, however, do not show a similar decline. This discrepancy has led to the proposition that summer monarch numbers are actually stable and that increasing mortality during migration has led to declining overwintering numbers. A competing hypothesis is that this discrepancy is due to a sampling bias in the summer counts and that the summer population has declined because of the eradication of milkweed habitat from crop fields that occurred in the late 1990s through mid‐2000s. We posit that the sampling bias occurred during the period when milkweeds were declining in crop fields and resulted from not sampling in the crop fields. We examined three measures of the size of the population made in the fall during migration and four summer survey measures and compared them to the overwintering measures. Counts of migrating monarchs are not expected to have this sampling bias since migrants come from all habitats, including crop fields. During the period of milkweed eradication, counts of the migrating population were correlated with the size of the overwintering population, whereas summer survey counts were not. After milkweed eradication from crop fields, all population measures were correlated with one another. These results indicate that during the pre‐eradication period, summer counts were not an accurate measure of summer population size. Population trends observed at the overwintering site reflect factors, principally milkweed loss, that affect summer population size.


INTRODUCTION
The iconic monarch butterfly (Danaus plexippus L., Lepidoptera: Nymphalidae, Danainae) in North America has received much attention from the public, the scientific community, private organizations and government agencies.Much of this attention has been generated by evidence of a severe decline in the size of the migratory eastern population since the mid-1990s, as measured by the area occupied by overwintering monarchs in México (Brower et al., 2012;Saunders et al., 2019;Vidal & Rend on-Salinas, 2014) and concerns about the future extinction probability of this population (Flockhart et al., 2015;Semmens et al., 2016).This concern has led to petitioning of the United States Fish and Wildlife Service (USFWS) for listing the monarch as a threatened species.In December 2020, after a status assessment, it was determined that listing the monarch under the Endangered Species Act was warranted but precluded by higher priority listing actions.Thus, the monarch is a candidate for listing with its status reviewed annually (USFWS, 2020).In July 2022, the International Union for the Conservation of Nature (IUCN) declared that the migratory monarchs are endangered (IUCN, 2022).Concerns about their monarch population decline led to efforts to determine its cause or causes so as to target conservation efforts.Climate change has been implicated as a driver of the decline (Saunders et al., 2018;Zylstra et al., 2021) as well as pesticide use (Grant et al., 2022).One causal candidate is the loss of milkweed habitat in the summer breeding area due to the use of glyphosate herbicide in crop fields (Pleasants & Oberhauser, 2013), referred to as the milkweed limitation hypothesis.In the late 1990s and early 2000s monarchs commonly laid eggs on milkweeds in agricultural fields, and because of the relative amounts of agricultural (ag) and non-agricultural (non-ag) land available to monarchs in the main breeding areas, these agricultural milkweeds were the source of most monarchs in the eastern migratory population (Oberhauser et al., 2001;Pleasants & Oberhauser, 2013).The management of these fields changed markedly beginning in 1996 with the introduction of glyphosate-tolerant corn and soybeans and subsequent widespread use of glyphosate herbicide (Figure S1).As a result, the abundance of milkweeds in crop fields declined and by the mid-2000s milkweeds had virtually disappeared from Midwestern US corn and soybean fields (Pleasants, 2017) (Figure S2).Using estimates of the density of milkweed in fields and non-agricultural areas outside of fields, and the area encompassed by both habitats, Pleasants (2017) estimated that there had been a 64% reduction in milkweed abundance on the Midwest landscape from 1999 to 2008.Similarly, Zaya et al. (2017) found a 68% reduction in milkweed abundance in Illinois from 1997 to 2016.Because the decline in the size of the overwintering area paralleled the decline in the abundance of milkweed and monarch egg production on the Midwest landscape, the milkweed limitation hypothesis was posited (Pleasants & Oberhauser, 2013).
However, some studies have noted that surveys of adult monarchs made during the summer do not show a decline (Crossley et al., 2022;Inamine et al., 2016;Ries et al., 2015).This disconnect between summer counts and overwintering area has prompted the hypothesis that an increase in mortality during fall migration has caused the decline seen on the overwintering grounds, referred to as the mortality migration hypothesis (Agrawal, 2019;Agrawal & Inamine, 2018).It has also led to the supposition that the monarch population is not actually in trouble because summer numbers appear to be stable, despite the observed decline on the overwintering grounds (Crossley et al., 2022).These alternative hypotheses have important conservation implications.The milkweed limitation hypothesis would direct conservation measures towards increasing the amount of milkweed habitat in the summer breeding range.The migration mortality hypothesis would focus conservation efforts on making nectar resources available along the migration flyways.
How can we reconcile this disconnect between summer counts and overwintering area?One possibility is that both surveys accurately measure the population in each of the two time periods.If this is the case then the migration mortality hypothesis would be a way to reconcile the difference between the two measures.We have previously asserted, however, that summer counts were not an accurate measure of summer population size during the period when milkweeds were declining in crop fields (Pleasants et al., 2016;Pleasants et al., 2017;Taylor et al., 2020).We posit that the lack of a decline seen in summer surveys is the result of a sampling bias resulting from a failure to conduct surveys in crop fields and account for the declining production of monarchs from that habitat.
Surveys are expected to provide a sample of the available habitat where monarchs may be found.A key assumption in ecological sampling is that the true size of the population is the density in the sampled area multiplied by the amount of available habitat.As long as the sampled area is representative of available habitat, and the amount of available habitat remains constant over the years, changes in density in the sampled area alone can be used to gauge changes in population size.But if the amount of available habitat decreases, as is the case with the loss of milkweed in agricultural row crops, then density measures conducted only in habitats where no decrease has occurred will not accurately reflect changes in population size.
As evidence of sampling bias, we previously showed that summer survey counts of adults and eggs were correlated with overwinter area when they were corrected by the proportion of milkweed habitat they represented (Pleasants et al., 2017).We also showed that the size of the migrating population, as estimated by number of butterflies tagged in the Monarch Watch Tagging Program (Monarch Watch, 2022), was correlated with the area occupied by the overwintering population (Taylor et al., 2020) and that monarch tag recovery rates did not decline as the migration mortality hypothesis would predict.
Despite having raised these concerns, summer counts during the milkweed eradication period continue to be used as a measure of population size in studies of monarch population dynamics.Here, we provide further evidence for sampling bias.We examined the size of the fall migrating population as an important way to tell if there is sampling bias in summer surveys.The landscape in the summer breeding region is a mosaic of agricultural fields and non-agricultural habitats.Here, we test the hypothesis that during the period when milkweeds were present but declining in agricultural fields (pre-eradication), summer counts of butterflies or their eggs and larvae made in non-agricultural habitats underestimated the true size of the population and, thus, did not detect the decline in summer population.To test this hypothesis, we analysed three data sets estimating the migrating population size, two of them new, and compared measures of population size based on summer survey counts, counts of migrating monarchs and overwintering area during the pre-milkweed eradication and post-milkweed eradication periods.
The predictions of our hypothesis are summarized in Table 1.
Primary Prediction: The primary prediction of the hypothesis is that during the pre-eradication period, the area occupied by overwintering monarchs in México will be correlated with estimates of migrating population size but not with summer survey counts.During the post-eradication period, summer, migrating and overwintering estimates of population size will all be correlated.This hypothesis focuses on whether summer counts during the pre-eradication period are an accurate reflection of the size of the monarch population.
There are also several corollary predictions, whether these predictions are supported does not affect the validity of the primary hypothesis.Corollary Prediction 1: The discrepancy between summer survey counts and overwintering area will be greatest in the early part of the eradication period and will decrease as milkweed eradication progresses and non-agricultural habitat becomes the primary milkweed habitat.Corollary Prediction 2: Counts of fall migrating monarchs and overwintering area will both show a decline during the pre-eradication period, reflecting the decline in milkweeds, whereas summer survey counts will not.Counts of migrating monarchs and overwintering area will be higher during the pre-eradication than the post-eradication period, whereas summer survey counts will not.These corollary predictions will be more difficult to prove statistically because they depend on regression relationships and there is a great deal of annual variation and a short time frame.However, we can determine whether the trends observed are consistent with the predictions.

