Predicting maize leaf area index by partial least square combined with wavelet transform

Abstract

AbstractThe leaf area index (LAI) is crucial for assessing and monitoring maize (Zea mays L.) vegetation status and photosynthetic ability. Predicting maize LAI by hyperspectral remote sensing technology is significant for managing agricultural production. In this study, three N rates (0, 120, and 240 kg N ha−1) and four drought stress treatments (60–70%, 45–55%, 30–40%, and 15–25% of field capacity), were imposed to provide the different environments for maize. The canopy spectral reflectance and LAI of maize were measured at the V6 and V12 stages. In this study, the main objectives were to investigate the performance of a new statistical method for monitoring LAI from canopy spectral reflectance. We used the canopy spectral reflectance to estimate the LAI and compared several methods of spectral analysis, including vegetation indices, wavelet functions, and the combination of continuous wavelet transform (CWT)–uninformative variable elimination (UVE)–partial least squares (PLS) (i.e., CWT–UVE–PLS), for spectral reflectance data analysis and model construction of maize LAI. Results showed that the model using the combination of db3–UVE–PLS achieved the best performance (with the highest coefficient of determination [R2] and lowest RMSE for the calibration [R2 = .979, RMSEC = .172] and validation [R2 = .861, RMSEP = .533] datasets, respectively) in estimating the maize LAI. The CWT–UVE–PLS exhibited a considerable advantage in avoiding redundant or noise information interaction and achieving excellent correlations with maize LAI.