Experimental Forecast of 5-20 days OLR Anomalies over Indian Region

Methodology & Data Used

An analog method for the experimental forecasting of OLR pentad anomalies over Indian region (15⁰ S -30⁰N,  50⁰E-110⁰E) has been adopted from Xavier and Goswami, 2007. The  method is  briefly described below. 

      Pentad OLR data were subjected to Empirical Orthogonal Function (EOF) decomposition and were reconstructed using first 10 EOFs. The EOF1 represented the pattern associated with the annual cycle of OLR variability. EOF2 represented the classical pattern of monsoon inraseasonal variations and the subsequent EOFs represented different phases of intraseasonal variability. The analog method was used for the forecasting of OLR pentad anomalies at lead time period of 1 pentad to 4 pentads.

      The basic steps used in the analog method are given below.

1. Identification of spatial analogs for the reference OLR pentad at an initial point (t= to) from all OLR pentads available prior to the reference point.  Only those pentads that having spatial correlation (R) > 0.7 and spatial RMSE < 20W/m2 with the reference pentad are retained for further analysis.

2. Identification of temporal analogs to the evolution of the temporal modes of first 10 PCs during the period from reference point (t= to) to 5 pentads behind.  The selection is done from the retained pentads in the step 1. For this purpose, for each PCs (1 to 10), the temporal correlation and RMSE of PC scores during 5 pentads just prior to the initial point and that of retained pentads were computed. The selection criteria used are; R > 0.5 and RMSE < unit standard deviation of PC scores of initial pattern.

3. Forecast of PCi (i= 1 to 10) at a lead time ‘Ł’ is computed as the average of the respective temporal analogs at that lead time. If Mi is the number of temporal analogs for the ith  PC, the forecast for the PCi at the lead time Ł is computed as

PCi (to + Ł) = 1/Mi Σ PCi (j+ Ł), where j=1,2,3,--------Mi.

4. Finally the OLR pattern at lead time Ł in respect to the initial time (to) is forecasted as the product of spatial EOF mode and the predicted temporal modes (PCs) in the earlier step.

OLR anomaly =  Σ EOFi  X  PCi(to + Ł)

Annual mean climatology is added with the forecast OLR anomaly to prepare OLR forecast. OLR forecast were prepared for the period 2001-2006.  OLR model climatology was computed from OLR pentad forecasts during these 6 years for each of the 4 lead times.  Similarly from the reconstructed OLR (based on first 10 EOFs) for the same period OLR climatology was also prepared for the actual OLR.   For preparing the model forecast OLR anomalies over Indian region, model climatology is used.  For preparing the actual OLR anomalies, actual OLR climatology is used.

The daily Outgoing Long wave Radiation (OLR) data from 1st Jan, 1979 onwards till the end date of pentad based on  which the forecast is being prepared. The data were obtained from the website of NOAA’s Earth System Research Laboratory, USA (www.cdc.noaa.gov).  Daily data were first converted to pentad averages.  For each year, 73 Pentads were obtained.  Before inserting in the model, annual mean climatology values were subtracted from the actual pentad OLR values to create pentad anomalies.