Data sets
All summer and fall surveys use citizen scientists to collect data.Adult butterfly surveys occur on both public and private land.All surveys attempt to incorporate different habitats although none include row T A B L E 1 Summary of predictions from primary and corollary hypotheses for summer survey counts and fall migration counts.Table 2 summarizes which surveys were used for summer, fall and overwintering.

Overwintering area
On the overwintering grounds in México monarchs form dense aggregations (hereafter, colonies) in stands of oyamel fir (Abies religiosa H.B.K., Pinaceae) and are virtually impossible to census.Consequently, observers delineate the perimeter of each colony and calculate the area occupied by monarchs, using this measure as an index of population size (Thogmartin, Diffendorfer, et al., 2017;Vidal & Rend on-Salinas, 2014).
We used the overwintering area as listed in Rend on-Salinas et al. (2022).

North American Butterfly Association
Participants in North American Butterfly Association (NABA, 2022) surveys monitor butterfly abundance at many sites across the United States.Each survey site consists of a 24-km (15-mile) diameter circle where a count is conducted during the spring or summer.
Several locations within the 24-km diameter circle may be surveyed by up to four observers.The data collected consist of the number of butterflies of each species observed and the number of party hours involved.There are 90 counting circles across the Midwestern states although only a subset was surveyed in the time frame we used.
Counts are typically collected in June and July leading to too few samples made in the fall for the NABA data set to be used in estimating the size of the population during migration.For each year we summed the number of monarchs counted across all sites as well as the number of party hours across all sites, and then calculated the mean number of monarchs per party hour.

