Generation changes¶

Things to watch when switching

• The first ERA6 release covers only 2007 onwards. Keep using ERA5 for long-term analyses for now, and do not join the two reanalyses into a single time series.
• When timely updates move from ERA5T to ERA6, workflows that extend ERA5 up to the present will stop working. Watch for ECMWF's transition-plan communications.
• ERA6 is delivered entirely in GRIB2, so check any GRIB1-only tools and any scripts that rely on current parameter names or units.
• Copernicus (March 2026) intends to keep the ERA5 operational workflow running uninterrupted for as long as necessary, so there is no need to move ongoing work in a hurry.

Things to watch when switching

• Resolution and production change around January–April 2011, so check for a step change there before joining 1979–present into one series. Take particular care with resolution-sensitive quantities such as precipitation, surface fluxes and extremes indices.
• NCAR and NCEI split the record differently. NCAR ends CFSR in December 2010 and starts its CFSv2 dataset (d094000) in January 2011, whereas NCEI runs CFSR to March 2011 and the operational analysis from April. Mixing the two archives can duplicate or drop January–March 2011.
• The operational analysis is not recomputed for long-term homogeneity the way a reanalysis is. For trends after 2011, check your results against long reanalyses such as ERA5 or JRA-3Q.
• The high-resolution output grid changes (1152 x 576 → 1760 x 880), so files with the same name have different grid definitions. Regrid both periods to a common grid (e.g. 0.5°) before combining them.

Things to watch when switching

• R2 starts in 1979, so 1948–1978 exists only in R1. Splicing R1 and R2 at 1979 builds the system differences into your series as a discontinuity — use one dataset per analysis.
• R2 is the same first-generation system as R1 in resolution and observational input, so switching does not change the upper-air circulation much. Agreement between R1 and R2 in a circulation study is not an independent confirmation.
• For soil moisture, surface fluxes and the water budget, the developers judge R2 to be better than R1. The R1–R2 difference itself can serve as a measure of analysis uncertainty.
• File names and folders differ on PSL (slp → mslp, surface_gauss → gaussian_grid), so R1 scripts will not find the R2 files without changes.
• Neither dataset has been updated since spring 2026; studies that need recent years should move to a current reanalysis such as ERA5 or JRA-3Q.

Things to watch when switching

• Do not splice R1 (or CDAS) and CORe into one time series. Model, observations and resolution all differ, so values differ even on the same date.
• CORe fields are ensemble means, which weakens strong winds such as those in tropical cyclones. For extremes and tropical cyclone work, look at the individual members (last 7 days on NOMADS).
• Scripts written for R1 (PSL NetCDF on the 2.5° grid) will not work as-is. You will need to read GRIB2 and regrid (e.g. with wgrib2).

Things to watch when switching

• The two reanalyses differ in climatology and trends, so do not splice JRA-55 and JRA-3Q into one time series at January 2024. If you need a long record, recomputing everything from JRA-3Q alone is the safer option.
• Precipitation and evaporation units changed from mm/day to kg m-2 s-1. Unchanged scripts will give values 86400 times too small, so update both the variable names and the unit conversion.
• JRA-3Q carries a non-commercial licence (CC BY-NC-SA 4.0), so it cannot be used for commercial projects or services, and derived data cannot be redistributed under other terms. For such uses, check the JMBSC route.
• The SST input switch in June 1985 (COBE-SST2 to MGDSST) and the sea-ice-edge quality problem from June 1985 to April 2013 are sources of discontinuity that JRA-55 did not have. For ocean boundary-layer or polar work, check JRA-3Q-COBE and the quality notices.
• JRA-3Q still underestimates stratospheric warming after volcanic eruptions, and its TC maximum 10 m winds are only 50–60% of best-track values. The improvements over JRA-55 do not extend to these issues.

Things to watch when switching

• Do not splice the two into one series. Climatologies and trends differ even over the 1979–2019 overlap, so long-term analyses are safest recomputed entirely with ERA5.
• The accumulation convention differs. ERA-Interim accumulates from the forecast start (3, 6, 9, 12 hours, etc.) and needs differencing, whereas ERA5 hourly data are 1-hour accumulations. Running old scripts unchanged will give wrong precipitation and radiation.
• ERA5 has a pronounced cold bias in the lower stratosphere in 2000–2006, for which a separate rerun, ERA5.1, exists. If you work on stratospheric temperature, humidity or ozone, use ERA5.1 for that period.
• ERA5 provides ensemble spread, but the ensemble has a different resolution and output interval (3-hourly) from the main (HRES) data.

Things to watch when switching

• Do not splice the two reanalyses into one time series at 2016. Observations, model and boundary conditions all differ, so climatologies and trends differ even in the overlap period.
• Variable names (T2M, SLP, PRECTOT, etc.) are largely the same, but collection names (MAT… to M2T…), file format (HDF to netCDF4) and grid (540 to 576 longitudes) changed, so reading and regridding scripts need updating.
• MERRA-2 starts in 1980, so 1979 exists only in MERRA. For analyses that include 1979, compare with another reanalysis or adjust the period.
• Precipitation is now split into model (PRECTOT) and observation-corrected (PRECTOTCORR) fields. When comparing with MERRA results, compare model precipitation with model precipitation.

Things to watch when switching

• The two reanalyses use different models and assimilation, so their climatologies and trends differ. Apart from comparisons over 1979–2004, do not splice them into one time series.
• JRA-25 holds CO2 fixed at 375 ppmv, which can make its long-term temperature and radiation trends differ from JRA-55. Keep this in mind when reproducing trend results from older papers.
• Scripts from the JRA-25 era often assume GrADS-style variable names such as TMPprs. JRA-55 files use a different naming scheme, so the reading code has to be updated first.
• JRA-55 also stopped updating in January 2024. If you need recent years, move to JRA-3Q, and check its licence (non-commercial) and the change in precipitation units at the same time.

