As modern astronomical surveys expand their reach, capturing millions of galaxies across unprecedented volumes of space, a fundamental shift is occurring in the nature of cosmological error budgets. Historically, the precision of cosmological parameters was limited by statistical noise—the simple variance associated with counting a finite number of objects. Today, however, the primary bottleneck is no longer statistical noise, but rather the systematic floors imposed by the process of observation itself.
As R. G. Bower notes, the challenge for next-generation dark energy experiments has fundamentally pivoted from gathering more data to controlling the subtle, non-statistical biases inherent in our instruments and environments. If left uncorrected, these observational systematic errors can easily mimic or obscure the faint signals of new physics.
How Observational Artifacts Mimic Fundamental Physics
To understand why observational systematics are so dangerous, consider how cosmologists extract information from the sky. By measuring the spatial distribution of galaxies or the subtle distortions in their shapes, researchers infer the properties of dark energy, the total mass of neutrinos, or the conditions of the early universe.
However, the universe is observed through a complex chain of filters: the Earth's atmosphere, galactic dust foregrounds, telescope optics, and digital detectors. Each link in this chain can introduce spatial or temporal patterns into the data.
For instance, variations in atmospheric seeing or the local density of stars within our own galaxy can systematically alter the probability of detecting a distant galaxy. If a patch of sky is observed under poor atmospheric conditions, fewer faint galaxies are detected. To a naive analysis pipeline, this artificial drop in galaxy density looks identical to a cosmic void.
Because these observational variations occur over large angular scales, they introduce spurious signals into the galaxy power spectrum. As Nishant Agarwal et al. demonstrate, failing to account for these unmodeled variations can lead to significant shifts in estimates for primordial non-Gaussianity and galaxy bias, potentially generating false-positive discoveries of new physics. Similarly, Shadab Alam et al. emphasize that extracting precise growth-rate measurements to test General Relativity requires meticulous isolation of these non-cosmological signals.
The Taxonomy of Survey Biases
Observational systematics generally fall into three categories based on where they originate in the data-collection pipeline:
1. Foreground and Atmospheric Modulations
Earth's atmosphere and the Milky Way act as dynamic screens. Variations in airmass, sky brightness, and stellar density modulate the depth of a survey across the sky. Antonio J. Cuesta et al. point out that these atmospheric and stellar variations introduce artificial patterns into large-scale structure maps, requiring complex corrections to prevent them from biasing distance measurements derived from Baryon Acoustic Oscillations (BAO).
2. Instrumental and Sampling Artifacts
The physical design of an observatory introduces hard constraints. For example, fiber-fed spectrographs cannot place two fibers arbitrarily close to one another, creating "fiber collisions" that artificially suppress the measured clustering of galaxies at small scales. A. G. Adame et al. highlight the necessity of aggressively controlling and propagating these fiber collisions and spectroscopic effects to ensure robust full-shape clustering analyses.
Even at the detector level, non-linearities like the "pocket effect" in charge-coupled devices (CCDs) can distort photometry, as discussed by M. Rigault et al. in the context of supernova light curves. Furthermore, the discrete temporal sampling of observations—the survey's "window function"—can introduce spurious periodicities, which Vinesh Rajpaul and S. Aigrain show can create entirely artificial signals that mimic physical phenomena.
3. Calibration and Selection Uncertainties
When surveys rely on broadband imaging rather than spectroscopy, they must estimate galaxy distances using photometric redshifts (photo-$z$). Nikhil Padmanabhan et al. establish that uncertainties and miscalibrations in photo-$z$ distributions propagate directly into systematic uncertainties in 3D clustering measurements.
In weak lensing surveys, this issue is compounded by shear calibration bias—minuscule errors in measuring the alignment of galaxy shapes. G. Efstathiou demonstrates that subtle errors in photo-$z$ calibration and the unmodeled intrinsic alignment of galaxies due to local gravitational fields can severely bias cosmological parameter estimates, potentially mimicking physics beyond the standard model.
[Distant Universe]
│
▼
[Galactic Foreground] ──► Modulates observed density (Stellar contamination)
│
▼
[Earth's Atmosphere] ──► Blurs images & varies depth (Seeing, airmass)
│
▼
[Telescope & Detector] ──► Introduces geometric & hardware limits (Fiber collisions)
│
▼
[Naver Analysis] ──► Spurious large-scale signals (Biased parameters)
Strategies for Mitigation
To unlock the true statistical power of modern datasets, cosmologists have developed a multi-layered toolkit designed to isolate, model, and remove these systematic footprints.
Angular Weighting and Polynomial Filtering
The most direct way to handle known spatial systematics is to map them explicitly. By creating templates of observational conditions across the sky (such as maps of stellar density or atmospheric clarity), researchers can calculate how these conditions correlate with observed galaxy numbers.
Beth Reid et al. describe the implementation of angular weighting schemes designed to counteract these non-cosmological fluctuations, ensuring that the final galaxy catalogues reflect true large-scale structure. Alternatively, as D. Thomas explains, polynomial models can be integrated directly into the clustering analysis to absorb these large-scale systematic patterns, though care must be taken to ensure they do not inadvertently absorb real cosmological signals.
Cross-Correlation and Null Tests
When the source of a systematic error is unknown, researchers exploit the mathematical properties of cosmic fields to isolate the bias. One powerful technique involves cross-correlating data from non-overlapping redshift slices. Because galaxies separated by vast cosmic distances should share no true physical clustering, any observed correlation between them must be driven by a shared observational contaminant (such as a patch of bad weather or dust that affected the entire line of sight).
Nishant Agarwal et al. leverage this property to automatically detect and discard contaminated data components. Similarly, Yongzhong Xu et al. demonstrate that directly comparing independent maps of the same sky area allows researchers to isolate hidden instrumental drifts and pointing offsets that indirect statistical measures miss.
Forward Modeling and Simulation-Based Inference
Rather than trying to perfectly clean or "debias" the observational data, an increasingly dominant paradigm is to accept the presence of systematics and simulate them from the ground up. This is known as forward modeling.
Instead of comparing a cleaned data vector to a pristine theoretical model, researchers apply the survey's exact observational imperfections—including masks, selection effects, and instrumental noise—directly to high-fidelity cosmological simulations.
ChangHoon Hahn et al. demonstrate this with the SIMBIG framework, using simulation-based inference to extract robust cosmological constraints from non-linear scales while implicitly accounting for observational systematics. By validating analysis pipelines against these realistic mock catalogs, as shown by Adrian E. Bayer et al., cosmologists can ensure that the final parameter constraints remain accurate, preventing experimental artifacts from being mistaken for new discoveries.