Illinois Butterfly Monitoring Network
An Illinois Butterfly Monitoring Network (IBMN, 2022) survey consists of one person conducting a Pollard Walk (Pollard, 1977), which is a walk along a route at a constant pace, stopping only to identify and record butterflies seen within 6 m of the route.Surveys are conducted on six occasions between 1 June and 7 August, with four of the six taking place before 20 July.Additional surveys are allowed as long as the required six are conducted between 1 June and 7 August.Samples are also collected in the fall.For analysis we used the total number of monarchs seen across all sites divided by the total number of hours of observation.

Ohio Lepidopterists
Ohio Lepidopterists (Ohio Lepidopterists, 2022) et al., 2015).As a measure of monarch abundance, we use the number of eggs per stem.This number varies over the season as the population builds up and then declines as late summer females complete their life span and newly emerged females enter reproductive diapause.For the analysis presented here, we used the average of the peak number of eggs per stem across all sites.

Monarch Watch
Monarch tagging kits are utilized from August-September by Monarch Watch (Monarch Watch, 2022) to volunteers broadly distributed across east of the Rocky Mountains in the United States and Canada.
Small, adhesive tags are affixed to the wing of migrating monarchs during a 3-month period from early August to November.The date, location and identity of the person tagging each butterfly are logged into a database.Starting in 2004, individuals were asked to indicate whether the butterfly was wild-caught or reared in captivity and released.We only used tagging data for wild-caught butterflies in this study.Prior to 2004, people who tagged reared butterflies usually noted this.Those entries were removed from the data.We used location information to distinguish individual taggers and we used the total number of tagged butterflies divided by the total number of tagging locations to estimate population size in each year.

Time frame
Years included: Surveys of the area occupied in winter in México began in 1994 (Vidal & Rend on-Salinas, 2014).Because we were interested in comparing overwintering area to counts made in the summer breeding region during the summer and fall migratory periods, we examined summer and fall data sets beginning in 1994.
However, some of those data sets did not start until a few years later (Table S1).We included all years up to 2021, where possible (Table S1).Monarch Watch tagging data extend only through 2015, the last year data from that program have been collated and analysed.
We did not have NABA data beyond 2016.
Summer sampling period: For the summer adult butterfly survey data sets we restricted counts to those made from 15 July to 20 August.There are two or more breeding generations each summer, somewhat overlapping, one roughly in late May and June and another in July and August.In the Midwest, egg-laying peaks in mid-to-late July and early August and declines from mid-to-late August (Nail et al., 2015).The eggs laid during this period give rise to migratory butterflies.Thus, the butterflies seen from 15 July to 20 August in the Midwest are primarily the penultimate generation and their offspring are the migratory generation.We cut off the count at 20 August, approximately when migration begins in the Midwest (Taylor et al., 2019), to keep the summer and migratory generations as separate as possible.For the MLMP data, the eggs counted during this period produce the migratory generation.
Fall sampling period: For the Illinois and Ohio fall surveys and Monarch Watch tagging data sets we used counts collected from 21 August to 10 October, corresponding to the primary migration period (Taylor et al., 2019).
Defining pre-and post-eradication periods: The fields monitored by Pleasants (Pleasants, 2017;

Geographic range
The Midwestern United States is the primary source of butterflies migrating to México (Flockhart et al., 2017;Wassenaar & Hobson, 1998).We restricted our analysis to data sets from this region.For the NABA data set we used count circles in Minnesota, Iowa, Missouri, Illinois, Wisconsin, Indiana, Ohio and Michigan (Figure S3); locations of survey sites for the Illinois and Ohio data sets are also shown in Figure S3.Midwestern MLMP survey sites are shown in Figure S4.Monarch tagging data were obtained from three 5 Â 5 latitude/longitude sectors in the Midwest (Figure S5).

Data analysis
Table S1 provides annual indices, calculated as described above, for all the data sets we examined.We divided each time series into R code for all analyses and figures is publicly available (Thogmartin et al., 2023).

