Global surface temperature
Global surface temperature denotes the spatially averaged temperature of Earth's near-surface air over land and sea surface over oceans, serving as a primary indicator of planetary thermal state in climatology.[1][2] Instrumental records, commencing reliably around 1850, derive from thermometers in weather stations, ship measurements, buoys, and Argo floats, with post-1979 satellite microwave data aiding sea surface components after calibration to in-situ observations.[3][4] Multiple independent datasets, including NASA's GISTEMP, NOAA's GlobalTemp, the UK's HadCRUT, Japan's JMA, and Berkeley Earth's analysis, converge on a long-term warming trend of approximately 1.1 to 1.3 °C from the 1850–1900 baseline to the present, with 2024 registering as the warmest year on record at about 1.55 ± 0.13 °C above pre-industrial levels per ensemble assessments.[5][6][7] This upward trajectory exhibits decadal variability influenced by phenomena such as El Niño-Southern Oscillation, volcanic eruptions, and solar cycles, superimposed on a prevailing linear increase, though recent analyses detect no statistically significant acceleration beyond the post-1970s rate in most series.[8] Uncertainties in these reconstructions stem from sparse early coverage, particularly in the Southern Hemisphere and oceans, station relocations, and homogenization adjustments for non-climatic biases like urban heat islands, with error margins narrowing from ~0.2 °C pre-1900 to ~0.05 °C in recent decades across datasets.[9][10] While peer-reviewed syntheses affirm the robustness of the observed warming, debates persist over the magnitude of adjustments—some amplifying trends—and the integration of proxy data for pre-instrumental context, revealing that current levels, though elevated relative to the 20th century, align within Holocene variability bounds when reconstructed from ice cores and sediments.[4][11] These records underpin attributions linking much of the post-1950 rise to anthropogenic greenhouse gas emissions, tempered by natural forcings, yet underscore the need for ongoing scrutiny of measurement fidelity amid institutional tendencies toward trend-favoring methodologies.[12]Definition and Fundamentals
Conceptual Definition
The global mean surface temperature (GMST), often referred to as global surface air temperature, is the area-weighted average of near-surface air temperatures over continental land masses and sea surface temperatures (SST) over oceanic areas, yielding an integrated metric of the planet's near-surface thermal state.[13] This composite index reflects the thermodynamic temperature at the atmosphere-surface interface, where land components derive from standardized air temperature readings and ocean components from water-skin or shallow-depth measurements, combined in proportion to Earth's surface coverage (roughly 29% land and 71% ocean).[1] GMST is typically analyzed as temporal anomalies relative to a reference period—such as 1850–1900 to approximate pre-industrial conditions or 1951–1980 for instrumental-era baselines—to isolate deviations from historical norms while accounting for incomplete spatial coverage in early records.[14][13] Near-surface air temperature over land is measured at heights of 1.5 to 2 meters above ground level using thermometers enclosed in ventilated shelters, such as Stevenson screens, to shield against direct sunlight, precipitation, and ground conduction while capturing representative atmospheric conditions.[14] SST, by contrast, captures the temperature of the ocean's upper boundary layer—ranging from the infinitesimal skin depth (influenced by evaporative cooling and infrared radiation) to bulk depths of 3–10 meters via ship engine intakes or moored buoys—introducing methodological variances that require harmonization for global averaging.[1] These measurements are aggregated via grid-based interpolation (e.g., 5° × 5° cells), with area weighting applied using latitude-dependent cosine factors to prevent polar overrepresentation, though the resulting GMST remains a statistical abstraction rather than a uniform physical observable, susceptible to undersampling in remote regions like the Arctic.[13][1] Conceptually, GMST serves as a diagnostic proxy for Earth's radiative energy imbalance, where sustained positive anomalies indicate net heat accumulation at the surface, but it does not equate to total planetary heat content, which includes deeper ocean layers and cryospheric storage.[14] Natural forcings like solar irradiance variations and volcanic aerosols, alongside internal variability (e.g., El Niño-Southern Oscillation), superimpose short-term fluctuations on any underlying trend, necessitating multi-decadal averaging—often 30-year periods—for robust trend assessment.[13] Despite its utility, the metric's reliance on heterogeneous observing networks underscores inherent uncertainties in pre-1950 estimates, where station density was sparse and ship-based SST dominated.[1]Measurement Principles and Challenges
