The case for using geography to understand phenology

 Alexandra Permar and Conor Flynn collaborated on this project.

Calochortus is an astonishingly diverse genus of flowering monocots, with common names that include Mariposa lily, Sego lily, and globe lily, as well as pussy ears, fairy-lantern, star-tulip, and others.  Their center of diversity is in CA, which has 48 out of the 78 total species in the genus.  In Arizona, there are 6 species, ranging in color from white, to pink, to yellow, and sometimes orange, with diagnostic darker markings on the inner base of the petals.  

    

    This genus in Arizona makes an interesting case study of biogeography and phenology because each species occurs in a relatively distinct (but overlapping) elevation and geographic location.  Also, because the flowers are very showy, there are more than 2,400 Research-grade observations documented on iNaturalist.   

    The plant grows quickly from underground tubers in the spring, flowers quickly, and then dies back; the growing stem and then the seed heads are relatively inconspicuous and are rarely photographed.  This means that almost all of the iNaturalist observations are of the flowers, which makes studying the phenology easier because there is less work needed to annotate images as flowering or not flowering.  Nevertheless, we still reviewed observations and annotated and removed any photographs that didn't show flowers from the phenology analysis below.  

Methods 

Research-grade Calochortus observations in AZ were exported from iNaturalist and processed in Excel and graphed in Excel and Tableau.  iNaturalist Research-grade observations are used for research, but observations with inaccurate locations can lead to misleading analyses and must be removed.  

A total of 2350 observations were exported.  Taxa were classified by species (even though subspecies IDs exist for several hundred observations).  Cleaned records with geopositional accuracy > 1000 m (excluding blanks; there are 594 blanks).  Removed 61 records with geoprivacy obscured. (In iNaturalist, obscured geoprivacy impacts the geopositional accuracy of an observation to +/- 500 square kilometers.)  8 observations were noted as not flowering and were removed from analysis.  After cleaning, the database contained 2066 records (observations).  This shows that about 88% had location precision that met our inclusion criteria.

 Calochortus kennedyi had the most observations (782),  C. ambiguus and C. flexuosus tied for 2nd most observations (534), while C. aureus (148) and C. nuttallii (61) had the least number of observations. 

Geocoding Elevation

iNaturalist does not record elevation of observations, so it is necessary to intersect the observation points with a Digital Elevation Model in GIS.  Observations were added to ArcGIS Pro and displayed using their XY coordinates.  Connect to elevation in Esri AGOL Living Atlas : Ground Surface Elevation - 30m (image service / raster DEM).  Use Geoprocessing tool Extract Multi Values to Points ( Need Spatial Analyst license).  Copy resulting data back to excel. 

        Tableau Public was used to graph data.  Calochortus Bio-Geo-Phenology | Tableau Public

Results: Biogeography and Elevation

    Ordered by average elevation, Calochortus kennedyi and C. flexuosus are the lowest elevation species, C. aureus, C. ambiguus, and C. nuttalli occur at middle elevations, and C. gunnisonnii occurs at the highest elevations.  

iNat allows visualization of biogeographic comparisons.  (Or new compare tool) 

C. kennedy (blue) is clearly visible at lower elevations along the foothills of the AZ mountains, and then extends to the West beyond the distribution of any other species.  C. flexuosus (green) is also visible to the NW of AZ, extending through central AZ in the lower elevation valleys of the Verde, Salt, and Gila rivers.  C. gunnisonnii (purple) is clearly visible in Colorado, barely extending down to the high country in AZ.  C. aureus (pink) stakes out a unique territory on the Navajo nation in NE AZ and SE Utah. C. ambiguus (red) is found throughout the mountains and high country of AZ.  

 Graphing both latitude and elevation help to illustrate this biogeographic comparison.  In the graph below, it can be seen that C. kennedyi is the most southerly of our Calochortus, and occurs at the lowest elevations.  C. flexuosus occurs farther north, but also at low elevations.  C. ambiguus marks out a consistent territory at higher elevations than other species, depending on elevation. Gunnisonii is a clear outlier, occuring only at high elevation in our study area.  There is a fair amount of overlap between C. aureus and C. nuttallii.

        This biogeographic comparison can help differentiate similar-looking species.  There are 4 white Calochortus in AZ.  C. gunnisonni is clearly only a high elevation species, although C. ambiguous can also occur at high elevations.  In the northern part of the state, C. flexulosus occurs at lower elevations, C. nuttallii at middle elevations, and C. ambiguous only occurs at the highest elevations. 

