Estimating chlorophyll content of rice under soil contamination stress using BPNN

Estimating chlorophyll content of plant exactly is meaningful in precision agriculture because of its ability in indicating photosynthesis activity, stress and nutritional state of plant. Heavy mental contamination in agricultural field has been one of vital ecological environment issues which threaten global environment quality, human being subsistence and food security as well. The soil contamination will damage the plant and change the chlorophyll content of rice. Sometimes the soil contamination stress is weak and there is no obvious and visual symptom, though the plant has been injured. In this study, a back propagation neural-network (BPNN) will be tried to estimate the subtle variations in leaf chlorophyll content of rice under potential contamination stress with hyperspectral data in field conditions. Field works were conducted in seven rice fields with different heavy mental contamination stress level in central part of northeast China during the summer of 2008 and 2009. Hyperspectral data, samples of rice and soil were acquired and analyzed. The hyperspectral data were processed using continuum removal, differential computation and binary encoding. Eleven hyperspectral variables were derived and the correlation between them and contamination was analyzed. Stepwise multiple linear regression methods were investigated to ascertain their performances in the prediction of rice leaf comparative chlorophyll content respectively. Models established by the linear regression analysis indicated the lower feasibility for estimating rice leaf chlorophyll content associated with soil contamination. Neural-network models provide more robust results for complicated system analysis than conventional mathematical models. The higher coefficients of determination (R² = 0.912) and lower prediction errors (RMSE = 2.34) was obtained using a BPNN model of 11-9-5-1 with an ideal performance.