Biogeography, Niche Modeling, and Climate-Disturbance Dynamics of North American Pinus

A Four-Module Course-Based Undergraduate Research Experience (CURE)

Author

J.Gibson, Arizona Western College

Published

November 18, 2026

Quaternay context for contempory ecology

The Quaternary is a geologic Period that began 2.6 million years ago and is characterized by ice age cycles. These cycles began somewhat erratically and became somewhat regular with about 90,000 years of glaciation followed by 10,000 years of inter-glacial conditions. These cycles are a powerful selective force on the evolution and ecology of many species, including our own. During this time the predominant conditions are glacial and punctuated by a rapid shift into interglacial conditions, which are relatively brief and soon begin a gradual and steady return to glacial conditions.

In North America the fossil records of plants from glacial periods are assembled into different communities than we know today. That is, the forests of the ice ages where composed of trees that today are differentiated into distinctly different forest types. Species of trees respond individualistically to climate changes. Furthermore, each tree has a unique way of dispersing into areas that become climatically favorably, and each tree has its own way of succumbing to a climate that now longer is favorably. Most tree species have not had a chance to grow everywhere in their region that might otherwise support them. They are limited by dispersal.

North American pine ecosystems of the current inter-glacial period are complicated by three major changes. First, with the end of the Pleistocene (the last glacial epoch) there was a vast dissapearance of megafauna that unravelled ecological interactions and nutrient cycles in ways science is barely becoming aware of. Second, throughout the Holocene (the current inter-glacial epoch), humans have been woven into most pine forests through cultural fire traditions that maintained biodiversity. Abrupt fire suppression through the 20th century led forests to build up fuel loads prone to catastrophic burns. Third, and finally, 21st century climate changes are changing where species can live.

—this intro is still in draft

Principles of Biogeography

To Get There and To Survive: The BAM Framework

A foundational maxim of biogeography states that a species occurs only where it can both get to and survive. In modern spatial ecology, this principle is formalized by the BAM framework (Soberón & Peterson, 2005), which defines a species’ occupied distribution (\(G_O\)) as the intersection of three distinct factors:

  1. \(\mathbf{A}\) (Abiotic Niche / Fundamental Niche): The multidimensional envelope of temperature, precipitation, seasonality, and topographic conditions within which a species can physiologically survive, grow, and reproduce.
  2. \(\mathbf{B}\) (Biotic Interactions): The set of positive and negative interactions with other organisms—including resource competition with faster-growing plants, mutualisms with mycorrhizal fungi and vertebrate seed dispersers, and trophic pressure from herbivores and bark beetles—that expand or contract the abiotic niche into the Realized Niche.
  3. \(\mathbf{M}\) (Movement / Accessible Geographic Domain): The geographic region accessible to the species via seed dispersal over relevant ecological and evolutionary time scales. Even if a distant mountain range possesses ideal climate (\(A\)) and favorable competitors (\(B\)), the species will be absent if it cannot cross the intervening desert or marine barrier (\(M\)).

Throughout this four-module laboratory suite, you will test how physiological tolerances (\(A\)), biotic competition, mutualisms, and pests (\(B\)), and dispersal barriers (\(M\)) interact to shape the past, present, and future distributions of North American pines (Pinus spp.).


Natural History of Pinus

The genus Pinus (Pinaceae) comprises roughly 115–120 extant species that dominate boreal, montane, Mediterranean, and coastal plain ecosystems across the Northern Hemisphere. Following the taxonomic synthesis of The Gymnosperm Database (https://www.conifers.org/pi/Pinus.php; Earle, 2026) and nuclear/plastome phylogenomic reconstructions (Gernandt et al., 2005; Eckert & Hall, 2006; Willyard et al., 2007; Parks et al., 2012; Jin et al., 2021), the genus is divided into two deeply diverged subgenera that split during the Early to Late Cretaceous (~85–125 Ma):

1. Subgenus Strobus (The White or Soft Pines — Haploxylon)

  • Anatomical & Physiological Traits: Needles contain a single fibrovascular bundle (haploxylon), usually arranged in fascicles of 5 (or 1–4 in pinyon pines), with deciduous fascicle sheaths and soft, flexible wood with minimal resin canal density.
  • Taxonomic Sections in North America:
    • P. sect. Quinquefoliae (P. subsect. Strobus): Five-needle white pines (P. albicaulis, P. flexilis, P. strobiformis, P. ayacahuite, P. lambertiana, P. monticola, P. strobus).
    • P. sect. Parrya: Subdivided into P. subsect. Cembroides (the pinyon pines: P. edulis, P. monophylla, P. cembroides, P. quadrifolia, P. remota, P. culminicola), P. subsect. Balfourianae (the ancient high-alpine bristlecone and foxtail pines: P. longaeva, P. aristata, P. balfouriana, plus Mexican relicts P. pinceana, P. maximartinezii, P. rzedowskii), and P. subsect. Nelsoniae (P. nelsonii).
  • Evolutionary Strategy — Abiotic Stress Tolerance & Vertebrate Mutualisms: As lineages within subgenus Strobus radiated across western North America during Cenozoic mountain building and aridification (Jin et al., 2021), many species specialized into abiotically extreme environments where interspecific plant competition is low: either the lower aridity treeline bordering warm deserts (P. subsect. Cembroides) or the upper alpine treeline battered by freezing winds and skeletal soils (P. subsect. Balfourianae and high-elevation P. albicaulis / P. flexilis). In these harsh, rocky, or arid environments, tiny wind-blown seeds quickly desiccate or starve before establishing a deep taproot. Consequently, white and pinyon pines repeatedly evolved large, heavy, wingless, lipid- and protein-rich seeds coupled with indehiscent or upward-angled cones—forming tight mutualistic dispersal syndromes (zoochory) with scatter-hoarding corvids (e.g., Clark’s nutcracker, Nucifraga columbiana, and the Pinyon jay, Gymnorhinus cyanocephalus) and rodents (Lanner, 1996). Corvids harvest thousands of seeds in sublingual pouches and cache them 1–3 cm deep in mineral soil up to 10–30 km away, flying directly across desert basins to sky-island mountain ranges.

