what's this ?
what's this ?
excerpts here
excerpts out
peer review
Click on a tag above to see relevant excerpts from this site.
Click on a tag above to see relevant excerpts from other articles in the mesh.
Search this article for any word:

Study of specimen images is a convenient way to collect and enter data that allows for crowdsourcing of data entry and expanding data sets by including new fields. Darwin Core database standards include phenological state, however, this field was only added to the MAINE database recently, and most specimens have no data. This paper describes a representative sampling for the potential insights that could be gained in expanding this survey to a larger portion of the collection.

Global climate change is having an observable impact on phenology, i.e., the timing of annual natural events like leaf emergence, flowering, and fruit set. Herbarium specimens collected during these stages provide documentation of phenology that can be evaluated to assess climate change. In this paper, I assess and implement methods used in published studies of phenological change using herbarium specimens. Using the University of Maine Herbarium's digital collection, I selected a handful of commonly collected species, and surveyed the specimen images to evaluate their phenological state. 

Phenological change is a leading indicator of climate change (Bertin, 2015). Like the canary in the coalmine, the timing of annual biological events is precisely tied to climatic conditions and provides an early warning of changing conditions (Everill, Primack, Ellwood, & Melaas, 2014). Herbarium specimens provide a chronological sample of phenological events that can be analyzed for detection of change (Miller-Rushing, Primack, Primack, & Imbres, 2004). However, evaluation of phenological state requires some knowledge of the organism itself and evaluation of phenological state may be inaccurate when recorded inconsistently or by inexperienced observers (Everill et al., 2014). Unlike most other fields used in the MAINE database, phenological state is not a standardized Darwin Core Term (Darwin Core Terms,

The central question in this project is to assess the potential for increased resolution of phenological change using consistent observation. To accomplish this, I have evaluated the database and image records for six species in the MAINE herbarium. At the MAINE herbarium, initial data entry has been an ongoing effort since the early 1990s. Over these 25 years, at least 15 MAINE herbarium workers have been involved in database entry of more than 50,000 specimen records. Variation in individual training, experience, and judgment have lead to inconsistencies in how specimens of the same species have been scored for the phenological state field, "repro_status" in the MAINE database.

The below details workflow for extracting accurate and informative phenological data from database and specimen images. Field names are italicized for emphasis.
1. Criteria for selection of species for further study
~100 specimens
short flowering periods preferable
long history of collection
must be records which have specimen images
2. extract species records to new sheet
3. sort by CDATE
4. convert CDATE to separate columns YEAR, MONTH, DAY
5. calculate DAY OF THE YEAR (DOTY) from MONTH, DAY, by adding the total number of days in the sum of the previous months to the day of the month in which the specimen was collected.
This method of calculating DOTY does not account for Leap Years.

Month #Days/Month #Days total
1 31
2 28 31
3 31 59
4 30 90
5 31 120
6 30 151
7 31 181
8 31 212
9 30 243
10 31 273
11 30 304
12 31 334
6. sort by repro_status
Standardize coding of repro_status field. Field state definitions are: "Fl": flowering, "Fr": fruiting, "Fl/Fr": flowering and fruiting, "Veg": Vegetative, not reproductive material. Eliminate variation in spelling, abbreviation, and capitalization of field states.
7. Record summary data: Family, Genus, species, total # of specimen records, first and late recorded date of collection, #of specimens with phenology field scored, #of specimens with phenology field unscored, #of specimens in flowering state.
8. Survey specimens with no data for repro_status by examining specimen images. Increase quality and quantity of metadata on each specimen by filling in new data for repro_status.
9. Evaluate quality of existing metadata by comparing database records of repro_status with the specimen itself.

Six species were selected for initial screening based on the criteria in step 1: Amelanchier laevis, Euphrasia nemorosa, Prunus pumila, Rosa virginiana, Rubus allegheniensis, and Sibbaldia tridentata (Table 1).

Table 1. Summary metrics for species screened in step 1.
Genus species #records first date last date #scored #Fl
Amelanchier laevis 106 1856-05-01 1999-05-07 105 28
Euphrasia nemorosa 82 1892-08-01 1993-10-11 77 74
Prunus pumila 70 1878-09-26 1983-08-04 70 13
Rosa virginiana 135 1880-09-08 1993-10-02 128 83
Rubus allegheniensis 101 1891-06-01 2003-09-04 101 45
Sibbaldia tridentata 98 1862-07-01 1998-06-26 97 74

Amelanchier laevis and Prunus pumila were not considered further due to a lack of flowering material. The remaining four species were extracted from the database, and for each specimen record, the Gregorian calendar day of the year was calculated following the methods previously outlined. Preliminary scatterplots of change in flowering over time were calculated for each. In order to assess a large and long-term history of samples, additional consideration was given to the number of specimens available for study from before 1920. With 20 collections from before 1920, and 83 flowering specimens available for study, Rosa virginiana was selected for further analysis. 

