SPECIES DISTRIBUTION PROJECTIONS UNDER INTERNAL CLIMATE VARIABILITY REVEAL MULTIPLE PLAUSIBLE FUTURES REQUIRING FLEXIBLE CLIMATE-READY DECISIONS

This study demonstrates that internal climate variability can produce qualitatively different and even reversed species distribution outcomes across marine and terrestrial ecosystems, necessitating flexible, stress-tested decision-making strategies that account for multiple plausible climate futures rather than relying on single best-estimate projections.

The Gist

The Big Idea: Why "One Best Guess" Isn't Enough

Imagine you are planning a massive outdoor wedding for 2050. You want to know: Will it rain? Will it be too hot? Where should we set up the tents?

Traditionally, climate scientists have acted like a single, very confident weather forecaster. They run a computer model once (or average out many runs) and say, "Based on our best guess, it will be 5 degrees warmer, and the rain will fall here." They treat the natural ups and downs of the weather (like a random storm or a heatwave that happens just by chance) as "noise" to be smoothed out.

This paper argues that smoothing out the noise is a mistake.

The authors say that even if we know exactly how much carbon humans will emit, the Earth's climate has its own internal "personality" or "mood swings" (called Internal Climate Variability). These mood swings are real, unpredictable, and can completely change the outcome for nature.

The Experiment: The "100 Different Futures"

To prove this, the researchers didn't just look at one future. They took a super-powerful climate computer model (CESM2-LENS2) and ran it 100 times.

• The Setup: They kept the "rules" the same (same amount of pollution, same computer code).
• The Twist: They changed the starting point of the simulation by a tiny, invisible amount (like changing the position of a single atom in the atmosphere).

Because the climate system is chaotic (like a butterfly flapping its wings), those tiny changes grew into 100 completely different, yet equally plausible, weather histories for the year 2050 and 2100.

They then asked: "If we use these 100 different weather histories to predict where 34 different animals and plants will live, do we get the same answer?"

The Answer: No. Sometimes, the answers were wildly different.

The Four "Futures" (The Decision Cases)

The authors found that for any given species, the 100 different futures usually fell into one of four categories. Think of this like a Decision Compass for policymakers:

1. Case 1: The "Sleeping Giant" (No Change)

   • The Analogy: You check the weather for 100 different scenarios, and they all say, "It's going to be a mild Tuesday."
   • Meaning: The species isn't moving much, no matter what the climate does. We can just keep doing what we're doing.

2. Case 2: The "Confident March" (Clear Direction & Size)

   • The Analogy: All 100 weather reports agree: "It will be a hot summer, and the heatwave will be exactly 5 days long."
   • Meaning: We know the species is moving north, and we know exactly how far. We can build a fence or a park right where they will go.

3. Case 3: The "Uncertain Magnitude" (Clear Direction, Unknown Size)

   • The Analogy: All 100 reports agree: "It will be a hot summer." But some say it will be a mild heatwave, while others say it will be a scorching, record-breaking inferno.
   • Meaning: We know the species is moving, but we don't know if they will move a little or a lot. This is dangerous for planning. If you build a small park, you might miss them if the heat is extreme. If you build a huge park, you might waste money if the heat is mild.
   • Solution: We need flexible plans that can grow or shrink.

4. Case 4: The "Coin Flip" (Total Reversal)

   • The Analogy: 50 reports say, "It will be a tropical paradise," and the other 50 say, "It will be a frozen wasteland."
   • Meaning: The model can't even agree on the direction. One future says the species will expand; another says they will go extinct.
   • Solution: We cannot make a specific plan. We must use "No-Regrets" strategies (things that are good to do regardless of the outcome) and watch closely.

The Big Surprise: Land vs. Sea

The study found a massive divide between animals on land and animals in the ocean.

• Ocean Animals (The "Steady Ship"): The ocean is huge and moves slowly. The "mood swings" (variability) are smaller. Most marine species fell into Case 2. We can be fairly confident about where they will go.
• Land Animals (The "Wild Horse"): The land heats up and cools down faster and more erratically. The "mood swings" are huge. Most land animals fell into Case 3 or 4.
  • Why? A small change in temperature or rain on land can push a plant over a cliff edge, causing it to die or move instantly. The ocean is more forgiving; the land is more sensitive.

The "Magic 8-Ball" Problem

The paper warns us that if we only look at the "average" of all 100 futures (the "Best Guess"), we might get a Magic 8-Ball answer that is completely wrong.

• Example: If 50 futures say a forest will grow, and 50 say it will burn down, the "average" says the forest will stay exactly the same size.
• Reality: The forest is either going to be huge or gone. The average hides the danger.

What Should We Do?

The authors say we need to stop trying to predict the one perfect future. Instead, we need Flexible, Climate-Ready Decisions.

• Don't build a rigid plan based on one map.
• Do stress-test your plans against all 100 possibilities.
  • If your plan works in the "hot" future, the "cold" future, and the "rainy" future, then it's a good plan.
  • If your plan only works for one specific scenario, it's too risky.

Summary in One Sentence

Nature is too chaotic to be predicted by a single "best guess"; instead of trying to find the one right answer, we must build flexible plans that work no matter which of the many possible futures actually happens.

Technical Summary

1. Problem Statement

Traditional Species Distribution Models (SDMs) typically rely on a single "best-estimate" climate trajectory derived from either a single Earth System Model (ESM) simulation or an ensemble mean. This approach collapses Internal Climate Variability (ICV)—the divergence in climate trajectories arising solely from sensitivity to initial conditions (ICs)—into a single path.

• The Gap: While the existence of ICV is well-established, its specific impact on ecological inference remains uncharacterized. Current practices assume that averaging ICs removes noise, but this study posits that ICV is a source of irreducible uncertainty that can qualitatively alter SDM outcomes (e.g., reversing the direction of range shifts).
• The Question: How does ICV translate into divergent ecological projections, and can single-best estimates be trusted for management decisions?

2. Methodology

The authors developed a fully ICV-aware species distribution modeling framework to explicitly propagate internal variability into ecological projections without averaging it away.

• Data Sources:
  • Species: 34 species (12 marine, 22 terrestrial) selected for data sufficiency, prior SDM usage, and diversity in life history, biogeography, and ecological niches (including vectors, fisheries, and conservation priorities).
  • Climate Projections: 100 initial-condition members from the CESM2-LENS2 (Community Earth System Model 2 Large Ensemble) under the SSP3-7.0 emissions scenario.
  • Time Horizons: Mid-century (2031–2060) and Late-century (2071–2100).
• Bias Correction: To preserve ICV while aligning with observed baselines, the authors applied delta-method bias correction separately for each of the 100 IC members against ERA5 (terrestrial) and CMEMS (marine) climatologies (1981–2014).
• Modeling Approach:
  • A single SDM (XGBoost via biomod2) was trained for each species using observational data and static predictors (elevation, bathymetry).
  • This fixed model was projected across all 100 bias-corrected IC realizations. Thus, the only variable changing between projections was the climate trajectory.
• Ecological Metrics: Four decision-relevant metrics were derived for each realization:
  1. PAC: Potential Area Change.
  2. PAG: Potential Area Gain.
  3. PAL: Potential Area Lost.
  4. PSRC: Potential Suitability Ratio Change.
• Analysis Framework:
  • Consensus Mapping: Calculated the percentage of ensemble members projecting presence at each grid cell.
  • Case Classification: Grouped outcomes into four cases based on the spread and direction of the ensemble.
  • Signal-to-Noise Ratio (SNR): Quantified forced habitat change relative to IC-driven spread.
  • Energy Distance (ED): Measured the separation in multivariate environmental space between ICs projecting presence vs. absence.

3. Key Contributions

1. Demonstration of Qualitative Divergence: Proved that ICV alone is sufficient to generate fundamentally different ecological outcomes, including reversals in projected range shifts (e.g., expansion vs. contraction) for the same species under identical emissions forcing.
2. Four-Case Decision Framework: Introduced a classification system for projection reliability:
   • Case 1: Robust absence of change (high agreement, near-neutral).
   • Case 2: Robust directional response (high agreement on direction and magnitude).
   • Case 3: Consistent direction but high uncertainty in magnitude (agreement on sign, disagreement on severity).
   • Case 4: Directional reversals (disagreement on the sign of change; ensemble mean is misleading).
3. Marine-Terrestrial Divide: Identified a systematic structural difference where marine species exhibit robust, predictable responses (mostly Case 2), while terrestrial species are highly susceptible to ICV (frequently Cases 3 and 4).
4. Nonlinearity Insight: Demonstrated that SDM divergence is not simply proportional to the magnitude of local climate variability but is driven by nonlinear ecological responses to small differences in climate trajectories.

4. Key Results

• Prevalence of Uncertainty: Across all metrics and time periods, 44% of species classifications fell into Case 3 or Case 4. This indicates that nearly half of all species projections are substantially altered by ICV, rendering single-realization or ensemble-mean maps potentially overconfident or directionally inaccurate.
• The Marine-Terrestrial Divide:
  • Marine Species: 80% occupied Case 2 (robust, predictable). The forced climate signal reliably dominates IC-driven spread in oceans.
  • Terrestrial Species: 60% occupied Cases 3 and 4. Terrestrial systems show higher ICV amplitude, leading to greater uncertainty in both magnitude and direction of change.
• Temporal Dynamics: While 75–83% of marine species retained their case classification from mid- to late-century, only 31–40% of terrestrial species did so. Terrestrial projections remain susceptible to ICV throughout the century.
• Metric Dependence: Projection divergence is not intrinsic to the species alone but depends on the metric used. Sessile/specialist species showed cross-metric stability, while mobile generalists showed high instability.
• Mechanism of Divergence:
  • Local climate variability magnitude (e.g., temperature spread) did not systematically predict where SDM disagreement occurred.
  • Energy Distance (ED) analysis revealed that ICs projecting presence vs. absence are systematically separated in multivariate environmental space, even within bounded climate ranges.
  • Unsupervised clustering of raw environmental predictors recovered less than 60% of SDM outcomes, confirming that nonlinear SDM response surfaces amplify small IC differences into divergent binary outcomes.

5. Significance and Implications

• Paradigm Shift: The study argues against the "best-guess" approach in biodiversity planning. Relying on a single trajectory or an ensemble mean can obscure irreducible uncertainty, potentially leading to maladaptive management (e.g., protecting areas that may not be suitable in specific IC realizations).
• Decision-Making Strategy: The authors propose a flexible, stress-tested decision framework:
  • Case 1 & 2: Support targeted, specific interventions.
  • Case 3: Requires flexible policy design (direction is clear, magnitude is not).
  • Case 4: Warrants "no-regrets" measures and enhanced monitoring (direction is unresolved).
• Robustness: Effective climate-ready decisions must be stress-tested across the full envelope of plausible futures rather than optimized for a single trajectory. This is particularly critical for terrestrial systems and mid-century planning horizons where ICV dominates the signal.
• Tooling: The authors provide an interactive Shiny application to explore species-specific projections and the full distribution of IC outcomes, moving beyond static maps to dynamic uncertainty visualization.

In summary, this paper establishes that Internal Climate Variability is a primary driver of irreducible uncertainty in ecological projections, necessitating a shift from deterministic modeling to probabilistic, decision-relevant frameworks that account for multiple plausible futures.