Papers
Topics
Authors
Recent
Search
2000 character limit reached

KiDS-1000: Cosmological Data Release

Updated 6 July 2026
  • KiDS-1000 is the fourth data release of the Kilo-Degree Survey, covering 1006 deg² in nine optical-to-near-IR bands and providing calibrated weak-lensing measurements.
  • The survey employs deep r-band imaging with precise PSF modelling and tomographic binning to achieve robust redshift calibration and shape estimation for 21 million galaxies.
  • Advanced analyses using KiDS-1000 data constrain the structure-growth parameter S8 and address systematics via techniques like MetaCalibration and simulation-based inference.

Searching arXiv for recent KiDS-1000 papers to ground the article. KiDS-1000 is the fourth data release of the Kilo-Degree Survey, covering 1006 deg2^2 in nine optical-to-near-infrared bands, ugriZYJHKsugriZYJHK_s, from VST/OmegaCAM and VIKING/VISTA, and optimized for weak lensing. For weak-lensing analyses, the release is typically described through an effective unmasked area of $777.4$ deg2^2, a “gold sample” of 21 million galaxies with calibrated redshift distributions, and five tomographic source bins spanning 0.1<zB≤1.20.1<z_B\le1.2 (Giblin et al., 2020, Hildebrandt et al., 2020). KiDS-1000 has been used for cosmic shear, galaxy-galaxy lensing, spectroscopic clustering, higher-order shear statistics, density split statistics, peak counts, cluster abundance and weak lensing, CMB-lensing cross-correlation, and simulation-based inference, with repeated emphasis on the structure-growth parameter S8≡σ8Ωm/0.3S_8\equiv \sigma_8\sqrt{\Omega_{\rm m}/0.3} and on the control of shear, redshift, intrinsic-alignment, and baryonic systematics (Heymans et al., 2020, Burger et al., 2023, Wietersheim-Kramsta et al., 2024).

1. Survey definition and catalogue construction

KiDS-1000 combines deep rr-band imaging for shape measurement with matched nine-band photometry for photometric redshifts. The rr-band weak-lensing data were observed under the requirement that the PSF full-width at half maximum is <0.8′′<0.8'', yielding a mean seeing of $0.7''$ and a median 5σ5\sigma point-source depth of r=25.02±0.13r=25.02\pm0.13 mag in a $2''$ aperture (Giblin et al., 2020). Shape estimation is based on the model-fitting algorithm lensfit, applied to the set of unstacked rr-band exposures, with each galaxy assigned an ellipticity ϵ=ϵ1+iϵ2\epsilon=\epsilon_1+i\epsilon_2 and a weight ww (Giblin et al., 2020).

The release is often summarized through a small set of survey-level quantities.

Quantity Value Source
Total imaging footprint $1006$ deg2^2 (Giblin et al., 2020)
Effective unmasked weak-lensing area $777.4$ deg2^2 (Giblin et al., 2020)
Gold-sample size 21 million galaxies (Giblin et al., 2020)
Gold-sample neffn_{\rm eff} $6.17$ arcmin−2^{-2} (Giblin et al., 2020)
Pre-redshift-selection neffn_{\rm eff} $8.43$ arcmin−2^{-2} (Giblin et al., 2020)
Tomographic binning five bins, 0.1<zB≤1.20.1<z_B\le1.2 (Hildebrandt et al., 2020)

PSF modelling was constructed on a 32×3232\times32 pixel grid per exposure, with each pixel represented by a 2D polynomial of order n=4n=4 across the focal plane and of order nc=1n_c=1 per CCD chip, so as to capture discontinuities at chip boundaries (Giblin et al., 2020). The shear catalogue paper states that the PSF model meets the requirement to induce less than a 0.1σ0.1\sigma change in the inferred cosmic-shear constraints on S8S_8, and the catalogue-level validation found no evidence for significant non-lensing B-mode distortions in the data (Giblin et al., 2020).

