Search

• A novel model (LabTNet) is proposed to estimate crop biochemical traits. • LabTNet jointly estimates multiple wheat traits by hyperspectral data. • LabTNet exhibits superior predictive performance than existing methods. • Grad-CAM analyzed hyperspectral attention regions of LabTNet. Accurate estimation of crop leaf and canopy biochemical traits, such as leaf dry matter content (Cm), leaf equivalent water thickness (Cw), leaf area index (LAI), dry leaf biomass (DLB), leaf total water content (LW), and fresh leaf biomass (FLB), is essential for monitoring crop growth accurately. The vegetation spectral feature technique combined with statistical regression methods is widely employed for remote sensing crop biochemical traits mapping. However, the crop canopy spectral reflectance is influenced by various crop biochemical traits and uncertainties in geometric changes of light and soil background effects. Consequently, the remote-sensing estimation of crop biochemical traits is limited. A potential solution involves training a deep learning model to understand the physical relationship between crop biochemical traits and canopy spectral reflectance based on a physical radiative transfer model (RTM). The primary focus of this study is to propose a winter-wheat leaf and canopy biochemical traits analysis and mapping method based on hyperspectral remote sensing, utilizing a deep learning network for leaf area index and leaf biochemical traits deep learning network (LabTNet). This study consists of four main tasks: (1) Field-based measurements of winter-wheat spectra and biochemical traits were conducted in two growing seasons. A PROSAIL RTM was also employed to generate a simulated dataset representing comprehensive and complex winter-wheat field conditions. (2) The LabTNet deep learning model was pre-trained using the simulated spectra dataset to acquire knowledge of the physical relationship between crop biochemical traits and canopy spectral reflectance derived from the RTM. Subsequently, the model was re-trained using the field-based spectra dataset from two growing seasons, employing a transfer learning technique. (3) An analysis was conducted to assess the performance of LabTNet against traditional statistical regression methods in estimating crop leaf and canopy biochemical traits. The study used the gradient-weighted class activation mapping (Grad-CAM) technique to analyze the attention regions of input spectra (454:8:950 nm, 960:10:1300 nm, 1450:10:1750 nm, 2000:10:2350 nm) by different convolutional neural network layers in LabTNet, aiming to enhance the interpretability of deep learning models. (4) Winter-wheat leaf and canopy biochemical traits (Cw, Cm, LAI, DLB, LW, and FLB) were mapped using the LabTNet deep learning model. Our research has the following conclusions: (1) Combining the RTM and deep learning techniques yields higher winter-wheat biochemical trait estimates than traditional statistical regression methods. (2) Different LabTNet deep learning model layers focus on distinct areas of canopy reflectance, corresponding to the sensitive regions for various winter-wheat biochemical traits. (3) LabTNet demonstrates similar winter-wheat leaf and canopy biochemical traits estimation performance using visible and near-infrared (VNIR) reflectance data and full-spectral (FS) range hyperspectral reflectance as inputs (Cw: R 2 = 0.603–0.653, RMSE = 0.0015–0.0015 cm; Cm: R 2 = 0.511–0.560, RMSE = 0.0006–0.0007 g/m2; LAI: R 2 = 0.773–0.793, RMSE = 0.65–0.66 m2/m2; LW: R 2 = 0.842–0.847, RMSE = 67.93–70.73 g/m2; DLB: R 2 = 0.747–0.762, RMSE = 21.10–21.89 g/m2; FLB: R 2 = 0.831–0.840, RMSE = 86.26–90.30 g/m2). The combined use of UAV hyperspectral remote sensing and the LabTNet model proves effective in providing high-performance winter-wheat leaf and canopy biochemical trait maps, offering valuable insights for agricultural management. [ABSTRACT FROM AUTHOR]

