Artificial neural network model for downscaling of temperature forecast over Western Himalaya

Piyush Joshi, M. S. Shekhar, Jitender Kumar Quamara, Ajay Kumar

MAUSAM · 01 July 2018

In one sentence. Artificial neural network statistically downscales MM5 temperature forecasts at three Western Himalaya stations, reducing RMSE to about 2 °C for day-2 forecasts.

Abstract

The western Himalaya receives good amount of precipitation in the form of rain and snow during winter months from November to April.Heavy snow fall during winter over these regions causes avalanches.Temperature plays a vital role in the process of metamorphism within the snow pack which leads to avalanches.Accurate prediction of temperature over specific location helps to predict the avalanche accurately.In this paper artificial neural network (ANN) method has been used for statistical downscaling of the temperature forecast.Mesoscale model MM5 temperature forecast on 10 × 10 km grids has been taken over three locations Kanzalwan, Stage II and Gulmarg for winter from 2003-2008 to study the performance of the ANN model for its ability to forecast temperature over these locations.Results show that the performance of ANN model for temperature forecast over these three location is good up to day 2. Root mean square error (RMSE) decreases significantly for all the three stations and is also less than the standard deviation (SD) which shows good performance of the ANN model.RMSE is close to 2 °C in case of ANN model, in spite of large variability in temperature over the region.The coefficients of determination between observed and predicted temperature by MM5 and ANN model for Kanzalwan, corresponding to validation set are 0.28 and 0.66 which shows improvement in forecast skill after downscaling.Predicted temperatures at particular location, after downscaling of MM5 model output are closer to the observed values.

Labels

Labels and summary are generated by Claude and reviewed weekly. Wrong? Edit this entry.