Characteristics of Seasonal Forecasts

Technical Foundations and Operational Implementation, the Characteristics of Seasonal Forecasts
Written by Tyler Anderson, R&D Data Scientist at Nebbo.

Introduction

Seasonal forecasts bridge the gap between short-term weather predictions and long-term climate projections. These forecasts provide expected conditions over the coming months, typically 1-7 months ahead. Contrary to short-term weather predictions, seasonal forecasts provide forecasts on the monthly mean instead of predictions for each day. This post examines the technical characteristics that define seasonal forecasting systems, focusing on resolution, ensemble methods, and model structure.

Temporal Resolution

Temporal resolution defines the time intervals at which forecast data is provided. Seasonal forecasts use different temporal aggregations than short-term weather forecasts and long-term climate projections. Short-term weather forecasts generally focus on temporal resolutions ranging from hourly to daily, while seasonal forecasts deal with monthly means. That is, they try to capture what the month will look like, on average, instead of trying to capture short-term signals or long-term signals.

This means that seasonal forecasts give information about what the weather conditions will be, on average, for each month. This larger temporal resolution allows for greater skill when making predictions from 1 to about 6 months ahead when compared to taking daily or weekly averages.

Spatial Resolution

Spatial resolution specifies the geographic grid size at which models calculate atmospheric and oceanic variables. Current operational systems use:

• Horizontal resolution:
  • Atmosphere: TL255 (~36 km) to T126 (~100 km) grid spacing.
  • Ocean: 0.25° to 1° grid spacing.
• Vertical resolution:
  • Atmosphere: 30-91 levels with non-uniform spacing (higher density near surface).
  • Ocean: 30-75 levels with increased resolution in the upper ocean.

Specific model configurations:

• ECMWF SEAS5: O320 Gaussian grid (~36 km) with 91 vertical levels; NEMO ocean at 0.25° with 75 levels.
• NCEP CFSv2: T128 (~100 km) with 64 vertical levels; MOM4 ocean at 0.5° with 40 levels.
• UK Met Office GloSea5: N216 (~60 km) with 85 vertical levels; NEMO ocean at 0.25° with 75 levels.
• JMA: TL159 (~110km) with 60 vertical levels; MRI.COM v3 ocean at 1° with 52 levels.

These resolutions describe the resolution with which the physics calculations are done for each of the models. The available data for these models, however is not always on the same grid with which the calculations were done. Most of the models provide data on a standard 1° latitude by 1° longitude grid for surface variables (except for the JMA model, which provides its data on a 1.5° latitude by 1.5° longitude grid). This transformation from the computation grid to the output grid is generally done through an interpolation method.

Ensemble Forecasting

Ensemble forecasting addresses uncertainty in long-range predictions by generating multiple forecast realizations. Key technical aspects include:

• Ensemble size:
  • Operational range: 10-120 members per forecast cycle.
  • Theoretical minimum: ~10 members to detect signal.
• Perturbation methods:
  • Initial condition perturbations:
    • Singular vectors targeting maximum error growth.
    • Ensemble Data Assimilation (EDA) using perturbed observations.
    • Analysis differences from multiple data assimilation cycles.
  • Stochastic physics schemes:
    • Stochastically Perturbed Parameterization Tendencies (SPPT).
    • Stochastic Kinetic Energy Backscatter (SKEB).
    • Stochastic Parameter Perturbations (SPP).
    • Random Parameters (RP) schemes.
  • Multi-model approaches:
    • Direct combination of different modeling systems.
    • Perturbed parameter ensembles varying uncertain model constants.
  • Lagged initialization:
    • Time-lagged ensemble members from different initialization dates.
    • Combines forecasts of different lead times into single ensemble.

Using ensembles when forecasting is a way of addressing the uncertainty when trying to forecast on the seasonal time scale because it allows for models to perturb the initial conditions slightly to see how these would affect the forecasts.

Because each of the models has some bias, however, there are often processing steps that must be done to properly use the seasonal forecasts. These include:

• Bias correction using model hindcasts (typically 20-30 years).

• Calibration to adjust ensemble spread.

• Multi-model combination through equal weighting or Bayesian methods.