• Introduction of a novel model that integrates OBIA and the attention mechanism for precise orchard mapping. • The contribution of SAR is studied in orchard classification. • The model obtains more accurate classification results than RNN-based and CNN-based deep learning models. • The uncertainty of the super-pixel size on classification accuracy is investigated. Reliable and accurate classification of orchards is important for the dynamic monitoring of large-scale orchards and food security evaluation. At present, the very similar spectral profiles of different fruit trees and the high susceptibility of optical data to interference by weather conditions limit the resolution of orchard classification. Synthetic Aperture Radar (SAR) imagery provides an advanced solution to this problem, with the advantages of being immune to weather conditions and advances in deep learning techniques. This paper presents an orchard classification model (STCM) with optical and SAR fusion, which integrates the advantages of the Simple Non-Iterative Clustering (SNIC) super-pixel algorithm with a deep learning algorithm based on a multi-headed attention mechanism, and it achieves the best classification accuracy of above 0.82 in comparison with existing deep learning models. Meanwhile, the model has exceptional classification accuracy and robustness in a study area with fragmented plots, many fruit tree species, and various distributions of optical images. In the absence of optical data, the classification accuracy of the STCM model using only SAR data is around 0.70, which makes the model have a promising potential application value. This study provides a technical solution for accurately obtaining different orchard categories in high-resolution remotely sensed orchard images. The study helps to improve the management and operation of orchards and provides a strong basis for decision-making in the fruit industry to enhance the sustainability of the global fruit industry. [ABSTRACT FROM AUTHOR]

• We introduce leaf N profile characteristic coefficient K m on its remote estimates. • Diverse datasets are used for developing robust models for remote estimates of K m. • Four regression algorithms are tested with various possible sets of predictors. • A robust method is proposed to estimate K m from canopy reflectance for winter wheat. The characteristic coefficient of vertical leaf nitrogen (N) profile is a canopy parameter that indicates the attenuation steepness of leaf N from the top of canopy downward. It is sensitive not only to crop production, grain yield and quality, and light- and N-use efficiency but also to N deficiency in the crop. We introduce this coefficient by exploring a robust method to estimate it from canopy spectral reflectance. We analyze comprehensively various approaches based on diverse datasets of winter wheat. We test and compare the accuracy and stability of models by using the adjusted and weighted coefficient of determination (wR adj 2), the mean absolute error (MAE), and the mean and coefficient of variation (CV) over multiple seasons. The analysis focuses mainly on the coefficient of mass-based leaf N profile (K m); nevertheless, a comparison with the coefficient of area-based leaf N profile is presented. The results indicate that the most robust model to estimate K m of winter wheat is K m = (1.8037 R G V I - 0.9702 + 0.0786 e x p (0.6315 / D A S F)) / 2 , where RGVI is the red and green ratio vegetation index, and DASF is the directional area scattering factor. The mean wR adj 2 is 0.663 with CV = 8.2% and the mean MAE is 0.117 with CV = 12.8% over three seasons including various situations. This makes it possible to assess K m at large areas and follow its dynamics over multiple periods in a timely and nondestructive manner. [ABSTRACT FROM AUTHOR]

• Multispectral data are effective in discriminating tea plant varieties. • Texture information are essential elements in discriminating tea plant varieties. • Data fusion of multispectral image can improve tea plant discrimination rate. Discrimination of tea plant (Camellia sinensis L.) varieties is of significant value for the efficient management of cultivation and resources optimization of tea industry. Phenotypic and spectral characteristics of tea plants are important indicators for determining the quality of tea cultivars to a certain extent. However, few studies have used spectral image information to identify tea plant varieties. The aim of this research was the discrimination of 16 types of high-yield tea plant varieties using a multispectral camera. This methodology involved image registration, calibration, segmentation, information extraction, and data fusion. The hue (H), saturation (S), and value (V), texture information, and several spectral vegetation indices were acquired from the multispectral image of the tea plant canopy. The successive projection algorithm (SPA) was used to analyze the original parameters. Three classification methods were applied to tea plant variety discrimination: Bayes discriminant analysis (BDA), support vector machine (SVM), and extreme learning machine (ELM). The results indicated that SPA based on fusing data combined with the SVM classification model, achieved a feasible method to identify tea plant varieties. Additionally, the method achieved accuracy in the training, test, and validation sets, reaching 97.00%, 90.52%, and 88.67%, respectively. This study proposed a new perspective on multispectral image information as an identifier of tea plant varieties. The explored model will be helpful for the development of portable instruments for commercial applications in variety identification and phenotype recognition of tea plants. [ABSTRACT FROM AUTHOR]