Global surface temperature combines near-surface air temperatures over land, measured approximately 2 meters above the ground, with sea surface temperatures representing the upper ocean layer, typically the top 10 meters.[3][15] Land measurements rely on thermometers housed in Stevenson screens, ventilated enclosures with louvered sides that shield instruments from direct sunlight, precipitation, and ground radiation while permitting free airflow to capture representative air temperatures.[16][17] These stations, part of networks like the Global Historical Climatology Network (GHCN), number around 7,000 to 10,000 active sites globally, though coverage varies by region and era.[1] Sea surface temperatures historically derived from ship-based methods, including canvas buckets (pre-1940s, prone to evaporative cooling) and engine intake water (post-1940s, potentially warmer due to depth), have transitioned to moored and drifting buoys since the 1970s for greater consistency.[18][19] To derive a global metric, datasets compute temperature anomalies relative to a baseline period (e.g., 1951–1980), interpolate values onto a latitude-longitude grid (typically 5° × 5° or finer), apply area weighting to account for varying grid cell sizes toward the poles, and average land (30%) and ocean (70%) components.[20][21] Key challenges include sparse spatial coverage, particularly in the Southern Hemisphere, Arctic regions, and remote oceans before the mid-20th century, where data voids necessitate interpolation and amplify uncertainties—estimated at ±0.2°C for 19th-century globals versus ±0.05°C recently.[9][5][22] Measurement inhomogeneities arise from transitions in instrumentation and practices, such as shifts from non-standard exposures to Stevenson screens in the late 19th century or bucket-to-intake SST methods, introducing biases up to 0.3°C that require post-hoc corrections.[23][18] Urban heat island effects, where station proximity to expanding cities elevates readings by 0.1–1°C locally, pose ongoing issues; while adjustments aim to mitigate this, analyses indicate residual contributions to land trends, potentially 20–25% in U.S. records and smaller globally, with debates over adjustment efficacy due to increasing urbanization near stations.[24][25] Additional complications stem from time-of-observation biases, elevation variations, and siting changes (e.g., airport relocations), which homogenization algorithms address but cannot fully eliminate without metadata gaps, leading to divergent trends across datasets like NOAA, NASA GISTEMP, and HadCRUT.[24][26] These factors underscore the reliance on statistical infilling and the inherent limits of surface networks in capturing a truly global, homogeneous signal.[5]Data Sources and Methods
Instrumental Surface Records (1850-Present)
Instrumental surface temperature records comprise in-situ measurements of near-surface air temperatures over land and sea surface temperatures (SST), with systematic global compilation beginning in 1850. These records draw from thermometer readings at weather stations and ship-based observations, initially concentrated in the Northern Hemisphere's mid-latitudes. Coverage in 1850 encompassed approximately 57% of Earth's surface, primarily Europe and North America, expanding to over 75% by later decades through additional stations and maritime routes.[27] Land measurements utilize thermometers housed in standardized shelters, such as Stevenson screens, recording daily maximum and minimum temperatures converted to monthly averages. The Global Historical Climatology Network (GHCN) aggregates data from thousands of stations, with early years featuring fewer than 1,000 reporting sites worldwide, rising to over 2,000 by 1900 and exceeding 5,000 from the 1950s onward. SST data prior to the mid-20th century relied on ships collecting water samples via buckets, subject to potential underestimation from evaporative cooling during hauling, shifting to engine room intake measurements after World War II that could overestimate due to warm seawater exposure.[28][29] Prominent datasets include HadCRUT from the UK Met Office and University of East Anglia, NOAA GlobalTemp, NASA GISTEMP, and Berkeley Earth, each integrating land and ocean components to produce gridded global temperature anomalies relative to baselines like 1850–1900 or 1961–1990. Berkeley Earth, for instance, processes over 1.6 billion temperature reports from multiple archives into a land-ocean product spanning 1850 to present. These records exhibit close agreement on long-term trends despite methodological differences, though early sparse sampling, particularly in the Southern Hemisphere and remote oceans, necessitates interpolation and introduces higher uncertainty in pre-1900 estimates.[30][14][7]Ocean and Buoy Measurements