RESULTS
Primary Prediction: Summer survey data prior to 2006 from four different monitoring programmes were not correlated with overwinter abundance (r's < 0.55, p-values > 0.12), whereas fall survey data from three monitoring programmes were (r's > 0.68, p-values < 0.04) (Table 3).In the post-eradication period, all summer and migration surveys were significantly correlated with overwinter area (r's > 0.50, p-values < 0.02).Figure S6 shows the correlations among all survey measures, pre-and post-eradication.In the pre-eradication period, all summer survey counts were correlated with each other, indicating that the same sampling bias applies to all.In the post-eradication period, all survey indices were correlated with each other.A Fisher's exact test ( p = 1.0) indicated these results are concordant with our prediction.
Corollary Prediction 1: For each of the migration or summer surveys, the difference between standardized counts and standardized overwintering area each year indicates how well the survey counts match the overwintering area.The hypothesis predicts that migration survey counts will match overwintering area equally well in the preand post-eradication periods, that is, the median standardized difference will not be different between the two periods.It also predicts, though, that the summer counts will underestimate the measure of overwintering area in the pre-eradication period such that the median difference pre-eradication will be significantly lower than posteradication.Table 4 indicates that this is the case, supporting corollary prediction 1.
An examination of the trends in the differences between the overwinter area and summer survey indices showed that summer surveys underestimated overwintering area in the pre-eradication period compared to post-eradication (Figure 1).Consistent with this prediction, the trends in Figure 1 showed that for summer surveys, the discrepancy with overwintering area decreased as the pre-eradication period progressed (positive slopes).This result indicates that counts made in the beginning of the pre-eradication period underestimated the size of the population more than counts made near the end.However, none of the trends were statistically significant (Table S2), ostensibly because of small number of years elapsed relative to inter-annual variation in monarch abundance.The pre-eradication trends for fall surveys were mixed, including a positive trend (for Illinois fall), no trend (tagging) and negative trend (Ohio fall); none were significant (Table S2).Post-eradication trend lines for differences were flat for all surveys except for Illinois fall, Ohio fall and MLMP, with positive slopes in the direction expected (Table S3).
Corollary Prediction 2: Regression analyses of the time series of annual survey indices showed differences during the pre-eradication period (Figure 2, Table S4).Overwintering area declined (Figure 2a) as did the migration population estimates for Monarch Watch tagging (Figure 2b), Ohio fall surveys (Figure 2d) and Illinois fall surveys  S4).The results were similar when the break point between pre-and post-eradication was 2007 or 2008 (Figure S7).The downward trend during the pre-eradication period for tagging and fall surveys in Ohio and Illinois was statistically significant only for Ohio (Table S4).The monarch butterfly population declined between the pre-and post-eradication periods (Table 5) but the within-period slopes were not significant (Figure 2a, Table S4), ostensibly due to the short period of time relative to the inter-annual variation (F values >1.65).There were no statistically significant trends for any of the  Note: Median and interquartile range (IQR) for pre-and post-period differences is provided.Bold highlights the tests depicting a significant difference between the median deviation from overwinter hectares pre-and post-eradication (pre-and post-2006).Significant differences in bold.Abbreviations: MLMP, Monarch Larva Monitoring Project; NABA, North American Butterfly Association.
other indices during the post-eradication period, except for Monarch Watch tagging which showed a significant decline (β = À1.501)(Table S4).
Overwintering area was significantly higher during preeradication compared to post-eradication (Table 5, Figure S8).For migration surveys, pre-eradication counts were significantly higher than post-eradication for tagging data (Mann-Whitney U test W = 7, p = 0.002), and near significance for Ohio fall (W = 49, p = 0.11) and Illinois fall (W = 60, p = 0.10) (Table 4, Figure S8).For all summer survey counts there were no significant differences in population estimates between the pre-and post-eradication periods ( p-values > 0.24).
F I G U R E 1 Time series plot of overwinter area occupied differenced from summer and fall survey counts for the pre-and post-2006 periods.All survey data were normalized (subtraction of the mean and division by the SD).Regression fits are accompanied by 95% (outer, lighter grey) and 80% (inner, darker grey) confidence intervals.