Results: Phenology

Overall, for Calochortus species in AZ, flowering begins at low elevations around week 13 (last week of March) and progresses to higher elevations, generally wrapping up around week 29 (third week of July).  Flowering species composition changes from C. flexuosus and C. kennedyi to C. ambiguus, with C. aureus and C. nuttalli thrown into the mix.  C. gunnisonni (barely visible at the far upper right) is, of course, the last to flower.  This chart also highlights how C. aureus is mainly restricted to 1600-1800 meters ( 5200-5900 feet).  

    Calochortus ambiguus shows the strongest relationship between elevation, latitude, and flowering date.  Based on this scatterplot, we were able to confirm that the location is not accurate for the observation at the far left side of the plot. 

Calochortus kennedyi, C. aureus, and C. flexulosus do not show as strong of a pattern, and C. gunnisonnii did not have enough observations in AZ to analyze.  

Example of C. kennedyi:

Graphing flowering week of C. kennedyi separately for Latitude and for elevation shows that there is some pattern, but it is clearly quite week.  The R squared values for these charts are 0.22 and 0.12.  Interesting, it looks like flowering actually starts at middle elevations (around week 11), then flowers appear at lower and higher elevations (week 16), until finally only the higher elevations are still flowering (week 20).  C. aureus and C. flexulosus show somewhat similar patterns (data not shown).  

Discussion

    There are many possible ecological and data collection reasons for weak r values:

1) Micro-site variation: south-facing aspect may have more impact than elevation.  Whether a site is forested, if there are shadows from cliffs, etc.

2) Genetic variation:  if a single species showed up with 2 distinct populations, that would be evidence for possible speciation or subspecies, but all populations showed continuous variation.  There may still be genetic variation between meta populations, or within single individuals.

3) Flowering period not known: our records only show flowering presence, not the start of flowering or the total period of flowering.  Correlations might be better if we had complete data, but our data is not complete: we only have point observations.  For example, if species A flowers from April 1-30 and Species B flowers from April 28-May 28, a visitor on April 28 would record both species as flowering at the same time, even though they do have distinct flowering periods.  

Joshua Trees Flower Episodically in Arizona

Intro

Joshua trees are large, visually-striking trees in the Mojave desert.  Most of them are in California and Nevada, but there is also a population in the western part of Arizona.  They are frequently photographed and there are more than 17,000 observations (1,000 in AZ) saved on iNaturalist, a website and app used to document biodiversity.  

Arizona Joshua Tree locations observed on iNaturalistiNaturalist.

Joshua trees produce showy clusters of white flowers at the tips of their branches in March, but this year I noticed they were not flowering.  I wondered how often they produce flowers and decided to answer this question using data from iNat.

53 observations in Arizona showed evidence of flowering, always in March and April.  

Methods

iNat observations of Joshua Trees (Yucca brevifolia) in Arizona were marked using the Plant Phenology option in the Annotation Field.  Plant phenology (flower budding, flowering, or fruiting) was determined based on visual inspection of the saved photos in iNat.

The iNat website has a nice phenology visualization tool.  This view is filtered to show only Arizona observations.  However, it cannot show differences in phenology from one year to the next.  

Observations were then filtered on the iNat Explore page by adding &term_id=12&term_value_id=13 to the URL and downloaded for analysis in Excel.

More info about using iNat search URLs.

Results

I graphed the results starting in 2017, because there are fewer iNat observations before 2017.  It appears Joshua Trees flower episodically.  According to this data, Joshua Trees flowered in four out of the last 7 years.  They appear to alternate years from 2017-2020, but then skipped 2021 and flowered in both 2022 and 2023.  

It is interesting to note that the Arizona Joshua Trees didn't flower in 2020, which was one of the wettest springs in the last 10 years. I wonder whether the spring moisture determines flowering, or if perhaps other climate variables, such as moisture in the fall, are more important.  Another possibility could be that the trees are only able to flower every other year, and that the trees were inhibited from flowering in 2020 because of the large number that flowered in 2019.  

According to this data they were able to flower in both 2022 and 2023, but at reduced numbers both years.  Perhaps the plants flowering in 2023 were different from the trees that had already flowered in 2022?  I can't answer that question with this iNat data.  While it appears that the areas of flowering in 2022 and 2023 were the same, not all of the Joshua trees in a given area necessarily flower, even in good years...