Ocean surface temperatures, which cover approximately 71% of Earth's surface, are primarily measured using ship-based observations, moored and drifting buoys, and profiling floats such as those from the ARGO array.[31] Historical ship measurements relied on methods like canvas or wooden buckets hauled from depth, uninsulated buckets, and later engine room intakes (ERIs), each introducing systematic biases due to heat exchange with air or hulls.[32] Bucket methods typically recorded cooler temperatures than ERIs because of evaporative cooling during hauling, with differences up to 0.5°C depending on insulation and wind conditions.[33] Modern buoy measurements, including moored buoys from networks like the National Data Buoy Center (NDBC) and drifting buoys, use hull-mounted or subsurface sensors that minimize exposure biases, providing more consistent skin-layer or bulk temperature readings.[34] Collated comparisons show ship observations averaging 0.12°C warmer than nearby buoys during dawn conditions, attributed to residual heat from ship structures or measurement timing.[34] The ARGO program, deploying over 3,900 autonomous profiling floats since 2000, contributes surface data during float ascents, enhancing spatial coverage in remote areas but primarily excels in subsurface profiles up to 2,000 meters.[35] ARGO data have refined estimates of upper-ocean heat content, indicating accelerated warming rates post-2004 at approximately 0.4–0.6 W/m² globally.[35] Datasets such as NOAA's Extended Reconstructed SST (ERSST) and the Hadley Centre's HadSST apply adjustments to homogenize these sources, correcting for method transitions like the shift from buckets to ERIs in the mid-20th century and the increasing buoy fraction since the 1980s.[36] In ERSST version 4, buoy data are adjusted upward by about 0.1°C to align with ERI records, based on paired observations, to mitigate underestimation of warming from the buoy cool bias.[37] However, these adjustments assume ship data as a baseline despite known ship warm biases, potentially amplifying long-term trends; independent buoy-only analyses still confirm post-1980 warming of 0.7–0.8°C in adjusted records.[38] Recent evaluations highlight uncertainties in pre-1940 data, with evidence of a cold bias in early ship SSTs leading to underestimated early-20th-century warming by up to 0.2°C after statistical corrections.[31] The transition to buoy dominance, now comprising over 50% of observations in some regions, improves data quality by reducing method-specific errors but introduces challenges in blending with sparser historical ship data, affecting global surface temperature reconstructions.[39] Peer-reviewed assessments affirm that post-adjustment SST trends from 1979–2015 align across datasets at 0.08–0.10°C per decade, with buoys and ARGO supporting robust ocean warming signals amid natural variability like El Niño.[40][38]Satellite and Upper-Air Observations
Satellite-based measurements of atmospheric temperature utilize microwave radiometers, such as the Microwave Sounding Unit (MSU) and Advanced MSU (AMSU), deployed on NOAA polar-orbiting satellites starting with NOAA-4 in December 1978. These instruments detect thermal emissions from atmospheric oxygen at specific frequencies, enabling the derivation of brightness temperatures for distinct vertical layers: the lower troposphere (TLT, weighted from near-surface to approximately 8 km altitude), mid-troposphere (MTT), and tropospheric trends (TTT). The TLT product serves as a bulk indicator of lower atmospheric warming, influenced by surface conditions but extending aloft, and requires corrections for orbital decay, sensor drift, and diurnal sampling biases.[41][42] Principal datasets include the University of Alabama in Huntsville (UAH) version 6.0/6.1, processed by Roy Spencer and John Christy, and Remote Sensing Systems (RSS) version 4.0, with a third from the Center for Satellite Applications and Research (STAR). From January 1979 to December 2024, UAH reports a global TLT trend of +0.15 °C per decade (+0.22 °C per decade over land, +0.14 °C per decade over oceans). RSS estimates a steeper trend of approximately +0.21 °C per decade over the same period. These rates are derived after merging data across satellite platforms and applying empirical adjustments, though methodological differences—such as handling of NOAA-14 instrument degradation and tropical hot spot amplification—contribute to variances between datasets.[43][41][44] Comparisons with surface records reveal tropospheric trends generally lower than 2-meter air temperature increases (around +0.18 °C per decade in datasets like NOAA or HadCRUT5), especially in the tropics where greenhouse forcing predicts enhanced warming aloft due to moist convection. This discrepancy persists in UAH data, showing near-zero tropical TLT trends in some analyses (+0.08 °C per decade to 2004), while RSS aligns more closely with surface amplification. Independent validations, including reanalyses like ERA5, support satellite warming signals but highlight uncertainties in pre-1990s data from satellite handover artifacts.