Editing IPCC:AR6/WGI/Chapter-1 (section)

=== 1.5.2 New Developments in Reanalyses ===

<div id="h2-28-siblings" class="h2-siblings"></div>

Reanalyses are usually the output of a model (e.g., a numerical weather prediction model) constrained by observations using data assimilation techniques, but the term has also been used to describe observation-based datasets produced using simpler statistical methods and models (Annex I: Observational Products). This section focuses on the model-based methods and their recent developments.

Reanalyses complement datasets of observations in describing changes through the historical record and are sometimes considered as ‘maps without gaps’ because they provide gridded output in space and time, often global, with physical consistency across variables on sub-daily time scales, and information about sparsely observed variables (such as evaporation; [[#Hersbach--2020|Hersbach et al., 2020]] ). They can be globally complete, or regionally focussed and constrained by boundary conditions from a global reanalysis (Section 10.2.1.2). They can also provide feedback about the quality of the observations assimilated, including estimates of biases and critical gaps for some observing systems.

Many early reanalyses are described in Box 2.3 of [[#Hartmann--2013|Hartmann et al. (2013)]] . These were often limited by the underlying model, the data assimilation schemes and observational issues ( [[#Thorne--2010|Thorne and Vose, 2010]] ; [[#Zhou--2018|Zhou et al., 2018]] ). Observational issues include the lack of underlying observations in some regions, changes in the observational systems over time (e.g., spatial coverage, introduction of satellite data), and time-dependent errors in the underlying observations or in the boundary conditions, which may lead to stepwise biases in time. The assimilation of sparse or inconsistent observations can introduce mass or energy imbalances ( [[#Valdivieso--2017|Valdivieso et al., 2017]] ; [[#Trenberth--2019|Trenberth et al., 2019]] ). Further limitations and some efforts to reduce the implications of these observational issues are detailed below.

The methods used in the development of reanalyses have progressed since AR5 and, in some cases, this has important implications for the information they provide on how the climate is changing. [[IPCC:Wg1:Chapter:Annex-i|Annex I]] includes a list of reanalysis datasets used in AR6. Recent major developments in reanalyses include the assimilation of a wider range of observations, higher spatial and temporal resolution, extensions further back in time, and greater efforts to minimize the influence of a temporally varying observational network.

<div id="1.5.2.1" class="h3-container"></div>

<span id="atmospheric-reanalyses"></span>
==== 1.5.2.1 Atmospheric Reanalyses ====

<div id="h3-24-siblings" class="h3-siblings"></div>

Extensive improvements have been made in global atmospheric reanalyses since AR5. The growing demand for high-resolution data has led to the development of higher-resolution atmospheric reanalyses, such as the Modern-Era Retrospective Analysis for Research and Applications, version 2 (MERRA-2; [[#Gelaro--2017|Gelaro et al., 2017]] ) and ERA5 ( [[#Hersbach--2020|Hersbach et al., 2020]] ). There is a focus on ERA5 here because it has been assessed as of high enough quality to present temperature trends alongside more traditional observational datasets ( [[IPCC:Wg1:Chapter:Chapter-2#2.3.1.1|Section 2.3.1.1]] ) and is also used in the Interactive Atlas.

Atmospheric reanalyses that were assessed in AR5 are still being used in the literature, and results from ERA-Interim (about 80 km resolution, production stopped in August 2019; [[#Dee--2011|Dee et al., 2011]] ), the Japanese 55-year Reanalysis (JRA-55; [[#Ebita--2011|Ebita et al., 2011]] ; [[#Kobayashi--2015|Kobayashi et al., 2015]] ; [[#Harada--2016|Harada et al., 2016]] ) and Climate Forecast System Reanalysis (CFSR; [[#Saha--2010|Saha et al., 2010]] ) are assessed in AR6. Some studies still also use the NCEP/NCAR reanalysis, particularly because it extends back to 1948 and is updated in near-real time ( [[#Kistler--2001|Kistler et al., 2001]] ). Older reanalyses have a number of limitations, which have to be accounted for when assessing the results of any study that uses them.

ERA5 provides hourly atmospheric fields at about 31 km resolution on 137 levels in the vertical, as well as land-surface variables and ocean waves. It is available from 1979 onwards and is updated in near-real time, with plans to extend back to 1950. A 10-member ensemble is also available at coarser resolution, allowing uncertainty estimates to be provided (e.g., [[IPCC:Wg1:Chapter:Chapter-2#2.3|Section 2.3]] ). MERRA-2 includes many updates from the earlier version, including the assimilation of aerosol observations, several improvements to the representation of the stratosphere, including ozone, and improved representations of cryospheric processes. All of these improvements increase the usefulness of these reanalyses (Section 7.3; [[#Hoffmann--2019|Hoffmann et al., 2019]] ).