More Springtime NDVI Observations

 I recently took a driving tour from Prescott to Cottonwood and back.  Most areas have dried out already.  This post reviews the phenology, NDVI greenness maps, and rainfall patterns in the area.

Verde Valley  - wildflowers, hedgehog cactus flowers

2,437 32 AGDD 

501 50 AGDD

I-17 and General Crook  - mesquite just starting to flower, grasses dry

2,202 32 AGDD 

394  50 AGDD

Verde Valley native desert grassland with scattered juniper.  Some wildflowers are present, but most areas are dry.

Dugas - mesquite just starting to flower, grasses dry

2,414 

462 50 AGDD

169 and I-17 - Mesquite still leafing out, some grasses still green

2,224  32 AGDD 

400 50 AGDD

Hills near Dugas with dry invasive annual grasses.  Few to no flowers present.

89A at base of Mingus mountain grasses greening up, still early in the growing season

1,773  32 AGDD 

89A at base of Mingus mountain powerline ROW showing early spring green up of cool season grasses.

NDVI

DroughtView is still showing anomalously green areas between Cordes and Flower Pot along I-17, but these hills around Dugas are already quite dry (see image above).

Most recent (4/6-4/21) NDVI difference map

It matches the NDVI variance from that time period.  At this time there was still anomalous green up around Dugas, for example.  

MODIS NDVI (Near Real Time 8 Day) 4/6-4/21

DroughtView also shows straight NDVI, with a nice mask that only shows areas that are green, and has much more recent data.  For example, the current Near Real time NDVI shows greenness only in the mountains and areas of mesquite that have greened up, shows brown over much of the grasslands.  This is what I observed driving through the area: it has already browned out.  This tool allows visualize of current state of greenness on the landscape.

The difference a few weeks can make for springtime greenness.  Note the difference along I-17 from Cordes to Flower Pot.

MODIS NDVI (Near Real Time 8 Day) 4/22-5/7

QuickDRI (updated 5/6) is still showing "improvement" in drought stress around Congress and South of Cordes lakes, but NDVI (above) shows no greenness there.  I think it could be mesquite leaf out, but interesting that NDVI NRT doesn't show it.  Both maps continue to agree that the area around Yava still looks good.  QuickDRI shows "stress intensification" in the Bradshaw and Mingus  mountain areas that NDVI NRT shows as green.  Both could be true.  QuickDRI shows "improvement" in Chino Valley, an area that NDVI NRT doesn't show any greenness.  This area is still somewhat greening up, so QuickDRI may be more accurate there.

QuickDRI (updated 5/6)

Phenology and Accumulated Precipitation

Most areas are below normal 90 day precipitation, except a small area around Perkinsville.  This area doesn't show up as being anomalously green, or having any greenness in NDVI NRT, or having lower stress in QuickDRI.

Feb-May rainfall

Based on AGDD, it is possible that this area is still too early in Spring and needs more time to green up.  

The area around Congress has received some moisture in the last 60 days, but it may not be enough to compensate for the low rainfall in February, or the high temperatures and complete lack of precipitation over the last 30 days.  

March-May rainfall

April rainfall

The Knife Edge of Spring

"To what purpose, April, do you return again?

Beauty is not enough.

You can no longer quiet me with the redness

Of little leaves opening stickily."

-from Spring by Edna St. Vincent Millay

In AZ, the spring green up of grasses and trees makes use of a narrow window of time between declining winter moisture and increasing summer heat.  Our redbud trees blooms for three or four days (this year, from April 26-29), the lilac bushes hardly last much longer, and the spring green up of weedy lawns is over by the end of May.  June is the dry season in Arizona.  June is when spring dies.

The knife edge of spring in Arizona can be seen in the biomass production.  For example, the Rangeland Analysis Platform shows that peak spring growth in Dugas, AZ in 2020 lasted approximately 3 weeks, from 4/1 to Earth Day 4/22.  By May, the growth of the annual green up was already in freefall.  More examples can be seen in this previous blog post using RAP to investigate biomass production variability from year to year.  

Phenology Mapping

The National Phenology Network's Visualization Tool can be used to follow the spring green up via Accumulated Growing Degree Days (AGDD), which work by adding up all of the days and temperatures above some minimum threshold for growth, usually 32 F for cool season plants (winter annuals, cool season grasses, willows and cottonwoods and other early spring plants) or 50 F for warm season plants (mesquite, acacia, and other warm season trees like Chilopsis, walnut, and warm season grasses and forbs).