Later reanalyses refined the basic catalogue products rather than replacing the survey definition. The “v2” cosmic-shear analysis updated lensfit from v309c to v321, applied an empirical PSF-leakage correction, and reported that the overall additive bias c2c_2 was halved relative to the v1 catalogue, now at c2∼3×10−4c_2\sim3\times10^{-4} (Li et al., 2023). A still later reanalysis implemented MetaCalibration, obtained neff=7.77n_{\rm eff}=7.77 arcmin−2^{-2} versus $6.69$ arcmin−2^{-2} for lensfit in the cosmology sample, and found multiplicative biases ∣m∣<0.017|m|<0.017 in all five cosmology bins (Yoon et al., 1 Oct 2025).

2. Redshift calibration and tomographic binning

KiDS-1000 tomography is defined from BPZ photometric-redshift point estimates zBz_B, with bins (0.1,0.3],(0.3,0.5],(0.5,0.7],(0.7,0.9],(0.9,1.2](0.1,0.3], (0.3,0.5], (0.5,0.7], (0.7,0.9], (0.9,1.2] (Hildebrandt et al., 2020). The baseline calibration of the source redshift distributions used deep spectroscopic reference catalogues re-weighted with a self-organising map (SOM) in nine-dimensional magnitude space. If Np(c)N_p(c) and Ns(c)N_s(c) denote the photometric and spectroscopic occupancies of SOM cell cc, the cell weight is

w(c)=Np(c)Ns(c),w(c)=\frac{N_p(c)}{N_s(c)},

and the reweighted redshift distribution is

nSOM(z)=∑c=1Ncellsw(c) nspec,c(z).n_{\rm SOM}(z)=\sum_{c=1}^{N_{\rm cells}} w(c)\,n_{{\rm spec},c}(z).

A “gold” selection removes cells without spectroscopic support or with a large spectroscopic-photometric discrepancy (Hildebrandt et al., 2020).

Validation on 100 independent KiDS-like MICE realizations found ∣Δ⟨z⟩jSOM∣≲0.01|\Delta\langle z\rangle_j^{\rm SOM}|\lesssim0.01 in all five bins, with σ(Δ⟨z⟩SOM)≈0.01\sigma(\Delta\langle z\rangle^{\rm SOM})\approx0.01 (Hildebrandt et al., 2020). An independent clustering-redshift approach fitted

nCZ(z)≈A⋅nSOM(z+δzCZ),n_{\rm CZ}(z)\approx A\cdot n_{\rm SOM}(z+\delta z^{\rm CZ}),

and found offsets consistent with zero, with combined uncertainties of order $0.01$–$0.02$ across the five bins (Hildebrandt et al., 2020). This dual calibration is central to the claim that redshift errors remain a subdominant part of the KiDS-1000 weak-lensing error budget (Hildebrandt et al., 2020).

The redshift calibration was later expanded in a dedicated cosmic-shear reanalysis. That work compiled 17 spectroscopic campaigns across six $1$ deg2^2 “KiDZ” fields, enlarging the spectroscopic sample from ∼25,000\sim25{,}000 to 52,90052{,}900, then to 61,20061{,}200 with PAUS and to 112,400112{,}400 with COSMOS2015 photo-zzs (Busch et al., 2022). The fraction of KiDS-1000 source galaxies with reliable calibration rose from ∼80%\sim80\% to ∼89%\sim89\% with the “spec-z fiducial” sample and to ∼95%\sim95\% when PAUS and COSMOS2015 were included, with shifts in S8S_8 of at most 0.5σ0.5\sigma across the tested calibration subsets (Busch et al., 2022).

A more recent development replaced summary redshift calibration with full posterior inference of galaxy properties. The pop-cosmos analysis performed SED fitting for 4 million KiDS-1000 galaxies, validated photometric redshifts against ∼185,000\sim185{,}000 DESI DR1 spectroscopic matches, and argued that physically selected source samples can mitigate intrinsic-alignment systematics while preserving statistical power (Halder et al., 3 Feb 2026). This suggests a transition from tomographic samples defined only by zBz_B to source samples defined by posterior physical properties.

3. Shear estimation and summary statistics

The basic KiDS-1000 shear observables are the tomographic two-point correlation functions

ξ±(ij)(θ)=⟨γtγt⟩ij(θ)±⟨γ×γ×⟩ij(θ),\xi_\pm^{(ij)}(\theta)=\langle \gamma_t\gamma_t\rangle_{ij}(\theta)\pm \langle \gamma_\times\gamma_\times\rangle_{ij}(\theta),

estimated from weighted galaxy pairs (Giblin et al., 2020). Multiple compressed representations of these two-point functions were developed for KiDS-1000. In the COSEBI basis, the pure-E mode on a finite angular interval is

En(ij)=12∫θmin⁡θmax⁡dθ θ[T+n(θ) ξ+(ij)(θ)+T−n(θ) ξ−(ij)(θ)],E_n^{(ij)}=\frac12\int_{\theta_{\min}}^{\theta_{\max}} d\theta\,\theta \left[T_{+n}(\theta)\,\xi_+^{(ij)}(\theta)+T_{-n}(\theta)\,\xi_-^{(ij)}(\theta)\right],

with T±nT_{\pm n} chosen so that only pure E-modes contribute (Burger et al., 2023). KiDS-1000 COSEBI analyses used θmin⁡=0.5′\theta_{\min}=0.5', θmax⁡=300′\theta_{\max}=300' in earlier work and θmin⁡=2′\theta_{\min}=2', θmax⁡=300′\theta_{\max}=300' in the improved cosmic-shear analysis (Busch et al., 2022, Li et al., 2023).

A parallel line of work used Fourier-space statistics. The pseudo-CℓC_\ell analysis divided 21,262,011 galaxies into five tomographic bins and measured eight logarithmic bandpowers in the multipole range 76<ℓ<150076<\ell<1500 for all auto- and cross-spectra (Loureiro et al., 2021). The method forward-modelled the survey mask through a mixing matrix Mℓℓ′M_{\ell\ell'}, and the B-mode bandpowers were reported to be consistent with zero signal, with no significant residual contamination from E/B-mode leakage (Loureiro et al., 2021). The joint weak-lensing and clustering methodology paper similarly emphasized band powers and related Fourier-space statistics because they are insensitive to the survey mask and display low levels of mode mixing (Joachimi et al., 2020).

KiDS-1000 also supported genuinely non-Gaussian lensing statistics. The third-order aperture-mass statistic is defined from the compensated filter UU and shear-space partner QQ, with

M(θ;θap)=∫d2θ′ Uθap(∣θ′∣) κ(θ+θ′)=∫d2θ′ Qθap(∣θ′∣) γt(θ+θ′),M(\theta;\theta_{\rm ap})=\int d^2\theta' \,U_{\theta_{\rm ap}}(|\theta'|)\,\kappa(\theta+\theta') =\int d^2\theta' \,Q_{\theta_{\rm ap}}(|\theta'|)\,\gamma_t(\theta+\theta'),

and third moments ⟨M3⟩(ijk)(θap,1,θap,2,θap,3)\langle M^3\rangle^{(ijk)}(\theta_{{\rm ap},1},\theta_{{\rm ap},2},\theta_{{\rm ap},3}) probe the bispectrum of the convergence field (Burger et al., 2023). These higher-order quantities were later combined with COSEBIs in a tomographic cosmology analysis.

4. Modelling, covariance, and inference pipeline

The standard KiDS-1000 cosmology pipeline couples weak-lensing projection kernels, calibrated source redshift distributions, non-linear matter modelling, nuisance-parameter marginalization, and validated covariance models. In the joint weak-lensing and clustering methodology, linear power spectra were computed with CAMB, non-perturbative non-linear matter power with HMCode, and galaxy clustering with a hybrid model that blends one-loop renormalised perturbation theory with halo-model ingredients (Joachimi et al., 2020). In the later cosmic-shear and higher-order analyses, the non-linear matter power spectrum was modelled with HMcode2020, while third-order statistics used BiHalofit for the bispectrum (Li et al., 2023, Burger et al., 2023).

Intrinsic alignments are generally treated with the non-linear alignment model. In the higher-order shear analysis,

δI=fIA(z) δ,fIA(z)=−AIA C1 ρˉ(z)/D+(z),\delta_I=f_{\rm IA}(z)\,\delta,\qquad f_{\rm IA}(z)= -A_{\rm IA}\,C_1\,\bar\rho(z)/D_+(z),

which induces the familiar PδIP_{\delta I}, PIIP_{II}, and bispectrum analogues (Burger et al., 2023). Baryonic feedback is incorporated either through HMCode nuisance parameters such as AbaryA_{\rm bary} or TAGNT_{\rm AGN}, or through multiplicative responses measured from hydrodynamical simulations such as Magneticum (Heymans et al., 2020, Li et al., 2023, Burger et al., 2023). Across these analyses, photo-zz shifts δzi\delta z_i and shear-calibration uncertainties mim_i are propagated as nuisance parameters or additive covariance contributions (Joachimi et al., 2020, Li et al., 2023).

Covariance construction was a major methodological component of KiDS-1000. The joint weak-lensing and clustering methodology used a dedicated suite of more than 20,000 mocks to assess covariance performance and to quantify the impact of survey geometry and spatial variations of survey depth on signals and errors (Joachimi et al., 2020). The same work found that standard point estimates of S8S_8 from a marginal posterior can under-cover the true value and introduced the projected joint highest-posterior-density (PJ-HPD) interval around the multivariate MAP point to improve calibration (Joachimi et al., 2020). This inference convention recurs in later KiDS-1000 analyses.

The simulation-based inference analysis replaced an analytic likelihood by a learned likelihood. KiDS-SBI used 18,000 forward realizations, score compression from 120 data points down to 7 summaries, non-Limber projection via Levin’s method, and log-normal random matter fields on the curved sky (Wietersheim-Kramsta et al., 2024). The forward model included variable depth, PSF anisotropy, shear calibration, and redshift calibration. A key methodological result was that neglecting variable depth and PSF anisotropies can cause S8S_8 to be overestimated by ∼5%\sim5\%, and that fixing the covariance at a fiducial cosmology underestimates uncertainties on S8S_8 by ∼10%\sim10\% (Wietersheim-Kramsta et al., 2024).

5. Principal cosmological constraints and the S8S_8 issue

Published KiDS-1000 cosmic-shear analyses report closely related but not identical S8S_8 constraints. The pseudo-CℓC_\ell analysis found

S8=0.754−0.029+0.027,S_8=0.754_{-0.029}^{+0.027},

and, when combined with SDSS BAO, RSD, and Lyα\alpha, obtained S8=0.771−0.032+0.006S_8=0.771^{+0.006}_{-0.032} (Loureiro et al., 2021). The enhanced redshift-calibration COSEBIs analysis reported

S8=0.748−0.025+0.021,S_8=0.748_{-0.025}^{+0.021},

while the improved cosmic-shear measurements paper reported

S8=0.776−0.027−0.003+0.029+0.002,S_8=0.776_{-0.027-0.003}^{+0.029+0.002},

where the second uncertainty quantifies systematic shear-calibration uncertainty (Busch et al., 2022, Li et al., 2023). The MetaCalibration reanalysis found

S8=0.789−0.024+0.020,S_8=0.789_{-0.024}^{+0.020},

with about 28%28\% improved constraining power relative to the lensfit analysis, while concluding that the difference with Planck remains at a similar level and is not caused by the shear measurements (Yoon et al., 1 Oct 2025).