Models of atmospheric composition and emissions sources and sinks allow the forecast and reanalysis of constituents such as O <sub>3</sub> , carbon monoxide (CO), nitrogen oxides (NOx) and aerosols. The Copernicus Atmosphere Monitoring Service (CAMS) reanalysis shows improvement against earlier atmospheric composition reanalyses, giving greater confidence for its use to study trends and evaluate models (Section 7.3; e.g., [[#Inness--2019|Inness et al., 2019]] ).

The intercomparison of reanalyses with each other, or with earlier versions, is often done for particular variables or aspects of the simulation. ERA5 is assessed as the most reliable reanalysis for climate trend assessment ( [[IPCC:Wg1:Chapter:Chapter-2#2.3|Section 2.3]] ). Compared to ERA-Interim, the ERA5 forecast model and assimilation system, as well as the availability of improved reprocessing of observations, resulted in relatively smaller errors when compared to observations, including a better representation of global energy budgets, radiative forcing from volcanic eruptions (e.g., Mt. Pinatubo: [[#Allan--2020|Allan et al., 2020]] ), the partitioning of surface energy ( [[#Martens--2020|Martens et al., 2020]] ), and wind ( [[#Kaiser-Weiss--2015|Kaiser-Weiss et al., 2015]] , 2019; [[#Borsche--2016|Borsche et al., 2016]] ; [[#Scherrer--2020|Scherrer, 2020]] ). In ERA5, higher resolution means a better representation of Lagrangian motion convective updrafts, gravity waves, tropical cyclones, and other meso- to synoptic-scale features of the atmosphere ( [[#Hoffmann--2019|Hoffmann et al., 2019]] ; [[#Martens--2020|Martens et al., 2020]] ). Low-frequency variability is found to be generally well represented and, from 10 hPa downwards, patterns of anomalies in temperature match those from the ERA-Interim, MERRA-2 and JRA-55 reanalyses. Inhomogeneities in the water cycle have also been reduced ( [[#Hersbach--2020|Hersbach et al., 2020]] ).

Precipitation is not usually assimilated in reanalyses and, depending on the region, reanalysis precipitation can differ from observations by more than the observational error ( [[#Zhou--2017|Zhou and Wang, 2017]] ; [[#Sun--2018|Sun et al., 2018]] ; [[#Alexander--2020|Alexander et al., 2020]] ; [[#Bador--2020|Bador et al., 2020]] ), although these studies did not include ERA5. Assimilation of radiance observations from microwave imagers which, over ice-free ocean surfaces, improve the analysis of lower-tropospheric humidity, cloud liquid water and ocean-surface wind speed have resulted in improved precipitation outputs in ERA5 ( [[#Hersbach--2020|Hersbach et al., 2020]] ). Global averages of other fields, particularly temperature, from ERA-Interim and JRA-55 reanalyses continue to be consistent over the last 20 years with surface observational data sets that include the polar regions ( [[#Simmons--2015|Simmons and Poli, 2015]] ), although biases in precipitation and radiation can influence temperatures regionally ( [[#Zhou--2018|Zhou et al., 2018]] ). The global average surface temperature from MERRA-2 is far cooler in recent years than temperatures derived from ERA-Interim and JRA-55, which may be due to the assimilation of aerosols and their interactions ( [[IPCC:Wg1:Chapter:Chapter-2#2.3|Section 2.3]] ).

A number of regional atmospheric reanalyses (Section 10.2.1.2) have been developed, such as COSMO-REA ( [[#Wahl--2017|Wahl et al., 2017]] ), and the Australian Bureau of Meteorology Atmospheric high-resolution Regional Reanalysis for Australia (BARRA; [[#Su--2019|Su et al., 2019]] ). Regional reanalyses can add value to global reanalyses due to the lower computational requirements, and can allow multiple numerical weather prediction models to be tested (e.g., [[#Kaiser-Weiss--2019|Kaiser-Weiss et al., 2019]] ). There is some evidence that these higher-resolution reanalyses better capture precipitation variability than global lower-resolution reanalyses ( [[#Jermey--2016|Jermey and Renshaw, 2016]] ; [[#Cui--2017|Cui et al., 2017]] ). They are further assessed in Section 10.2.1.2 and used in the Interactive Atlas.

In summary, the improvements in atmospheric reanalyses, and the greater number of years since the routine ingestion of satellite data began, relative to AR5, mean that there is increased confidence in using atmospheric reanalysis products alongside more standard observation-based datasets in AR6 ( ''hi'' ''gh confidence'' ).