2. Subgenus Pinus (The Yellow or Hard Pines — Diploxylon)

  • Anatomical & Physiological Traits: Needles contain two fibrovascular bundles (diploxylon), persistent fascicle sheaths (usually in fascicles of 2 or 3), and dense, resinous wood (hard pines).
  • Taxonomic Sections in North America:
    • P. sect. Pinus (P. subsect. Pinus): Principally Eurasian, represented in North America only by Red Pine (P. resinosa) in the Northeast/Great Lakes and Tropical Pine (P. tropicalis) in western Cuba.
    • P. sect. Trifoliae: The dominant North American radiation, comprising three major subsections:
      • P. subsect. Ponderosae: Western and Mexican yellow pines (P. ponderosa, P. jeffreyi, P. arizonica, P. engelmannii, P. coulteri, P. sabiniana, P. hartwegii, P. montezumae, P. pseudostrobus, P. devoniana).
      • P. subsect. Contortae: Boreal and eastern nutrient-poor/disturbance specialists (P. contorta, P. banksiana, P. virginiana, P. clausa).
      • P. subsect. Australes: Southern yellow pines, Caribbean pines, and California/Mesoamerican closed-cone pines (P. taeda, P. palustris, P. elliottii, P. echinata, P. rigida, P. radiata, P. attenuata, P. muricata, P. patula, P. oocarpa).
  • Evolutionary Strategy — Biotic Competition, Wind Dispersal & Fire Mutualism: Unlike the stress-tolerant white pines, most yellow pines occupy warmer, more mesic, or seasonally productive environments where abiotic stress is moderate but biotic competition from broadleaved angiosperms and shade-tolerant conifers (Abies, Pseudotsuga, Tsuga) is intense. Yellow pines prevent competitive exclusion through an evolutionary mutualism with wildfire (Keeley, 2012):
    1. Fire-Tolerant Resisters (e.g., P. ponderosa, P. palustris, P. jeffreyi): Shed highly flammable, resinous needle litter that fuels frequent, low-severity surface fires every 2–15 years. These surface fires kill thin-barked competing saplings while mature yellow pines survive via thick insulating bark, self-pruned lower branches, high foliar moisture, and (in P. palustris) a fire-resistant “grass stage.”
    2. Fire-Embracing Evaders (e.g., P. contorta var. latifolia, P. banksiana, P. attenuata, P. clausa): Retain seeds inside resin-sealed serotinous cones for decades until a stand-replacing crown fire melts the resin and releases millions of small, winged, wind-dispersed seeds (anemochory) onto competition-free, nutrient-rich ash beds.

Pre-Lab Challenge: Formulating Phytogeographical Hypotheses

Before opening the Shiny applications, complete the following conceptual table and questions in your lab report:

  1. Hypothesis 1 (Climatic & Topographic Breadth): Based on the physiological split between subgenus Strobus (P. sect. Parrya and P. sect. Quinquefoliae) and subgenus Pinus (P. sect. Trifoliae), which clades do you predict will occupy the cold/dry extremes of environmental PCA space and the highest Terrain Ruggedness (TRI) values in Module 1?
  2. Hypothesis 2 (Evolutionary Convergence in CART): In Module 2, both P. resinosa (P. sect. Pinus) and P. banksiana (P. subsect. Contortae) inhabit glaciated northern forests, while P. edulis (P. subsect. Cembroides) and P. sabiniana (P. subsect. Ponderosae) inhabit hot, dry woodland foothill ecotones. How might a CART decision tree reveal ecological convergence between distantly related sections?
  3. Hypothesis 3 (Corvid vs. Wind Dispersal in the Quaternary): In Module 4, when hindcasting to the Last Glacial Maximum (~21,000 BP) and tracking Holocene migration across the Greenlandian and Northgrippian stages, how might the long-distance, targeted caching of white/pinyon pines by corvids produce different range-expansion patterns compared to wind-dispersed yellow pines?

Module 1 (pinus_app1.R): Fundamental vs. Realized Niche & Ordination

Launch pinus_app1.R in RStudio. As you work through each tab, use the Download (PNG) buttons to save figures for your report.

Tab 1 (Overview) & Tab 2 (Geography Occurrence): Spatial Scale & Sampling Bias

Statistical First-Principles

Raw biodiversity records from GBIF reflect human observation effort (proximity to roads, trails, and universities) as much as true biological abundance. To reduce spatial autocorrelation: 1. All raw occurrences were thinned to 1 km grid cells and filtered using a 3rd-nearest-neighbor (\(\le 100\text{ km}\)) outlier test so isolated arboretum plantings are excluded. 2. In Tab 2, occurrences are aggregated into 100 km equal-area grid cells to map continental Species Richness (count of unique species per cell) and Abundance (total records per cell). Because raw record counts span four orders of magnitude (\(1\) to \(>10,000\)), applying a base-10 logarithmic transformation (\(\log_{10}(\text{Abundance} + 1)\)) compresses extreme outliers so regional density patterns become visible.

Lab Exercise 1.1

  1. On Tab 1 (Overview), inspect the hierarchical taxonomy table. Identify which subsection has the highest number of species and which species have the fewest occupied 1 km cells.
  2. Switch to Tab 2 (Geography Occurrence). Click through the four sections (P. sect. Parrya, P. sect. Quinquefoliae, P. sect. Pinus, and P. sect. Trifoliae) and toggle between Log10 Abundance and Total Abundance (Raw).
  3. Report Deliverable 1A: Export the Species Richness map (PNG) for P. sect. Parrya and P. sect. Trifoliae. Explain why the biodiversity hotspot of P. sect. Parrya is centered in the intermountain Southwest and Sierra Madre Oriental, whereas P. sect. Trifoliae exhibits high richness in both the western cordilleras and the southeastern coastal plain.

Tab 3 (Climatic Occurrence): Univariate Niche Separation & Microclimate Resolution

Statistical First-Principles

Our dataset pairs 1 km point extractions (used for boxplots, PCA, and model training) with 5 km continental rasters (used for map rendering). In steep mountain ranges where elevation changes by \(1,000\text{ m}\) over \(5\text{ km}\), temperature drops by \(\sim 6.5^\circ\text{C}\) via the environmental lapse rate. Extracting training points at 1 km prevents subalpine ridges from being averaged with warm desert valley floors, while aggregating TRI to 5 km using the pixel maximum preserves steep canyon and sky-island microhabitats.

Lab Exercise 1.2

  1. On Tab 3 (Climatic Occurrence), select all four Pinus sections under 2. Sections and select Outgroups under 1. Genus Level.
  2. Keep Match occurrence record counts checked (Max Observations = 300). Compare their 1D boxplots across:
    • Annual Mean Temp (BIO1) and Min Temp Coldest (BIO6)
    • Annual Precipitation (BIO12) and Precip Driest Qtr (BIO17)
    • Terrain Ruggedness (TRI)
  3. Now clear the Sections and select three individual species under 4. Species: bristlecone pine (P. longaeva), Colorado pinyon (P. edulis), and loblolly pine (P. taeda), alongside L. tridentata (creosote bush) and B. nana (dwarf birch).
  4. Report Deliverable 1B: Export the spatial map and horizontal boxplot (PNG) for BIO6 or TRI. Using the median lines and interquartile boxes (25th–75th percentiles), explain how the six non-pine Outgroups bracket the upper thermal, lower thermal, and aridity limits of the genus Pinus.

Tab 4 (Hyperdimensional Niche Space): Principal Component Analysis (PCA)

Statistical First-Principles

Because the 17 Bioclim variables and TRI are measured in incompatible units (\(^\circ\text{C}\), \(\text{mm}\) of rain, dimensionless indices) and are strongly collinear, Principal Component Analysis (PCA) first standardizes every variable \(X_j\) to a \(Z\)-score (\(\mu = 0, \sigma = 1\)): \[Z_{ij} = \frac{X_{ij} - \bar{X}_j}{s_j}\] PCA then computes the eigenvalues (\(\lambda_k\)) and eigenvectors (loadings) of the \(18 \times 18\) correlation matrix, rotating the 18 axes into orthogonal Principal Components (PC1, PC2, …) ordered by the percentage of total variance explained. The Convex Hull polygons trace the outer boundary of each clade’s realized niche in 2D ordination space.

