Introduction to CPT Tool

CPT, created by IRI (The International Research Institute for Climate and Society), provides a Windows-based statistical tool used for

• (Step 1) Seasonal climate forecast (regression) model construction,
• (Step 2) Performing model validation, and
• (Step 3) Producing forecasts given updated data.

Step 1: For seasonal climate model construction, 3 complex types of linear regression models are available, namely, the Principal Components Regression (PCR),  the Canonical Correlation Correlation Analysis (CCA), and the Multiple Linear Regression Model (MLR). The type of linear regression that is being used here to generate seasonal models and the corresponding seasonal forecasts is the PCR technique. More on the PCR technique is discussed here.

Step 2: Before forecasts can be made using the constructed model, the user might want to first evaluate the performance of the model, i.e. its ability to generate quality forecast with the training data used to construct it. This is where model validation comes in. Two methods are available in CPT to do model validation: retroactive forecasts validation method and cross-validation method. With fewer years of training data available to construct the model, the cross-validation method has been recommended, and it is the method that has been adopted for validation of models on this website. The cross-validation method and the corresponding measure of forecast quality (Pearson's Correlation Coefficient) is discussed here.

Step 3: Once the model has been validated to be good enough to be used, the next step will be to make the forecast itself. CPT provides two types of forecast outputs, the absolute/deterministic value of the rainfall rate forecasts and also the corresponding probabilistic forecasts i.e. whether they are expected to be above normal, normal or below normal. Both of these forecasts types are presented on this website and explanations as to how they are arrived at are given here.