Simple Multiple Regression Model for long range forecasting of Indian Summer Monsoon Rainfall

• Published in: Meteorology and atmospheric physics Vol. 99; no. 1-2; pp. 17 - 24
• Main Authors: Sadhuram, Y., Ramana Murthy, T. V.
• Format: Journal Article
• Language: English
• Published: Vienna Springer-Verlag 01.02.2008; Springer; Springer Nature B.V
• Subjects: Aquatic Pollution; Atmospheric Sciences; Earth and Environmental Science; Earth Sciences; Earth, ocean, space; Exact sciences and technology; External geophysics; Math. Appl. in Environmental Science; Meteorology; Rain; Regression analysis; Summer; Surface water; Temperature; Terrestrial Pollution; Waste Water Technology; Water in the atmosphere (humidity, clouds, evaporation, precipitation); Water Management; Water Pollution Control; Weather analysis and prediction; Weather forecasting; Indian Ocean; Asia; India; Summer Monsoon; Indian Ocean; Indian Ocean Dipole; Root Mean Square Error; Monsoon Rainfall; rainfall; Weather forecast; atmospheric precipitation; Linear regression; Multiple linear regression; Sea surface temperature; Forecast model; Statistical forecasting; Summer monsoon; Long range forecast; Rainfall rate; Correlation analysis; Regression model
• ISSN: 0177-7971; 1436-5065
• DOI: 10.1007/s00703-007-0277-0

Summary The relationship between the Indian Ocean Sea-Surface Temperature Anomalies (SSTA) and the Indian Summer Monsoon Rainfall (ISMR) have been examined for the period, 1983–2006. High and positive correlation (0.51; significant at >99% level) is noticed between ISMR and SSTA over southeastern Arabian Sea (AS) in the preceding January. Significant and positive correlation (0.61: significant at >99% level) is also observed with the SSTA over northwest of Australia (NWA) in the preceding February. The combined SSTA index (AS + NWA) showed a very high correlation of 0.71 with the ISMR. The correlation between East Asia sea-level pressure (average during February and March in the region, 35° N–45° N; 120° E–130° E) and ISMR is found to be 0.62. The multiple correlation using the above two parameters is 0.85 which explains 72% variance in ISMR. Using the above two parameters a linear multiple regression model to predict ISMR is developed. Our results are comparable with those obtained from the power regression (developed with 16, 8 and 10 parameters) and ensemble models (using 3 to 6 parameters) of the Indian Meteorological Department (IMD) (Rajeevan et al. 2004; 2006). The rainfall during 2002 and 2004 could be predicted accurately from the present model. It is well known fact that most of the dynamical/statistical methods failed to predict the rainfall in 2002. However, as for associations between SST and ISMR, the index is quite susceptible to inter decadal fluctuations and markedly reduced skill is found in the decades preceding 1983. The RMS error for 24 years is 5.56 (% of long period average, LPA) and the correlation between the predicted and observed rainfall is 0.79.