[45][46] Upper-air observations complement satellites via radiosondes—instruments attached to weather balloons that ascend to 30-40 km, measuring temperature, pressure, and humidity at high vertical resolution. The Integrated Global Radiosonde Archive (IGRA) compiles data from over 2,800 stations worldwide, with consistent coverage since the 1950s, though global averages emphasize Northern Hemisphere land biases. Homogenization is essential to correct for instrument shifts (e.g., from carbon hypsometers to thermocouples in the 1990s), time-of-day changes, and urban effects, as raw data often exhibit spurious cooling trends from sensor upgrades.[47][48] Homogenized radiosonde datasets, such as RAOBCORE and RICH, indicate global tropospheric warming of +0.1 to +0.2 °C per decade since 1979, broadly consistent with satellite TLT after adjustments. For instance, lower tropospheric trends from select tropical stations average +0.08 to +0.15 °C per decade to 2004, aligning with UAH but below RSS. Stratospheric cooling (-0.3 °C per decade) contrasts with tropospheric warming, as expected from ozone depletion and greenhouse gas stratification. Critics, including analyses of adjustment methodologies, note that corrections frequently amplify post-1979 warming without fully independent metadata verification, potentially mirroring surface record issues where algorithmic homogeneity assumes error patterns favoring recent increases.[45][49][42] Integration of satellite and radiosonde data enhances global coverage, particularly over oceans and remote regions where surface stations are sparse, but vertical weighting differences preclude direct equivalence to surface air metrics. Radiosondes validate satellite layer assignments, with post-2000 convergence in trends amid improved instrumentation, though pre-satellite radiosonde records (1958-1979) show muted warming (+0.3 °C globally in 850-300 hPa layer to 1987), underscoring adjustment sensitivities. Ongoing efforts, like merged reanalyses, aim to resolve residual uncertainties for attribution studies.[50][51]Proxy-Based Reconstructions
Proxy-based reconstructions estimate past surface temperatures using indirect environmental indicators preserved in geological and biological archives, extending records beyond the instrumental era beginning around 1850. These proxies include tree-ring widths and maximum latewood density for temperature-sensitive growth in boreal forests, oxygen-18 isotope ratios in ice cores reflecting air temperature at precipitation sites, Mg/Ca ratios in foraminiferal shells from ocean sediments indicating sea surface temperatures, and borehole thermometry from heat diffusion in continental crust.[52][53] Each proxy type responds to temperature with varying sensitivity, seasonal bias, and potential confounding influences from precipitation or CO2 fertilization, necessitating calibration against overlapping instrumental data via transfer functions derived from regression or machine learning approaches.[54][55] Statistical methods aggregate multiple proxies into hemispheric or global means, often employing principal component analysis to identify common signals amid noise, or optimal information extraction techniques to weight proxies by reliability. Multi-proxy ensembles, such as those from the PAGES 2k Consortium's database of 692 records across 648 sites covering continents and oceans, aim to enhance robustness by diversifying sources and reducing type-specific errors.[56][57] Reconstructions for the past two millennia typically show Northern Hemisphere temperatures varying within ±0.5°C of the 1850-1900 baseline during the Medieval Warm Period (roughly 950-1250 CE) and cooling during the Little Ice Age (1450-1850 CE), with 20th-century warming exceeding these fluctuations by 0.6-1.0°C in median estimates.[53][58] However, regional proxy evidence indicates the Medieval Warm Period approached or matched modern warmth in parts of the North Atlantic and Asia, challenging claims of global uniformity in pre-industrial variability.[59] A persistent challenge is the divergence problem in dendrochronological proxies, where tree-ring indices from high-latitude sites fail to capture post-1960 instrumental warming despite tracking earlier 20th-century rises, potentially stemming from increased winter warming, drought stress, or elevated CO2 suppressing density signals.[60] This discrepancy has prompted truncation of calibration periods or exclusion of divergent series in some reconstructions, raising concerns over inflated pre-calibration variability or underestimated modern anomalies; critics argue it undermines the reliability of tree rings as centennial-scale thermometers without independent validation.[61][62] Spatial coverage remains skewed toward the Northern Hemisphere and land areas, with sparse Southern Hemisphere and ocean data amplifying uncertainties in global means, estimated at ±0.2-0.5°C for millennium-scale reconstructions.[56] Methodological sensitivities, including proxy selection and network density, can alter amplitude of past variations by up to 50%, as demonstrated in ensemble tests.