Combining hyperspectral imagery and LiDAR pseudo-waveform for predicting crop LAI, canopy height and above-ground biomass

Abstract

Crop Leaf area index (LAI), canopy height and above-ground biomass (AGB) are important structural parameters. Accurate predictions of these three parameters are required for improving the applications of crop growth monitoring, health status assessment and yield prediction. Airborne Light Detection and Ranging (LiDAR) system is the most reliable technique for accurately predicting vegetation structure parameters. LiDAR technique has been broadly applied to estimate vegetation LAI, height and biomass, and reliable prediction results have been obtained. However, LiDAR data lack the spectral information of vegetation. The combination of LiDAR data and hyperspectral imagery can achieve complementary advantages of two data sources and improve the prediction accuracies of vegetation parameters. In this research, we aim to estimate maize LAI, canopy height and AGB using the combined hyperspectral imagery and LiDAR pseudo-waveforms. We constructed the LiDAR pseudo-waveforms through discrete-return point clouds and extracted pseudo-waveform variables. The prediction models of maize LAI, canopy height and AGB were established with a random forest (RF) regression algorithm using the traditional statistical variables derived from discrete-return point clouds, the pseudo-waveform variables, the combined hyperspectral vegetation indices and pseudo-waveform variables, respectively. Moreover, the comparative analyses of the three prediction models were conducted to determine the optimal prediction model and explore the potential of the combined hyperspectral vegetation indices and pseudo-waveform variables for predicting maize crop structural parameters. The results showed the strong relationships between LiDAR pseudo-waveform variables and maize LAI, height, and biomass (R2 = 0.799, 0.832 and 0.871, respectively). Moreover, the pseudo-waveform variables produced better results than the results estimated from traditional statistical variables of discrete-return LiDAR (R2 = 0.772, 0.812 and 0.811, respectively). Therefore, it is a viable method for predicting maize LAI, canopy height and AGB using the LiDAR pseudo-waveforms created from discrete-return LiDAR data. Nevertheless, we found that the combined pseudo-waveform variables and vegetation indices derived from hyperspectral imagery produced a better prediction result (R2 = 0.829, 0.892 and 0.909, respectively) when compared to LiDAR pseudo-waveform data alone, and the prediction accuracies improved by 3.8%, 7.2% and 4.4%, respectively. The combined hyperspectral imagery and LiDAR pseudo-waveform data provided complementary information and therefore improved prediction accuracies of these parameters. Although small improvements were observed, the combined data have potential for improving predictions of crop parameters. Our study will provide valuable information for predicting vegetation LAI, canopy height and AGB using the combined hyperspectral imagery and pseudo-waveform constructed from discrete-return LiDAR data.