<div id="1.5.2.2" class="h3-container"></div>

<span id="sparse-input-reanalyses-of-the-instrumental-era"></span>
==== 1.5.2.2 Sparse Input Reanalyses of the Instrumental Era ====

<div id="h3-25-siblings" class="h3-siblings"></div>

Although reanalyses such as ERA5 take advantage of new observational datasets and present a great improvement in atmospheric reanalyses, the issues introduced by the evolving observational network remain. Sparse input reanalyses, where only a limited set of reliable and long-observed records are assimilated, address these issues, with the limitation of fewer observational constraints. These efforts are sometimes called centennial-scale reanalyses. One example is the atmospheric 20th century Reanalysis ( [[#Compo--2011|Compo et al., 2011]] ; [[#Slivinski--2021|Slivinski et al., 2021]] ) which assimilates only surface and sea-level pressure observations, and is constrained by time-varying observed changes in atmospheric constituents, prescribed sea surface temperatures and sea ice concentration, creating a reconstruction of the weather over the whole globe every three hours for the period 1806–2015. The ERA-20C atmospheric reanalysis (covering 1900–2010; [[#Poli--2016|Poli et al., 2016]] ) also assimilates marine wind observations, and CERA-20C is a centennial-scale reanalysis that assimilates both atmospheric and oceanic observations for the 1901–2010 period ( [[#Laloyaux--2018|Laloyaux et al., 2018]] ). These centennial-scale reanalyses are often run as ensembles that provide an estimate of the uncertainty in the simulated variables over space and time. [[#Slivinski--2021|Slivinski et al. (2021)]] conclude that the uncertainties in surface circulation fields in version 3 of the 20th century Reanalysis are reliable and that there is also skill in its tropospheric reconstruction over the 20th century. Long-term changes in other variables, such as precipitation, also agree well with direct observation-based datasets (Sections 2.3.1.3 and 8.3.2.8).

<div id="1.5.2.3" class="h3-container"></div>

<span id="ocean-reanalyses"></span>
==== 1.5.2.3 Ocean Reanalyses ====

<div id="h3-26-siblings" class="h3-siblings"></div>

Since AR5, ocean reanalyses have improved due to: increased model resolution ( [[#Zuo--2017|Zuo et al., 2017]] ; [[#Lellouche--2018|Lellouche et al., 2018]] ; [[#Heimbach--2019|Heimbach et al., 2019]] ); improved physics ( [[#Storto--2019|Storto et al., 2019]] ); improvements in the atmospheric forcing from atmospheric reanalyses (see [[#1.5.2.1.3|Section 1.5.2.1.3]] ); and improvements in the data quantity and quality available for assimilation (e.g., [[#Lellouche--2018|Lellouche et al., 2018]] ; [[#Heimbach--2019|Heimbach et al., 2019]] ), particularly due to Argo observations (Annex I; [[#Zuo--2019|Zuo et al., 2019]] ).

The first Ocean Reanalyses Intercomparison project (ORA-IP; [[#Balmaseda--2015|Balmaseda et al., 2015]] ) focussed on the uncertainty in key climate indicators, such as ocean heat content ( [[#Palmer--2017|Palmer et al., 2017]] ), thermosteric sea level ( [[#Storto--2017|Storto et al., 2017]] , 2019), salinity ( [[#Shi--2017|Shi et al., 2017]] ), sea ice extent ( [[#Chevallier--2017|Chevallier et al., 2017]] ), and the AMOC ( [[#Karspeck--2017|Karspeck et al., 2017]] ). Reanalysis uncertainties occur in areas of inhomogeneous or sparse observational data sampling, such as for the deep ocean, the Southern Ocean, and western boundary currents ( [[#Lellouche--2018|Lellouche et al., 2018]] ; [[#Storto--2019|Storto et al., 2019]] ). Intercomparisons have also been dedicated to specific variables such as mixed-layer depths ( [[#Toyoda--2017|Toyoda et al., 2017]] ), eddy kinetic energy, globally ( [[#Masina--2017|Masina et al., 2017]] ) and in the polar regions ( [[#Uotila--2019|Uotila et al., 2019]] ). [[#Karspeck--2017|Karspeck et al. (2017)]] found disagreement in the AMOC variability and strength in reanalyses over observation-sparse periods, whereas [[#Jackson--2019|Jackson et al. (2019)]] reported a lower spread in AMOC strength across an ensemble of ocean reanalyses of the recent period (1993–2010), linked to improved observation availability for assimilation. Reanalyses also have a larger spread of ocean heat uptake than data-only products and can produce spurious overestimates of heat uptake ( [[#Palmer--2017|Palmer et al., 2017]] ), which is important in the context of estimating climate sensitivity ( [[#Storto--2019|Storto et al., 2019]] ). The ensemble approach for ocean reanalyses provides another avenue for estimating uncertainties across ocean reanalyses ( [[#Storto--2019|Storto et al., 2019]] ).

