Overview¶
Reanalysis products are comprehensive datasets that combine historical observations with models simulation to create a consistent, spatially and temporally complete representation of the Earth’s past climate and weather. They are one of the most valuable resources in Earth system science for understanding climate variability, trends, and dynamics.
Reanalysis vs. Analysis vs. Simulations vs. Observations¶
Understanding the differences between these fundamental data types is crucial for Earth system science research.
Reanalysis vs. Analysis 👈
Analysis refers to the real-time, operational data assimilation product produced by weather forecasting centers:
Uses the current version of the forecast model and assimilation system
The model and assimilation methods change over time as improvements are made
Optimized for short-term weather forecasting
Creates discontinuities when the system is upgraded
Reanalysis retrospectively processes historical observations:
Uses a fixed model and assimilation system for the entire time period
Ensures temporal consistency and homogeneity
Not updated in real-time; produced in multi-year projects
Better suited for climate studies and long-term trend analysis
More computationally expensive due to reprocessing decades of data
Key Distinction: Analysis prioritizes current forecast skill; reanalysis prioritizes long-term consistency.
Reanalysis vs. Model Simulations 👈
Model Simulations (also called free-running simulations or climate projections):
Run forward in time without observational constraints
Initial conditions may come from observations, but the model evolves freely
Used for future climate projections and sensitivity experiments
Can drift from observed climate due to model biases
Examples: CMIP6 models, CESM simulations
Reanalysis:
Continuously constrained by observations through data assimilation
Cannot drift far from observed atmospheric state
Represents the “best estimate” of what actually happened
Limited to historical periods where observations exist
Blends model physics with observational evidence
Key Distinction: Simulations show what the model thinks should happen; reanalysis shows what actually happened (constrained by observations).
Reanalysis vs. Observations 👈
Direct Observations (in-situ and remote sensing):
Actual measurements from instruments
Highest accuracy at measurement location and time
Spatially and temporally incomplete (gaps between stations, satellite swaths)
Different instruments have different biases and uncertainties
No information about unobserved variables or locations
Examples: Weather station data, satellite retrievals, radiosonde profiles
Reanalysis:
Gridded, gap-filled product combining observations with model physics
Provides estimates even where/when no observations exist
Spatially and temporally complete
Includes hundreds of variables, many not directly observed
Less accurate than direct observations at observed locations
Smooths out small-scale features
Subject to both observational and model uncertainties
Key Distinction: Observations provide ground truth but are incomplete; reanalysis provides complete coverage but is less accurate than direct observations.
How Reanalysis Works¶
Data Assimilation Process¶
Reanalysis uses a process called data assimilation to blend:
Observational Data: Including satellite measurements, weather stations, radiosondes (weather balloons), ship and buoy observations, and aircraft reports
Numerical Models: Physics-based models that simulate atmospheric, oceanic, and land surface processes
The data assimilation system statistically combines these elements, weighing observations and model predictions based on their respective uncertainties to produce the “best estimate” of the atmospheric state at any given time.
Key Characteristics¶
Temporal Consistency: Uses a fixed, modern assimilation system and model throughout the entire reanalysis period.
Spatial Completeness: Fills gaps where observations are sparse or unavailable using model physics
Regular Grid: Produces data on uniform spatial and temporal grids, making it easier to analyze
Multiple Variables: Provides hundreds of atmospheric, oceanic, and land variables that are physically consistent with each other
Types of Reanalysis¶
Atmospheric Reanalysis¶
ERA5 (ECMWF): Currently the most widely used, covering 1940-present
JRA-3Q (JMA): Japanese reanalysis with high-quality precipitation estimates
Ocean Reanalysis¶
ORAS5: Ocean reanalysis system
Coupled Reanalysis¶
CERA-20C: 20th century coupled reanalysis
Applications in Earth System Science¶
Reanalysis products are used for:
Climate Monitoring: Tracking long-term temperature, precipitation, and circulation patterns
Extreme Event Analysis: Studying hurricanes, droughts, heat waves, and other extreme weather
Model Validation: Evaluating climate and weather models against a consistent reference
Forcing Data: Driving regional models, hydrological models, and impact assessments
Process Studies: Understanding physical mechanisms and teleconnections
Trend Analysis: Identifying climate change signals and natural variability
Deep Learning/Machine Learning: Training and validating AI-based weather and climate models
Limitations and Considerations¶
Observational Network Changes¶
Early periods (pre-satellite era) have fewer observations and higher uncertainty
Introduction of new satellite systems can create discontinuities
Model Biases¶
The underlying numerical model has systematic biases that affect the reanalysis
Variables directly constrained by observations (temperature, wind) are more reliable than derived quantities (precipitation, evaporation)
Spurious Trends¶
Some trends may be artifacts of changing observational systems rather than real climate signals
Multiple reanalysis products should be compared for robust conclusions
Resolution Limitations¶
Even high-resolution reanalysis cannot resolve all small-scale features
Convective processes and orographic effects may be inadequately represented
Best Practices¶
When using reanalysis data:
Choose the Right Product: Consider temporal coverage, spatial resolution, and which variables are best represented
Validate Carefully: Compare with independent observations when possible
Use Multiple Products: Cross-validate findings across different reanalysis systems
Understand Uncertainty: Be aware of which variables are observation-constrained vs. model-derived
Check Documentation: Review known issues and dataset updates from producing centers
Resources¶
Major reanalysis centers provide extensive documentation: