Written by Yazmina Zurita, physicist and R&D Data Scientist at Nebbo.
Large meteorological centers worldwide provide seasonal forecasts using their unique forecasting system. These systems are complex entities composed of interacting components—models for the atmosphere, ocean, land surface, and sea ice—that collaboratively simulate climate dynamics. Due to diverse configurations, forecasts for the same month from different centers can differ significantly. These discrepancies stem from the distinct ways each system is designed and operated. Below are key differences contributing to these forecast variations:
Model Components and Parametrizations
Seasonal forecasting systems depend on carefully configured model components to capture Earth’s complexity. These components vary in horizontal and vertical resolutions, physics schemes, and numerical methods. Balancing detail and computational efficiency often causes differences among forecasting systems. Some comparisons include:
- Atmospheric Model: For instance, SEAS5 (ECMWF) uses high horizontal resolution (36 km) and detailed physics for precise simulations, albeit at high computational costs. Conversely, CFSv2 (NCEP) employs coarser grids (~100 km) to optimize performance in simulating atmospheric dynamics. [1][2]
- Oceanic Model: Systems like System8 (Météo-France) utilize fine grids (0.25º) to capture ocean currents and thermal dynamics in detail. In contrast, others, like GCFS 2.1 (DWD), use a coarser grid (0.4°), which can be more efficient for modeling broader oceanic processes. [3][4]
- Land Surface Model: Systems represent land-atmosphere interactions with varying levels of complexity. For instance, the CMCC-CM2 (CMCC) system accounts for eight distinct crop types, whereas GloSea6-GC3.2 (UK Met Office) simplifies this by categorizing crops into two grass types. [5][6]
- Sea Ice Model: Crucial for polar forecasts, systems like JMA/MRI-CPS3 (JMA) use models with detailed sea ice dynamics. Simpler systems may prioritize computational efficiency over such detail. [7]
Coupling
Coupling refers to the integration of the different components to simulate their interactions.
- Coupling Schemes: The way model components exchange information and operate in coordination can vary from one system to another.
- Frequency: Model components can exchange information at different rates. While exchanges typically occur every hour, systems like CMCC-CM2 (CMCC) perform data exchanges every 30 minutes between certain components. [8]
Initialization of Ensemble Members
Forecasting systems produce predictions by advancing a set of initial conditions through time. To address the uncertainty inherent in these initial conditions, seasonal forecasting systems generate multiple predictions by introducing slight variations in these starting points. This process results in a collection of potential outcomes for a specific target date, known as an ensemble. The number of ensemble members, initialization timing, and the reference data for initial conditions can vary significantly across different systems.
- Number of Initializations: The CFSv2 (NCEP) system performs ~120 monthly initializations whereas SEAS5 (ECMWF) performs 51. [2][1]
- Timing of Initialization: The CFSv2 (NCEP) system performs some initializations every day, whereas SEAS5 (ECMWF) triggers them all once a month. [2][1]
- Reference Data: The choice of the reference data influences the accuracy and reliability of the initial conditions. The CFSv2 (NCEP) system uses the Climate Forecast System Reanalysis, while the SEAS5 (ECMWF) system uses ERA-Interim. [2][1]
Post-Processing
These systems apply statistical corrections to improve the accuracy and reliability of forecasts. For instance, bias correction methods, such as those in CanESM5.1p1bc (ECCC), refine forecasts by adjusting model outputs to align with reference data. Different methodologies are employed across systems to address specific aspects of the forecasts. [9]
Climate and Forcings
Forcings refer to external factors that can influence or alter Earth’s climate system. These include natural elements like solar radiation and volcanic aerosols, as well as human-induced factors such as greenhouse gas emissions. Each system incorporates these forcings using distinct methods and specific databases, often tailored to meet modeling requirements. [1][9]
The discussed differences between forecasting systems highlight their diversity in modeling, integration, and data handling. Each decision in the multistage forecasting process influences the outcome and the level of agreement among systems. While one might consider selecting a single system as superior, it is crucial to recognize that each system has unique capabilities, and their combination often creates a more robust forecasting framework than relying on just one.
References
- Description of SEAS5-v20171101 C3S contribution
- Description of CFSv2-v20110310 C3S contribution
- Description of System8-v20210101 C3S contribution
- Description of GCFS2.1-v20200320 C3S contribution
- The EC-Earth3 Earth system model and its coupling components
- JULES (Joint UK Land Environment Simulator) Overview
- Description of cps3-v20220201 C3S contribution
- Description of CMCC-CM2-v20191201 C3S contribution
- Description of CanESM5.1p1bc-v20240611 C3S contribution