Lab Exercise 1.3

  1. On Tab 4 (Hyperdimensional Niche Space), select three subsections from subgenus Strobus (P. subsect. Balfourianae, P. subsect. Cembroides, P. subsect. Strobus) and three subsections from subgenus Pinus (P. subsect. Ponderosae, P. subsect. Contortae, P. subsect. Australes). Click Run PCA.
  2. Inspect the percentage of variance explained by PC1 and PC2 and examine the red arrows in the Climatic & Topographic Drivers (Loadings) plot.
  3. Report Deliverable 1C: Export both the PCA Niche Space plot and the Loadings plot (PNG). Which environmental variables load most strongly onto PC1 versus PC2? Do the vertebrate-dispersed white/pinyon pine subsections occupy a distinct region of hyperdimensional space compared to the wind-dispersed yellow pine subsections?

Module 2 (pinus_app2.R): Interpretable Rules, Clade Radiations & The Meaning of Absence (CART)

Launch pinus_app2.R in RStudio.

First-Principles of CART: White-Box Partitioning & Gini Impurity

Classification and Regression Trees (CART) are readily interpretable, “white-box” models. At each node \(t\), CART searches every selected predictor variable and every possible numerical threshold to find the binary Yes/No split that maximizes the reduction in Gini Impurity (\(I_G\)): \[I_G(t) = 1 - \sum_{k=1}^{K} p(k|t)^2\] where \(p(k|t)\) is the proportion of observations belonging to taxon \(k\) at node \(t\). If a node contains only a single taxon, \(I_G(t) = 0\) (pure). The Complexity Parameter (cp) acts as a regularizer: any split that does not decrease overall relative error by at least cp is pruned away.

Tab 2 (Climatic Niche Partitioning): Clade Radiations & Convergence

Lab Exercise 2.1

  1. On Tab 2 (Climatic Niche Partitioning), set Taxonomic Level to All Subsections (Pinus spp.), uncheck Include 'Outgroups', select BIO1, BIO4, BIO6, BIO12, BIO15, and TRI, and set cp = 0.015. Click Run CART Model.
  2. Follow the decision tree from the root node (top) to the terminal leaves (bottom).
  3. Now test for ecological convergence: change Taxonomic Level to Species and select four species from different subsections that occupy either boreal/cold forests (P. resinosa, P. banksiana, P. albicaulis) or summer-dry/winter-wet Mediterranean/foothill climates (P. sabiniana, P. monophylla, P. coulteri).
  4. Report Deliverable 2A: Export the multi-class CART tree and the Predicted Clade Dominance map (PNG). Explain how the top split of the tree reflects a major biogeographical divide across North America, and identify an example where two species or subsections end up in adjacent or shared environmental rules despite belonging to different subgenera.

Tab 3 (Anatomy of a Split) & Tab 4 (Simple SDM): The Meaning of Absence & The Domain Trade-Off

Ecological & Statistical Theory

In presence-background modeling, your model does not learn “what a pine likes” in isolation; it learns what distinguishes Taxa 1 (Presence) from Taxa 2 (Absence / Background). Furthermore, the Geographic Domain Radius (km) slider exposes a fundamental modeling trade-off: * Large / Unconstrained Continental Domain (Risk of Overfitting): Including background points from thousands of kilometers away provides huge climatic contrast and inflates apparent accuracy, but many distant absences are caused by dispersal limitation (\(M\)) rather than unsuitable climate (\(A\)). The tree may overfit spurious geographic proxies (e.g., separating a Californian pine from eastern forests using summer rainfall even if the pine could physiologically grow in the East). * Small / Tightly Buffered Domain (Biological Meaning vs. Risk of Underfitting): Restricting Taxa 2 to a tight radius around Taxa 1 ensures every absence comes from an accessible adjacent habitat where the pine is genuinely excluded by local microclimate or competition (\(B\)). However, shrinking the radius too far removes macroclimatic contrast and underfits the species’ broader thermal tolerances. * Historical Precedent for the 100 km Modeling Domain in Pinyon Pines: While the Geographic Domain Radius slider allows you to test radii from 100 km to 3,000 km, a 100 km buffer around target occurrences has a well-established history in western North American conifer modeling. In regional Forest Inventory and Analysis (FIA) studies of Colorado pinyon (Pinus edulis) and singleleaf pinyon (P. monophylla), a 100 km buffer around known presences was established as an optimal modeling extent for selecting biologically meaningful absences, defining expansion and contraction tension zones, and limiting unwarranted model extrapolation across continental barriers (Gibson et al., 2013). * Random Forests in Ecological Classification: Random Forest has a long and influential history in ecology as a non-parametric classifier capable of modeling complex, high-order interactions among collinear environmental predictors while providing unbiased Out-of-Bag (OOB) accuracy metrics, variable importance rankings, and partial dependence curves (Cutler et al., 2007).

Lab Exercise 2.2

  1. On Tab 3 (Anatomy of a Split), set Taxa 1 to P. edulis (Colorado pinyon) and Taxa 2 to Outgroups. Isolate Annual Precipitation (BIO12) and compare the 1D density split with the Geographic Domain Radius set to 300 km versus unchecking the domain constraint (Continental).
  2. Switch to Tab 4 (Simple SDM). Set Taxa 1 to P. ponderosa (or P. albicaulis). Run the model three times while changing Taxa 2 and the Domain Radius:
    • Run 1 (Biome Boundary): Taxa 2 = Outgroups, Domain Radius = 500 km.
    • Run 2 (Intrageneric Competition): Taxa 2 = Pinus spp. (Rest of Genus), Domain Radius = 500 km.
    • Run 3 (Continental Overfitting Test): Taxa 2 = All Background, Domain Constraint = Unchecked (Full Continent).
  3. Report Deliverable 2B: Export the CART Tree and SDM Map (PNG) for two of your runs. Compare how changing Taxa 2 and the Geographic Domain Radius altered (a) the primary split variable at the top of the tree, (b) Sensitivity and Specificity, and (c) the biological meaning of the predicted habitat map.

Module 3 (pinus_app3.R): Ensemble Probabilities & Non-Linear Niches (Random Forest)

Launch pinus_app3.R in RStudio.

First-Principles of Random Forest: Bootstrapped Ensembles of CART Trees

While a single CART tree (Module 2) is transparent and human-readable, its rectangular axis-aligned splits produce blocky step-function maps, and small changes in the training sample can alter the entire tree structure (high variance).

A Random Forest (randomForest) solves this by growing an ensemble of a set number of CART trees (here, \(N_{\text{tree}} = 500\) unpruned trees) using two randomization mechanisms: 1. Bootstrap Aggregation (Bagging of Response/Occurrence Records): Each of the 500 CART trees is trained on an independent bootstrap sample drawn with replacement (~63.2% of the records). The remaining ~36.8% of records not seen by that tree form its Out-of-Bag (OOB) validation set, providing an unbiased internal test of accuracy, Sensitivity, Specificity, and the ROC / AUC curve. 2. Random Predictor Subsetting (mtry): At every split node inside each tree, the algorithm randomly selects only \(m_{\text{try}} = \lfloor\sqrt{p}\rfloor\) predictor variables (out of \(p\) total selected variables) as candidates for splitting. This prevents a single dominant variable from masking secondary ecological drivers and allows the 500-tree ensemble to approximate smooth, non-linear physiological response surfaces.