AGDD Details

A "growing degree day" (GDD) is calculated by subtracting the threshold temperature (T) from the average temperature for each day when the minimum temperature is above the threshold temperature of 32 or 50 degrees. 

Average T = (Maximum T - Minimum T)  / 2

GDD = Average T - Threshold T

 Accumulated growing degree days (AGDD) simply adds up all of the GDD since the beginning of the year.

AGDD = GDD (January 1) + GDD (January 2) + … + GDD (yesterday)

To use the NPN Visualization Tool for AGDD, choose Map>Base Layer>Category: Daily temperature accumulations>Layer:Current Day.

AGDD Example - 32 F

The figure below shows AGDD around Prescott, AZ from a threshold temperature of 32 as 5/2/24.  Most of the map shows AGDD of about 2,000; since there have been about 120 days since the beginning of the year, that works out to an average daily temperature of 48 degrees F (2,000/120 + 32 =  48).  Of course, some areas have been warmer than that and some have been cooler:  

Phenology AGDD from 32 F 

1,600 - cottonwoods leafed out (Prescott)

1,800 - still winter grasses not green (Kirkland junction)

2,200 -  annual grasses very green, early spring wildflowers blooming (Dugas)

2,245 - mesquite not leafed out yet, annual grasses still somewhat green (Yava)

2,500 - mesquite leafed out, annual grasses brown (Date)

2,800 mesquite flowering (Congress)

AGDD Example 50 F

Or maybe a 50 degree threshold better shows mesquite and grass green up? The average daily temperatures, of course, are the same, but the different threshold yields much lower AGDDs, mostly in the low 100's.  On this map, light green area are grasses already brown and mesquite leafed out, green areas actually are green fields of annual grasses, whereas white and blue are areas where willows, cottonwoods, and elms have leafed out, but herbaceous plants are just barely getting started.

Phenology AGDD from 50 F 

240 - oaks turning brown (Kirkland)

350 - mesquite not greened up yet, annual grasses and forbs green (Dugas)

400 - mesquite not leafed out yet, annual grasses still somewhat green (Yava)

550 - mesquite leafed out, annual grasses brown (Date)

800 - mesquite flowering (Congress)

AGDD Anomalies

The examples above of specific phenology AGDD values can be used in conjunction with NPN's Visualization Tool to predict plant growth stage.  Also, it is possible to use NPN's Visualization Tool to highlight geographic areas that may be ahead or behind the usual spring green up using the "Anomaly" visualization.  In the figure below, anomaly from 32 F AGDD, it can be seen that the Kirkland valley, Black Canyon City, and the Verde valley are behind (blue) normal phenology, whereas Prescott valley and especially the area north of Bagdad are ahead (red) normal phenology.  

"Life in itself 

Is nothing,

An empty cup, a flight of uncarpeted stairs.

It is not enough that yearly, down this hill,

April

Comes like an idiot, babbling and strewing flowers."

-from Spring by Edna St. Vincent Millay

Rangeland Analysis Platform

New data source for In-Season NDVI:  Rangeland Analysis Platform. (RAP) https://rangelands.app/rap/ 

RAP allows mapping of Cover and Biomass, and generates reports for an Area of Interest for Cover, Annual biomass, and 16-day biomass.  I'm hopeful they will upgrade the map to include 16-day biomass.  If they did, I could add it to the comparisons of the other NDVI sources.  Mapping would allow in-season management decisions based on forage production.

Case Example: Dugas, AZ

This series of years from 2018-2023 shows the variability in biomass production by season in a desert grassland at mid-elevation (4,000 ft) in AZ:

2018 shows a drought year, when there was little to no spring green-up due to a lack of winter precipitation, and a low green-up in response to summer monsoons.

2019 and 2020 were the "nonsoon" years, when the summer monsoons failed to materialize.  However, because the winter rains were good in 2019 and exceptional in 2020, total production was high.

2021 and 2022 show the potential for growth in years of good monsoon rains.  2023 shows a "normal" year with bimodal peaks in production corresponding to the spring green-up peaking in late March, and the summer monsoons peaking in mid-August.  However, for some reason this year had almost no annual biomass production associated with the monsoon.  Each year is different!

----

Case Example: Congress, AZ

This series from around Congress, AZ shows the extreme variability of plant growth in the Sonoran desert (2,500 ft).

