Estimating leaf chlorophyll content of rice under heavy metal stress using neural network model

In plants, chlorophyll content (CC) is an important indicator of photosynthetic activity, stress, and nutritional status. In this study, hyperspectral indices were used to create a neural network model to detect subtle variations in leaf chlorophyll content of rice under heavy metal stress. We selected four experiment rice farmlands with different levels of heavy metal contamination. The four sites were located in Changchun City, Jilin Province, China. Hyperspectral, leaf chlorophyll content and heavy metal content data were collected. Then, continuum removal (400-760 nm) and first derivative spectra were applied to obtain spectral indices. In a sensitivity study, we showed that variations in leaf chlorophyll content were related to twelve spectral indices; green peak reflectance, green peak position, red valley reflectance, red edge position, red edge slope, red edge margin, red edge area, ratio index, ratio composite index, absorption depth, absorption width, and absorption area. Further, our results verified that the BP neural network using these twelve spectral indices estimated variations in chlorophyll content in heavy metal-stressed rice crops with reasonable accuracy. The most accurate estimation was obtained using the model with two hidden layers, that is, 20 nodes in first hidden layer and 7 nodes in the second hidden layer (R²=0. 9371).