Lab Exercise 3.1: Multi-Class Ensembles & Binary SDM Diagnostics

  1. On Tab 2 (Niche Explorer Multi-Class), replicate your All Subsections (Pinus spp.) comparison from Module 2 using the same predictor variables. Click Run Forest (500 Trees).
  2. Compare the smoothed 500-tree ensemble dominance map and OOB confusion matrix against your single-tree CART results from Module 2, and inspect the Variable Importance (Mean Decrease Gini) bar chart.
  3. Switch to Tab 3 (Binary SDM). Select a focal white pine (e.g., P. albicaulis or P. edulis) and a focal yellow pine (e.g., P. ponderosa or P. taeda). Set Taxa 2 to All Background (Pinus spp. + Outgroups) with a 750 km Domain Radius.
  4. Report Deliverable 3A: Export the RF Suitability Map, the OOB ROC Curve (AUC), the Variable Importance plot, and the 4-panel Partial Dependence Plots (PDPs) (PNG). In your report:
    • Explain why the Partial Dependence Plots (which trace the marginal log-odds of presence across each variable while integrating over all other variables) show smooth unimodal optima or sigmoidal thresholds rather than single vertical steps.
    • Test the Domain Radius trade-off in Random Forest: shrink the Domain Radius from 2,500 km down to 250 km and record how the OOB AUC score and the top-ranked Gini variable shift.

Module 4 (pinus_app4.R): Quaternary Migrations, 21st-Century Forecasts & Disturbance Ecology

Launch pinus_app4.R in RStudio. Notice that the left-hand sidebar is shared across both projection tabs: when you click Train Baseline SDM, your model is trained once on the Late 20th Century Climate Normal (1970–2000 CE) and retained in memory so you can project it into the Quaternary (Tab 2) and the 21st Century (Tab 3) without retraining.


Sub-Tab 1 (Quaternary Climates): Fossil Validation & Corvid vs. Wind Expansion Rates

Paleoclimatic & Dispersal Context

When you project a baseline SDM onto Quaternary paleoclimate grids (Beyer2020), the Range Shift Map uses asymmetric domain clipping: the 1970–2000 CE baseline range is clipped to your chosen Geographic Domain Radius (\(M\)), while paleoclimate predictions are allowed to project beyond the modern domain to capture distant glacial refugia. Overlaying radiocarbon-dated macrofossils and packrat middens (Neotoma Paleoecology Database) tests whether the species occupied its predicted Pleistocene niche and reveals the paradoxical behavior of post-glacial range expansions: * Following the Last Glacial Maximum (~21,000 BP), rapid warming during the Greenlandian (11,700–8,300 BP) and Northgrippian (8,300–4,200 BP) stages saw pines migrate hundreds of kilometers poleward and upslope across western North America (Reid’s Paradox). * Dispersal Syndrome Comparison: Large-seeded white and pinyon pines (Subgenus Strobus, such as P. monophylla, P. edulis, and P. flexilis) rely on corvid birds (Clark’s nutcrackers and Pinyon jays) that actively fly 10–30 km across inhospitable desert valleys to cache seeds directly into newly suitable montane woodland bands. In contrast, small-seeded yellow pines (Subgenus Pinus, such as P. ponderosa and P. contorta) rely on wind dispersal (anemochory), where 95% of seeds fall within 50–100 meters of the parent tree unless rare convective updrafts or stepping-stone corridors connect mountain ranges.

Lab Exercise 4.1: Tracking Post-Glacial Migration Across the Holocene

  1. In the shared left-hand sidebar, select Taxa 1 = P. edulis (or P. monophylla — a corvid-dispersed pinyon pine), Taxa 2 = All Background, Domain Radius = 750 km, Algorithm = Random Forest (500 Trees), and click Train Baseline SDM.
  2. On the Quaternary Climates tab, step chronologically from past to present through all four stages while keeping Auto-sync slider to stage checked:
    • Last Glacial Maximum (~21,000 BP) (18,000–25,000 cal yr BP)
    • Greenlandian: Early Holocene (11,700–8,300 BP)
    • Northgrippian: Mid Holocene (8,300–4,200 BP)
    • Meghalayan: Late Holocene (4,200–0 BP)
  3. Record the Mean Paleo Suitability at Fossil Sites, the % of Fossils Inside Predicted Range, and the distribution of fossils in the Neotoma Fossils by Range Shift Category bar plot (Modern Only, Stable Refugia, Paleo Range, Outside Niche).
  4. Now train a baseline SDM for a wind-dispersed yellow pine (P. ponderosa or P. contorta) using the same sidebar settings and repeat the four-stage Quaternary progression.
  5. Report Deliverable 4A: Export the Quaternary Range Shift map (PNG) at the LGM (lgm_21ka) and Mid-Holocene (northgrippian) for both your white/pinyon pine and your yellow pine. Compare their fossil validation hit rates and explain how corvid-mediated directed dispersal versus wind dispersal influences post-glacial colonization across fragmented Basin-and-Range or Rocky Mountain topography.

Sub-Tab 2 (21st Century Climates): Fire Suppression, Beetle Voltinism & Range Contractions

Historical & Physiological Context for 21st-Century Disturbances

Climate envelope models predict where abiotic conditions will cross a species’ tolerance threshold (Range Contraction), persist (Stable Climate Refugia), or become newly suitable (Range Expansion) across Early (2021–2040), Mid (2041–2060), and Late (2081–2100) 21st-century horizons. However, mature trees rarely die slowly from a \(2^\circ\text{C}\) rise in mean annual temperature; instead, wildfire and bark beetles act as the proximate ecological catalysts of range shifts. * Expansion and Contraction Tension Zones: When projecting conifer distributions under 21st-century climate scenarios, trailing-edge Contraction Zones and leading-edge Expansion Zones act as ecological tension zones—landscapes where adult tree persistence, seedling recruitment, interspecific competition, and disturbance agents (wildfire and bark beetles) interact most intensely to govern the rate of biome transition (Gibson et al., 2013).

  1. The Legacy of 20th-Century Fire Suppression: Beginning in the early 20th century, federal land management policies mandated total wildfire exclusion across western North American forests. In dry yellow pine forests (P. ponderosa, P. jeffreyi) historically maintained as open parklands by frequent, low-severity surface fires every 5–15 years, a century of fire suppression caused dense infilling of shade-tolerant saplings (“ladder fuels”) and deep surface fuel beds. When combined with 21st-century warming and aridification, fires that once stabilized yellow pine dominance now erupt into high-severity, stand-replacing megafires that kill mature seed trees across thousands of hectares, preventing wind-dispersed seeds from reaching the interior of burn scars.
  2. 21st-Century Drought Stress & Bark Beetle Generation Time (Voltinism): Pine-killing bark beetles in the genera Dendroctonus (e.g., Mountain Pine Beetle, D. ponderosae; Western Pine Beetle, D. brevicomis; Southern Pine Beetle, D. frontalis) and Ips (e.g., Pinyon Ips, I. confusus; Pine Engraver, I. pini) bore into the phloem to lay eggs and inoculate blue-stain fungi, girdling the tree. Under historical climates, healthy pines repelled beetles by pitching them out with pressurized oleoresin, while extreme winter cold (\(<-30^\circ\text{C}\)) killed overwintering larvae and restricted high-elevation beetles to a 1-year (univoltine) or 2-year (semivoltine) life cycle. Under 21st-century warming and drought:
    • Elevated Vapor Pressure Deficit (VPD) and soil drought induce xylem cavitation and carbon starvation, collapsing resin defenses—especially along the warm, dry trailing edge (Contraction Zone).
    • Warmer winters eliminate cold-mortality events, and longer, hotter growing seasons accelerate beetle development from semivoltine/univoltine to univoltine or bivoltine/multivoltine (multiple generations per year), triggering exponential landscape-scale outbreaks that spill over into Stable Climate Refugia and high-elevation whitebark/bristlecone stands.