Things to watch when switching

• Do not splice the two into one time series. Climatologies and trends can differ even over the 1979–2002 overlap.
• ERA-40 has 23 pressure levels against ERA-Interim's 37; compare on the common levels only.
• ERA-Interim starts in 1979. For 1957–1978, use ERA5 (from 1940) rather than ERA-40.
• ERA-Interim still has its own limitations, such as too strong a hydrological cycle over the oceans and warm, moist low-level biases in the Arctic. Do not overstate the improvement.

Things to watch when switching

• Model, assimilation and observation usage all differ, so climatologies and variability can differ. Do not mix the two, even over their 1979–1993 overlap.
• ERA-15 has only 17 pressure levels up to 10 hPa. When reproducing older stratospheric analyses, compare only the pressure levels common to ERA-40.
• Uppala et al. (2005) judged ERA-40 generally better than ERA-15, but noted excessive tropical oceanic precipitation in the later years and an overly strong Brewer–Dobson circulation in ERA-40. Switching to ERA-40 does not remove the limits for hydrological-cycle and stratospheric studies.
• Both are discontinued earlier-generation products; use ERA5 for new work.

Things to watch when switching

• Do not join the two products into one time series. Different model physics and land-surface schemes can give different climatologies; ECMWF staff have said on the forum that each system is better in some respects and that a full comparative evaluation is in preparation.
• The grid is entirely new (two Lambert domains replaced by one polar stereographic domain), so grid-point indices, masks and orography files must all be regenerated.
• CARRA2 pressure levels stop at 50 hPa; 30, 20 and 10 hPa are only in CARRA1. Height levels above 500 m (up to 3000 m) are only in CARRA2.
• The CARRA1 recipe for 24-hour precipitation (+30 h minus +6 h) does not work with CARRA2's 18-hour forecasts. Follow the accumulation guidance in the CARRA2 user guide or use the daily/monthly entry.
• Any workflow that needs data after March 2027 must switch to CARRA2 before then.

Things to watch when switching

• Versions 2c and 3 differ in boundary conditions, observations and model, so values differ even for the same year. Do not mix versions; when reproducing v2c results with v3, recompute climatologies and trends from scratch.
• In v3 the SST input switches from SODAsi.3 to HadISST2.2 between 1980 and 1981, and files are split into SI (1806–1980) and MO (1981–2015) folders. Check for a discontinuity in any trend analysis that crosses this boundary.
• Variable abbreviations and folder layout changed (e.g. sea-level pressure prmsl, 3-hourly accumulated precipitation apcp), so v2c scripts will not run unchanged.
• The v3 ensemble spread is more reliable, so when interpreting 19th-century results where observations are sparse, report uncertainty from the spread files rather than relying on the ensemble mean alone.

Things to watch when switching

• Do not splice the two versions into one time series. V3.3 is generally wetter, especially over the oceans.
• V3.3 monthly starts in 1983, so for 1979–1982 only V2.3 is available. If you use it, state the version mix in your paper.
• Climatology and index code written for the 2.5° grid must be adapted to the 0.5° grid, including the grid-centre positions.
• NOAA advises against using V2.3 except in rare cases, such as reproducing earlier studies.

Things to watch when switching

• Do not splice V06 and V07 into one time series; download the whole period again as V07 (V06 is no longer distributed).
• Scripts with '06' in paths or precipitationCal in variable names must change to '07' and precipitation.
• High-latitude and winter results come from areas that are newly filled in V07, so check the Quality Index alongside them.

Things to watch when switching

• If you need research-grade (Final) data after September 2025, wait for V08 Final, or use the Late Run (V07C) as a stopgap and state that it is a different run.
• When V08 Final is released, all values from 1998 will change, so do not mix it with climatologies or trends built from V07.

Things to watch when switching

• Do not splice values from the two products into one series; recompute the full period with a single product.
• File names, URLs and variable layout have changed, so automated download scripts need updating.
• Daily data exist only in the discontinued old product, so recent periods require another dataset.

Things to watch when switching

• Values for March 2000 to March 2022 are identical in both editions. For anything that uses April 2022 onward, use Ed4.2.1 throughout and do not mix editions.
• The differences are small, but results computed with Ed4.2 for April 2022 to July 2024 are worth re-checking with Ed4.2.1.
• The CERES team recommends checking the edition and release date in your files against the Data Quality Summary (DQS).

Things to watch when switching

• Even once ORAS6 is released, do not join it to ORAS5 as one time series. Forcing, model and assimilation all differ, so climatologies and trends may differ.
• If you plan to analyse alongside ERA6, ORAS6 is the natural choice because it is consistent with ERA6's ocean boundary conditions.
• The ECMWF Newsletter notes that Arctic sea ice can be biased thin in some regions, so validate against observations before using ORAS6 sea-ice thickness.
• Release dates, distribution channels and variable names are not yet fixed, so do not assume existing ORAS5 download scripts will work unchanged.

Things to watch when switching

• V4r4 and V4r5 are different optimized solutions, so do not splice them into one time series. If you need 2018–2019, use V4r5 for the whole period.
• V4r5 can only be read with ecco_access inside the AWS cloud (us-west-2 recommended); downloading from outside incurs transfer charges.
• The ECCO team has said V4r6 will supersede V4r5, so check when V4r6 is released before starting a new long-term analysis.

Things to watch when switching

• Specific model and data assimilation changes have not yet been documented officially. Read the synopsis once it is released before switching.

Things to watch when switching

• The two products differ in resolution and level structure; do not splice them into one time series or compare grid points directly.
• SODA4 files are much larger for the same period. Plan to download only the regions and variables you need.
• The SODA4 paper is still a preprint; when citing it, give the EGUsphere DOI and check whether it has since been published.