[63] Over the Holocene epoch (last 11,700 years), proxy syntheses reveal a Holocene Thermal Maximum peaking 9,000-5,000 years ago, with Northern Hemisphere summer temperatures 0.5-2°C above late 20th-century levels in continental interiors, driven by orbital forcing and ice-sheet retreat, followed by Neoglacial cooling toward the Little Ice Age.[64][65] Global annual mean reconstructions using data assimilation of pollen, chironomid, and non-marine proxies indicate peak warmth regionally asynchronous, with mid-Holocene averages comparable to or exceeding recent decades in Eurasia and North America, though Southern Hemisphere trends differ due to Antarctic influences.[66][67] Uncertainties in Holocene records stem from proxy-specific seasonal biases—e.g., tree rings favoring summer, ice cores winter—and dating imprecision exceeding 100 years in sediments, complicating precise global synchrony assessments.[68] Despite advances in multiproxy integration, debates persist over whether current warming rates or levels are unprecedented, given evidence of comparable Holocene excursions under natural forcings alone.[69]Data Processing and Potential Biases
Homogenization Techniques
Homogenization techniques detect and adjust for artificial breaks in station temperature records caused by non-climatic factors, such as instrument replacements, station relocations, or changes in observation times, to ensure the series reflects climatic variability alone. These methods generally rely on relative comparisons with neighboring stations, assuming spatial coherence in true climate signals, and proceed in two phases: changepoint detection via statistical tests on difference series and adjustment by shifting segments to align with references. Absolute methods, referencing theoretical standards, are less common for surface air temperatures due to sparse historical baselines.[70] The pairwise homogenization algorithm (PHA) exemplifies relative methods, constructing difference time series between a candidate station and multiple nearby references, then identifying breaks where the mean difference shifts significantly using a two-phase regression test. Potential breaks are ranked by a quality metric balancing detection power and spatial coherence, with adjustments applied iteratively from largest to smallest to avoid propagating errors; this automation handles large networks efficiently, as validated in benchmark tests reducing trend biases under simulated inhomogeneities. NOAA applies PHA to monthly data in the U.S. Historical Climatology Network (USHCN version 2.5, last updated May 2008) and upstream for Global Historical Climatology Network (GHCN) contributions.[71][72][73] The Standard Normal Homogeneity Test (SNHT) provides an alternative detection tool, normalizing the series to zero mean and unit variance, then computing a test statistic from the cumulative sum of deviations to flag shifts, with critical values adjusted for sample size and break location (e.g., central breaks yield higher power). SNHT excels for additive offsets but assumes Gaussian residuals and may miss gradual changes; it has been used in European networks like the Centenary dataset for monthly means from 1901–1990.[74][75] Berkeley Earth integrates homogenization into a Bayesian framework, fitting each station's series with piecewise linear trends and changepoints estimated via kriging against hundreds of neighbors, simultaneously modeling the regional climate signal to derive spatially informed adjustments without sequential break processing. This approach, detailed in their averaging process, processes over 39,000 stations and yields trends aligning closely with raw data after corrections.[76] NASA's GISTEMP relies on NOAA's pre-homogenized GHCN and USHCN land data without further station-level corrections, instead smoothing anomalies over 1200 km radii and applying fixed urban-rural offsets derived from pairwise rural comparisons to mitigate site-specific biases. HadCRUT5, combining CRUTEM5 land and HadSST4 sea data, incorporates homogenization performed by national services prior to submission, with central steps emphasizing infilling via kriging and ensemble uncertainty rather than raw adjustments.[77]Adjustments for Station Issues and Urban Heat