The flagship multi-probe KiDS-1000 3×23\times2pt analysis combined cosmic shear, spectroscopic galaxy clustering from BOSS, and galaxy-galaxy lensing from the KiDS-BOSS and KiDS-2dFLenS overlaps. Its fiducial MAP+PJ-HPD result was

S8=0.766−0.014+0.020,σ8=0.760−0.023+0.021,Ωm=0.305−0.015+0.010,S_8=0.766^{+0.020}_{-0.014},\qquad \sigma_8=0.760^{+0.021}_{-0.023},\qquad \Omega_{\rm m}=0.305^{+0.010}_{-0.015},

with S8S_8 lower than Planck by 8.3±2.6%8.3\pm2.6\% (Heymans et al., 2020). In the companion beyond-flat-Λ\LambdaCDM study, the same 3×23\times2pt data were found to be fully consistent with ΩK=0.011−0.057+0.054\Omega_K=0.011^{+0.054}_{-0.057}, ∑mν<1.76\sum m_\nu<1.76 eV at 95%95\% CL, and w=−0.99−0.13+0.11w=-0.99^{+0.11}_{-0.13}, with no clear preference for the fiducial flat Λ\LambdaCDM model or the tested extensions (Tröster et al., 2020).

A recurrent theme of the KiDS-1000 literature is the “S8S_8 tension” with Planck. The quoted significance depends on probe combination, modelling choices, and the tension metric. The improved cosmic-shear measurements paper reported ∼2.3σ\sim2.3\sigma (Li et al., 2023), the enhanced redshift-calibration analysis reported ∼3.4σ\sim3.4\sigma (Busch et al., 2022), the pseudo-CℓC_\ell analysis reported ∼3σ\sim3\sigma (Loureiro et al., 2021), and the 3×23\times2pt analysis found ∼3σ\sim3\sigma for one-dimensional S8S_8 comparisons but ∼2σ\sim2\sigma in the full multidimensional parameter space (Heymans et al., 2020). The beyond-Λ\LambdaCDM analysis further showed that one-dimensional S8S_8 tension can disappear in wwCDM while persisting in the joint (S8,w)(S_8,w) space (Tröster et al., 2020).

6. Non-Gaussian, multi-probe, and methodological extensions

KiDS-1000 has been a testbed for moving beyond Gaussian two-point statistics. The combined second- and third-order shear analysis performed the first cosmological parameter analysis of KiDS-1000 with COSEBIs and ⟨Map3⟩\langle M_{\rm ap}^3\rangle, using HMcode2020 for the power spectrum, BiHalofit for the bispectrum, an analytic intrinsic-alignment model, and hydrodynamical simulations for baryonic feedback (Burger et al., 2023). A key technical step was the equal-filter-radii ansatz, with θap={4′,8′,14′,32′}\theta_{\rm ap}=\{4',8',14',32'\}, which reduced the data-vector dimension from ∼775\sim775 to 215 while losing only ≃8%\simeq8\% of the joint Ωm\Omega_{\rm m}–S8S_8 figure of merit (Burger et al., 2023). The resulting combined constraint was

Ωm=0.248−0.055+0.062,S8=0.772±0.022,\Omega_{\rm m}=0.248^{+0.062}_{-0.055},\qquad S_8=0.772\pm0.022,

with negligible validation bias and smaller errors than the second-order-only case (Burger et al., 2023).

Other non-Gaussian statistics were also applied to KiDS-1000. Density split statistics yielded

Ωm=0.27±0.02,S8=0.731−0.018+0.030,\Omega_{\rm m}=0.27\pm0.02,\qquad S_8=0.731^{+0.030}_{-0.018},

and were described as competitive with two-point cosmic shear while additionally constraining galaxy bias and shot-noise deviations (Burger et al., 2022). Peak count statistics obtained

Σ8≡σ8[Ωm/0.3]0.60=0.765−0.030+0.030,\Sigma_8\equiv \sigma_8\left[\Omega_{\rm m}/0.3\right]^{0.60}=0.765^{+0.030}_{-0.030},

in the KiDS-only analysis and Σ8joint=0.759−0.017+0.020\Sigma_8^{\rm joint}=0.759^{+0.020}_{-0.017} in the combined KiDS-1000 and DES-Y1 analysis (Harnois-Deraps et al., 2024). A halo-model joint analysis of the stellar mass function, projected clustering, and galaxy-galaxy lensing found