Things to watch when switching

• Do not splice v5 and v6 into one time series. The interpolation methods differ, so values and spatial patterns differ even for the same month.
• The anomaly baselines differ, so v5 and v6 ssta cannot be compared directly. Recompute both against the same climatology period first.
• NOAA ENSO indices based on v6 (such as RONI) may differ from historical ONI values computed from v5. Recheck your list of El Niño and La Niña events.
• Without the 3-month running mean, v6 can show more month-to-month variability. Account for the smoothing when comparing variance or extremes.
• Uncertainty estimates for v6 are still in development (announced for release in 2026). If your analysis needs them, check their availability.

Things to watch when switching

• COBE-SST3 is not updated on a schedule and has no data after 2024. Stay with COBE-SST2 for analyses that must run to the present.
• SST changes around the 1940s are smoother than in COBE-SST2, which can change mid-20th-century trends and multidecadal variability.
• The two differ in resolution and time step, so do not splice them or difference them directly; regrid to a common grid and monthly means first.
• The authors report that COBE-SST3 daily values do not match buoys and Argo better than OISST v2.1. For recent daily analyses, also consider satellite-based products such as OISST.

Things to watch when switching

• Global and regional mean SST around 1941–1945 differ from the previous version. Recompute any long-term trend or variability analysis that includes this period with 4.2.0.0.
• The Met Office advises downloading the full dataset again at every update and always using the latest version. Quote the exact version number (HadSST.4.2.0.0) in publications.

Things to watch when switching

• Do not mix v2.1 and v3.0 in one series; download the whole period again as v3.0. Values in the 1980s and after 2016 in particular may have changed.
• For a 1991–2020 baseline, use the v3.0 climatology product.
• The citation DOIs and the data paper have changed.

Things to watch when switching

• Periods with the same name (e.g. decav) cover different years in WOA18 and WOA23. Check the years before switching a model initial state or an anomaly baseline.
• WOA23 has no density, conductivity or mixed layer depth files. Compute them from WOA23 temperature and salinity, or keep using WOA18 for those fields.
• Added observations and QC changes mean values can differ even for the same period. Do not mix the two versions and read the difference as a real change.

Things to watch when switching

• Do not combine DT-2021 and DT-2024 files in one time series; download the whole record from 1993 again in DT-2024.
• Scripts using the 0.25° dataset IDs (my, myint, nrt) must switch to the 0.125° IDs. The grid size changes, so rebuild any precomputed masks or regional indices.
• The new MDT changes the mean fields of ADT and the absolute geostrophic currents (ugos, vgos). Take care when comparing mean circulation with earlier results.
• From DT-2024 the C3S and Copernicus Marine products differ in mapping method (OI vs MIOST) and grid, so do not mix them or use one as a drop-in replacement for the other.

Things to watch when switching

• If you are starting OMIP-style experiments now, JRA55-do v1.6.0 (1958–2023 cycle) is still the standard. There is no official reason yet to delay experiments while waiting for CFORCE.
• Once CFORCE is out, its parent reanalysis, resolution and adjustment methods will all differ, so runs forced with JRA55-do cannot be compared directly. Run the same model with both forcings to quantify the differences.
• Variable list, bulk formulae and freshwater runoff for CFORCE have not been announced. Do not commit to adapting your JRA55-do preprocessing code until it is released.

Things to watch when switching

• The single SSMIS-series record from 1978 stops at 2025-10-16. If you fill 2021 onwards with the AMSR ICDR (OSI-438), use the matching AMSR CDR (OSI-458) for consistency — which means starting in 2002.
• Joining different series into one time series (e.g. OSI-450-a1 + OSI-438) can introduce a step at the junction because of differences in sensor resolution and coastal contamination. If you must, first quantify the difference between the two series over their 2002–2020 overlap.
• Automated download scripts need the new FTP directories and product IDs in file names (icdr-v3p0 → icdr-v3p0-amsr, etc.). For 2-day-latency files after 2026-09-01, switch to OSI-438-a.
• After the switch to AMSR3 on 2026-09-01, OSI-438-a uses a different sensor from OSI-438 (AMSR2). Use OSI-438-a for the most recent weeks and the nominal OSI-438 files for climate analyses, and document it if you mix the two.
• Cite the DOI of each product you actually used (OSI-458, OSI-438, OSI-438-a).

Things to watch when switching

• If you have OSI-450-a files, replace them with OSI-450-a1. Check the affected dates (PUM Appendix B), especially if you work on concentration or area near coasts.
• Update the DOI you cite to 10.15770/EUM_SAF_OSI_0023.

Things to watch when switching

• V5 files are no longer distributed. For new work, download the full record from 1978 onward as V6, and never mix V5 and V6 in one time series.
• The input sensor changes from SSMIS to AMSR2 on 2025-01-01. NSIDC reports the differences are within the data uncertainty, but for anomaly or trend analyses that straddle 2025, state the sensor change and test the sensitivity of your results to it.
• Scripts that read the prototype_am2 group, or hard-code F17 in file names, will fail on 2025+ data. Switch to the main CDR variables and the am2 satellite code.
• The land spillover fix makes SMMR-era (1978–1987) extent slightly larger than in V5. Keep this in mind when comparing 1980s baselines or long-term trends with earlier V5-based results.
• The DOIs have changed. Cite V6 (10.7265/b18j-z797), and also V4 (10.7265/tm2n-1m33) if the most recent period comes from the near-real-time product.

