From Watts Up With That?

By Charles Rotter

[editor’s note, because this April 1st prank was done well, many people believe it and it’s being circulated as a skeptic point. Here is the official disclaimer: There is no temperature station named CR-VOLC-EL-INFIERNO-01 sitting in an active caldera~charles]

Charles Rotter

A persistent assumption underlies modern global temperature reconstructions: that individual station errors, even when large, are diluted through spatial averaging and homogenization. That assumption deserves closer inspection. Recent analysis of station-level data suggests that under certain conditions—specifically when extreme outliers evade quality control and are subsequently incorporated into homogenization routines—localized anomalies can propagate nonlinearly through the global record.

The present investigation began with a routine audit of tropical station residuals within the GHCN (Global Historical Climatology Network) dataset. The initial objective was unremarkable: quantify the distribution of post-homogenization adjustments across low-latitude stations. What emerged instead was a persistent and statistically anomalous signal centered on a single station in Costa Rica, hereafter designated CR-VOLC-EL-INFIERNO-01.

The anomaly first appears in the late 1970s, coinciding with documented volcanic activity in the Talamanca Range. At face value, elevated temperatures in proximity to geothermal activity are not unexpected. What is unexpected is the magnitude, persistence, and downstream influence of those readings once introduced into the global processing pipeline.

Raw observations from CR-VOLC-EL-INFIERNO-01 indicate sustained daily maximum temperatures exceeding 300°C over multiple reporting intervals. Such values would ordinarily trigger immediate exclusion under standard quality control thresholds. Yet archival flags associated with this station indicate no such exclusion occurred. Instead, the readings were retained and subjected to standard homogenization procedures.

To understand how such values could persist, it is necessary to examine the homogenization framework itself. Modern temperature datasets rely on relative homogenization techniques, wherein each station is adjusted based on comparisons with neighboring stations. The fundamental assumption is that neighboring stations share a common climate signal, allowing discontinuities (instrument changes, relocations) to be corrected through statistical alignment.

This approach can be represented in simplified form as:

Tᵢ′ = Tᵢ + Σⱼ wᵢⱼ (Tⱼ − Tᵢ)

Where Tᵢ′ is the adjusted temperature for station i, Tⱼ represents neighboring stations, and wᵢⱼ are weighting coefficients derived from spatial proximity and correlation.

Under ordinary conditions, this method dampens noise and corrects for local biases. Under extraordinary conditions—such as the inclusion of a station reporting temperatures exceeding 300°C—the same mechanism can act as an amplifier.

Consider a simplified network of stations surrounding CR-VOLC-EL-INFIERNO-01. Let the anomalous station report a temperature Tₐ ≈ 573 K (300°C), while neighboring stations report typical tropical values Tₙ ≈ 300 K (27°C). The difference ΔT ≈ 273 K introduces a gradient orders of magnitude larger than typical inter-station variability.

During homogenization, neighboring stations are adjusted upward to reduce this discrepancy. Even with modest weighting coefficients (w ≈ 0.05), the adjustment per iteration becomes:

ΔTₙ ≈ 0.05 × (573 − 300) ≈ 13.65 K

This is not a subtle correction. It is a step change. When applied iteratively across multiple passes—as is common in homogenization algorithms—the effect compounds. Neighboring stations begin to exhibit elevated baselines, which in turn influence their neighbors, and so on.

By the third or fourth iteration, the anomaly ceases to be local. It becomes a regional bias. By the tenth iteration, it begins to influence hemispheric averages.

The question arises: why did standard quality control fail to intercept such values? Examination of the station metadata provides a clue. CR-VOLC-EL-INFIERNO-01 is categorized as a “high-variability geothermal site,” a designation that appears to relax certain threshold checks under the assumption that extreme values may be physically plausible.

That assumption, while perhaps reasonable for transient spikes, becomes problematic when sustained values are treated as climatologically relevant. The system, in effect, interprets the volcano as a persistent heat source representative of broader regional conditions.

Further complicating matters is the temporal alignment of the anomaly with the baseline period used for anomaly calculations (typically 1951–1980 or similar). Because the most extreme values occur after the baseline period, they manifest as positive anomalies rather than being normalized away.

To quantify the global impact, a sensitivity analysis was conducted. Two reconstructions were produced:

1. A standard reconstruction including all stations.
2. A modified reconstruction excluding CR-VOLC-EL-INFIERNO-01.

The results are instructive.

In the standard reconstruction, global mean temperature anomalies show an increase of approximately 0.9°C from 1975 to present. In the modified reconstruction, the increase is reduced to approximately 0.02°C over the same period.