S8=0.773−0.030+0.028,Ωm=0.290−0.017+0.021,S_8=0.773^{+0.028}_{-0.030},\qquad \Omega_{\rm m}=0.290^{+0.021}_{-0.017},

demonstrating that small-scale clustering and galaxy-galaxy lensing can deliver constraints comparable to 3×23\times2pt analyses without including cosmic shear (Dvornik et al., 2022). A cluster-count and stacked weak-lensing analysis of about 8000 AMICO clusters reported

Ωm=0.218−0.021+0.024,σ8=0.86−0.03+0.03,S8=0.74−0.03+0.03,\Omega_{\rm m}=0.218^{+0.024}_{-0.021},\qquad \sigma_8=0.86^{+0.03}_{-0.03},\qquad S_8=0.74^{+0.03}_{-0.03},

together with an average mass precision of ≃8%\simeq8\% for the log⁡λ∗−log⁡M200\log\lambda^*-\log M_{200} relation (Lesci et al., 18 Jul 2025).

Analysis Parameter Representative result
Improved cosmic shear measurements S8S_8 0.776−0.027−0.003+0.029+0.0020.776_{-0.027-0.003}^{+0.029+0.002}
Multi-probe 3×23\times2pt S8S_8 0.766−0.014+0.0200.766^{+0.020}_{-0.014}
COSEBIs +⟨Map3⟩+\langle M_{\rm ap}^3\rangle S8S_8 0.772±0.0220.772\pm0.022
Density split statistics S8S_8 0.731−0.018+0.0300.731^{+0.030}_{-0.018}
Peak counts Σ8\Sigma_8 0.765−0.030+0.0300.765^{+0.030}_{-0.030}
AMICO clusters: counts + weak lensing S8S_8 0.74−0.03+0.030.74^{+0.03}_{-0.03}

KiDS-1000 has also supported cross-correlation and model-building studies. The Planck CMB-lensing cross-correlation analysis used an intrinsic-alignment self-calibration method and obtained Alens=0.84−0.22+0.22A_{\rm lens}=0.84^{+0.22}_{-0.22} and AIA=0.60−1.03+1.03A_{\rm IA}=0.60^{+1.03}_{-1.03}, while stressing the importance of boost-factor, cosmic-magnification, and photometric-redshift modelling (Yao et al., 2023). A model-agnostic reconstruction of the three-dimensional power spectrum found that a Planck-consistent reference spectrum requires 20%20\%–30%30\% suppression on non-linear scales to match KiDS-1000, whereas a lower-S8S_8 reference avoids suppression; the authors explicitly noted that this could indicate spurious systematic errors, inaccuracies in the intrinsic-alignment model, or potentially a non-standard cosmological model with delayed structure growth (Simon et al., 6 Feb 2025). At the methodological level, KiDS-1000 has further been used for map-level emulation with conditional generative adversarial networks (Yiu et al., 2021) and for accelerated constraints on Dark Scattering with ReACT and CosmoPower emulators (Carrion et al., 2024).

Taken together, these analyses show that KiDS-1000 is not a single cosmological measurement but a survey framework in which the same imaging and calibrated source sample support multiple estimators of late-time structure growth. A consistent pattern across the literature is that the dominant limitations are no longer purely statistical. Intrinsic alignments, baryonic feedback, redshift calibration, and forward-modelled observational anisotropies recur as the leading modelling issues, and several KiDS-1000 papers present specific methodological responses: improved SOM calibration, physical SED-based source characterization, MetaCalibration, equal-filter-radii compression for third-order statistics, and SBI with catalogue-level systematics (Busch et al., 2022, Halder et al., 3 Feb 2026, Yoon et al., 1 Oct 2025, Burger et al., 2023, Wietersheim-Kramsta et al., 2024).

Definition Search Book Streamline Icon: https://streamlinehq.com
References (20)

Topic to Video (Beta)

No one has generated a video about this topic yet.

Whiteboard

No one has generated a whiteboard explanation for this topic yet.

Follow Topic

Get notified by email when new papers are published related to KiDS-1000.