Things to watch when switching

• Do not mix v3.0 and v4.0 in one time series. If you need CryoSat-2 data after 2020-05, the safest option is to download the whole period again as v4.0.
• Until v4.0 has formal documentation and a DOI, give the download date, the CEDA catalogue URL and the file's product_version attribute in papers, and check again whether a DOI has been issued.
• Sentinel-3 and CryoSat-2 differ in orbit and sensor, so values differ even for the same month. Before combining satellites, check their differences over the overlap period (from 2016-10).

Things to watch when switching

• Do not splice freeboard or thickness series built from v6 or earlier onto v7 data; recompute the full record with v7.
• Scripts that assume v6 file names (…_006_xx.h5) or IS2SITMOGR4 v3 names need the version numbers updated.

Things to watch when switching

• v206 (2010–2025) and v300 (2025/26 onward) differ in resolution, input satellites and uncertainty model, so a series joined across the switch may show a step. Until a consistent record in a single version is available, treat the v206 and v300 periods separately in long-term analyses.
• File names (SMOS_CS2 → SMOS_CS2_S3A_S3B, 25KM → 12P5KM, v206 → v300) and variables have changed, so existing readers will break. Check the variable names in the new files before updating your code.
• Larger v300 uncertainties reflect the new uncertainty model, not lower quality. Do not compare uncertainties across versions directly or filter both versions with the same threshold.
• Antarctic winter thickness is new; the CS2SMOS record is Arctic only. For Antarctic work, check when the record starts and how far it has been validated.

Things to watch when switching

• Both the satellite and the wind inputs change on 2025-01-01. If you build long-term trends or climatologies from 1979, check separately for systematic differences across this boundary.
• Scripts with v4.1 hard-coded in the file name will not find the 2025 files. Replace the version part with a wildcard.
• In your data description, state that 2025 onward uses AMSR2 and ERA5 inputs (v4.2), and if you use that period, also cite Hersbach et al. (2020), the ERA5 reference given in the user guide.

Things to watch when switching

• If you build a record that spans OSI-405-c and OSI-405-d, note the input changes in July 2025 and September 2026; everything from September 2026 onward is AMSR3-based.
• EUMETCast ingest scripts that expect NetCDF3 or gunzip a .nc.gz file need updating.
• Users of the single-sensor files should check the FTP listing directly: the PUM (v1.9, December 2024) predates the SSMIS removal (2025-07-17) and the AMSR3 switch (2026-09-01), so the current per-sensor file set and names are not documented there.

Things to watch when switching

• Values from 2025 onwards are slightly smaller than a continued SSMIS record would give, which tilts the extent trend since 1979 very slightly more negative. Allow for this difference (tens of thousands to over a hundred thousand km²) when ranking "record lows" from 2025 on.
• NSIDC describes the intercalibration with SSMIS as preliminary and plans further improvements, so values from 2025 onwards may change again in future versions.
• V3 files are no longer distributed. Do not mix previously downloaded V3 CSVs with V4 CSVs; download the full record again as V4 (values through 2024 are identical).
• Daily values can change slightly the next day because of gap filling, so re-download the most recent few days before using them.

Things to watch when switching

• GLDAS-1 data can no longer be downloaded. Reproducing older studies requires recomputing with GLDAS-2.0/2.1, and the numbers may differ.
• Do not mix GLDAS-1 and GLDAS-2 values in one time series.

Things to watch when switching

• Because the model differs, soil moisture and runoff differ even for the same dates. Do not splice the two into one series; download Noah-MP data from October 2000 instead.
• Anomaly indices for Central Asia are stated to use a 2000-10-01 to 2022-09-30 baseline (water years 2001–2022).

Things to watch when switching

• Daily data for Africa are not available from the public GES DISC products. Work with the monthly data or contact FEWS NET or NASA LDAS.

Things to watch when switching

• GRIB scripts need both variable names and file paths changed.
• The climatology baseline changed, so anomaly values change too.

Things to watch when switching

• Beta data may change substantially before release, so keep using NLDAS-2 for long-term analyses in publications for now.
• When NLDAS-3 arrives, resolution, model and forcing all change; do not splice it with NLDAS-2 into one series.
• The schedule in the April 2025 workshop slides (open loop in late summer 2025, data assimilation in late fall 2025) was not met. Check the NASA LDAS NLDAS-3 page and GitHub for release news.

Things to watch when switching

• Do not mix climatologies or trends computed from v1.0 with v1.1 results; download the full record again as v1.1.
• Because the inputs are now ERA5, studies comparing or validating against ERA5 should keep in mind that the two are not independent.
• Earth Engine and THREDDS copies may differ in version and period, so state the data source in your paper.

Things to watch when switching

• When switching versions, download the past record again; values for the same date differ between versions.
• The definition of sm_uncertainty changed, so do not compare it directly with error estimates from v08 or earlier.

Things to watch when switching

• Replace time series built from the old versions by downloading the full record again; values for the same date differ slightly.
• Update the version value in earthaccess calls and scripts to '009' (SPL3SMP) and '006' (SPL3SMP_E).

Things to watch when switching

• Do not mix V7 and V8; download the full record again as V8. The new precipitation forcing changes root-zone soil moisture climatologies as well.
• The user guide notes that the SPL4SMGP V8 variable depth_to_water_table_from_surface is, despite its name, the water table depth in peat.

Things to watch when switching

• Do not mix v3 climatologies or trends with GLEAM4, especially for component-level analyses such as transpiration and bare-soil evaporation.
• The grid changed, so regrid to a common grid before cell-by-cell comparison.

Things to watch when switching

• Do not join 4.2 and 4.3 into one series; download the full record again as 4.3.
• In irrigated regions (e.g. the North China Plain, northern India) 4.3 evaporation can differ markedly from 4.2, so state the version in regional analyses.

Things to watch when switching

• Do not mix C6 and C6.1; download the full period again from the C6.1 GF products (C6 has not been distributed since 2023-07-31).
• The C6.1 forward product has fewer gaps thanks to the backup inputs, but pixels filled with climatological LAI/FPAR do not reflect the actual vegetation state of that year.