Adjustments for non-climatic station issues, such as relocations, instrumentation changes, and variations in observation timing, are applied through homogenization algorithms that detect and correct discontinuities in temperature records. The NOAA's Pairwise Homogenization Algorithm (PHA), used in the Global Historical Climatology Network monthly (GHCNm) dataset, identifies abrupt shifts by pairwise comparisons between a target station and highly correlated neighbors, assuming shared regional climate signals while isolating local artifacts.[78] Detected breaks are adjusted to minimize differences across the network, with the algorithm favoring corrections that align earlier periods to the most recent instrumentation and location standards; for instance, in GHCNm version 3.2.0, this process reduced artificial cooling biases from undocumented changes.[79] Similar techniques in the U.S. Historical Climatology Network (USHCN) address time-of-observation biases, which can inflate daily maximum temperatures by up to 0.3°C if readings shift from morning to afternoon without correction.[73] Urban heat island (UHI) effects, arising from localized warming due to impervious surfaces, reduced vegetation, and waste heat in populated areas, introduce positive biases estimated at 0.1–1.0°C in urban stations relative to rural ones.[25] Homogenization partially mitigates UHI by regressing urban records toward rural neighbors, as urban discontinuities manifest as step changes detectable against less-affected stations. Berkeley Earth's methodology explicitly models UHI by classifying stations via nighttime lights data and satellite-derived population density, reconstructing global land trends with rural subsets that yield warming rates within 0.01°C per decade of full-network results, indicating UHI contributes negligibly (<0.05°C total since 1950) to global averages due to the dominance of rural and oceanic coverage.[80][81] Critics contend that homogenization can inadvertently blend UHI signals into rural records when urban stations comprise a significant fraction of neighbors, a phenomenon termed "urban blending" observed in analyses of over 800 European GHCN stations where post-adjustment rural trends aligned more closely with urban ones than raw data suggested.[82] Independent validations, however, demonstrate convergence between adjusted land trends and satellite-derived lower-troposphere records, with U.S. rural stations post-1895 showing summer warming of 0.7–1.0°C after UHI corrections, comparable to national averages.[25] These adjustments, while peer-reviewed and iteratively refined, remain debated for their reliance on statistical assumptions over metadata, with fewer than 20% of NOAA shifts corresponding to documented station events in some audits.[83]Criticisms of Adjustment Methodologies
Critics contend that homogenization and adjustment procedures applied to surface temperature records, intended to correct for non-climatic inhomogeneities such as station relocations, instrument changes, and time-of-observation biases, often fail to eliminate systematic errors or introduce new ones that exaggerate warming trends. These methods, employed by datasets like NOAA's Global Historical Climatology Network (GHCN) and NASA's GISS, rely on algorithms such as pairwise comparison, which assume nearby stations share similar climatic signals; however, when neighboring stations are influenced by correlated anthropogenic factors like urbanization, corrections can propagate biases across the network.[82] A prominent critique focuses on inadequate mitigation of the urban heat island (UHI) effect, where urban development artificially elevates local temperatures. In their 2007 analysis of gridded global land temperatures from 1979 to 2002, Ross McKitrick and Patrick Michaels regressed temperature anomalies against socioeconomic variables proxying for development (e.g., GDP per capita and energy consumption); they found these factors accounted for roughly 50% of the observed warming trend after adjustments, indicating residual contamination not addressed by standard homogenization.[84] This suggests that adjustments, by blending urban and rural records without sufficient isolation of UHI signals, underestimate the non-climatic contribution to reported trends.[85] Further evidence of methodological flaws emerges from examinations of the "urban blending" phenomenon in homogenization. A 2023 study by Ronan Connolly and colleagues applied synthetic data tests to U.S. and global records, demonstrating that pairwise algorithms inadvertently transfer UHI-induced warming from urban breakpoints to rural stations during offset calculations, creating spurious upward trends in homogenized series by up to 0.2–0.5°C per century in affected regions.[82] The authors argued this arises because algorithms prioritize spatial proximity over land-use differences, failing to distinguish climatic from developmental breaks, and recommended incorporating explicit UHI diagnostics absent in current NOAA and Hadley Centre procedures.[82] Station siting quality represents another vector of criticism, with many legacy networks featuring sensors in suboptimal locations (e.g., near asphalt, air conditioning exhausts, or buildings), introducing warm biases estimated at 0.1–1.0°C locally. Evaluations of the U.S. Historical Climatology Network (USHCN) revealed that over 80% of stations prior to 2010 rated poorly (Class 3–5) under NOAA's own siting metadata, yet adjustments primarily target temporal breaks rather than spatial exposure issues. A 2021 assessment concluded that this uncorrected siting bias inflates U.S. temperature estimates by approximately 0.3–0.5°C since 1950, with homogenization exacerbating rather than resolving the discrepancy when compared to pristine networks like the U.S. Climate Reference Network (USCRN), which records lower warming rates post-2005.