Monitoring maize growth conditions by training a BP neural network with remotely sensed vegetation temperature condition index and leaf area index

Abstract

Crop water stress and vegetation status are critical parameters and should be proposed as input variables of an integrated model for crop productivity and yield estimation. In this study, to improve the monitoring of the regional maize growth conditions in the North China Plain, PR China, the remotely sensed vegetation temperature condition index (VTCI) and leaf area index (LAI) at five growth stages of maize (the seeding, jointing, heading, milk and mature stages) during 2010–2015 were generated as inputs of three-layer back propagation (BP) artificial neural networks (ANNs) with different numbers of nodes in the hidden layer to estimate the crop growth. Among these BP ANN models, an architecture with 12 nodes in the hidden layer provided the best training (RMSE = 755.7 kg/ha, MSE = 0.023) and testing (RMSE = 644.3 kg/ha, MSE = 0.037) performance and was selected to simulate values of the integrated growth monitoring index of maize (IGMIM) and to map the regional maize growth conditions pixel by pixel in the North China Plain during 2010–2018. The spatiotemporal characteristics displayed by the maize growth maps based on the IGMIM showed that the best year was 2016, the worst year was 2015, and maize growth in different parts of the plain varied accordingly with variations in the meteorological conditions. Thus, the information reflected by the IGMIM was in good agreement with the actual results. To further validate the accuracy of the integrated index, the correlations between the values of the IGMIM and several growth-related variables, including the measured yield, planting density, plant height and relative soil humidity at the 0–10 cm layer, at thirteen meteorological stations from 2010 to 2012 were analyzed, and the results were meaningful and presented a significant linear relationship. Thus, the BP ANN-based model has the ability to integrate information reflected by multiple maize growth-related factors at each growth stage and provides a better quantification of the monitoring results of regional maize growth conditions.