Things to watch when switching

• If you already downloaded the withdrawn VIIRS ET V2 files, do not use them for research. NASA has stated V2 is not of science quality.
• Even after V2.1 is released, check the differences between the two records over their overlap before joining them. In particular, no VIIRS satellite continues Terra's morning observations.

Things to watch when switching

• When joining the two records, check pixel- and season-level differences over the overlap (2012 onward, 2018 onward for NOAA-20) and state in the paper that there is a seam.
• The QC bit layout differs, so do not reuse MODIS quality-filter code on VIIRS. The _500m suffix has also gone from the variable names.
• Analyses built on the MODIS 4-day product (MCD15A3H) can only be continued at 8 days with VIIRS.
• Suomi NPP products (VNP15A2H) stop being distributed in November 2026, so the reference satellites going forward are NOAA-21 (VJ215A2H, primary) and NOAA-20 (VJ115A2H, secondary).

Things to watch when switching

• Re-check results made with 006 using 061. The 006 files can no longer be downloaded.

Things to watch when switching

• Re-check results made with V1 using V2. V1 files are no longer available from Earthdata Search, the Data Pool or AppEEARS.

Things to watch when switching

• Analyses that need 250 m (small fields, urban green space) cannot be continued at that resolution with VIIRS. Compare at a common 500 m.
• Long EVI series may show a seam because of the blue-band difference, so consider NDVI or EVI2 first when joining the two sensors.
• Fill values, scale notation and pixel-reliability classes differ, so do not reuse MODIS quality-filter code as is.
• Suomi NPP products (VNP13) stop being distributed in November 2026, so the reference going forward is NOAA-21 (VJ213, primary) and NOAA-20 (VJ113, secondary).

Things to watch when switching

• Re-check results made with 006 using 061. The 006 files can no longer be downloaded.

Things to watch when switching

• Re-check results made with V1 using V2. V1 files are no longer available from Earthdata Search, the Data Pool or AppEEARS.
• The definition of the VI Quality usefulness bits changed, so review any quality filters written for V1.

Things to watch when switching

• Do not mix climatologies or trends computed from V1 with V2.0 results; re-download the full record from 2014 as V2.0.
• Change datasetIdentifier and file-name patterns in scripts from _v1 and V1.1 to _v2 and V2.0. V1 remains on CDSE as an archive.
• Cite the V2.0 DOI in papers and state whether only RT6 was used.

Things to watch when switching

• To combine the 1 km V2 and 300 m V2.0 for analyses starting in 1999, first assess their differences over the 2014–2020 overlap.
• The 1 km V2 will not be corrected any more, so any known issues remain.

Things to watch when switching

• Do not mix NDVI3g v1 and GIMMS-3G+ in one time series; re-download from 1982 as GIMMS-3G+. Winter high-latitude NDVI and post-2006 values differ most.
• File names say v1_1 up to 2014 and v1_2 from 2015, but ORNL distributes them as one dataset (V1.2). Do not split the record by this label.
• GIMMS-3G+ cannot extend the record beyond 2022 either.

Things to watch when switching

• The PKU version is a derived product separate from NASA's GIMMS-3G+. Trends can differ between the two, so compare them rather than resting conclusions on one.
• The consolidated version uses MODIS values from 2003, carrying Terra MODIS characteristics (including orbital drift). Use the AVHRR-only version (to 2015) if AVHRR-only consistency is needed.
• No extension beyond 2022 has been announced (V1.2 was the latest Zenodo version on 2026-10-13).

Things to watch when switching

• Do not mix C6 and C6.1 maps in one time series; download the whole period again as V061.
• Cropland or wetland areas computed from the UMD legend (LC_Type2) will differ in C6.1.

Things to watch when switching

• Pick the version that matches the year in CDS requests (v2_0_7cds for 1992–2015, v2_1_1 from 2016). Joining the two is the officially intended use.
• Input sensors and production change at the 2015→2016 boundary; if change in that year looks anomalous, check whether it is real.

Things to watch when switching

• Do not read differences between the two maps as real land cover change (they are mostly algorithmic).
• If only one year is needed, the more accurate 2021 v200 map is the usual choice.

Things to watch when switching

• LCFM is a separate product; check legend, definitions and accuracy before joining it to WorldCover in one series.
• For 10 m global maps after 2021, follow the LCFM release schedule.

Things to watch when switching

• Do not mix C6 and C6.1; download the full period again from C6.1 (C6 has not been distributed since 2023-07-31).

Things to watch when switching

• Joining a MOD11 (split-window) series to VIIRS introduces a discontinuity from the algorithm difference (e.g. the 3–5 K arid-land bias). To join records, compare MOD21 (TES) and VIIRS over their overlap.
• No VIIRS observation continues the Terra (morning) series. Even when joining Aqua (afternoon) to VIIRS, Aqua's drift period (2021 onward) samples a different time of day from VIIRS at 13:30.
• VIIRS satellites also differ from each other in overpass time by 50 minutes to an hour, and a summer difference of about 0.5 K has been reported (VNP21 User's Guide).

Things to watch when switching

• Do not mix V1 and V2; download the full period again in V2 (V1 has not been distributed since 2025-04-08).

Things to watch when switching

• Before joining MODIS and VIIRS albedo into one series, compare the two over their overlap (2012 onward) at the same sites and bands. Band centres differ, so spectral albedos do not map one-to-one; comparing broadband (vis, nir, shortwave) values is safer.
• MCD43 pools morning and afternoon observations, whereas the VIIRS products use a single afternoon satellite. Fewer observations can raise the share of magnitude inversions, so look at the quality-flag distribution too.
• Delivery of Suomi NPP data (VNP43) ends on 2 November 2026. Going forward, use VJ143 (NOAA-20) or VJ243 (NOAA-21).
• HDF4 code (pyhdf, etc.) must be rewritten for HDF5 (h5py, xarray).

