Modelling Species Distribution for 4 Species of Maple (Acer)

{Introduction}

Manmade climate change will continue to cause drastic changes in temperature and rainfall, ultimately causing great disruption to current forested areas. The biodiversity of forests [1], woodland species relationships and dependence on forests [2], and economical services (ie. timber and pulp) will experience detrimental impacts [3]. There is a great need to understand how and when these changes will occur and begin planning for management, mitigation, and conservation.

University of British Columbia Botanical Garden (UBCBG) is one institution dedicated to understanding future climatic impacts over the species they currently house and determine whether or not their geographic location is the best area for these species to thrive.

This report examined 4 maple trees (pictured above): Five-finger maple (Acer pentaphyllum Diels), Bloodbark/Paperbark maple (Acer griseum (Franch.) Pax), Vine maple (Acer circinatum Pursh), and Bigleaf maple (Acer macrophyllum Pursh). Recent science has shown that many maple trees actually migrated to North America during warm conditions over the Eocene (roughly 56 - 34 million years ago) [4] and of the 91 species that originally came to exist across 28 distinct areas of North America, only 5 species in 5 areas remain today [5,6]. One study theorizes this decline of biodiversity was a result of the Quaternary glacial cycles (~2.6 million years ago) [4] as maple trees struggle to succeed in areas where the average annual temperature is at or below 0°C [5]. The study by Jake J. Grossman[5] looked at 184 maple species to determine the evolutionary journey, relationships, and similar ecological areas maples trees may share. They concluded that nearly every species of maple that exists today has its own unique preference of climatic conditions, and making an assumption that one maple's temperature/precipitation preference is the same as another's is wrong. This means that, if UBCBG wants to succeed in their inquiry of having a suitable location for maple trees, investigations into each species will need to take place; making a broad assumption for all maple trees would be inappropriate.

{Study Area}

University of British Columbia Botanical Garden is located at 49.2478293° N 123.246107° W on xʷməθkʷəy̓əm (Musqueam) territory within the University of British Columbia, Vancouver, British Columbia (Canada). This is considered the Lower Mainland Ecoregion, which can experience warm drought-like temperatures during the summer and cold air from the Arctic over winter[7]. There are air currents from the mountains in the North and East that largely control the weather and suitable vegetation areas; the ‘normal’ average annual temperature for this area is 6.7°C, and average annual precipitation is 1664 mm.

{Methods}

In order to situate UBC Botanical Garden most appropriately, I elected to examine these species under two different climate change scenarios. These are new pathways recently introduced in the latest International Panel for Climate Change (IPCC) report and are now referred to as 'shared socioeconomic pathways' or SSPs. These not only address the carbon-equivalent greenhouse gas warming that certain emissions will cause, but also consider the nation-to-nation relationships.

SSP245 is defined as 'the middle of the road' [8] and has a projected warming of 2.7-3.4°C higher than that of temperatures from 1800-1900 [17], and comes with roughly 1.5-8.3% more precipitation than the average over 1995-2014. SSP370 is defined as 'rocky road / rivalries' [8] and has a projected warming 2.0-3.7°C higher and precipitation 0.5-9.6% more abundant [17].

From a shared socioeconomic lens, SSP245 (the 'middle of the road') models a scenario that is not much different than the current historical trend humanity has been on. Some countries adopt green technology, whilst others fall behind; there is slow progress towards meeting sustainable targets, and mitigation/adaptation will be a great challenge [8]. SSP370 (the 'rocky road/rivalries'), however, models a scenario where increasing hostility between nations causes rivalry towards resources and hinders technological growth [8]. Consumption prevails and the inequality gap between developed and underdeveloped countries grows large; there is a focus on self perseverance and very little cooperation between nations, making mitigation/adaptation a very great challenge for the globe [8]. Both scenarios were downloaded from WorldClim at a resolution of 2.5 minutes; future SSP scenarios were derived from the MIROC-ESM2 global circulation model.

