Downscaling of MM5 model output using artificial neural network over western Himalaya

Abstract

iwoZ dh vksj tkus okyh flukWfIVd ekSle iz.kkyh vFkkZr if’peh fo{kksHk ¼MCY;w-Mh-½ dh otg ls if’peh fgeky; esa fge ds :i  esa cgqr vf/kd ek=k esa o"kZ.k gksrk gSA ,e- ,e- 5] la[;kRed ekSle iwokZuqeku ¼,u- MCY;w- ih-½ ekWMy fxzM esa o"kZ.k iwokZuqeku miyC/k djkrk gSA ,u-MCY;w-ih- vkmViqV dh lkaf[;dh;  MkmuLdsfyax ls iwokZuqeku dh lVhdrk esa lq/kkj vk ldrk gSA if’peh fgeky; esa ,e- ,e- 5 ekWMy iwokZuqeku dks LFkku fof’k"V o"kZ.k iwokZuqeku esa MkmuLdsy djus ds fy, bl v/;;u esa ,d vjSf[kd i)fr]  —f=e raf=dk latky ¼,- ,- ,u-½ dk mi;ksx fd;k x;k gSA bl v/;;u ds fy,  o"kZ 2003 ls 2008 rd ds 'khrdkyhu eghuksa ¼uoEcj ls ekpZ½ ds vk¡dM+sa fy, x, gSaA izf’k{k.k ds fy, 2003 ls 2007 ds vk¡dM+sa vkSj oS/krk iz;kstu ds fy, o"kZ 2007&2008 ds 'khrdkyhu vk¡dM+sa fy, x, gaSA izf’k{k.k dh izfØ;k esa vkxr&fuxZr laca/k dk irk yxk;k x;k vkSj vafre Hkkj eSfVªDl dk vkdyu fd;k x;kA

 Western Himalaya receives enormous amount of precipitation in the form of snow due to eastward moving synoptic weather system called western disturbance (WD). MM5 a numerical weather prediction (NWP) model, provides precipitation forecast over a grid. Statistical downscaling of NWP output can improve forecast accuracy. In this study artificial neural network (ANN), a non linear method is used to downscale MM5 model forecast to location specific precipitation forecast over western Himalaya. Data of winter months (November to March) from 2003 to 2008 are considered for the study. Data from 2003 to 2007 is used for training and data of winter 2007-2008 is used for validation purpose. In the training process the input-output relationship is extracted and final weight matrix are computed.