Lab Exercise 4.2: Beetle Guild Turnover & Fire Across Climatic Gradients

  1. Train your baseline SDM on Taxa 1 = P. edulis (Colorado pinyon, lower-treeline woodland) with a 750 km Domain Radius. Switch to the 21st Century Climates tab (SSP5-8.5, 2090 Late-Century).
  2. Testing Bark Beetle Guild Turnover Across Range Shift Zones:
    • Keep the Bark Beetle Time Window at 2000–2025.
    • First, set Bark Beetle Taxon Overlay to Ips spp. (which includes the Pinyon Ips, I. confusus, a specialist of drought-stressed pinyon woodlands). Inspect the Range Shift Map and the right-hand panel of the 4-Horizon Trajectory Bar Plot.
    • Next, switch Bark Beetle Taxon Overlay to D. ponderosae (Mountain Pine Beetle) or D. brevicomis (Western Pine Beetle) (beetles that attack montane P. ponderosa and P. contorta forests at higher elevations).
    • Compare where the two beetle groups fall: Are Ips observations concentrated in the pinyon’s Contraction / Refugia Zones, while Dendroctonus observations are concentrated in the pinyon’s upslope/poleward Expansion Zone (where montane yellow pines currently live)?
  3. Testing Wildfire Across Climatic Gradients & Time Windows:
    • Enable Overlay Classified MTBS Wildfire Grid and slide the Wildfire Time Window across 5-year/decadal increments (1985–1995, 2000–2010, 2010–2025, and 1985–2025).
    • Compare the classified fire pixels (Dark Red = Fire in Contraction Zone; Dark Orange = Fire in Stable Refugia; Bright Yellow = Fire in Expansion Zone; Dark Grey = Fire Outside Range) and the 4-period trajectory bar chart between a fire-sensitive, corvid-dispersed pinyon/white pine (P. edulis or P. albicaulis) and a fire-adapted, wind-dispersed yellow pine (P. ponderosa or P. contorta).
  4. Report Deliverable 4B: Export the Range Shift + Disturbance Map (PNG) for both P. edulis and P. ponderosa under SSP5-8.5 (2090). In your report, synthesize:
    • How bark beetle guilds (Ips vs. Dendroctonus) partition across a shifting pine’s Contraction, Refugia, and Expansion zones, and whether a pinyon migrating upslope into ponderosa habitat escapes beetle pressure or encounters a new beetle guild.
    • How wildfire functions differently at the trailing edge (Contraction Zone) versus the leading edge (Expansion Zone) for thin-barked white/pinyon pines versus fire-adapted yellow pines.

Place-Based Synthesis & Citizen Science Validation with iNaturalist

Connecting Models to the Living Landscape

Statistical models and 5 km raster grids are abstractions of real trees rooted in real soil. In this capstone synthesis, you will choose a specific geographic place—preferably your home region, a nearby mountain range or watershed, or a landscape you have visited or hope to visit—and investigate a native Pinus species that grows there. You will justify a customized Species Distribution Model in Module 4 (pinus_app4.R), validate its historical behavior against the fossil record, project its 21st-century range shift, and test your model’s predictions against dated field photographs from the iNaturalist citizen science platform (https://www.inaturalist.org/).


Step 1: Place Selection & Biological/Statistical Modeling Rationale

Select your focal landscape and identify a native Pinus species that occurs in or adjacent to that region (use Module 1, Tab 2 [Geography Occurrence] if you want to explore which Pinus species occur near your chosen location). Open Module 4 (pinus_app4.R) and configure your baseline model in the shared left-hand sidebar. In your lab report, explicitly justify your three core modeling decisions:

  1. Absence / Background Selection (Taxa 2):
    • Did you contrast your focal pine against Outgroups (the six non-pine biome anchors), Pinus spp. (Rest of Genus), a specific sister Section / Subsection, or All Background?
    • Biological & Statistical Rationale: Explain what an “absence” means in your chosen setup. Are you modeling the abiotic physiological boundary between pine woodland and non-forest biomes (e.g., desert scrub or grassland), or are you modeling intrageneric competitive sorting between your focal pine and neighboring Pinus clades?
  2. Geographic Model Domain Radius (M):
    • Which Geographic Domain Radius (km) did you select around your focal species’ occurrences?
    • Biological & Statistical Rationale: Explain how your chosen radius balances the statistical risk of overfitting (if the domain is too broad and includes distant regions where the pine is absent only because of continental dispersal barriers) against the risk of underfitting (if the domain is so narrow that it truncates the climatic gradient and fails to capture the species’ thermal and aridity limits). How does your species’ seed dispersal mechanism (corvid caching vs. wind dispersal) inform what constitutes an ecologically accessible domain?
  3. Environmental & Topographic Variable Selection (BIO1–BIO19, TRI) & Algorithm:
    • Which subset of temperature, precipitation, seasonality, and Terrain Ruggedness (TRI) variables did you select, and did you use the 500-tree Random Forest ensemble or a single CART tree?
    • Biological & Statistical Rationale: Connect your chosen variables to the known physiological vulnerabilities of your focal pine (e.g., summer drought stress, winter freezing minimums, monsoonal vs. winter precipitation, or steep micro-topographic refugia).

Step 2: Quaternary Hindcast & Fossil Validation (Tab 2: Quaternary Climates)

Before trusting your model to forecast the 21st century, test its transferability across the past 21,000 years on the Quaternary Climates tab:

  1. Step through the Last Glacial Maximum (~21,000 BP), Greenlandian (11,700–8,300 BP), Northgrippian (8,300–4,200 BP), and Meghalayan (4,200–0 BP) stages.
  2. Inspect the Neotoma Fossils by Range Shift Category bar plot and the validation summary readout (Mean Paleo Suitability and % Inside Paleo Range).
  3. In Your Report: Does the hindcast capture the known Pleistocene packrat midden or macrofossil sites for your taxon (or its subsection)? If fossils fall within the predicted Paleo Range or Stable Refugia, explain how this supports your model’s parameterization. If fossils fall in Outside Niche, discuss whether topographic micro-refugia (TRI), non-analog Pleistocene climates, or domain truncation could explain the mismatch.