Things to watch when switching

• Do not mix C6 and C6.1; download the whole period again as C6.1.
• Code that assumes quality values of only 0/1 must be updated to handle 2–7.

Things to watch when switching

• Do not mix C1 and C2; download the whole period again as C2.
• Values near coastlines and over bright high-latitude surfaces may differ from C1.

Things to watch when switching

• Hall et al. (2024) found that VIIRS (VNP10A1F) maps slightly more snow than MODIS (MOD10A1F) in the US Great Basin (a less conservative cloud mask and different overpass times). The mean daily difference in snow-covered area in water year 2023 was 0.69%, while VIIRS snow seasons in the Great Salt Lake Basin were about two days longer.
• Analyses that used the MODIS 8-day or monthly products must composite VIIRS daily data themselves. Match the compositing rule (maximum extent, mean) to MODIS.
• Terra (morning) and VIIRS (afternoon) observe at different times, so diurnal melt and morning cloud differences show up in the values. Before joining, compare Aqua with VIIRS over their overlap (2012 onward).
• Delivery of Suomi NPP data (VNP10) ends on 2 November 2026; from then on use VJ110 (NOAA-20) and VJ210 (NOAA-21).

Things to watch when switching

• When C7 appears the whole record will change, so do not mix it with C6.1.
• Code using the 8-day or monthly products or snow albedo will need an alternative in C7 (for example compositing the daily data yourself).

Things to watch when switching

• Do not mix V1 and V2; download the whole period again as V2.
• Code that reads the mask_values attribute must switch to flag_values.

Things to watch when switching

• Do not mix GlobSnow v3.0 and CCI SWE in one time series. Grids and algorithms differ, so values differ even on the same day.
• The GlobSnow "bias-corrected monthly" product is a separate product corrected with snow course data for the underestimation of deep snow. Do not compare it directly with monthly means of CCI daily data.
• Like GlobSnow, CCI SWE excludes mountains, Greenland and glaciers. It cannot be used for mountain snow studies.

Things to watch when switching

• SCFV and SCFG differ only in forests and are identical elsewhere. For forested regions, always state which one you used.
• MODIS (1 km) and AVHRR (5 km) differ in resolution and sensor, so values differ even on the same day. Pick one to suit the purpose, e.g. AVHRR for long-term trends and MODIS for regional detail.
• Earlier versions (MODIS v1.0–v3.0, AVHRR v1.0 and v2.0) are marked "superseded" at CEDA. Download the whole period again in v4.0.

Things to watch when switching

• 2017-07 to 2018-05 has no observations. Interpolated or reconstructed values are not GRACE data; if you use them, state the method.
• The Tellus Level-3 land grids come in separate GRACE (RL06) and GRACE-FO (RL06.3) collections, so you must join them yourself. The mascons contain everything from 2002 in one file.
• No bias across the gap was reported, but when estimating trends use methods that account for the gap and the missing months (for instance, regression that skips missing months).

Things to watch when switching

• When RL07 mascons and Level-3 grids appear, every value from 2002 onward will change, so do not mix them with trends or climatologies computed from RL06.x — download the full period again.
• Only GRACE L2 RL07 is out so far, so joining GRACE (RL07) directly with GRACE-FO (RL06.3) would mix releases. Take care if you work with Level-2 coefficients.

Things to watch when switching

• If GRACE-FO stops before GRACE-C launches, another gap could open. Long-term trend studies should state the mission boundaries.

Things to watch when switching

• Do not append C6.1 to a time series built from C6 files; download the whole record again as C6.1.

Things to watch when switching

• Do not splice the two records naively. Check regional differences over the overlap period (2012 onward) first, and if you do join them, report how in the paper.
• Regional differences range from 0.1% in Europe to 33.1% in Southeast Asia, so a single global offset (about 4%) is not an adequate correction.
• The file structure and layer names match, so MCD64A1 code works almost unchanged. Without a morning overpass, though, VIIRS burn dates agree slightly less often with active fire dates (same-day agreement 38% vs 46%).
• Because Suomi NPP stops delivering data in November 2026, VIIRS burned area after that date depends on whether a NOAA-20-based product (planned Collection 3) is released.

Things to watch when switching

• Recompute results built on V001 with V002; V001 can no longer be downloaded.

Things to watch when switching

• Burned area after November 2026 must wait for a Collection 3 or NOAA-20 release; once available, check its differences from C2 over the overlap period.

Things to watch when switching

• Switching atmospheric chemistry model emissions or carbon budgets tuned to GFED4s over to GFED5 substantially increases fire emissions; re-evaluate model results accordingly.
• Do not combine GFED4.1s post-2016 beta values with GFED5.1 in a single time series. Use one version for the whole period.
• GFED5.1 (through 2022) and GFED5.1NRT (from 2023) use different methods; when joining them, cite both papers and describe the change in method.

Things to watch when switching

• Do not splice v7 and v8 into one time series; download the whole period again as v8. Values in the v7 period (2014 onward) also differ from v8.
• Change v7 to v8 in FTP paths (/standard/v8/, /realtime_ver/v8/). On G-Portal, product version V05 is v8.
• The reprocessed standard data first released in July 2023 (V8.0000.0) were regenerated as V8.0000.1 in September 2023, so files downloaded in between must be fetched again.

Things to watch when switching

• When V06 appears, read the algorithm and format documents, and do not mix V05 and V06.
• HDF5 reading code needs to change to NetCDF4. Check the new specification for file naming as well.

