Temporal uncertainty in fossil records can distort distributions and ecological niches during periods of climatic instability

This study demonstrates that while fossil-based species distribution models remain robust to temporal uncertainty during stable climatic periods, high environmental variability—particularly during deglacial transitions—significantly distorts inferred ecological niches and distributions when age errors exceed a few thousand years.

The Gist

The Big Idea: The "Time Travel" Problem in Fossils

Imagine you are a detective trying to solve a crime that happened 10,000 years ago. You find a single clue (a fossil) buried in the ground. You know where it was found, but you aren't 100% sure exactly when it was left there. Maybe it was 10,000 years ago, maybe 10,200, or maybe 9,800.

This paper is about what happens when scientists try to use these fuzzy dates to figure out what the climate was like back then and where animals lived.

The authors asked a simple question: If we get the date of a fossil slightly wrong, does it completely mess up our picture of the past?

The Experiment: Virtual Animals and Time Machines

Since we can't actually go back in time to check the dates, the researchers built a "virtual world" inside a computer.

1. The Cast: They created four "virtual" small mammals (like virtual mice and rabbits) and gave them specific "personalities" regarding what climates they liked (e.g., one loved cold snow, another loved warm deserts).
2. The Truth: They knew exactly where these virtual animals lived and what the weather was like at three specific moments in history:
   • The Holocene (6,000 years ago): A time of stable, warm weather (like a calm summer day).
   • The Deglacial (13,500 years ago): A time of wild, chaotic weather swings (like a stormy day with sudden temperature spikes).
   • The Last Glacial Maximum (18,000 years ago): A time of deep ice and cold, but relatively steady (like a frozen winter).
3. The Mistake: Then, they pretended to be "bad detectives." They took the virtual fossils and randomly shifted their dates forward and backward by different amounts (from a few years to thousands of years).
4. The Test: They tried to rebuild the map of where the animals lived using these "wrong" dates and compared it to the "True" map they started with.

The Results: It Depends on the Weather

The study found that the answer isn't a simple "yes" or "no." It depends entirely on how much the weather was changing at that time.

1. The Calm Days (The Holocene)

Analogy: Imagine you are taking a photo of a still pond. If you take the photo a few minutes early or late, the picture looks almost exactly the same. The water hasn't moved.

• What happened: During the stable Holocene period, even if the researchers were off by 2,500 years, the computer models still got the animal's home range mostly right. The climate didn't change fast enough for a dating error to matter much.

2. The Stormy Days (The Deglacial)

Analogy: Now imagine taking a photo of a rollercoaster at the exact moment it's flipping upside down. If you take the photo just a split second too early or too late, the picture looks completely different. One second the rider is up, the next they are down.

• What happened: During the deglacial period, the climate was changing wildly. Here, even a small dating error (like being off by 600 years) caused the computer models to get confused. They thought the animals lived in the wrong places because the climate shifted so fast that a "wrong" date meant matching the animal to the wrong weather.

3. The Frozen Days (The Last Glacial Maximum)

Analogy: Imagine a frozen lake. It's cold and steady. If you take a photo a few days early or late, the ice looks the same. But if you wait too long (thousands of years), the ice melts and the whole landscape changes.

• What happened: During the Ice Age, the models were okay with small errors, but if the dating error was huge (spanning the whole last 22,000 years), the models fell apart. They tried to match animals to environments that didn't exist anymore (like matching a cold-loving animal to a warm, melted landscape).

The Takeaway: Why This Matters

This paper is a warning label for scientists who use fossils to predict the future.

• The Trap: If you use a fossil to say, "This animal lived here 15,000 years ago," and you assume the climate was exactly what it was at that moment, you might be wrong. If the climate was changing rapidly (like during the deglacial), your "wrong" date leads to a "wrong" climate picture.
• The Consequence: If we get the past wrong, we might misunderstand how animals adapt to climate change. This could lead to bad decisions about how to protect endangered species today.

The Bottom Line

Fossils are like time capsules, but sometimes the clock inside them is a little broken.

• If the world was calm when the animal died, a broken clock doesn't matter much.
• If the world was chaotic when the animal died, a broken clock ruins the whole story.

The authors suggest that scientists need to be extra careful when studying periods of rapid climate change. They should test their models by "jittering" the dates (shaking them up a bit) to see if their conclusions hold up, rather than assuming they have the perfect date.

Technical Summary

1. Problem Statement

Ecological Niche Models (ENMs) are increasingly used with fossil data to reconstruct past species distributions and understand evolutionary responses to climate change. However, these models rely on the precise alignment of occurrence data with environmental conditions. Fossil records inherently suffer from temporal uncertainty due to:

• Limitations in radiometric dating precision.
• Time-averaging within stratigraphic units.
• The assignment of fossils to broad time bins rather than specific dates.

When a fossil's age is uncertain, the environmental data assigned to it (e.g., temperature, precipitation) may correspond to a different time period than the actual occurrence. This "mismatch" can distort inferred ecological niches and distribution predictions. While temporal uncertainty is well-studied in phylogenetics, its specific impact on ENM performance—particularly how it interacts with environmental variability across different climatic epochs—remains largely unexplored.

2. Methodology

The authors employed a Virtual Ecology approach to isolate the effects of temporal uncertainty from other biases (like preservation potential) by using simulated data where the "true" niche and distribution are known.

