Chutia, Sumpi2022-09-172022-09-172017-07https://krishikosh.egranth.ac.in/handle/1/5810187792Crop yield forecasting is an art of predicting crop yields and production before the harvest actually takes place, typically a couple of months in advance. It is very much crucial for the sound planning and policy making in the agricultural sectors of the country. Different types of models viz, crop simulation models, crop weather analysis models and empirical statistical models are generally used to develop district, state and national level yield forecast. Keeping this in view, a study was carried out to develop yield forecast models of winter rice using modified Hendricks and Scholl technique in 14 districts of the Brahmaputra valley of Assam at vegetative (F1) and mid season (F2) stage of the crop. To develop the model, long-term yield data (kg/ha) and weather data on daily basis (maximum temperature, minimum temperature, rainfall, relative humidity morning and evening) were collected for the period 1990-2015. In addition, one extra parameter - BSSH from two locations (Jorhat and Tezpur) for the same period were also collected and utilized to develop forecast models for Golaghat, Jorhat and Sonitpur districts. The daily data were grouped into weekly basis as per requirement of the model. Weekly data were used to prepare simple and weighted weather indices for individual weather variables as well as for interaction of variables. Among the 25 years of yield and weather indices, 22 years data (1990-2012) were used to develop the forecast models and remaining three years data (2013, 2014 and 2015) were used for validation of the models developed. Stepwise regression analysis was executed by trial and error method to obtain the finest combination of predictors at 5% significant level. Result revealed that the model developed for Kamrup district showed good performance compared to other models with highest value of R2 (0.88 & 0.92 in F1 & F2) and with acceptable limit of per cent error, RMSE, nRMSE, MAE and MBE during the process of validation. On the other hand, yield forecast model developed for Bongaigaon district showed poor performance during validation and recorded the highest value of per cent error, RMSE, nRMSE, MAE and MBE compared to other districts during both the forecast (F1 & F2). Interaction of weather variables like Tmax & RH-II, Tmin & RF and Tmin & RH-1 were mainly found to influence the rice yield during F1 and F2 forecast in most of the districts. Forecast model developed after inclusion of BSSH has shown improvement in R2 except Sonitpur district during F1 forecast compared to the model developed without BSSH. Better result was observed in Golaghat district with highest R2 and lowest per cent error, RMSE, nRMSE MAE and MBE compared to Jorhat and Sonitpur. Yield forecast models developed in these three districts showed their dependency on the interaction of BSSH with rainfall as an important weather variable in influencing the winter rice yield. Thus, BSSH data may be included in developing the crop yield forecast models wherever available for better accuracy of forecast.EnglishThesis