The data were cleaned for duplicate points and preserved for only latitude and longitude information (there were other data about name, sponsorship, where the seed came from, and more that I didn't need). After cleaning, bloodbark maple had 66 observations (from 71) and five-fingered maple had 29 observations (from 47) to work with. Bigleaf and vine maple data were downloaded from the  global biodiversity information facility  and cleaned by removing data containing unavailable latitude/longitude information, duplicate points, and being considered a 'PRESERVED_SPECIMEN' as that indicates a tree is contained in an herbarium and potentially not representative of the true distribution. As both of these North American trees happen to be elsewhere in the world, I further filtered the points by being between 130°W and 90°W - the edges of North America. This meant working with 4550 observations for vine maple, and 10879 for bigleaf maple.

Before we get into what went wrong, a better understanding of Maxent is needed. Maxent is a favourable machine learning algorithm many scientists have utilised for creating species distribution models, like you've seen above, because it only needs 'presence' data. That means you only need to know where the trees are, and not also where they are not - which other models (like generalized linear models) often require [9]. As absence data is rarely collected in field missions, Maxent has become the ideal way to conduct species distribution models. The principle behind Maxent lies in the ‘maximum entropy principle’, defined by Jaynes in 1957 that theorizes finding an unknown distribution of probabilities (for existence) is best done by understanding all the constraints around that probability [10]. This leaves a maximum amount of entropy (or 'chaos' or 'disorder') that can then be used to make inferences about incomplete data [9]. In its essence, Maxent estimates the probability that a species will occur within a geographical space given some change of parameter [11]. It really should not be used to find probability of occurrence for the entire world, as that means making an assumption that the entire world experiences an identical climate to that of where the species truly occur. Due to the small amount of data for the five-fingered maple (29 points) and this presumption, we see the result above.

How exactly did I do this?

I used a 'bootstrap method' with Maxent to find which of the variables contribute the most to predicting the distribution of both species in China. For five-fingered maple, I told Maxent to develop predictions based on 28 of the data points, and used the 29th point to test how well the model developed. This involved running Maxent 29 times for the five-fingered maple, and then 66 times for bloodbark maple. For the North American species, bigleaf and vine maple, I divided the data arbitrarily and used 90% of the points to train Maxent, and test the efficiency with the remaining 10% of the points. I did this 30 times over.

{Results}

For the most part, Maxent determined anywhere from six and only 2 variables of the 20 tested would predict the species distribution you see below. Given this information, only five-fingered maple will find the climate at UBCBG suitable, whilst the other 3 species of maples will not. Differences between the climate scenarios (SSP245 and SSP370) were not apparent over UBCBG and the greater area, and projecting to 2050 and 2090 showed nearly identical maps. The table below shows the final variables that predict each species' distribution, and these variables were used to compose the maps you see below this. For the sake of space and being excessively repetitive, I've only included side-swipes of the endangered maple species (five-fingered maple and bloodbark maple).

As you can see from the above table, only 2 variables were included in predicting the distribution for five-fingered maple: Mean Temperature of the Coldest Quarter (55.75%) and Precipitation Seasonality (39.93%). This indicates five-fingered maple's distribution depends on cold temperature over winter, and the amount of deviation from a ‘normal’ amount of precipitation that each month exhibits over the course of the year [12]. The model predicts a 100% probability of UBCBG being suitable for A. pentaphyllum given the climatic variables represented in the model. The following 3 maps show the very slight differences between 2050 and 2090 in each scenario, and then the very slight differences between both SSPs in 2050.

SSP245 'The Middle of the Road' 2050 vs 2090

SSP370 'Rocky Road & Rivalries' 2050 vs. 2090

SSP245 vs SSP370 in 2050

Unlike five-fingered maple, there was a total of 6 variables used to predict the distribution of Bloodbark maple: Mean Temperature of the Coldest Quarter (25.60%), Elevation (17.75%), Minimum Temperature of the Coldest Month (12.93%), Mean Diurnal Range (10.54%), Precipitation of the Driest Month (9.57%), and Mean Temperature of the Driest Quarter (8.39%). This species was the only one to have elevation included in it's predictive model, and the variables determined can be summed up as being 'dry' and 'cold'. In the end, Maxent predicted a 14% probability of UBCBG being suitable for Bloodbark maple, given the variables represented in the model. The following 3 maps show the very slight differences between 2050 and 2090 in each scenario, and then the very slight differences between both SSPs in 2050.