[86] Critics, including those analyzing USHCN version upgrades, note that iterative adjustments have progressively cooled early-20th-century U.S. records (e.g., by 0.4°F in maximum temperatures from 1930–2000), enhancing apparent century-scale warming by up to 40% relative to raw data in some analyses, though agencies maintain such changes align with independent validations.[86][87] These concerns are compounded by limited independent auditing and reliance on automated algorithms without routine raw-versus-adjusted trend comparisons against unadjusted high-quality subsets. While proponents cite pairwise method benchmarks showing reduced error variance, detractors argue real-world tests against reference networks like USCRN reveal persistent divergences, underscoring the need for greater transparency in adjustment metadata and breakpoint rationales to verify claims of trend enhancement.[88]Observed Instrumental Trends
Overall Temperature Anomalies Since 1850
Instrumental records of global surface temperature anomalies, primarily from land stations and sea surface temperatures, begin reliably around 1850 and reveal a net warming of about 1.5°C from the 1850-1900 baseline to 2024.[5] This estimate draws from independent analyses by NASA GISTEMP, HadCRUT, NOAA GlobalTemp, and Berkeley Earth, which converge on a linear trend of roughly 0.06-0.08°C per decade over the full period, with acceleration to 0.18-0.20°C per decade since 1970.[1][29][89] Early 20th-century anomalies fluctuated around -0.3 to -0.4°C relative to pre-industrial levels, influenced by natural variability including volcanic eruptions and ocean cycles, before rising steadily post-1950.[77] Spatial coverage was sparse before 1900, concentrated in Europe and North America, leading to larger uncertainties estimated at ±0.1-0.2°C for global means in that era.[77] Datasets employ infilling techniques for unsampled regions, such as oceans and polar areas, which constitute over 70% of Earth's surface; these methods assume anomalies propagate from observed points, introducing potential smoothing of extremes.[90] Despite variations in processing—NASA and NOAA apply homogenization for station relocations and urban effects, while Berkeley Earth uses more stations with minimal adjustments—the post-1950 trends align closely, supporting a robust signal of multi-decadal warming.[30] Annual anomalies reached +1.54°C above 1850-1900 in 2023, the first year to surpass the 1.5°C threshold, with 2024 exceeding it further.[91][5] Decadal smoothing highlights the progression: 1850-1900 averaged near zero by definition, 1901-1950 showed minimal net change (~+0.1°C), 1951-2000 warmed by ~0.5°C, and 2001-2020 by another ~0.5°C, culminating in the 2020s as the warmest decade.[92] Interannual variability, driven by El Niño-Southern Oscillation, overlays the trend, with peaks in 1998, 2016, and 2023-2024 amplifying recent highs.[1] Uncertainties diminish over time with denser networks, but pre-1880 estimates carry higher error bars due to fewer than 1,000 stations globally by 1850.[29] Cross-validation with independent buoy and ship data post-1980 confirms the upward trajectory without systematic divergence.[89]Recent Developments (Post-2000 Including 2024-2025)
Global surface temperature anomalies have risen steadily since 2000, with major datasets reporting an average increase of approximately 0.18°C per decade from 2000 to 2020, accelerating in the 2010s and early 2020s.[93] [94] This trend reflects contributions from greenhouse gas forcings amid natural variability, including El Niño-Southern Oscillation (ENSO) cycles. The 1998-2013 interval exhibited slower surface warming rates (about 0.05°C per decade in some records), prompting debate over a potential "hiatus" in the rate of rise relative to prior decades; analyses attributed this to internal variability such as enhanced ocean heat uptake in deeper layers rather than a cessation of anthropogenic warming, with subsequent data confirming resumed acceleration.[95] [96] [97] The decade 2015-2024 marked the warmest on record across datasets from NASA, NOAA, HadCRUT, Berkeley Earth, and Copernicus, surpassing the 2011-2020 period by 0.2-0.3°C on average.[6] [98] Eighteen of the nineteen warmest years since 1850 have occurred since 2000, with 2024 confirmed as the hottest globally at 1.28°C above the 1951-1980 baseline (NASA) or approximately 1.55°C above pre-industrial levels (WMO aggregate).[99] [100] [6] This record was influenced by a strong El Niño event peaking in 2023-2024, which amplified surface temperatures, alongside persistent anthropogenic forcings; 24% of Earth's surface set local annual records in 2024 per Berkeley Earth.[5] [101] In 2025, through August, global anomalies ranked second-warmest on record behind 2024, with January-March averaging among the highest starts to a year.[102] [103] April 2025 recorded 1.49°C above the 1850-1900 baseline (Berkeley Earth), the second-highest for that month.[104] Transition to La Niña conditions by mid-2025 has moderated extremes, yet anomalies remain elevated above the 1991-2020 average across 91% of the globe as of early 2025.[101] These developments align across independent datasets, though discrepancies arise from differing baseline periods and adjustment protocols; for instance, Copernicus ERA5 reanalysis shows 2024 at 1.6°C above 1850-1900, slightly higher than land-ocean index estimates.[105]| Year | Anomaly (°C, relative to 1850-1900, Berkeley Earth example) | Notes |