Ensemble Generation Frequency

Operational centers employ different strategies for ensemble generation timing, balancing computational resources against forecast freshness and ensemble size:

• Burst approach:
  • ECMWF SEAS5: 51 members on 1st of month.
  • Météo-France System 7: 51 members on 1st of month
  • DWD: 30 members on 1st of month.
  • CMCC-SPS3.5: 50 members on 1st of month.
  • Advantages: Consistent lead times, simplified verification, concentrated computing resources.
  • Disadvantages: Forecasts become stale by end of month, single initialization state.
• Lagged approach:
  • NCEP CFSv2: 4 members daily (00Z, 06Z, 12Z, 18Z) = ~120 monthly members.
  • UK Met Office GloSea5: 2 members daily = ~60 monthly members
  • Advantages: Continuous updates, diverse initialization states, resilience to initialization shocks.
  • Disadvantages: Mixed lead times, complex verification, continuous computing demand.

It is important to understand that each of the models initialize their ensemble members in different ways. The burst models have the advantage that they are a bit easier to compare each member because they were initialized at the same time, but their predictions do generally become a bit stale by the end of the month. On the contrary, lagged models have the advantage that their ensembles are constantly being updated, but it is difficult to compare the members because they each had access to different information depending on when they were initialized.

Physical Model Characteristics

The seasonal forecast systems use coupled Earth system models that combine atmospheric models, ocean models and other climate models.

• Atmospheric component:
  • Dynamical cores: Semi-Lagrangian (ECMWF IFS), finite-volume (GFDL), spectral (NCEP GFS).
  • Physics parameterizations:
    • Convection schemes (mass flux, CAPE-based).
    • Radiation (shortwave/longwave, direct/diffuse).
    • Cloud microphysics (1-2 moment schemes).
    • Planetary boundary layer (K-profile, TKE-based).
  • Resolution: TL255-T126 horizontal, 30-91 vertical levels.
• Ocean component:
  • Models: NEMO (European systems), MOM4/6 (US systems), HYCOM, MRI.COM.
  • Dynamics: Primitive equations, Boussinesq approximation.
  • Vertical mixing schemes: K-profile, TKE-based.
  • Resolution: 0.25°-1° horizontal, 30-75 vertical levels.
  • Data assimilation: 3D-Var, EnKF, or hybrid schemes for initialization.

The coupled approach captures essential feedback mechanisms:

1. Ocean-atmosphere: ENSO, IOD, Atlantic Niño (3-9 month predictability).
2. Land-atmosphere: Soil moisture-temperature/precipitation (1-3 month predictability).
3. Cryosphere-atmosphere: Sea ice-circulation patterns (1-4 month predictability).

Summary

The characteristics of seasonal forecast systems—temporal and spatial resolution, ensemble methods, generation frequency, and coupled model components—define system capabilities and limitations. These technical specifications balance scientific requirements against computational constraints.

Understanding these characteristics helps users interpret forecast information and uncertainties. Ongoing developments in higher resolution, improved physics, advanced initialization methods, and ensemble techniques continue to enhance forecast skill for various regions and variables.

References

• Doblas-Reyes, F.J., García-Serrano, J., Lienert, F., Biescas, A.P., & Rodrigues, L.R.L. (2013). Seasonal climate predictability and forecasting: status and prospects. Wiley Interdisciplinary Reviews: Climate Change, 4(4), 245-268.
• Johnson, S.J., Stockdale, T.N., Ferranti, L., Balmaseda, M.A., Molteni, F., Magnusson, L., Tietsche, S., Decremer, D., Weisheimer, A., Balsamo, G., Keeley, S.P.E., Mogensen, K., Zuo, H., & Monge-Sanz, B.M. (2019). SEAS5: the new ECMWF seasonal forecast system. Geoscientific Model Development, 12(3), 1087-1117.
• Saha, S., Moorthi, S., Wu, X., Wang, J., Nadiga, S., Tripp, P., Behringer, D., Hou, Y.T., Chuang, H.Y., Iredell, M., Ek, M., Meng, J., Yang, R., Mendez, M.P., van den Dool, H., Zhang, Q., Wang, W., Chen, M., & Becker, E. (2014). The NCEP Climate Forecast System Version 2. Journal of Climate, 27(6), 2185-2208.