A ground-based real-time remote sensing system for diagnosing nitrogen status in cotton plants

A ground-based real-time remote sensing system has been developed for diagnosing nitrogen status in cotton plants. Study on spectral reflectance characteristics of cotton leaves was conducted at The University of Tennessee Milan Agricultural Experiment Station for three consecutive years (1996, 1997, and 1998). Spectral reflectance from cotton canopies with different nitrogen treatments was measured using a fiber optic spectrometer. It was found that useful spectral features could be extracted from the spectral reflectance spectra in the blue, green, red, and near infrared wavebands. A mathematical model was developed to calculate the spectral index of the measured plants. The spectral index showed strong correlation with nitrogen application rate, petiole nitrate content, and yield at pinhead square. Based on the relationship between the spectral reflectance characteristics and the nitrogen status in the cotton plants, a multispectral plant health sensor was developed and field tested. The sensor has four wavebands (blue, green, red, and near infrared). Each output of the sensor represents the intensity measurement of spectral reflectance in that waveband. To avoid measurement error that may be induced by variations in spectral characteristic under natural illumination of the plant canopy; a modulated light source was used to illuminate the canopy. A demodulating electronic system was employed in the sensor to measure the spectral reflectance of the canopy resulting from the modulated light. Field test results demonstrated that the plant health index, calculated using the data from the plant health sensor, strongly correlated with the nitrogen application rate and the yield (R² = 0.99 and 0.81, respectively). An artificial neural network (ANN) was trained using the spectral reflectance data. The ANN models had three layers, five inputs, and two outputs. Inputs were spectral reflectance measurements in blue, green, red, and near infrared wavebands and the stage of plant development (days after planting). Nitrogen-deficiency and non-nitrogen-deficiency were the two outputs. ANN models were tested and results showed that the trained ANNs could diagnose the nitrogen status in cotton plants with an accuracy rate greater than 95%. The ground-based remote sensing system with ANNs has shown great potential for real-time variable-rate control of nitrogen in cotton.