Leaf area index (LAI) and biomass are important indicators that reflect the growth status of maize. The optical vegetation indices and the synthetic-aperture radar (SAR) backscattering coefficient are commonly used to estimate the LAI and biomass. However, previous studies have suggested that spectral features extracted from a single pixel have a poor ability to describe the canopy structure. In this paper, we propose a method for estimating LAI and biomass by combining spectral and texture features. Specifically, LAI, biomass and remote-sensing data were collected from the jointing, trumpet, flowering, and filling stages of maize. Then we formed six remote-sensing feature matrices using the spectral and texture features extracted from the remote sensing data. Principal component analysis (PCA) was used to remove noise and to reduce and integrate the multi-dimensional features. Multiple linear regression (MLR) and support vector regression (SVR) methods were used to build the estimation models. Tenfold cross-validation was adopted to verify the effectiveness of the proposed method. The experimental results show that using the texture features of both optical and SAR data improves the estimation accuracy of LAI and biomass. In particular, SAR texture features greatly improve the estimation accuracy of biomass. The estimation model constructed by combining spectral and texture features of optical and SAR data achieves the best performance (highest coefficient of determination (R2) and lowest root mean square error (RMSE)). Specifically, we conclude that the best window sizes for extracting texture features from optical and SAR data are 3×3 and 7×7, respectively. SVR is more suitable for estimating the LAI and biomass of maize than MLR. In addition, after adding texture features, we observed a significant improvement in the accuracy of estimation of LAI and biomass for the growth stages, which have a larger variation in the extent of the canopy. Overall, this work shows the potential of combining spectral and texture features for improving the estimation accuracy of LAI and biomass in maize.

Heterogeneity exists in the vertical distribution of the biochemical components of crops. A leaf chlorophyll deficiency occurs in the bottom- and middle-layers of crops due to nitrogen stress and leaf senescence. Some studies used multi-angular remote sensing data for estimating the vertical distribution of the leaf chlorophyll content (LCC). However, these studies performed LCC inversion of different vertical layers using a fixed view zenith angle (VZA), but rarely considered the contribution of the components of the non-target layers to the spectral response. The main goal of this work was to determine the LCC of different vertical layers of the canopy of winter wheat (Triticum aestivum L.), using multi-angular remote sensing and spectral vegetation indices. Different combinations of VZAs were used for obtaining the LCC of different layers. The results revealed that the responses of the transformed chlorophyll in reflectance absorption index (TCARI) and modified chlorophyll absorption in reflectance index (MCARI)/optimized soil-adjusted vegetation index (OSAVI) to the upper-layer LCC were strongest at VZA 10°. For the middle-layer LCC, the response was strongest at 30°, but the response was significantly lower than that of the upper-layer. For the bottom-layer LCC, the responses were weak due to the obscuring effect of the upper- and middle-layer; thus, the LCC inversion of the bottom-layer data was not optimal for a single VZA. The optimal VZA or VZA combinations for LCC estimation were VZA 10° for the upper-layer LCC (TCARI with coefficient of determination (R²) = 0.69, root mean square error (RMSE) = 4.80 ug/cm², MCARI/OSAVI with R² = 0.73, RMSE = 4.17 ug/cm²), VZA 10° and 30° for the middle-layer LCC (TCARI with R² = 0.17, RMSE = 4.81 ug/cm², MCARI/OSAVI with R² = 0.17, RMSE = 4.76 ug/cm²), and VZA 10°, 30°, and 50° for the bottom-layer LCC (TCARI with R² = 0.40, RMSE = 6.29 ug/cm², MCARI/OSAVI with R² = 0.40, RMSE = 6.36 ug/cm²). The proposed observation strategy provided a significantly higher estimation accuracy of the target layer LCC than the single VZA approach, and demonstrated the ability of canopy multi-angular spectral reflectance to accurately estimate the wheat canopy chlorophyll content vertical distribution.

Remote sensing approaches have several advantages over traditional methods in determining information on physical and chemical parameters, including timely data acquisition, low costs, and wide coverage. Thus, remote sensing is widely used in crop growth monitoring. Unlike vertical observations, multi-angular remote sensing technology can obtain the vertical distribution information of the central and lower leaves of a crop. Furthermore, applications of remote sensing on the vertical distribution of maize canopy components is complicated, and related research is limited. In the current paper, we employed multi-angular spectral data, measured by a self-designed multi-angular observation instrument at view zenith angles (VZAs) of 0°, 10°, 20°, 30°, 40°, 50°, and 60°, to explore the monitoring strategy and monitoring precision of the vertical distribution of chlorophyll content in the maize canopy. This was then used to determine the optimal monitoring method for the chlorophyll content (soil and plant analyzer development (SPAD) value) of each layer. The correlation between SPAD value and chlorophyll sensitivity indices at different growth stages was used as the basis for screening indices and VZAs. The correlation between the selected EPI (eucalyptus pigment index) and REIP (red edge inflection point) indices and chlorophyll content indicated view zenith angles (VZAs) of 0°, 30°, and 40° as optimal for the early growth stage monitoring of chlorophyll content in the 1st, 2nd, and 3rd layers, respectively. These values were associated with RMSEs of 4.14, 1.71, and 1.11 for EPI, respectively; and 4.61, 2.31, and 1.00 for REIP, respectively. In addition, a VZA of 50° was selected to monitor the chlorophyll content of the 1st, 2nd, 3rd, and 4th layers at the late growth stage, with RMSE values of 2.97, 3.50, 2.80, and 4.80 for EPI, respectively; and 3.16, 5.02, 4.55, and 7.85 for REIP, respectively. The results demonstrated the ability of canopy multi-angular spectral reflectance to accurately estimate the maize canopy chlorophyll content vertical distribution, with the VZAs of different vertical layers varying between the early and late growth stages.