Step 3: 21st-Century Range Shift & Disturbance Forecast (Tab 3: 21st Century Climates)

Switch to the 21st Century Climates tab (using your retained baseline model) and project your focal pine’s distribution to Late-Century (2081–2100, SSP5-8.5 or SSP2-4.5) with the MTBS Wildfire and Bark Beetle overlays enabled:

  1. Locate your chosen geographic place on the Classified Range Shift & Disturbance Overlay map. Identify specific mountain ranges, canyons, or elevational bands near your focal place that are classified as:
    • Range Contraction (Lost Habitat — Trailing Edge)
    • Stable Climate Refugia (Persisting Habitat)
    • Range Expansion (Gained Habitat — Leading Edge)
  2. Note where classified wildfire pixels (Dark Red in Contraction, Dark Orange in Refugia, Bright Yellow in Expansion) and bark beetle observations (Dendroctonus or Ips) intersect your focal region.
  3. Report Deliverable 5A: Click Export Range Shift + Disturbance Map (PNG) and embed this figure in your report alongside the area (km²) and wildfire/beetle exposure statistics.

Step 4: Demographic & Forest Health Investigation on iNaturalist

Because mature conifers can persist for centuries after the local climate becomes unsuitable for seedling establishment (“extinction debt”), raw presence points on a map cannot tell us whether a population is thriving, recruiting, or dying out. However, the photographs attached to citizen science observations contain vital demographic and forest-health clues.

  1. Open iNaturalist (https://www.inaturalist.org/), click Explore, and type the scientific name of your focal Pinus species into the Species box.
  2. Pan and zoom the iNaturalist map to your focal geographic region, switching the base layer to Terrain or Satellite so you can match valleys, ridgelines, and latitude/elevation bands to your Module 4 Range Shift Map.
  3. Compare the Contraction Zone vs. Expansion Zone:
    • Inspect 5–10 observations located inside your model’s predicted Contraction Zone (low-elevation, southern, or arid trailing edge): Click into the observation photos. Are the photographed trees old, gnarled relict individuals growing on exposed rock outcrops, or do they show signs of crown dieback, drought-browned needles, bark beetle pitch tubes/boring dust, or standing dead snags inside recent burn scars? Do you see any young seedlings or saplings in the understory?
    • Inspect 5–10 observations located inside or near your model’s predicted Expansion Zone or Stable Refugia (high-elevation, north-facing, or poleward leading edge): Do the photographs show healthy, dense foliar crowns, active pollen/seed cone production, and young regenerating seedlings or saplings establishing in meadows or subalpine ecotones?
  4. Investigate the iNaturalist Time Filter:
    • Click the Filters button in the top-right corner of the iNaturalist Explore page and use the Date Observed filter (Range or Exact Date) to compare observations across different time windows (for example, pre-2015 vs. 2015–2020 vs. 2021–present, or before and after a major regional megafire or multi-year drought). Also explore the Annotation filters (such as Alive or Dead* or Plant Phenology).

Step 5: Synthesis Reflection — Building Better Models & Better Empirical Tests

Conclude your report by addressing the following three questions based on your iNaturalist investigation and your Module 4 outputs:

  1. Field-Truthing Your Range Shift Map: What specific observations could you personally make and upload to iNaturalist on your next visit to your chosen place that would provide the highest scientific value for testing your range-shift model?
  2. Temporal Dynamics & The Time Filter: How did filtering iNaturalist observations across time windows change the spatial pattern or volume of records? Why must spatial ecologists account for the rapid post-2015 exponential growth of citizen science uploads when using time-sliced occurrence data to test for range shifts?
  3. Next-Generation SDMs: Traditional Species Distribution Models treat every GBIF/iNaturalist coordinate as an identical binary 1 (present), whether that point is a 400-year-old dying snag in a burned contraction zone or a thriving 5-year-old sapling in an expansion zone. How could researchers extract structured information from iNaturalist observations—such as demographic stage (seedling/sapling vs. mature adult), tree vigor/mortality status, active cone production, bark beetle/fire damage, and repeat temporal photographs—to build more mechanistic niche models and stronger empirical tests of range contraction and expansion?

Appendix: Introduction to R & Optional Tab-by-Tab Coding Exercises

This appendix introduces the R data structures powering the Shiny applications and provides standalone code blocks corresponding to each module so you can inspect the raw data, compute summary statistics, and perform raster algebra directly in R.

A.1 Setting Up Your R Environment & Inspecting Data Structures

In R, our pipeline relies on two primary data structures: 1. data.frame / tibble (dplyr): Tabular spreadsheets where rows are 1 km grid cell occurrences and columns are taxonomic ranks, Albers coordinates (x, y in meters), and extracted 1 km environmental values. 2. SpatRaster (terra): Multi-layer gridded matrices aligned to a 5 km North American Albers Equal-Area projection (+proj=aea ... +units=m).

library(dplyr)
library(tidyr)
library(terra)
library(rpart)
library(randomForest)

# 1. Load tabular 1km occurrence extractions and Gymnosperm Database taxonomy
df_raw       <- read.csv("data/pinus_aggregated.csv") %>% drop_na()
pinus_groups <- read.csv("data/Pinus_species_groups.csv") %>% distinct(species, .keep_all = TRUE)
df           <- df_raw %>% left_join(pinus_groups, by = c("species", "genus"))

# Inspect tabular structure and column names
str(df)
head(df[, c("species", "sp_abbrev", "section", "subsection", "x", "y", "BIO1", "BIO12", "TRI")])

# 2. Load 5km Late 20th Century Climate Normal (1970-2000 CE) SpatRaster stack
r_pres_5km <- rast("data/bioclim_na_albers_5km_present.tif")
print(r_pres_5km) # Check dimensions, resolution (5000 x 5000 m), CRS, and layer names

A.2 Optional Coding Assignment — Module 1 (Summary Statistics & PCA)

corresponds to Module 1, Tabs 1–4: compute subsectional niche means/IQRs and run a Principal Component Analysis (prcomp) from scratch.

# 1. Tab 1 & 3 Equivalent: Summary statistics by Section and Subsection
subsect_summary <- df %>%
  group_by(section, subsection) %>%
  summarize(
    n_species       = n_distinct(species),
    n_1km_cells     = n(),
    total_records   = sum(occurrence_count),
    median_BIO1_C   = median(BIO1, na.rm = TRUE),
    iqr_BIO1_C      = IQR(BIO1, na.rm = TRUE),
    median_BIO12_mm = median(BIO12, na.rm = TRUE),
    median_TRI      = median(TRI, na.rm = TRUE),
    .groups = "drop"
  ) %>%
  arrange(section, subsection)

print(subsect_summary)

# 2. Tab 4 Equivalent: Standardized Principal Component Analysis (PCA)
pca_vars <- c("BIO1", "BIO4", "BIO6", "BIO10", "BIO12", "BIO15", "BIO18", "TRI")

set.seed(42)
df_pca_sample <- df %>%
  group_by(subsection) %>%
  slice_sample(n = 200) %>%
  ungroup()

pca_fit <- prcomp(df_pca_sample[, pca_vars], center = TRUE, scale. = TRUE)
summary(pca_fit)          # Proportion of variance explained by PC1, PC2, ...
round(pca_fit$rotation[, 1:2], 3) # Variable loadings on PC1 and PC2

A.3 Optional Coding Assignment — Module 2 (CART Rules & Domain Buffering)

Corresponds to Module 2, Tabs 3–4: filter background points to a geographic domain buffer around P. edulis and fit a pruned CART decision tree (rpart).

library(FNN)

# 1. Isolate Taxa 1 (P. edulis) and Candidate Taxa 2 (Outgroups)
d1 <- df %>% filter(sp_abbrev == "P. edulis")
d2 <- df %>% filter(genus_label == "Outgroups")