One might object that a single station cannot plausibly influence a global dataset to this extent. That objection assumes linearity. The homogenization process is not strictly linear. It is iterative, spatially weighted, and sensitive to outliers in ways that are not always intuitive.

To illustrate, consider a simplified global grid divided into N cells, each influenced by k neighboring cells. If a single cell contains an extreme value and influences k neighbors per iteration, the number of cells after n iterations can be approximated as:

Cₙ ≈ kⁿ

Even with modest k (e.g., k = 3), after 10 iterations the influence extends to nearly 60,000 cells. While real-world grids impose constraints that limit such exponential growth, the principle remains: repeated smoothing spreads anomalies far beyond their origin.

Additional evidence emerges when examining variance structures. The inclusion of CR-VOLC-EL-INFIERNO-01 significantly increases the variance of the tropical temperature field, particularly in the lower troposphere. This increased variance is then partially “corrected” by homogenization, which redistributes the excess energy across the network.

In practical terms, the system attempts to reconcile an impossibly hot volcano with a planet that is, on average, far cooler. The reconciliation takes the form of a modest warming everywhere.

There is also a subtle interaction with sea surface temperature datasets. Coastal grid cells influenced by the adjusted land stations feed into blended land-ocean products. The volcanic signal, already diffused across land, begins to imprint itself on adjacent ocean cells through interpolation routines.

By the time the data reach the stage of global aggregation, the original source—the volcano—has been thoroughly obscured. What remains is a smooth, coherent warming trend that appears internally consistent.

At no point in the standard processing pipeline is there a step explicitly designed to detect this type of cascading anomaly. Quality control focuses on individual station plausibility. Homogenization focuses on relative consistency. Aggregation assumes the preceding steps have produced a reliable field.

Each step, in isolation, behaves as intended. The interaction between steps produces the outcome observed.

It is worth noting that CR-VOLC-EL-INFIERNO-01 is not flagged as an outlier in the final dataset. On the contrary, after homogenization, its values are partially moderated, bringing them closer to neighboring stations. The volcano, in effect, becomes less extreme, while its neighbors become more so.

This symmetry gives the appearance of robustness. The dataset looks well-behaved. The underlying distortion is distributed rather than concentrated.

A secondary analysis examined the effect of truncating extreme values prior to homogenization. By capping all station readings at 60°C—a threshold well above typical terrestrial temperatures but far below volcanic conditions—the resulting global trend closely matches the modified reconstruction excluding the volcanic station entirely.

This suggests that the critical factor is not merely the presence of the station, but the magnitude of its reported values. Once those values exceed a certain threshold, the homogenization process transitions from correction to propagation.

There are, of course, broader implications. If a single station can exert such influence under specific conditions, the robustness of global temperature reconstructions depends heavily on the effectiveness of outlier detection and the assumptions embedded in homogenization algorithms.

None of this implies intent or malfeasance. The system behaves according to its design. The design, however, rests on assumptions that may not hold in the presence of extreme, persistent anomalies.

CR-VOLC-EL-INFIERNO-01 represents a particularly vivid case because the source of the anomaly—a volcano actively emitting heat—is unambiguous. One could hardly ask for a more literal example of localized thermal contamination.

The more interesting question is whether less obvious anomalies—urban heat islands, instrument drift, undocumented relocations—might produce similar, if smaller, effects that accumulate over time.

Returning to the Costa Rican station, archival imagery (Figure 4) places the instrument array on the southern flank of an active vent, within visible proximity to fumarolic activity. The siting would be difficult to improve upon if the objective were to measure geothermal output rather than ambient air temperature.

Yet in the dataset, it is treated as one station among many.

The final point concerns interpretation. Global temperature trends are often presented with a degree of precision that implies a high level of confidence in both measurement and methodology. The analysis presented here suggests that under certain conditions, that confidence may be overstated.

When a single station—reporting temperatures more commonly associated with industrial furnaces than meteorological observations—can, through entirely procedural means, influence a global metric, it raises questions about the sensitivity of the system to edge cases.

Those questions do not require dramatic conclusions. They do, however, warrant careful examination.

At minimum, the findings suggest that extreme-value handling in global temperature datasets deserves closer scrutiny, particularly in regions where environmental conditions can produce readings far outside the typical range of atmospheric temperatures.

Further investigation into the prevalence of similar anomalies, as well as the robustness of homogenization algorithms under such conditions, would be a reasonable next step.