|---|---|---|
| 2024 | ~1.62 | Warmest on record; El Niño peak |
| 2023 | ~1.54 | Second-warmest; record prior to 2024 |
| 2020 | ~1.02 | Third-warmest pre-2023 streak |
| 2025 (Jan-Aug) | ~1.50 (preliminary) | Second to 2024; La Niña onset |
Warmest Years, Decades, and Periods
According to multiple independent datasets, including those from NOAA, NASA, and Berkeley Earth, 2024 was the warmest year on record since instrumental measurements began in 1850, with a global surface temperature anomaly of approximately 1.29°C above the 20th-century average.[106][5][107] This surpassed the previous record set in 2023, which had an anomaly of about 1.17°C.[106] The third warmest year was 2016, influenced by a strong El Niño event.[93] The ten warmest years in the instrumental record have all occurred within the last decade (2015–2024), reflecting accelerated warming in recent times.[93] This ranking holds across major datasets such as NOAA's GlobalTemp, NASA's GISTEMP, and Berkeley Earth's global temperature series, despite minor variations in exact anomalies due to differences in baseline periods and coverage.[1][5] For instance, NOAA reports the ranking as 2024, 2023, 2020, 2016, 2022, 2019, 2015, 2017, 2010, and 2005.[106]| Rank | Year | Anomaly (°C, relative to 20th-century average, NOAA) |
|---|---|---|
| 1 | 2024 | +1.29 |
| 2 | 2023 | +1.17 |
| 3 | 2020 | +1.02 |
| 4 | 2016 | +0.99 |
| 5 | 2022 | +0.89 |
| 6 | 2019 | +0.86 |
| 7 | 2015 | +0.84 |
| 8 | 2017 | +0.83 |
| 9 | 2010 | +0.72 |
| 10 | 2005 | +0.68 [106] |
Influencing Factors and Attribution
Natural Climate Variability
Natural climate variability encompasses fluctuations in global surface temperature arising from internal dynamics of the climate system, such as ocean-atmosphere interactions, and external natural forcings like variations in solar irradiance and volcanic aerosol emissions.[112] These processes generate short-term to multidecadal variations superimposed on longer-term trends.[113] The El Niño-Southern Oscillation (ENSO) represents a primary mode of interannual variability, with El Niño phases typically elevating global mean surface temperatures by 0.1 to 0.25 °C relative to neutral conditions, while La Niña phases produce cooling of similar magnitude.[114] For instance, the strong 2023 El Niño contributed to a rapid 0.29 ± 0.04 K rise in global-mean surface temperature from 2022 to 2023.[115] ENSO influences arise from altered heat redistribution in the Pacific Ocean, affecting atmospheric circulation and global energy balance.[116] On decadal to multidecadal timescales, oscillations such as the Pacific Decadal Oscillation (PDO) and Atlantic Multidecadal Oscillation (AMO) drive low-frequency variations in sea surface temperatures that partially imprint on global averages.[113] The AMO, characterized by North Atlantic SST anomalies, and PDO, involving Pacific basin-wide patterns, exhibit phases lasting 20-60 years, with positive phases associated with hemispheric warming influences that can contribute up to 0.1-0.2 °C to global temperature anomalies during their peaks.[117] These modes reflect internal ocean dynamics and do not exhibit a net trend over the instrumental period, unlike the observed century-scale warming.[118] Solar variability, primarily through the 11-year sunspot cycle, modulates total solar irradiance by approximately 1 W/m², corresponding to global temperature perturbations of about 0.1 °C.[119] However, reconstructions indicate that solar forcing has been stable or slightly declining since the mid-20th century, exerting minimal influence on recent warming.[120] Volcanic eruptions provide episodic cooling through stratospheric sulfate aerosols that reflect sunlight, with major events like the 1991 Mount Pinatubo eruption inducing a global temperature drop of roughly 0.5 °C persisting for 1-3 years.[121] Such effects are transient, as aerosols settle out within a few years, and do not offset multidecadal trends.[122]| Variability Mode | Timescale | Approximate Global Temperature Impact |
|---|---|---|
| ENSO | Interannual | ±0.1–0.3 °C[115][114] |
| PDO/AMO | Decadal–Multidecadal | ~0.1–0.2 °C phase-dependent[117] |
| Solar Cycle | ~11 years | ~0.1 °C[119] |
| Major Volcanic Eruption | Episodic (1–3 years) | –0.5 °C[121] |