# 2. Apply a 500 km (500,000 m) Geographic Domain Radius around P. edulis
nn_dist_m <- FNN::get.knnx(
  data  = as.matrix(d1[, c("x", "y")]),
  query = as.matrix(d2[, c("x", "y")]),
  k     = 1
)$nn.dist[, 1]

d2_buffered <- d2 %>% filter(nn_dist_m <= 500000)

# 3. Balance sample sizes (300 cells per group) and fit CART
set.seed(42)
d_train <- bind_rows(
  d1 %>% slice_sample(n = 300) %>% mutate(target = "P. edulis"),
  d2_buffered %>% slice_sample(n = 300) %>% mutate(target = "Outgroups")
) %>%
  mutate(target = factor(target, levels = c("Outgroups", "P. edulis")))

cart_mod <- rpart(
  target ~ BIO1 + BIO4 + BIO10 + BIO12 + TRI,
  data    = d_train,
  method  = "class",
  control = rpart.control(cp = 0.01)
)

# Print human-readable decision rules and variable importance
print(cart_mod)
print(cart_mod$variable.importance)

A.4 Optional Coding Assignment — Module 3 (Random Forest Ensembles)

Corresponds to Module 3, Tab 3: train a 500-tree Random Forest on d_train, inspect Out-of-Bag (OOB) confusion matrices, and project continuous suitability onto the 5 km raster stack.

model_vars <- c("BIO1", "BIO4", "BIO10", "BIO12", "TRI")

set.seed(42)
rf_mod <- randomForest(
  x          = d_train[, model_vars],
  y          = d_train$target,
  ntree      = 500,
  importance = TRUE
)

# Inspect OOB error rate, confusion matrix, and Mean Decrease Gini importance
print(rf_mod)
importance(rf_mod)

# Predict continuous habitat suitability (0 to 1) across the 5km Albers grid
r_prob_pres <- predict(r_pres_5km[[model_vars]], rf_mod, type = "prob", na.rm = TRUE)[[2]]
plot(r_prob_pres, main = "P. edulis RF Suitability (1970-2000 CE)")

A.5 Optional Coding Assignment — Module 4 (Raster Algebra for Range Shifts & Wildfire Overlap)

Corresponds to Module 4, Tabs 2–3: use terra raster algebra to compute a 21st-century Range Shift raster (1 = Contraction, 2 = Stable Refugia, 3 = Expansion) and cross-tabulate burned MTBS wildfire pixels across each range-shift zone.

# 1. Load Late-Century (2090, SSP5-8.5) 5km stack and predict future suitability
r_fut_2090  <- rast("data/bioclim_na_albers_5km_ssp585_2090.tif")
r_prob_2090 <- predict(r_fut_2090[[model_vars]], rf_mod, type = "prob", na.rm = TRUE)[[2]]

# 2. Raster Algebra: Threshold at P >= 0.50 and classify Range Shift
# (pres_bin * 1) + (fut_bin * 2) -> 0 = Absent, 1 = Contraction, 2 = Expansion, 3 = Stable Refugia
pres_bin <- r_prob_pres >= 0.50
fut_bin  <- r_prob_2090 >= 0.50
raw_code <- (pres_bin * 1) + (fut_bin * 2)

# Reclassify so 1 = Contraction, 2 = Stable Refugia, 3 = Expansion
r_shift <- classify(raw_code, rcl = matrix(c(0, NA,  1, 1,  3, 2,  2, 3), ncol = 2, byrow = TRUE))
plot(r_shift, col = c("#d73027", "#1a9850", "#3288bd"),
     main = "P. edulis Range Shift (1970-2000 vs. SSP5-8.5 2090)")

# 3. Quantify Area (km^2) per Range Shift Zone (each 5km x 5km pixel = 25 km^2)
shift_vals <- values(r_shift)[, 1]
area_km2   <- table(factor(shift_vals, levels = 1:3,
                           labels = c("Contraction", "Stable_Refugia", "Expansion"))) * 25
print(area_km2)

# 4. Raster Algebra: Cross-tabulate 2000-2025 MTBS Wildfire Frequency with Range Shift Zones
r_fire_5yr <- rast("data/fire_5yr_na_albers_5km.tif")
# Sum the 5-year bands from Fire_2000 through Fire_2020
r_fire_2000_2025 <- sum(r_fire_5yr[[c("Fire_2000", "Fire_2005", "Fire_2010", "Fire_2015", "Fire_2020")]], na.rm = TRUE)
burned_bin       <- values(r_fire_2000_2025 >= 1)[, 1]

fire_by_zone_km2 <- c(
  Burned_Contraction_km2 = sum(shift_vals == 1 & burned_bin, na.rm = TRUE) * 25,
  Burned_Refugia_km2     = sum(shift_vals == 2 & burned_bin, na.rm = TRUE) * 25,
  Burned_Expansion_km2   = sum(shift_vals == 3 & burned_bin, na.rm = TRUE) * 25
)
print(fire_by_zone_km2)

Acknowledgments & AI Collaboration Disclosure

Conceptual Design, Scientific Framework & Authorship

The pedagogical architecture, ecological hypotheses, taxonomic framework, spatial modeling domain parameterization (including the 3rd-nearest-neighbor spatial outlier filter, hybrid 1 km microclimate training / 5 km continental projection design, and asymmetric domain clipping), and integration of Quaternary macrofossil validation with 21st-century wildfire and bark beetle disturbance ecology were conceived and directed by Jacob Gibson (Arizona Western College).

AI-Assisted Software Engineering & Curriculum Co-Development

This four-module R Shiny application suite (data_prep.R, pinus_app1.R–pinus_app4.R), Quarto quickstart guide, and CURE laboratory manual were co-developed in an iterative human–AI collaboration with Gemini (Google, Large Language Model, November 2026). Specifically, Gemini contributed to:

  1. Geospatial & Data Pipeline Engineering: Co-authoring the automated terra, pastclim, rgbif, neotoma2, and FNN data preparation scripts, including coordinate projection workflows, 5-year MTBS wildfire rasterization, and reactive Shiny state caching.
  2. Interactive Visualization & Statistical Scaffolding: Implementing base R and terra spatial plotting routines, stratified multi-clade downsampling functions, range-shift raster algebra, and student-facing R coding exercises.
  3. Verification: All statistical algorithms, ecological interpretations, taxonomic assignments, and spatial outputs were independently tested, verified, and curated by the human author.
  • Suggested Software & Curriculum Citation Format: Gibson, J. (2026). Biogeography, Niche Modeling, and Climate-Disturbance Dynamics of North American Pinus: A Four-Module R Shiny CURE Suite (Version 1.0, co-developed with Google Gemini) [Computer software and curriculum]. Arizona Western College / Northern Arizona University.