DISCUSSION
The primary prediction that in the pre-eradication period, summer counts would not match the area occupied in winter, but that migration counts would is well supported.The results in Table 3 show a lack of correlation between summer counts and overwintering area, but a positive correlation between migration counts and overwintering area.
This difference indicates that summer surveys were not an accurate estimate of the size of the monarch population during the preeradication period when there was a significant, but declining, amount of milkweed habitat in corn and soybean fields.For the posteradication period, when milkweeds in crop fields were gone and the sampling bias of summer counts no longer existed, the results in Table 3 show that, as predicted, summer counts were now correlated (2021) also found a correlation between summer survey counts and overwintering area for a period they analysed that corresponds to the post-eradication period.These results, in sum, indicate that summer surveys were not a good estimator of population size during the preeradication period.Summer survey data are only accurate when sampling biases are corrected (Pleasants et al., 2017) or in the posteradication period (ca.>2006).
Unlike summer surveys that only count butterflies in non-agricultural habitats, migration surveys, while also made in non-agriculturalhabitats, count butterflies that have come from both agricultural fields and non-agricultural habitats.As such they represent a more complete picture of the size of the population.It should be noted that the Ohio and Illinois migration surveys count not only butterflies that have begun their journey near the sampling locations (see Figure S3), but also butterflies passing through those sampling locations from points to the northeast.The fact that these two state-level surveys, which sample only a portion of the entire migrating population, were so well correlated with the size of the overwintering population shows the highly mobile nature of the monarch butterfly and indicates that there is a relatively even spring colonization of its broad summer breeding habitat and similar population growth in all areas during the summer.
The butterflies tagged by Monarch Watch were similarly drawn not only from the sectors noted in Figure S5 but also from areas to the north of those sectors.
It is important to note that the fall migration counts were made early in the migration, before any significant mortality during migration would have occurred.Thus, migration counts are expected to reflect the size of the summer population, not a population winnowed by migration mortality.The fact that summer counts and migration counts were correlated with each other in the post-eradication period supports this.Although the size of the migrating population was correlated with the size of the overwintering population, this result does not mean that the level of migration mortality is the same every year; it just means that the variability in migration mortality among years is not sufficiently large to negate the correlation between the two.
Corollary Prediction 1 was that summer surveys would underestimate the true population size in the pre-eradication period.Table 4 indicated that the median standardized difference between summer counts and overwinter hectares was significantly lower for the preeradication period compared to the post-eradication period, but that they were not significantly different for the migration surveys.We also expected that the underestimate in the summer count would decrease as milkweed eradication progressed because the discrepancy between summer surveys and overwintering area would be greater when there was a larger amount of milkweed in agricultural fields and a greater amount of monarch production was unaccounted for.Fig- ure 1 does show trend for the summer surveys, but the slopes of the regressions were not statistically significant (Table S3).
There was a sizable discrepancy between fall counts and overwintering area in the pre-eradication period in some years (Figure 1).One unexplained pattern in Figure 1 is the upward trend for the post-eradication period in the discrepancy between overwintering numbers and MLMP summer counts, Ohio fall counts and Illinois fall counts.This upward trend was not seen for tagging numbers but there are six fewer years of tagging data.The upward trend indicates that these surveys are progressively overestimating the size of the overwintering population.The upward trend is driven by greater discrepancies in the years 2018-2021.For the three surveys with upward trends, those years had increasing numbers of participants (Table S1), which may be a possible explanation.
Corollary Prediction 2: We would expect that if milkweed decimation in crop fields was driving monarch population size, then any monarch population decline would be manifested during the preeradication period.Table 5 shows that the size of the overwintering population in the pre-eradication period was significantly higher than in the post-eradication period.The size of the monarch population as estimated by the migration surveys was also higher in the preeradication period (Table 5) with the estimate based on tagging statistically significant and estimates based on Ohio fall and Illinois fall, nearly so.None of the summer surveys showed a difference in population size between pre-and post-eradication (Table 5).
In addition to comparing population size estimates pre-and posteradication, we also examined the regression of estimates by year.
Although the regression between overwintering area and year during the pre-eradication period showed a decline (Figure 2a), the trend was not statistically significant ( p = 0.11, Table S4).While one could take this result as a lack of evidence for a decline during this period, there is a potential explanation.The monarch butterfly population at the end of the summer has undergone three to four generations since the previous overwintering period.The size of each generation is affected by the size of the previous generation and breeding conditions, principally temperature and precipitation, during each generation (Zylstra et al., 2021).The amount of milkweed habitat available sets a ceiling for the size of the summer population but temperature and precipitation, and other factors, determine whether that ceiling can be reached.Annual variation in monarch overwintering population size can be very large.For example, in 2000 and 2004 overwintering area was 2.83 and 2.19 hectares, respectively, whereas in the other years in the pre-eradication period overwintering area ranged from 5.56 to 18.19 ha (Table S1).Because of this large annual variation, we do not expect a monotonic decline in the monarch numbers, making it difficult to obtain a significant regression result when only 12 years are included (also, see Semmens et al., 2016).The declines in migratory population estimates for the pre-eradication period (Figure 2b,d, f) were also not significant, except for Ohio (Table S4).Again, large annual variation plus a limited number of years sampled in the period for some fall surveys make it difficult to obtain a significant regression result.Wepprich et al. (2019) analyzed data from the Ohio surveys and found a decline from 1996 to 2018.They includedsurvey data from April to October, which is heavily influenced by the large numbers observed in the fall migration.
The regression of overwintering area with year in the posteradication period also shows a decline (Figure 2a), but the decline was very shallow and not significant (Table S5).Although this trend is not statistically significant, it may be real.Although the majority of milkweed habitat loss occurred during the period of eradication of milkweeds from crop fields, the amount of milkweed habitat continued to be eroded by land conversion from habitat with milkweed to cropland and development (Lark et al., 2020(Lark et al., , 2022)).Additionally, weather conditions associated with low monarch numbers appear to be more frequent under climate change (Zylstra et al., 2022).None of the migration surveys had a downward trend (Figure 2) except Monarch Watch tagging which showed a significant decline (Table S5).The regression of summer surveys on years, post-eradication, generally did not show any trend except for NABA, which showed a decline, though not significant (Table S5).
In sum, for Corollary Prediction 2, the results are generally consistent with the hypothesis that declining milkweeds during the preeradication period led to a decline in the monarch population, although they do not constitute strong evidence.
One question that needs to be addressed is why a decline in population size is not evident by looking only at non-agricultural habitats.
The argument for expecting it to be evident is as follows: There are several generations in the summer so if there are fewer milkweeds in the summer in one year compared to the previous year there will be fewer monarchs produced.Female monarchs move extensively in their lifetime during the breeding season (10-15 km) encountering many habitat types, are drawn to milkweed patches and only briefly stay in any patch (Zalucki et al., 2016).Consequently, one would expect that a smaller population than the previous year would result in fewer monarchs observed in all habitats, including non-agricultural habitats.However, the number of monarchs and the number of milkweed patches have BOTH decreased, meaning the smaller number of monarchs is spread over a smaller number of patches.This means that the density of monarchs in the remaining milkweed patches may not go down.In Pleasants et al. (2017) we modelled this scenario (see This long-term data set did not show any decline over the 29-year period nor any decline over shorter, decadal periods.The number of monarchs counted was not correlated with overwintering area but was correlated with measures of the summer population in the Northeast. Besides the overcounting problem noted above, there are other reasons why Cape May counts might not match overwintering area.First, the Northeast landscape does not have the extensive agricultural land present in the Midwest.As such, the elimination of milkweeds in agricultural fields would be expected to have a lesser effect on numbers of monarchs in the northeastern United States.Second, the Northeast is a minor contributor to the overwintering population compared to the Midwest (Flockhart et al., 2017), because of its smaller geographic size and less overall monarch production, and also because of its poorer migration success due to its greater distance from México (Taylor et al., 2020).