Things to watch when switching

• Do not splice V0.x and V1.0 into one time series. Download the whole period again as V1.0 for long-term analysis.
• Scripts that use the old FTP path (global_CMORPH) should switch to the V1.0 paths at NCEI or AWS.

Things to watch when switching

• Do not splice v2 and v3 into one series; download the whole period again as v3. Because v3 is generally wetter, recompute climatologies and trends.
• The 50°–60° latitude bands are new in v3, so results there cannot be compared with v2.
• Earth Engine collection IDs change (UCSB-CHG/CHIRPS/DAILY → UCSB-CHC/CHIRPS/V3/DAILY_RNL or DAILY_SAT).

Things to watch when switching

• Do not splice CLARA-A2 and A3 into one time series. Calibration and algorithms differ, so values for the same month differ.
• The developers note that the visible-channel calibration in the ICDR (2021 onward) is not of CDR quality. Interpret trends across the 2020/2021 boundary with care.

Things to watch when switching

• The NCEI THREDDS 'PATMOS-X Cloud Properties' file listing still holds v05r03 files (through 2022). Always check the version in the file name and do not mix v5.3 with v6.0.
• Cloud-top height, temperature and pressure changed the most, so cloud-height results obtained with v5.3 are worth re-checking with v6.0.

Things to watch when switching

• Do not mix v1.2 and v2.0 in one time series; re-download the full record as v2.0.
• The most recent days to months are preliminary, and the final release changes the geostationary input, so values can change. Re-check results against the final data.
• NCEI advises using v1.2 only for cases such as reproducing earlier studies.

Things to watch when switching

• The sensors, algorithms and grids differ, so climatologies and anomaly amplitudes differ. Do not splice them; compare them over the 1979–2022 overlap first.
• Indices defined on Interpolated OLR (for example the OMI) need care when continued with another dataset because the baseline changes. PSL itself switched the OMI input to blended OLR.

Things to watch when switching

• The satellites and atmospheric inputs differ before and after March 2022; check for discontinuities in any cloud or surface-flux analysis that spans that boundary.
• When re-checking Ed4A results with Ed4B, confirm the edition string in the file name (Edition4B) and that the variable names match.
• Edition 5 is expected to reprocess the whole record, so always state the edition in publications.

Things to watch when switching

• Do not mix C6 and C6.1 in one time series; download the whole record again as C6.1.

Things to watch when switching

• MOD06 C6.1 and CLDPROP give different values even for the same Aqua observations (different refractive index, cloud top algorithm and QA). Do not join the MODIS standard record directly to VIIRS CLDPROP; use Aqua MODIS CLDPROP as the bridge.
• Platnick et al. (2021) report systematic differences in effective radius and phase because of the VIIRS 2.25 µm versus MODIS 2.13 µm band, part of which has not yet been separated from radiometric calibration differences between the sensors.
• No continuity product carries on the Terra (morning) cloud record. Do not mix observations from different times of day.

Things to watch when switching

• Do not mix V1 and V2 masks in one analysis. Check the file attributes to see which mask version CLDPROP V1.1 used as input.

Things to watch when switching

• Do not mix ISCCP-D climatologies or trends with H results. The resolution and calibration differ, so cloud amount differs even over the same region.
• H keeps the cloud retrieval algorithm largely unchanged, so NCEI notes that the biases known from D (viewing-angle and satellite-change effects) carry over to H.

Things to watch when switching

• Use AIRS-only (AIRS2RET, AIRS3STM) from the start for long time series. Splicing AIRX and AIRS at September 2016 mixes processing differences into the climate signal.
• According to the V7 L2 guide, the AIRS+AMSU stream has a larger share of QC 0/1 (best/good) cases. Even over the same period, the samples entering AIRS-only averages can differ, so check the count (_ct) fields.

Things to watch when switching

• V6 and V7 monthly means are computed differently, so values for the same month can differ. Do not mix versions.
• Code that used the old 8-day product (AIRS3ST8 V006) should be rewritten to aggregate V7 daily files (AIRS3STD) with count weighting.

Things to watch when switching

• Do not mix v5 and v6 in one time series. N2O, temperature (upper stratosphere to mesosphere), water vapour, HCN and upper-stratospheric HNO3 in particular differ between versions.
• Screening rules (Quality and Convergence thresholds, useful pressure ranges) differ by version. Re-check the tables in the v6 quality document when switching.
• The team only says v5 forward processing continues "for the foreseeable future", so starting new analyses with v6 is the safer choice.

Things to watch when switching

• Do not mix v6.0 and v6.1 values. They differ after 2021, so v6.0 numbers in earlier papers cannot be compared directly with v6.1.
• Since 2021 UAH relies on a single satellite, so it is more sensitive to that satellite's drift or failure.

Things to watch when switching

• Recompute trends obtained with v4.x using v5.0; both intercalibration and diurnal adjustment changed.
• Files released from January to May 2023 were missing February 1985 in TUT and some TMT grid points after 2018 (also affecting TLT); this was fixed after June 2023. Re-download files obtained in that window.

Things to watch when switching

• When the new version appears, join CDR and ICDR of the same version; do not mix, for example, a v1 CDR with a v2 ICDR.
• State the product ID (e.g. GRM-29-R1, GRM-29-I1) and version in publications.

Things to watch when switching

• Do not mix C6 and C6.1 in one time series; download the whole record again as C6.1.

Things to watch when switching

• MODIS and VIIRS AOD generally differ by fairly constant offsets. Sawyer et al. (2025) find that for DT, NOAA-20 is close to Aqua, Terra is about 0.015 higher and Suomi NPP is the highest. Check the difference over an overlap period before joining records.
• Sawyer et al. (2020) report that much of the DT difference comes from sampling (pixel size, cloud mask, swath width) rather than calibration. Comparing only cells retrieved by both sensors on the same day is the safest approach.
• Analyses that used the MODIS combined AOD must choose either DB (AERDB) or DT (AERDT) for VIIRS. If you need daily or monthly L3, only AERDB D3 and M3 exist at present.
• There is no successor for Terra's morning observations, so Terra-based records differ in time of day from Aqua and VIIRS afternoon observations.