SSP245 'The Middle of the Road' 2050 vs 2090

SSP370 'Rocky Road & Rivalries' 2050 vs. 2090

SSP245 vs SSP370 in 2050

Maxent determined only 3 variables were needed to predict the distribution of Bigleaf maple: Precipitation of the Coldest Quarter (73.76%), Minimum Temperature of the Coldest Month (11.23%), and Mean Temperature of the Driest Quarter (6.86%). The model predicted a 20% probability that Bigleaf maple will find UBCBG climate suitable in the future, and - much like the similarities you saw in Five-fingered and Bloodbark maple above - each climate scenario and future projection looked identical. The map to the right was made with QGIS and shows a local, regional, and global probability for Bigleaf maple.

Maxent also determined only 3 variables were needed to predict the distribution of Vine maple, but these differed from the variables for Bigleaf maple: Precipitation of the Wettest Quarter (63.13%), Temperature Annual Range (9.93%), and Precipitation of the Driest Month (6.47%). The model predicted a 30% probability that Vine maple will find UBCBG climate suitable in the future. As mentioned before, the map to the left (also made with QGIS) shows one scenario and one future, but all the prediction maps looked very similar.

{Discussion}

Due to their rarity, there is very little literature to compare these results to. One study done by Jake J. Grossman did examine Bloodbark maple, but noted that Five-fingered maple was deliberately not included in their study due to lack of publicly available data [5]. Most literature about either species is detailing collection expeditions [13,14] or genomic sequencing [4,15,16]. Sequencing studies reveal that these species (Bloodbark and Five-fingered maple) occupy 'sister clades' in the phylogenetic tree, meaning they share more genomic traits than other maple species [4,15]. Both species had bio11 (Average Temperature of the Coldest Quarter) in their final models in this study, but the study done by Grossman showed bio15 (Precipitation Seasonality) having significance in the Maxent model for Bloodbark maple [5] - the same biovariable that composes the rest of the prediction for Five-fingered maple in my study. While I cannot state for certain that these correlate and the reason they correlate is the phylogenetic proximity, it is still an interesting result that warrants further investigation.

Unlike the variable similarity observed with the other two species (Bloodbark and Five-fingered maple), Bigleaf and Vine maple have no similarities whatsoever. Bigleaf maple appears largely defined by 'coldness' as the most contributing variables to predicting its distribution are Precipitation in the Coldest Quarter and the Minimum Temperature of the Coldest Month. Vine maple, however, could be defined by quantity of precipitation (composing ~90% of the final model) and then by Temperature Annual Range (10%). It is worth noting that, unlike Bloodbark and Five-fingered maple, these North American species are no where near related to one another phyllogenetically; in fact, Grossman shows that the species converged from their shared ancestor roughly 50 million years ago [5].

I won't lie - these results are dismal. If Bigleaf maple is largely defined by cold times of the year, and all future climate scenarios anticipate a warming between 2.0 and 3.7°C [17], it makes sense why Maxent would estimate such a low probability of existence under this new climate. It is important to remember that Maxent only concludes this with the information it is given - which is climatic variables and elevation - and no tangential information, like human intervention, assisted migration, and adaptation that the tree is actively learning today. It is also worth mentioning that very few species have gone fully extinct due to global warming since the last ice age [18]. While Maxent provided odds that make a coin-flip look favourable, it did so with an assumption that there are no conservation plans nor migrating/adapting that the species makes. Future analysis and research would be wise to consider adaptation/migration into the analysis, as this can provide the best advice to gardens moving forward in creating comprehensive plans of action [19].