The results in this study indicate that summer counts cannot be used to discern population trends in the pre-eradication period, although they can be used in the post-eradication period.However, one could ask whether summer counts in the pre-eradication period could be used to examine the role of factors, like temperature and precipitation, affecting annual variation in population size?We would expect that annual increases and decreases in the population would be somewhat reflected in summer counts although the magnitude of the increases or decreases would not be accurate because the dwindling amount of available habitat is not accounted for.A direct measure of the amount of available habitat has been estimated (Pleasants et al., 2017).However, studies using summer counts to look at annual variation in monarch population size have used the amount of glyphosate applied to fields or percent of fields using glyphosate as a proxy variable for milkweed habitat loss (Crossley et al., 2022;Saunders et al., 2018;Zylstra et al., 2021).Glyphosate use, however, is not directly proportional to the amount of milkweed habitat available.There would have been a greater reduction in milkweeds in crop fields in the early years following the adoption of herbicide-tolerant crops, when glyphosate use was still low and many fields still had milkweeds, than in later years, when glyphosate use was high but most milkweeds had already been eliminated from fields.
Including glyphosate use in the post-eradication period will be uninformative since crop-field milkweeds have been largely eliminated.Some studies have found that glyphosate use is negatively associated with summer counts, although it does not explain a large amount of variation (Crossley et al., 2022;Saunders et al., 2018;Zylstra et al., 2021).We would expect that glyphosate use would provide some indication of habitat loss, but an imperfect one, and one that would not measure its full effect, but including glyphosate use could also confound the ability to gauge the effect of temperature and precipitation on population dynamics in the pre-eradication period.
The overall conclusion of the present study is that the size of the population as measured by overwintering area is an accurate representation of the size of the population that is produced every summer.
The decline seen in both the overwintering population and the migrating population in the pre-eradication period means that the summer population has also declined.That the decline occurred during the period when milkweeds were disappearing from corn and soybean fields strongly indicates that the two are related causally.Because summer surveys did not provide an accurate measure of the monarch population during the pre-eradication period, there is no need to advance the migration mortality hypothesis to explain the discrepancy between summer and winter estimates of population size.These findings support conservation efforts to increase monarch population size by increasing the number of milkweeds on the landscape (Thogmartin, L opez-Hoffman, et al., 2017).Buffering the population from the effects of climate change is a more difficult proposition (Zylstra et al., 2021(Zylstra et al., , 2022)), but increasing habitat availability across the summer breeding range could, at least for a while, provide a buffer against deleterious effects of a changing climate.Table S5.Simple linear regression of monarch butterfly survey programs for the years 2006 and after.Significant differences in bold.
Level of significance indicated by legend on bottom.
Summer surveys are made in habitats outside of crop fields and reflect monarch activity there.The monarchs in the migrating population are the offspring of the monarchs in the summer population which reflects monarch activity and egg production in all habitats, including crop fields.Thus, counts of migrants would not have the sampling bias of summer surveys.Migrants counted early in the migration journey would not yet have experienced the mortality posited by the migration mortality hypothesis.The number of migrating butterflies, therefore, should reflect the true size of the summer population.A correlation between numbers of migrating butterflies and overwintering area would indicate that the summer population determines the size of the overwintering population.If this is the case, then the observed lack of correlation between summer surveys and overwintering area would indicate that there is a sampling problem with summer surveys.If, however, the size of the migrating population is not correlated with overwintering area it would lend credence to the migration mortality hypothesis.Another indication of sampling bias would be if, after the period when milkweeds have disappeared from crop fields and the sampling bias problem no longer exists, summer counts are then correlated with overwinter area.
crop agricultural fields.Participants in the Monarch Larva Monitoring Project (MLMP, 2022) chose one or more milkweed patches to monitor, which may be on public or private land but not in agricultural fields.The area covered by overwintering monarchs in México is measured by personnel from the Monarch Butterfly Biosphere Reserve and World Wildlife Fund-México (Vidal & Rend on-Salinas, 2014).
surveys are made by a single individual along a fixed route.The survey consists of a Pollard Walk along the centre of a fixed path 4.5-m wide.Surveys are made once per week in the summer and fall.The data used are the total number of monarchs observed divided by the number of surveys.Monarch Larva Monitoring Project Individuals participating in the Monarch Larva Monitoring Project (MLMP, 2022) monitor one or more patches of milkweed, ideally on a weekly basis, throughout the spring and summer.They record the number of eggs and larvae and the number of milkweed stems observed.The data consist of weekly per stem numbers of monarchs at each life stage.Only data were included if more than 10 milkweeds were monitored and more than four visits to a milkweed patch occurred between 1 July and 31 August (Stenoien see Figure S2) were devoid of milkweeds by 2008, and nearly so by 2006 and 2007.In this analysis we defined the period from 1994 to 2005 as the pre-eradication period and 2006 and beyond as the post-eradication period.We repeated all analyses for post-eradication periods beginning in 2007 and 2008 and found negligible differences.
a pre-eradication period comprising indices before 2006 and a post-eradication period comprising indices from 2006 and thereafter.Primary Prediction 1: To determine whether summer counts in the pre-eradication period were an accurate measure of summer monarch population size we compared summer counts and migration counts with overwintering area.A correlation between fall migration estimates and overwintering area for the pre-eradication period, but not for summer surveys, would indicate support for Prediction 1.There would be additional support if summer surveys were correlated with migration surveys and overwintering area in the post-eradication period.We calculated Pearson correlations among all population estimates for pre-and post-eradication periods.We calculated the Fisher's exact test comparing expected number of significant correlations to observed number of significant correlations.Corollary Prediction 1: To examine the discrepancy between summer survey counts and overwintering hectares, we made the two variables commensurate by standardizing the entire time series for each survey by subtracting the mean and dividing by the SD.For each summer and migration survey we examined the standardized difference between counts and overwintering area for both the pre-and post-eradication periods.If a survey's count is equally good at matching overwintering hectares during the pre-eradication period as the post-eradication period, then the median difference for each period should not be significantly different.If summer surveys underestimate overwintering hectares in the pre-eradication period then the median difference should be lower in the pre-eradication period than the post-eradication period.We also examined the change in magnitude of the difference from the beginning to the end of the pre-eradication period using simple linear regression.A trend of decreasing discrepancy from the beginning to the end of the pre-eradication period for the summer counts would also provide support for Corollary Prediction 1. Corollary Prediction 2: To compare changes in population estimates based on summer surveys, fall migration surveys and overwintering area we separately regressed indices against year for the pre-and post-eradication time periods.During the pre-eradication period, a decline in the migrating population similar to that of the overwintering population, but no such decline for the summer surveys, would indicate support for Corollary Prediction 2. In addition, we compared the mean index values for all three surveys for the preand post-eradication periods using a Mann-Whitney U test.If index values were higher in the pre-eradication period than the posteradication period it would indicate that the population had declined.A higher index value for the pre-eradication period compared to the post-eradication period for the migrating population and the overwintering population but not the summer counts would indicate support Corollary Prediction 2.

