Remote sensing estimation of chlorophyll content in rape leaves in Weibei dryland region of China

Abstract

Listen

Translate article

Abstract Chlorophyll is an important factor in crop growth, and is a good indicator of plant nutritional stress, photosynthetic capacity, and growth status. Real-time and reliable crop nutritional diagnosis is the basis for scientific fertilizer management and one of the key technologies for practicing fine agriculture. Focusing on rapeseed crops within the northwest region, this study employed correlation analysis between Soil Plant Analysis Development (SPAD) values and spectral parameters of rape leaves to identify SPAD-sensitive spectral parameters. SPAD values are units of relative chlorophyll content and are used to determine the current relative amount of chlorophyll in a leaf. Subsequently, single-factor models, partial least squares regression models, Back Propagation neural network (BPNN) models, Genetic Algorithm (GA) optimization BPNNs, and BPNN models optimized through GAs based on multiple linear stepwise regression using spectral parameters (referred to as MLSR-GA-BP NN models) were constructed and compared. Findings revealed several significant observations: (1) Consistency in the spectral curves of rape leaves, with spectral reflectance diminishing as chlorophyll content increased; (2) Strong correlations among seven spectral parameters utilized in the modeling, all exceeding 0.770 and achieving significant correlations at the 0.01 level; (3) Across various growth periods, the BPNN model optimized through GAs based on multiple linear stepwise regression emerged as the optimal model. With modeling R 2 surpassing 0.77 and reaching a maximum of 0.91, validation further demonstrated R 2 exceeding 0.73, with a maximum of 0.92, root mean square error ranging between 1.32 and 3.22, and relative error between 2.50 and 4.49%. Hence, the BPNN model optimized by GAs based on multiple linear stepwise regression proves to be an effective inversion method for accurately and swiftly estimating SPAD values in rape leaves.