{Conclusion / Future Considerations}

I would perform a secondary analysis on these species before making firm conclusions that UBCBG will or will not be suitable for them. And I would do the analysis differently:

1. For a better idealized future, I would use more variables aside from the 19 climatic ones and elevation. Other variables could be: soil organic carbon, soil pH (acidity), vapor pressure, wet-day frequency, UV-B seasonality [20], surrounding vegetation and communities [21], as well as canopy cover, slope, aspect, soil compactness, and proximity to other landscapes like urban or water features.
2. Future studies should also strive to use presence data that is within the geographical area of interest. Having location points from Vancouver to use in a distribution model results in avoiding the grave assumption that climate here (in BC) is the same as climate over Hubei, China (where the tree locations were).
3. When dealing with small datasets, like Five-fingered and Bloodbark maple, one should use a method known as Ensembles of Small Models. This was demonstrated in a 2015 paper where series of variables were pitted against each other with Maxent, and the accuracy of final models was significantly better than that of normal generalized linear model, Maxent, gradient boosting, or expectation propagation [22].
4. Larger datasets (>1000 records) should be critically examined before analysis, as they may occupy multiple ecoregions. An accurate future prediction should only consider the species within the ecoregion, as making assumptions for climate occurring over a region in ~northern California matching that as interior British Columbia is just as inappropriate as the assumption made in point 2.
5. To realize the full extent of distribution and make plans accordingly, one should include scenarios that represent an adaptation/migration scenario. For example, in this study, it might reveal that Vine maple and Bloodbark maple are actually fine as they have historically adapted to changing climates, but Bigleaf maple will need assistance and human help to survive.
6. Lastly, this study only used one global circulation model to grab future climate scenario values (MIROC-ESM2). In reality, multiple global circulation models should be utilized in making future predictions of species distribution, lest a single model’s bias is overtly optimistic or pessimistic in its projection.

This project would not have been made possible without the assistance of UBC Botanical Gardens and SEEDS. Thank you Tara Moreau for being my mentor throughout this process and providing connections and resources wherever possible. Thank you Adriana Lopez for providing literature, data, and insight in the most timely manner. Thank you Daniel Mosquin for your objectivity, resourcefulness, vast botanical knowledge and imbued ‘friendly Manitoba’ approachability. Thank you Ben Scheufler for organizing and facilitating meetings between all members - it would be absolute chaos without you. Thank you Ginny Hang, Meagan Ng, and Anson Lau for being compatriots, ‘rubber ducks’ and other idea-entertainers throughout this research project. Thank you Ira Sutherland, Gaby Barragán, Martin Queinnec and Paul Pickell for answering my emails at obscene hours of the day/night, meeting with me as frequently as you have, and the constant reassurance that I’m on the right path. Thank you Peter Arcese for your invaluable insight and advice regarding Maxent methodology. Thank you Drew Kerkhoff for your SDM tutorial you’ve graciously uploaded to github and your timely email response. I could not have done any of this without any of you and deeply appreciate the time you have given me to explore the world of species distribution modeling and Maxent.

Photo Credit/Sources:

(1) Header composite: Bigleaf maple image taken by Alan Majchrowicz, pulled from https://fineartamerica.com/featured/big-leaf-maple-trees-iii-alan-majchrowicz.html; Five-fingered maple from the UBCBG forums posted by simongrant, from https://forums.botanicalgarden.ubc.ca/threads/acer-pentaphyllum.8864/page-2; Bloodbark maple pulled from https://www.gardenia.net/plant/acer-griseum-paperbark-maple; and Vine maple pulled from https://www.gardenia.net/plant/acer-circinatum (2) https://landscapeplants.oregonstate.edu/plants/acer-pentaphyllum (3) Adrienne Legault; https://www.thespruce.com/growing-the-paperbark-maple-acer-griseum-3269319 (4) https://www.gardenia.net/plant/acer-circinatum (5) Evgeniya Vlasova; https://www.thespruce.com/growing-big-leaf-maple-3269301 (6) https://landscapeplants.oregonstate.edu/plants/acer-pentaphyllum (7) http://www.tree-guide.com/bigleaf-maple (8) http://www.tree-guide.com/paperbark-maple (9) https://wildernessrim.org/vine-maple-leaf/ (10) https://landscapeplants.oregonstate.edu/plants/acer-pentaphyllum (11) https://www.flickr.com/photos/forestservicenw/34503806910