T A B L E 3
Correlation r, with sample size n and significance p (α = 0.05), between overwintering area and survey indices for the pre-and post-2006 periods.

F
I G U R E 2 Time series counts for all data sets with linearly regressed trends for pre-and post-eradication of milkweed from agricultural row crops.Time series are regressed independently for the pre-and post-2006 periods (with 95% confidence intervals).with overwintering area.Saunders et al. (2019) andZylstra et al.
These discrepancies have several possible explanations.Illinois and Ohio fall surveys only capture monarchs passing through a state-wide area.While overwintering monarchs are drawn primarily from the Midwest, state-wide surveys may not accurately represent the totality of the population.The Illinois survey is not weekly, like the Ohio survey, and may miss pulses in migration; we suggest counts in 1994 and 1995 were underestimates of the overwintering population.Ohio fall counts matched the largest overwinter population in 1996 but appears to have overestimated the 1997 population for unknown reasons.Additionally, the proportion of migrating monarchs reaching México varies among years, leading to discrepancies between fall and winter numbers.For example, Monarch Watch tagging overestimated overwinter hectares in 2000.The year 2000 had the lowest tag recovery rate of the years 2000-2015 indicating higher than average migration mortality, probably driven by extreme drought conditions inTexas(Taylor et al., 2020).Saunders et al. (2019) andZylstra et al. (2021) found that summer abundance explained the majority of the variation in overwintering area for the post-eradication period but that the greenness of the Texas region and the amount of dense cover at the overwintering colonies explained important additional variation.Although conditions during migration played a role in determining overwinter hectares,Taylor et al. (2020) andZylstra et al. (2021) found no indication that there had been a progressive deterioration in conditions during migration as the migration mortality hypothesis would predict.T A B L E 5 Statistical summary (medians pre-and post-2006) and differences (W = Mann-Whitney U test) between pre-and post-2006 period counts for all monarch butterfly survey data sets.