In drought years like 2018 and 2022, there is almost no plant growth, whereas the extreme winter precipitation year of 2020 annuals produced almost 130 pounds/acre of spring growth.  None of the years hadmuch perennial herbaceous production, and monsoons inconsistently produce up to 40 pound/acre of growth in good years.  

----

Case Example:  Grand Canyon Junction 

SR-64 and SR 180 intersection, just south of Grand Canyon high-elevation grassland (6,000 ft).

Maximum production compared to the lower elevation sites is lower, only reaching 50 pounds/acre in good years.   However, total annual production is usually more consistent.  There is still the potential for bimodal production peaking in the late spring (early June) (2023 and 2017, not shown) and in the monsoons.  The monsoon peak seems to be most consistent, except in 2019 and 2020 when the monsoons failed - luckily those years had relatively good spring growth.   

In contrast to the low desert site, annual production (red) is usually less important than perennial production (green) at this site:

Spring Update: NDVI Differences

Last September, I wrote about Finding the Greenest Place in AZ.  This Spring, we have continued to compare and evaluate the different NDVI difference mapping applications and compare them to the actual growth of wildflowers and grasses we see when we go out hiking.  

Methods

I conduct pre-field research to identify predicted greenness/moisture from UA's Droughtview, USGS VegDRI, and NWS Accumulated Precipitation.  I take a screenshot of each product and assign the proposed site a scale from 1 (driest) to 10 (greenest).  We then visit the area and evaluate the plant production, recording example photos of overall landscape greenness, as well as assigning a score.  The data are organized in a OneNote table.  I then compare the numerical scores in an excel table, adding up the differences between each model and the observations.  

Results

So far, the UA model seems to slightly overestimate greenness, the USGS model greatly underestimates greenness, and the NWS precipitation record comes out closest to observation.

For the UA model, I think it might be helpful to have a difference from maximum, instead of the difference from period. The latter overestimates early spring greenness when the denominator NDVI is very small, so any amount of NDVI in the numerator saturates the index.  Using the maximum NDVI for that pixel could help with this phenology issue.  Plus, % of maximum NDVI may be more intuitive than “difference from average”.

The USGS VegDRI index consistently estimates pre-drought to severe drought in areas that have above average precipitation this water year and have an NDVI above average.  This leads me to think that VegDRI 7-Day eVIIRS is either not well calibrated to the desert southwest, or perhaps that it is better used as a predictive index – perhaps these areas are drying out even though they currently appear green?  However, SWCC does not show significant vegetation drying yet in the areas I assessed.  

Examples

Wingfield Mesa:  UA Droughtview shows this area at maximum NDVI (for this time of year)(=10/10), USGS showed it as pre-drought to moderate drought (4/10) , and NWS shows 125-200% of normal year to date precipitation (9/10).  It is quite green, but it is still early in the growing season and the mesquite have not leafed out yet.  We rated it 6 out of 10.  

Dugas Rd:  UA shows above average (8/10), while USGS showed Moderate drought (3.5/10) and NWS showed slightly below average precipitation (75-90%) (3.5/10).  It is quite green right now, but again not quite at maximum greenness production.  We rated it 8 out of 10.   

Biodiversity Phenology

 There are many ways to explore biodiversity using iNaturalist.  One way to study the biodiversity of a geographic area is with phenology:  the science of when things happen.

For example, it is possible to modify the search URL to find observations of plants within 100 km of Dublin, OH, to search for specific months.  The count of observations for each species give some idea of their abundance in that month.

Central Ohio's most common flowering plants in March.

Central Ohio's most common flowering plants in April.

It is interesting to note that the number of observations for e.g. Bloodroot (Sanguinaria canadensis) increase from March to April, but its relative abundance appears to decrease as Dutchman's Breeches (Dicentra cucullaria), Trout Lilies (Erythronium americanum) and other species are photographed much more in April.

Phenology, Accumulated Growing Degree Days, and Soil Moisture

US Crop Calendar

Source: https://ipad.fas.usda.gov/countrysummary/Default.aspx?id=US

Arizona had a good year for NDVI

Source: https://glam1.gsfc.nasa.gov/

NASA SMAP data.  Data is global.

This mapped layer is delayed by 2 weeks.  I haven't found a layer that shows real-time moisture.

NPN Visualization tool can view Historical, Current, and Anomaly Accumulated Growing Degree Days. Data is only for USA.

Source: https://data.usanpn.org/vis-tool/#/explore-phenological-findings