Accurate estimation of the leaf or canopy chlorophyll content is crucial for monitoring crop growth conditions. Remote sensing monitoring of crop chlorophyll is a non-destructive, large-area, and real-time method that requires reliable retrieval models and satellite data. High-resolution satellite imagery generally has better object recognition capabilities. However, the influence of the spectral and spatial resolution of medium- and high-spatial-resolution satellite imagery on chlorophyll retrieval is currently unexplored, especially in conjunction with radiative transfer models (RTMs). This has important implications for the accurate quantification of crop chlorophyll over large areas. Therefore, the objectives of this study were to establish an RTM for the retrieval of maize chlorophyll and to compare the chlorophyll retrieval capability of the model using medium- and high-spatial-resolution satellite images. We constructed a hybrid model consisting of the PROSAIL model and the Gaussian process regression (GPR) algorithm to retrieve maize leaf and canopy chlorophyll contents (LCC and CCC). In addition, an active learning (AL) strategy was incorporated into the hybrid model to enhance the model’s accuracy and efficiency. Sentinel-2 imagery with a spatial resolution of 10 m and 3 m-resolution Planet imagery were utilized for the LCC and CCC retrieval, respectively, using the hybrid model. The accuracy of the model was verified using field-measured maize chlorophyll data obtained in Dajianchang Town, Wuqing District, Tianjin City, in 2018. The results showed that the AL strategy increased the accuracy of the chlorophyll retrieval. The hybrid model for LCC retrieval with 10-band Sentinel-2 without AL had an R² of 0.567 and an RMSE of 5.598, and the model with AL had an R² of 0.743 and an RMSE of 3.964. Incorporating the AL strategy improved the model performance (R² = 0.743 and RMSE = 3.964). The Planet imagery provided better results for chlorophyll retrieval than 4-band Sentinel-2 imagery but worse performance than 10-band Sentinel-2 imagery. Additionally, we tested the model using maize chlorophyll data obtained from Youyi Farm in Heilongjiang Province in 2021 to evaluate the model’s robustness and scalability. The test results showed that the hybrid model used with 10-band Sentinel-2 images achieved good accuracy in the Youyi Farm area (LCC: R² = 0.792, RMSE = 2.8; CCC: R² = 0.726, RMSE = 0.152). The optimal hybrid model was applied to images from distinct periods to map the spatiotemporal distribution of the chlorophyll content. The uncertainties in the chlorophyll content retrieval results from different periods were relatively low, demonstrating that the model had good temporal scalability. Our research results can provide support for the precise management of maize growth.

Accurate estimation of the maize leaf area index (LAI) and biomass is of great importance in guiding field management and early yield estimation. Physical models and traditional machine learning methods are commonly used for LAI and biomass estimation. However, these models and methods mostly rely on handcrafted features and theoretical formulas under idealized assumptions, which limits their accuracy. Deep neural networks have demonstrated great superiority in automatic feature extraction and complicated nonlinear approximation, but their application to LAI and biomass estimation has been hindered by the shortage of in situ data. Therefore, bridging the gap of data shortage and making it possible to leverage deep neural networks to estimate maize LAI and biomass is of great significance. Optical data cannot provide information in the lower canopy due to the limited penetrability, but synthetic aperture radar (SAR) data can do this, so the integration of optical and SAR data is necessary. In this paper, 158 samples from the jointing, trumpet, flowering, and filling stages of maize were collected for investigation. First, we propose an improved version of the mixup training method, which is termed mixup⁺, to augment the sample amount. We then constructed a novel gated Siamese deep neural network (GSDNN) based on a gating mechanism and a Siamese architecture to integrate optical and SAR data for the estimation of the LAI and biomass. We compared the accuracy of the GSDNN with those of other machine learning methods, i.e., multiple linear regression (MLR), support vector regression (SVR), random forest regression (RFR), and a multilayer perceptron (MLP). The experimental results show that without the use of mixup${}^{+}$, the GSDNN achieved a similar accuracy to that of the simple neural network MLP in terms of $R^2$ and RMSE, and this was slightly lower than those of MLR, SVR, and RFR. However, with the help of mixup${}^{+}$, the GSDNN achieved state-of-the-art performance ($R^2$ = 0.71, 0.78, and 0.86 and RMSE = 0.58, 871.83, and 150.76 g/m${}^2$, for LAI, Biomass_wet, and Biomass_dry, respectively), exceeding the accuracies of MLR, SVR, RFR, and MLP. In addition, through the integration of optical and SAR data, the GSDNN achieved better accuracy in LAI and biomass estimation than when optical or SAR data alone were used. We found that the most appropriate amount of synthetic data from mixup${}^{+}$ was five times the amount of original data. Overall, this study demonstrates that the GSDNN + mixup${}^{+}$ has great potential for the integration of optical and SAR data with the aim of improving the estimation accuracy of the maize LAI and biomass with limited in situ data.