GORS 2(15) Linear and Neural Modeling of Multi-Indices/Multi-Stages Spectral Data for Accurate Prediction of Yield Cotton Under Water and Nutrition Stresses

Abstract

Abstract: Crop yield estimation programs using remote sensing technology are based on spectral monitoring of plant growth during the main phenological phases that lead to productivity. Spectral modeling establishes processes that provide accurate yield estimation and is based on processing large amounts of spectral data that includes all growth stages under environmental growth factors using mathematical modeling with different methodologies. The research aimed to Multi-indices/multi-stage spectral modeling depending on linear and neural network models for prediction the yield cotton from early and advanced growth stages under water and nutrition stresses. Then, to select the best models in predicting the yield with less input factors at certain growth stages that achievement high accuracy. The results showed: regression coefficient values ​​- with a high statistical significance - for the three multiple regression models: estimated/comprehensive, predictive/detailed and predictive/abbreviated were the convergence in estimating the yield crop opposite the actual yield. The spectral predictive neural models achieved great accuracy - according to the statistical indicators, with an average predictive error of 0.3% and a relative error of 0.001% - in estimating and predicting the productivity of the cotton crop. The predictive spectral models designed by linear modeling and neural networks according to a certain number of spectral indices and at specific stages have given high accuracy in predicting yield crop than using a large number of indices for all stages under the influence of growth factors and water and/or nutrition stresses - compared to the yield field. The paper suggests applying spectral field modeling to predict productivity on other crops; And the application of the proposed models for the cotton crop from the spectral field level to the satellite image data in order to estimate the yield at the level of geographical agricultural areas, taking into account the associated processes when applying.