Impact of land surface initialization on seasonal forecasts of the extremes of Indian summer monsoon

Impact of land surface initialization on seasonal forecasts of the extremes of Indian summer monsoon

Subhadeep Halder¹, Paul Dirmeyer¹, Lawrence Marx¹, James L. Kinter III¹

¹Center for Ocean–Land–Atmosphere Studies, George Mason University

Rainfall received over India during the monsoon season (June-September) not only sustains its increasing population but also affects its agricultural output and economy. The Indian summer monsoon rainfall exhibits variability that extends from the synoptic and intraseasonal to seasonal and interannual time scales. Accurate prediction of the extremes of the monsoon season, namely droughts and floods are extremely important but also challenging at the same time. Land surface anomalies in the form of soil moisture and snow cover, besides sea surface temperature, impart memory to the climate system that can be potentially utilized to extend the prediction skill of temperature and rainfall over the Indian region beyond the deterministic limit of weather forecasts. The objective of the present study is to analyze and quantify for the first time the impact of land surface initialization and land state driven predictability on the seasonal extremes of the Indian summer monsoon during 1982 till 2009, using the National Centers for Environmental Prediction (NCEP) global Climate Forecast System Version 2 (CFSv2) model. For this purpose, ensemble coupled hindcast simulations are made in each of the 28 years (1982-2009), initialized on the 1^st of April, May and June and ending at the end of September. A comparison of the simulations with realistic versus random initialization of the land surface and atmospheric states for the extreme monsoon years gives us a measure of the hindcast skill for the model that may be attributed to land surface anomalies at the beginning of the monsoon season. However, biases in the representation of coupled land-atmosphere feedback processes in CFSv2 also appear to degrade the skill of the model forecasts with lead-time. Detailed results will be presented.