A Soft-computing Ensemble Approach (SEA) to forecast Indian Summer Monsoon Rainfall

Agriculture is the backbone of the Indian economy and contributes to ~16% of the Gross Domestic Product (GDP) and ~10% of the total exports. About 2/3 of this cultivated land depends on monsoon rainfall. As a result, the economic growth of the country largely depends on the accurate prediction of the monsoon rainfall. While predicting the rainfall, more emphasis is given to the southwest monsoon rainfall as 90% of the total rain falls in this season. Even a slight deviation from the average rainfall of 887.5 mm significantly influences the Indian economy. Gadgil and Gadgil (2006) report that the negative impact of drought, on food grain production, is far greater than the positive impact of surplus rain. In addition, heavy rainfall or floods can also damage the food production. Thus, both deficit and excess rainfall affects the Indian economy. Hence, accurate and timely forecasting of monthly Indian Summer Monsoon Rainfall (ISMR) is very much in demand for economic planning and agricultural practices.

Several methods and models, comprising of dynamical, statistical and combination of the two, exist for monsoon forecasting. With the availability of supercomputing facilities, the atmospheric modelers have been running the models with different initial conditions giving different forecasts. Their forecasts are then used to generate one ensemble forecast. An ensemble forecast consists of multiple runs of dynamical models with either different initial conditions or with different numerical simulations of the atmospheric phase (Gneiting and Raftery 2005). In the single model ensemble method, forecasts of a model with different initial conditions are used whereas in the multi-model ensemble approach, forecasts of many models with different initial conditions are used. The initial error, though very small, could grow very fast into different scales as the integration time increases (Lorenz 1982). Using ensemble forecast, instead of a single deterministic forecast, reduces this forecasting error (Zhu, 2005).

Krishnamurti et al. (1999) used multi-model super ensemble model for climate prediction using the multiple regression technique. They reported that their super ensemble model outperformed all model forecasts for different scales of weather and hurricane forecasts. Since Artificial Neural Networks (ANN) outperform multiple regression technique (Pao 2008, Swain et al. 2014), we used ANN techniques in place of multiple regression to develop a Soft-computing Ensemble Algorithm (SEA).

For this purpose, we used 57 members from 6 models from 1982 – 2016 from  IRI/LDEO Climate Data Library with 1⁰ x 1⁰ spatial resolution. Besides rainfall from model predictions, we also use measured rainfall from NOAA Earth System Research Laboratory (ESRL) from 1982 to 2014, having a spatial resolution of 0.5⁰ x 0.5⁰ degree. Then, the model forecasts and the observations are collocated, after bringing both the observations to the same resolution. Since our aim is to predict the ISMR, we analyze the data from June to September of every year. Initially, the model forecasts given in January for June, July, August and September (JJAS) are collocated with the ESRL measured rainfall of the JJAS on monthly basis, thus creating four sets of data for the four monsoon months.   

Out of six models, arithmetic averages of three models predicted more than normal and three predicted below normal rainfall, for 2016. To remove the model biases, we consider all the six models in developing a Soft-computing Ensemble Approach (SEA) using Artificial Neural Network (ANN) approach. We developed 4 ensemble models using different combinations of the 6 model forecasts and found that the estimated errors are minimal if all the 57 members of the six models are used as the independent variables and observation as the dependent variable.  The % deviations of rainfall predicted using SEA and from India Meteorology Department (IMD) for the season as a whole   during 1982 – 2015 (Figure), show a very good agreement between the estimations and the observations.    Since we predicted rainfall at each grid point, we studied the efficiency of the model at different locations, by computing the scatter index (SI: defined as the RMSE normalized to observed mean) at each grid point. The model performance is considered good if the SI is less than 100%. This criteria is met in 77.3%, 83.8%, 85.4% and 86.8% of the total grid points during June, July, August and September respectively.  

SEA predicted an excess rainfall of 108% for 2016 compared to the long period average of 887.5 mm. We also studied the distribution of rainfall over the Indian subcontinent for June- September, as a season, during 2016.   Highest rainfall is predicted for the west coast of India followed by the central India. South east, north east and north India is expected to receive less rainfall. The probability of ISMR being excess, above normal, normal, below normal and drought, following the criteria of IMD, is 52%, 23%, 22%, 2% and 0%. Similar exercise is carried out using the model forecasts given in February, March, April and May and the predictions are almost same. Assuming that the predictions from SEA are correct, we expect an above normal monsoon after two successive drought years.  

References:

Gadgil, S., and S. Gadgil, 2006: The Indian monsoon, GDP and agriculture. Economic and Political Weekly, 4887- 4895

Gneiting, T., and A. E. Raftery, 2005: Weather forecasting with ensembles methods. Science, 30, 248- 249.

Krishnamurty, T. N., C.M Kishtawal, E. L. Timothy, D. R. Bachiochi, Z. Zhang, C. E. Williford, S. Gadgil and S. Surendran, 1999: Improved weather and seasonal climate forecasts from multimodel super ensemble. Science, 285, 1548 – 1550.

Lorenz, E. N., 1982: Atmospheric predictability experiments with a large numerical model. Tellus, 34, 505 – 513.

Pao, H., 2008: A comparison of neural network and multiple regression analysis in modeling capital structure. Exp. Systems with App. 35, 720 – 727.

Swain, D., M. M. Ali, and R. A. Weller, 2006: Estimation of mixed layer depth from surface parameters. J. of Mar. Res., 64, 745 – 758.

Zhu, Y, 2005: Ensemble forecast: A new approach to uncertainty and predictability. Adv. In Atmos. Sci., 22(6), 781 – 788.