Remote sensing monitoring of winter wheat growth using physico-chemical composite parameter and neural network

Accurate and timely regional crop growth monitoring will be of great benefit to the establishment and adjustment of agricultural planning and policies. Remote sensing technology as an effective measure for collecting crop growth information across large areas has been receiving increasing attention nowadays. To enhance the accuracy and comprehensiveness of remote sensing monitoring winter wheat growth, a physico-chemical composite parameter (PCCP) was constructed using field measurements of aboveground fresh biomass (AFB), leaf area index (LAI), soil and plant analyzer development (SPAD) and leaf nitrogen content (LNC) of winter wheat at jointing stage. This construction was achieved through the utilization of the entropy weight method (EWM). Based on individual and community characteristics, winter wheat growth at jointing stage was divided into 3 levels in the study area, which were poor (Ⅰ), medium (Ⅱ) and well (Ⅲ). On this basis, the differences of each parameter under different growth levels were evaluated by Kruskal-Wallis test. To further validate the reliability of the composite parameter, linear regression models between grain yield of winter wheat and each parameter were constructed. Then, Sentinel-2A was used as the data source to analyze the correlation between different remote sensing indexes and LAI、SPAD、AFB、LNC、PCCP of winter wheat at jointing stage. Remote sensing indexes with high correlation were selected as inputs of back propagation (BP) artificial neural networks (ANN) to estimate PCCP. The 10-fold cross-validation method was used to obtain the optimal parameters of the BP-ANN model. The best model was selected to simulate values of the PCCP and to map the regional winter wheat growth conditions pixel by pixel at jointing stage. The weighting results showed that the weight of crop physical parameters was greater than biochemical parameters, among which LAI had the largest weight (0.387), followed by AFB and SPAD, and LNC had the least weight (0.105). The performance evaluation results of PCCP showed that the difference of PCCP under different growth levels was the most significant. The correlation between PCCP value and grain yield was closer than that between grain yield and LAI, SPAD, AFB, or LNC alone. The coefficient of determination was increased by 0.035 to 0.468, and the root-mean-square error is reduced by 46.2 kg/hm² to 520.0 kg/hm². During remote sensing monitoring, the correlation among the PCCP constructed by LAI, SPAD, AFB, and LNC and remote sensing indexes were all improved to different degrees compared with the single parameter. The accuracy of PCCP simulation by BP-ANN remote sensing monitoring model was high, which the coefficient of determination and the root mean square error were 0.830 and 0.080 in the test set, respectively. The overall growth of winter wheat at jointing stage in the study area was stable and concentrated, showing the spatial distribution characteristics of "the middle bad and the north-south well". Therefore, the construction of PCCP is an effective way to improve the reliability and accuracy of growth remote sensing monitoring, which can provide scientific basis for field management of winter wheat and serve the strategic needs of developing intelligent agriculture and building an agricultural power in China.