• Study Area & Data:

  • Region: North America.
  • Time Periods: Three distinct climatic epochs in the late Quaternary (~22,000 years ago):
    1. Holocene (6,000 y.b.p.): Warm and stable.
    2. Deglacial (13,500 y.b.p.): Highly variable transition.
    3. Last Glacial Maximum (LGM) (18,000 y.b.p.): Cold and moderately variable.
  • Environmental Variables: Six layers (elevation, slope, mean annual temperature, temperature seasonality, annual precipitation, precipitation seasonality) at 1km resolution, sourced from CHELSA-TraCE21K.

• Virtual Species Creation:

  • Four virtual small mammal species were modeled based on real-world analogs (Microtus pennsylvanicus, Neotoma albigula, Lepus californicus, Blarina carolinensis).
  • Real-world MaxEnt models were trained on GBIF occurrence data, and response curves were used to generate mathematical functions for the virtual species.
  • "True" distributions were generated for each species at the midpoints of the three time periods.

• Experimental Design (The "Jittering" Approach):

  • Baseline: 100 occurrences were sampled from the "true" distribution and matched with contemporaneous environmental layers to create a "True ENM."
  • Uncertainty Simulation: The same 100 occurrences were subjected to temporal jittering. Dates were randomly reassigned from uniform distributions centered on the true date, with uncertainty ranges ($\pm$) of: 200, 400, 600, 800, 1000, 1500, 2500, 4000 years, and the full Late Quaternary (~22,000 years).
  • Replication: This process was repeated 100 times for each uncertainty level to assess variability.
  • Models: Both Principal Component Analysis (PCA) for niche space and MaxEnt for distribution prediction were run on the jittered data.

• Metrics:

  • Niche Reconstruction Accuracy: Measured as the Euclidean distance between the centroid of the jittered PCA space and the true PCA centroid.
  • Distribution Prediction Error: Measured as the Root Mean Squared Error (RMSE) between the probability of occurrence predicted by the jittered MaxEnt model and the true probability.
  • Environmental Variability: Quantified as the Interquartile Range (IQR) of environmental differences within the uncertainty window, normalized to a 0–1 scale.

3. Key Contributions

• Quantification of Thresholds: The study identifies specific thresholds of temporal uncertainty beyond which ENM accuracy degrades significantly, showing these thresholds are not static but depend on climatic stability.
• Interaction with Environmental Variability: It demonstrates that the impact of temporal uncertainty is non-linear and heavily dependent on the rate of environmental change. High variability amplifies the negative effects of dating uncertainty.
• Methodological Framework: Provides a robust virtual ecology framework for testing ENM sensitivity to temporal biases, which can be applied to real-world fossil datasets.

4. Key Results

• Robustness in Stable Periods: During the Holocene (stable climate), niche reconstructions and distribution predictions remained accurate even with temporal uncertainties up to $\pm$ 2,500 years. Prediction errors only spiked when uncertainty spanned the entire Late Quaternary.
• Sensitivity in Variable Periods:
  • Deglacial: This period showed the highest sensitivity. Even low levels of temporal uncertainty ($\pm$ 600 years) caused rapid increases in prediction error and niche distortion. The "hump-shaped" pattern in centroid distance suggested that jittering during this transition period caused occurrences to oscillate between LGM and Holocene conditions.
  • LGM: Showed intermediate sensitivity. Errors increased gradually with uncertainty but reached high levels when the full Late Quaternary range was included.
• Environmental Variability as a Driver: There is a strong positive correlation between environmental variability and prediction error.
  • In stable periods, large temporal windows are required to introduce significant environmental mismatch.
  • In highly variable periods (Deglacial), even small temporal windows introduce large environmental mismatches, leading to immediate model degradation.
• Niche Distortion: Temporal uncertainty introduced systematic, non-random biases in niche centroids during stable periods (shifting them away from the true niche), whereas during the deglacial, the bias was more stochastic (orbiting the true centroid).

5. Significance and Implications

• Re-evaluating Past Studies: The findings suggest that previous studies reporting niche shifts or range changes based on fossil ENMs may have conflated true ecological responses with artifacts of temporal uncertainty, particularly for studies focusing on the deglacial transition or deep-time periods with poor dating resolution.
• Conservation and Prediction: For conservation planning using paleo-data, ignoring temporal uncertainty in periods of high climatic volatility can lead to erroneous conclusions about species' niche breadths and resilience.
• Strategic Dating Priorities: The study offers a framework for resource allocation in paleontology. It suggests that fossil specimens from periods of high environmental variability (e.g., deglaciation) or those with large initial age uncertainties should be prioritized for high-precision dating, as they have the greatest potential to distort ENM outputs.
• Best Practices: The authors recommend:
  • Conducting sensitivity analyses by varying fossil dates within their error bounds rather than assuming a single midpoint.
  • Avoiding the use of broad time-bin midpoints for environmental assignment in volatile periods.
  • Integrating independent lines of evidence (genomics, phylogenetics) to validate ENM inferences.

In conclusion, the paper establishes that while ENMs are robust to temporal uncertainty in stable climates, they are highly vulnerable during periods of rapid climatic change. Explicitly accounting for temporal uncertainty is therefore critical for accurate paleo-ecological reconstruction.