Surveying specimens with no data for repro_status in steps 8 and 9 revealed that the repro_status field has not been accurately or consistently evaluated in our database. The inaccuracy of the original scoring of repro_status is also evidenced in the scatterplot of flowering time in Rosa virginiana (Appendix, pages 6 and 7). Several specimens are recorded as "flowering" from October, which is phenologically improbable. Examination of specimen images revealed these scorings were incorrect. 

In order to correct these errors, a new field, "peak flowering" was added to the database. Specimens were reevaluated from the images and scored according to the following field states (states in bold). Following the phenological definition used by Callinger et al. (2013), peak flowering time (peak) is defined as having 50% or more of flower buds at anthesis (open). Expanding upon the Callinger et al. (2013) definition of peak bloom, specimens with less than 50% of flowers at anthesis, and more in bud than past bloom are defined as pre-flowering peak (pre). Specimens with less than 50% of flowers at anthesis, and more past bloom than in bud are defined as past-flowering peak (past). A specimen is fruiting (fruit), when fruit are present, but flower buds and spent flowers are absent. Lastly, a specimen is vegetative (veg) when no reproductive structures are present, or only moldy remnants of previous years fruit are present.

Preliminary scatterplots revealed potential differences between phenological responses of plants growing in coastal versus inland climates (Fig. 1). 

Fig. 1. Scatterplot of DOTY and Year, with a trend line depicting change in flowering time.

In order to assess these responses quantitatively, the collection location of each specimen was scored as either coastal (less than two miles from the ocean), or inland (more than two miles from the ocean). Specimen collection locations are mapped with coastal locations marked as red squares and coastal locations marked as blue diamonds (Fig. 2).

Fig. 2. Collection locations of MAINE specimens of Rosa virginiana. See text for further description. 

Comparing the flowering time of coastal and inland plants does not reveal observable differences between these geographical distributions (Fig. 3). Although many of the coastal plants are flowering later in the year than the inland plants, the two geographical groups are not equally represented over the time period being analyzed. Most of the coastal collections were made after 1930, and few inland collections were made after 1960. This incongruent sampling pattern obfuscates broader conclusions. 

Fig. 3. Scatterplot of DOTY and Year, with coastal plants as red squares and inland plants as blue diamonds. 

Although phenology data has the potential to be informative on how species are responding to climate change, the utility of phenology data is limited by a lack of quantity and low quality of collection metadata. Phenology data is available digitally through database records and the metadata can be curated to increase accuracy and add value by validation with corresponding specimen image vouchers. Evaluation of phenological condition requires an understanding and consistent application of the field state definitions, and some knowledge of the organism being studied. 

Comparing the scoring of the old "repro_status" field with the new "peak_flowering," reveals that the original scoring defined the state of the specimen correctly using the repro_status field definitions most of the time. Cells marked in red in Table 2 indicate 39 specimens (29%) where phenological state was scored erroneously in repro_status. Scoring incorrectly can dramatically mislead inferences of phenological change, such as observations recorded as flowering in October when the specimen was actually in fruit. Validating collection record metadata will be essential for making accurate conclusions about phenological change over time.

repro_status peak_flowering # specimens
Flowering Fruit 10
Flowering past 18
Flowering peak 50
Flowering pre 6
Flowering Veg 3
Flower & Fruit past 3
Flower & Fruit peak 4
Flower & Fruit pre 3
Fruit Fruit 5
Fruit past 6
Fruit peak 16
Fruit pre 3
Fruit Veg 1
Veg Fruit 2
Veg peak 1
Veg pre 3
Veg Veg 2
Total 136

Change in flowering time across years is an observable trend in assessment of herbarium specimens (Fig. 1, 3). Sample size, sampling frequency, and geographic distribution of sampling are critical variables in elucidating more fine-grained questions. In this paper, coastal plants were evaluated along with inland plants to assess whether these two subsets are experiencing different phenological trends. Although the sample size may have been large enough to detect a trend, these samples were not equally well represented across time and space. 

Future analyses will be empowered by the increasing number of herbarium collections available online. However, the potential for future inference of biological responses to climate change are limited by the lack of collection efforts being made in the present. Across all of the species assessed in this paper, the frequency of collections has declined precipitously over the last 30 years.

Bertin, R. I. (2015). Climate Change and Flowering Phenology in Worcester County, Massachusetts. International Journal of Plant Sciences, 176(2), 107”"119.

Calinger, K. M., Queenborough, S., & Curtis, P. S. (2013). Herbarium specimens reveal the footprint of climate change on flowering trends across north-central North America. Ecology Letters, 16(8), 1037”"1044.

Darwin Core Terms, A complete historical record. (2015), Accessed 13 December, 2015

Everill, P. H., Primack, R. B., Ellwood, E. R., & Melaas, E. K. (2014). Determining past leaf-out times of New England's deciduous forests from herbarium specimens. American Journal of Botany, 101(8), 1293”"1300.

Miller-Rushing, A. J., Primack, D., Primack, R. B., & Imbres, C. (2004). Herbarium specimens as a novel tool for climate change research. Arnoldia.