Figure S9 )
FigureS9) and found that summer counts remained relatively constant while the population declined, as is observed in the actual summer counts (Figure2c,g,e,h).Even if we observed a decrease in density in the non-agricultural habitats, we would underestimate the decrease in the population as a whole because we would not know the decrease in density in the crop fields which would certainly be greater, given many fields would no longer have milkweeds to draw monarch visitors.None of the data sets used to estimate annual monarch population size, including overwintering area, is without shortcomings.Specific shortcomings of each of the data sets are listed in Appendix A. In general, some of the fall surveys did not begin early enough to cover the years 1995 and 1996 and thus could not verify the very high number of overwintering hectares observed in those years.Also, the early years of surveys often had fewer participants.Some surveys were not conducted at a constant frequency throughout the summer or fall and, thus, may over-or under-sample certain periods of time.Despite shortcomings of all surveys, it is important to note that posteradication, all measures of monarch population size were highly correlated with each other (FigureS6), meaning that it is possible to use any of these measures to accurately gauge changes in monarch population size after ca.2006.

Figure S2 .
Figure S2.Decline of milkweeds in corn and soybean fields based on yearly monitoring of a 6 fields near Ames, Iowa (Pleasants, 2017) and surveys of fields in central Iowa in 1999 and 2009 by Hartzler and Buhler (2000) (H&B) and Hartzler (2010) (H). Figure from Pleasants (2017).The y-axis refers to what percentage of initial stems counted, or initial density, remains at each time interval, beginning with 100%.

Figure S3 .
Figure S3.Location of North American Butterfly Association (NABA) (blue dots) and Ohio Lepidopterists (orange dots in Ohio) and Illinois Butterfly Monitoring Network (IBMN) (orange dots in Illinois) survey sites used (figure from Zylstra et al., 2021).

Figure S4 .
Figure S4.Location of Monarch Larva Monitoring Project (MLMP) sample sites.We used data from sites in the North Central (NC) region (Figure from Stenoien et al., 2015).

Figure S5 .
Figure S5.Location of the three 5 Â 5 latitude/longitude Midwest sectors used for Monarch Watch tagging data (Monarch Watch, 2022).

Figure S6 .
Figure S6.Pearson correlation plots, pre-eradication (upper) and posteradication (lower) 2006.Upper portion of the panel provides the correlation estimate; the lower portion of the panel provides an ellipse depicting the direction of the correlation (positive in blue, negative in red) and the strength of the correlation (darker blue or red with a more line-like

Figure S7 .
Figure S7.Time series of monarch butterfly surveys conducted during summer, fall, and winter, with breakpoints at 2007 and 2008.

Figure S8 .
Figure S8.Half-violin and dotplots depicting distributions of monarch butterfly survey indices pre-and post-eradication.

Figure S9 .
Figure S9.(fromPleasants et al., 2017).In this scenario the monarch population declines as the number of milkweeds declines.The curve for the butterfly population decline is based on actual estimates of the number of milkweeds in the Midwest over those years.The butterfly count is based on the amount of the population that is in non-agricultural habitat (equivalent to density in a non-agricultural habitat).By the time we get to the post-eradication period, 2006 and beyond, the majority of the population is found in non-agricultural habitats.

Figure S10 .
Figure S10.Map with location of funnel points where migration counts occur.From west to east: Peninsula Point, Michigan, United States; Long Point, Ontario, Canada; Cape May, New Jersey, United States.
Summary of monarch butterfly surveys conducted in summer, fall and winter and abbreviations used in the text.
Wilcoxon Rank Sum Test of pre-and post-2006 period medians of standardized differences (fall or summer survey minus overwinter survey).
T A B L E 4