Things to watch when switching

• Do not mix V2.0 and V2.1 in one time series; download the whole record again as V2.1.
• DB V2.1 has less ocean coverage than V2.0 in some places (stricter sun-glint screening).

Things to watch when switching

• Analyses built on V22 files and code need rewriting for V23, since both field names and grid changed.
• The quality statement reports wider coverage and better agreement with AERONET-DRAGON deployments at 4.4 km, though formal validation of V23 AOD was still ongoing when it was written.

Things to watch when switching

• Re-download the full record as V004 for trends and climatologies. The NO2 README warns about discontinuities introduced by version changes and urges care in trend studies that mix versions.
• The V004 NO2, ozone and SO2 collection metadata state that the products are not yet fully validated and recommend checking with the algorithm leads about known problems before trend analysis.
• Code that read HCHO from V003 must change: V004 OMHCHO is netCDF4, with its variables under the key_science_data group.

Things to watch when switching

• Units differ. Multiply TROPOMI mol m-2 values by 6.02214×10^19 to get molecules cm-2 (file attribute multiplication_factor_to_convert_to_molecules_percm2).
• Algorithms, a priori profiles and cloud treatment differ, so NO2 and HCHO show systematic differences even for the same place and day. Check the offsets over the post-2018 overlap before joining the records.
• For a consistent multi-satellite NO2 record, consider NASA MEaSUREs MINDS NO2 (OMI_MINDS_NO2, TROPOMI_MINDS_NO2), which applies the OMI algorithm uniformly to TROPOMI and other sensors.

Things to watch when switching

• V11.2r values after 2024 retain the icing bias (up to about 0.5 ppm); switch to V11.3r. Biases of 0.2–0.5 ppm matter for growth-rate and seasonal-amplitude studies.
• There are no V11.3r files before 2024. Long time series must join the two versions; state the join (2024-01) in publications.
• According to the release documents, Lite SIF and L2 Standard are not part of this revision.

Things to watch when switching

• For long-term analysis the official advice is to use RPRO (2018-04-30 to 2022-07-25) and continue with the latest OFFL. Use RPRO orbits 24655–24779 for the overlap.
• Versions kept changing after RPRO: 02.06.00 (lower mid-latitude winter NO2), 02.07.01 (DLER update causing a discontinuity), then 2.8.0 and 2.9.1. Check the processor version in the file name (for example 020901) before trend analysis and wait for the next reprocessing if needed.
• OFFL and NRTI files from the V1 period (2018–2021) differ from the V2 RPRO; do not mix them.

Things to watch when switching

• XCO2 and XCH4 from the two satellites differ in instrument, algorithm and version and must not be joined directly. Check each against TCCON or other ground data over the overlap (2019 onward).
• GOSAT-2 L2 (SWFP) has a time-varying positive bias after 2024 from Band 1 sensitivity changes (not corrected in V02.30), so trend analyses after 2024 need particular care.

Things to watch when switching

• As of October 2026, TANSO-3 L2 is not yet publicly released. Use GOSAT and GOSAT-2 L2 until the 2027 release, and check validation results before relying on early TANSO-3 versions.
• Imaging gives very different sample numbers and spatial patterns from the earlier point-sampling satellites. Account for sampling differences when comparing site or regional averages.

Things to watch when switching

• The V5.00 Level 3 document reports that, for June–August 2015, global mean AOD rose by about 17% (all-sky) and 12% (cloud-free) relative to V4.20. Recompute climatologies and trends from scratch with V5.00.
• LEM sharply reduces samples near the SAA (by 80–100%), so those grid cells can be fill values. Check sample counts in South American and South Atlantic analyses.
• If V4.x code read the altitude array from file metadata, switch to the Lidar_Data_Altitudes dataset in V5.00.

Things to watch when switching

• Because the wavelengths differ, AOD, backscatter, depolarisation and lidar ratio cannot be compared one-to-one. The 355/532 nm ratio depends on aerosol type, so studies that bridge the two records should state their wavelength-conversion assumptions and uncertainties.
• The aerosol classification schemes (subtype names and number) differ. Compare classifications only after mapping both to common categories such as dust or smoke.
• No public spaceborne lidar data exist from July 2023 to August 2024, leaving a gap in any combined time series.
• Overpass time (13:30 versus 14:00) and altitude differ, so account for sampling-time differences for boundary-layer aerosol and clouds with strong diurnal cycles.

Things to watch when switching

• Pixel values and missing pixels differ between 5.2 and 5.3, so do not mix them in one series; download the whole period in 5.3.
• Quality level 0 is now included, so re-check your quality_level and pathfinder_quality_level filters if you want to match 5.2-era screening.
• The high-latitude solstice errors are not fixed in 5.3.

Things to watch when switching

• Do not splice V2.0 and V3.x into one time series. Where there are no observations, values differ by the difference between backgrounds.
• V2.0 assimilated buoys, which left artificial structures around buoy sites that stood out in derivatives such as stress curl. V3.x does not have this problem, but buoy comparisons are no longer comparisons with assimilated data.
• Only V2.0 covers the years before 1993, so studies using that period should state the version clearly and check differences from V3.x separately.

Things to watch when switching

• Values from July 2012 onward differ from V3.0, so do not mix results produced with V3.0 and V3.1. Earlier values are the same.
• The citation DOIs on the RSS page are still labelled Version 3.0. In papers, state the version you actually used (V03.1 in the file names).