References

Biogeography, Niche Theory & Species Distribution Modeling

  • Breiman, L. (2001). Random forests. Machine Learning, 45(1), 5–32. https://doi.org/10.1023/A:1010933404324
  • Clark, J. S., Fastie, C., Hurtt, G., Jackson, S. T., Johnson, C., King, G. A., Lewis, M., Lynch, J., Pacala, S., Prentice, I. C., Schupp, E. W., Webb, T., & Wyckoff, P. (1998). Reid’s paradox of rapid plant migration: Dispersal theory and interpretation of paleoecological records. BioScience, 48(1), 13–24. https://doi.org/10.2307/1313224 *Gibson, J., Moisen, G., Frescino, T., & Edwards, T. C. (2013). Using Publicly Available Forest Inventory Data in Climate-Based Models of Tree Species Distribution: Examining Effects of True Versus Altered Location Coordinates. Ecosystems, 17, 43–53. https://doi.org/10.1007/s10021-013-9703-y
  • Pulliam, H. R. (2000). On the relationship between niche and distribution. Ecology Letters, 3(4), 349–361. https://doi.org/10.1046/j.1461-0248.2000.00143.x
  • Reid, C. (1899). The Origin of the British Flora. Dulau & Co., London. *Cutler, D. R., Edwards, T. C., Beard, K. H., et al. (2007). Random forests for classification in ecology. Ecology, 88, 2783–2792. https://doi.org/10.1890/07-0539.1
  • Soberón, J., & Peterson, A. T. (2005). Interpretation of models of fundamental ecological niches and species’ distributional areas. Biodiversity Informatics, 2, 1–10. https://doi.org/10.17161/bi.v2i0.4
  • Therneau, T., & Atkinson, B. (2023). rpart: Recursive Partitioning and Regression Trees. R package version 4.1.23. https://CRAN.R-project.org/package=rpart

Pinus Taxonomy, Genomics, Physiology & Mutualisms

  • Earle, C. J. (Ed.). (2026). Pinus (Pine) description and taxonomy. The Gymnosperm Database. https://www.conifers.org/pi/Pinus.php
  • Eckert, A. J., & Hall, B. D. (2006). Phylogeny, historical biogeography, and patterns of diversification for Pinus (Pinaceae): Phylogenetic tests of fossil-based hypotheses. Molecular Phylogenetics and Evolution, 40(1), 166–182. https://doi.org/10.1016/j.ympev.2006.03.009
  • Gernandt, D. S., López, G. G., García, S. O., & Liston, A. (2005). Phylogeny and classification of Pinus. Taxon, 54(1), 29–42. https://doi.org/10.2307/25065300
  • Jin, W.-T., Gernandt, D. S., Wehenkel, C., Xia, X.-M., Wei, X.-X., & Wang, X.-Q. (2021). Phylogenomic and ecological analyses reveal the spatiotemporal evolution of global pines. Proceedings of the National Academy of Sciences, 118(20), e2022302118. https://doi.org/10.1073/pnas.2022302118
  • Keeley, J. E. (2012). Ecology and evolution of pine life histories. Annals of Forest Science, 69(4), 445–453. https://doi.org/10.1007/s13595-012-0201-8
  • Lanner, R. M. (1996). Made for Each Other: A Symbiosis of Birds and Pines. Oxford University Press, New York.
  • Parks, M., Cronn, R., & Liston, A. (2012). Separating the wheat from the chaff: Mitigating the effects of noise in a plastome phylogenomic data set from Pinus L. (Pinaceae). BMC Evolutionary Biology, 12, 100. https://doi.org/10.1186/1471-2148-12-100
  • Willyard, A., Syring, J., Gernandt, D. S., Liston, A., & Cronn, R. (2007). Fossil calibration of molecular divergence infers a moderate mutation rate and recent radiations for Pinus. Molecular Biology and Evolution, 24(1), 90–101. https://doi.org/10.1093/molbev/msl131

Disturbance Ecology: Wildfire Regimes, Drought Mortality & Bark Beetles

  • Allen, C. D., Macalady, A. K., Chenchouni, H., Bachelet, D., McDowell, N., Vennetier, M., Kitzberger, T., Rigling, A., Breshears, D. D., Hogg, E. H., Gonzalez, P., Fensham, R., Zhang, Z., Castro, J., Demidova, N., Lim, J.-H., Allard, G., Running, S. W., Semerci, A., & Cobb, N. (2010). A global overview of drought and heat-induced tree mortality reveals emerging climate change risks for forests. Forest Ecology and Management, 259(4), 660–684. https://doi.org/10.1016/j.foreco.2009.09.001
  • Bentz, B. J., Régnière, J., Fettig, C. J., Hansen, E. M., Hayes, J. L., Hicke, J. A., Kelsey, R. G., Negrón, J. F., & Seybold, S. J. (2010). Climate change and bark beetles of the western United States and Canada: Direct and indirect effects. BioScience, 60(8), 602–613. https://doi.org/10.1525/bio.2010.60.8.6
  • Eidenshink, J., Schwind, B., Brewer, K., Zhu, Z.-L., Quayle, B., & Howard, S. (2007). A project for Monitoring Trends in Burn Severity. Fire Ecology, 3(1), 3–21. https://doi.org/10.4996/fireecology.0301003
  • Raffa, K. F., Aukema, B. H., Bentz, B. J., Carroll, A. L., Hicke, J. A., Turner, M. G., & Romme, W. H. (2008). Cross-scale drivers of natural disturbances prone to anthropogenic amplification: The dynamics of bark beetle eruptions. BioScience, 58(6), 501–517. https://doi.org/10.1641/B580607

Climate, Paleoclimate, Fossil & Biodiversity Data Archives

  • Beyer, R. M., Krapp, M., & Manica, A. (2020). High-resolution terrestrial climate, bioclimate and vegetation for the last 120,000 years. Scientific Data, 7, 236. https://doi.org/10.1038/s41597-020-0552-1
  • Chamberlain, S., Oldoni, D., & Waller, J. (2024). rgbif: Interface to the Global Biodiversity Information Facility API. R package. https://CRAN.R-project.org/package=rgbif
  • Fick, S. E., & Hijmans, R. J. (2017). WorldClim 2: New 1-km spatial resolution climate surfaces for global land areas. International Journal of Climatology, 37(12), 4302–4315. https://doi.org/10.1002/joc.5086
  • Hijmans, R. J. (2024). terra: Spatial Data Analysis. R package. https://CRAN.R-project.org/package=terra
  • iNaturalist Community. (2026). Observations of Pinus in North America. Exported from https://www.inaturalist.org/.
  • Leonardi, M., Hallett, E. Y., Beyer, R., Krapp, M., & Manica, A. (2023). pastclim 1.2: An R package to easily access and use paleoclimatic reconstructions. Ecography, 2023(3), e06481. https://doi.org/10.1111/ecog.06481
  • Socorro, O., Goring, S., Dawson, A., & Williams, J. W. (2023). neotoma2: Working with the Neotoma Paleoecology Database. R package. https://CRAN.R-project.org/package=neotoma2
  • Williams, J. W., Grimm, E. C., Blois, J. L., Charles, D. F., Davis, E. B., Goring, S. J., Graham, R. W., Smith, A. J., Anderson, M., Arroyo-Cabrales, J., Ashworth, A. C., Betancourt, J. L., Bills, B. W., Booth, R. K., Buckland, P. I., Curry, B. B., Giesecke, T., Jackson, S. T., Latorre, C., … Takahara, H. (2018). The Neotoma Paleoecology Database, a multiproxy, international, community-curated data resource. Quaternary Research, 89(1), 156–177. https://doi.org/10.1017/qua.2017.105