While there are still limitations in their representation of oceanic features, ocean reanalyses add value to products based only on observation, and are used to inform assessments in AR6 (Chapters 2, 3, 7 and 9). Reanalyses of the atmosphere or ocean alone may not account for important atmosphere–ocean coupling, motivating the development of coupled reanalyses ( [[#Laloyaux--2018|Laloyaux et al., 2018]] ; [[#Schepers--2018|Schepers et al., 2018]] ; [[#Penny--2019|Penny et al., 2019]] ), but these are not assessed in AR6.

<div id="1.5.2.4" class="h3-container"></div>

<span id="reanalyses-of-the-pre-instrumental-era"></span>
==== 1.5.2.4 Reanalyses of the Pre-Instrumental Era ====

<div id="h3-27-siblings" class="h3-siblings"></div>

Longer reanalyses that extend further back in time than the beginning of the instrumental record are being developed. They include the complete integration of paleoclimate archives and newly available early instrumental data into extended reanalysis datasets. Such integration leverages ongoing development of climate models that can simulate paleoclimate records in their units of analysis (i.e., oxygen isotope composition, tree ring width, etc.), in many cases using physical climate variables as input for so-called proxy system models ( [[#Evans--2013|Evans et al., 2013]] ; [[#Dee--2015|Dee et al., 2015]] ). Ensemble Kalman filter data assimilation approaches allow for combining paleoclimate data and climate model data to generate annually resolved fields (Last Millenium Reanalysis, [[#Hakim--2016|Hakim et al., 2016]] ; [[#Tardif--2019|Tardif et al., 2019]] ) or even monthly fields ( [[#Franke--2017|Franke et al., 2017]] ). This allows for a greater understanding of decadal variability ( [[#Parsons--2019|Parsons and Hakim, 2019]] ) and greater certainty around the full range of the frequency and severity of climate extremes. This, in turn, allows for better-defined detection of change. It also helps to identify the links between biogeochemical cycles, ecosystem structure and ecosystem functioning, and to provide initial conditions for further model experiments or downscaling (Chapter 2).

<div id="1.5.2.5" class="h3-container"></div>

<span id="applications-of-reanalyses"></span>
==== 1.5.2.5 Applications of Reanalyses ====

<div id="h3-28-siblings" class="h3-siblings"></div>

The developments in reanalyses described above mean that they are now used across a range of applications. In AR6, reanalyses provide information for fields and in regions where observations are limited. There is growing confidence that modern reanalyses can provide another line of evidence in describing recent temperature trends (Tables 2.4 and 2.5). As their spatial resolution increases, the exploration of fine-scale extremes in both space and time becomes possible (e.g., wind; [[#Kaiser-Weiss--2015|Kaiser-Weiss et al., 2015]] ). Longer reanalyses can be used to describe the change in the climate over the last 100 to 1000 years. Reanalyses have been used to help post-process climate model output, and drive impact models; however, they are often bias adjusted first (Cross-Chapter Box 10.2; e.g., [[#Weedon--2014|Weedon et al., 2014]] ). Copernicus Climate Change Service (C3S) provides a bias-adjusted dataset for global land areas based on ERA5 called WFDE5 ( [[#Cucchi--2020|Cucchi et al., 2020]] ) which, combined with ERA5 information over the ocean (W5E5; [[#Lange--2019|Lange, 2019]] ), is used as the AR6 Interactive [[IPCC:Wg1:Chapter:Atlas|Atlas]] reference for the bias adjustment of model output.

The growing interest in longer-term climate forecasts (from seasonal to multi-year and decadal) means that reanalyses are now more routinely being used to develop the initial state for these forecasts, such as for the Decadal Climate Prediction Project (DCPP; [[#Boer--2016|Boer et al., 2016]] ). Ocean reanalyses are now being used routinely in the context of climate monitoring, (e.g., the Copernicus Marine Environment Monitoring Service Ocean State Report; [[#von%20Schuckmann--2019|von Schuckmann et al., 2019]] ).

In summary, reanalyses have improved since AR5 and can increasingly be used as a line of evidence in assessments of the state and evolution of the climate system ( ''high confidence'' ). Reanalyses provide consistency across multiple physical quantities, and information about variables and locations that are not directly observed. Since AR5, new reanalyses have been developed with various combinations of increased resolution, extended records, more consistent data assimilation, estimation of uncertainty arising from the range of initial conditions, and an improved representation of the atmosphere or ocean system. While noting their remaining limitations, this Report uses the most recent generation of reanalysis products alongside more standard observation-based datasets.

<div id="1.5.3" class="h2-container"></div>

<span id="climate-models"></span>