Probabilistic Monthly and Seasonal Outlook by a Hybrid Prediction (Dynamical and Machine Learning Models) System

Data and Documentation

With over three decades of continuous development, the Famine Early Warning Systems Network (FEWS NET) provides early warning and analysis of acute food insecurity based on an evidence basis that includes agroclimatic forecasts. These monthly and seasonal outlook products are generated to support the FEWS NET food security outlooks.

About These Products

To improve agroclimatic forecasts for FEWS NET, we developed an experimental probabilistic multi-model ensemble using a hybrid system based on Dynamical models and Machine learning for Global and 12 regions. This webpage shows forecast products and skill scores in hindcast (1993-2016).

Product Descriptions

Forecast Products and Reference Climatology

  • Tercile Probability (Probability for three categories : Below Normal Near Normal and Above Normal)
  • Probability of lowest 20% Climatology (Probability of not-exceeding 20th percentile of the climatology)
  • Probability of highest 20% Climatology (Probability of exceeding 80th percentile of the climatology)
  • Climatology (Long period average of observed precipitation during 1993-2016)

Skill Scores Estimated in Hindcast During 1993-2016 using Leave-One-Out Cross Validation

  • Rank Probability Skill Score (RPSS)
  • Logarithm Skill Score (LSS)
  • Reliability Diagram
  • Relative Operating Characteristic (ROC) Curve
  • Generalized Relative Operating Characteristic Curve Score (GROCS)

More about each skill scores can be found at the WWRP/WGNE Joint Working Group on Forecast Verification Research page.

Note: The white shade in the forecast map indicates the climatological odds. The light pink indicates a dry mask (no forecast). Dry mask defined as when 10% or more of years are dry. Then the grid point is dry and no forecast generated on that grid. In other words,forecasts are only produced when more than 90% of the training sample is non-zero.

Method

These monthly and seasonal forecasts are based on the outputs of initialized dynamical forecasts from Copernicus Climate Change Service (C3S) multi-system predictions. To construct the probabilistic multi-model ensemble, we propose the use of the Extreme Learning Machine (ELM), a novel machine learning approach. ELM is a state-of-the-art generalized form of single-hidden-layer feed-forward neural network. However, since the traditional ELM network only produces a deterministic outcome, we use a modified version of ELM called Extended Probabilistic Output Extreme Learning Machine (EPO-ELM). Climate Hazards Group InfraRed Precipitation with Stations (CHIRPS) is used as a reference data set to evaluate the model skill. All the forecast and hindcast skill products generated by using the XCast tool.

For more technical descriptions of the methods and XCast tool, see:

  • Acharya N and Hall, K.J.C (2023): A Machine Learning Approach for Probabilistic Multi-Model Ensemble Predictions of Indian Summer Monsoon Rainfall. MAUSAM https://doi.org/10.54302/mausam.v74i2.5997
  • Hall, K.J.C and Acharya N (2022): XCast: A python climate forecasting toolkit. Frontiers in Climate https://doi.org/10.3389/fclim.2022.953262
  • Acharya N, Srivastava N.A, Panigrahi B.K. and Mohanty U.C. (2014): Development of an artificial neural network based multi-model ensemble to estimate the northeast monsoon rainfall over south peninsular India: an application of extreme learning machine.Climate Dynamics. 43(5):1303-1310.https://doi.org/10.1007/s00382-013-1942-2

Data Source

Forecast products from Copernicus Climate Change Service (C3S) Precipitation from the Climate Hazards InfraRed Precipitation with Stations (CHIRPS) dataset is used in this forecast system.

Page Credits

Real-time forecast and skill scores code development: Kyle Hall and Nachiketa Acharya.

Referencing Forecasts

To reference forecast plots, we ask that you acknowledge PSL as in ”image is provided by the NOAA Physical Sciences Laboratory, Boulder, Colorado, USA, from their website at https://psl.noaa.gov/”. You should also reference the publication:

Acharya N and Hall, K.J.C (2023): A Machine Learning Approach for Probabilistic Multi-Model Ensemble Predictions of Indian Summer Monsoon Rainfall. MAUSAM, 74 (2), 421-428,https://doi.org/10.54302/mausam.v74i2.5997

Note

These forecasts are experimental. NOAA/PSL is not responsible for any loss occasioned by the use of these forecasts.