• Vegetation indices show a strong correlation with nitrogen status in tea plants. • The combination of multispectral data and hyperspectral data is effective in monitoring nitrogen level. • Fusing data of image information and spectral information has remarkable ability to predict nitrogen content in tea plants. Nitrogen (N) plays a pivotal role in management of tea plantation, with significant impacts on the growth, productivity, and nutrition status of tea plants. The existing methods for N content monitoring of tea leaves are complicated and can not realize in suite and in real time way. This study proposed a method for estimating the N content of tea plants in field conditions based on a combination of a multispectral imaging system and hyperspectral data. A total of 32 parameters were extracted from five tea gardens using calibrated multispectral images of the tea plant canopy, and 27 indices were selected by Pearson correlation analysis. A total of 28 wavelengths selected by competitive adaptive reweighted sampling from hyperspectral data were combined with 27 multispectral indices as the original data. Subsequently, five variables of fused data (H, VOG, BGI, 1664 nm and 1665 nm) were selected by variable combination population analysis based on the 55 combination parameters. Partial least squares regression, random forest regression, and support vector machine regression (SVR) models all showed excellent performance for both the calibration and prediction sets. The overall results indicated that the infused data of multispectral and hyperspectral data combined with SVR are effective in monitoring the N level under field conditions, and the R2 (coefficient of determination) and root mean square error values of the prediction were 0.9186 and 0.0560, respectively. The findings of this study are important in retaining the nutritional and quality attributes of agricultural commodities. [ABSTRACT FROM AUTHOR]

• Accurately forecast the first flowering period of two typical apple producing areas in China. • It fills in the research of forecasting the first flowering period of apple from point scale to regional scale. • The phenological survey method of replacing time with space is very effective for the calibration of sequential model. • The prediction of the apple first flowering on a regional scale has important application value for optimizing orchard management. China is one of the largest apple-producing countries in the world, with large orchards and diverse climates. Accurately forecasting the first-flowering time of apple trees can assist orchard managers in their deciding when to apply anti-freeze. The temperature-driven sequential model from previous studies can be used to forecast the flowering phenology of deciduous fruit trees. However, this model requires many years of observational data for calibration, so flowering forecasts based on traditional phenological models cannot be implemented in areas that lack such historical data. To overcome this problem, the present work combines a spatial rather than a temporal phenological survey method with 1-km-gridded temperature products to calibrate the chill and heat requirement parameters of the sequential model. We then use the model to forecast the first-flowering on a regional scale for Luochuan and Linyi, which are two main apple-producing areas of China. The results show that the proposed method accurately forecasts regional flowering. The root mean squared errors (RMSE) for Luochuan and Linyi were 4.7 and 4.4 days, respectively, and the normalized RMSEs were all less than 5.19%. We expect the proposed regional first-flowering forecast method to be an important aid to optimize orchard management. [ABSTRACT FROM AUTHOR]

• Synthetic Minority Oversampling Technique and Edited Nearest Neighbors are applied. • A SMOTE-ENN-XGBoost model is developed to detect lodging in imbalanced datasets. • The developed model is explained with SHAP values. • Deeper insight is obtained into how the model behaves with these explanations. • Complex and interrelated impacts of the selected features are captured and analyzed. Remote sensing image is becoming an increasingly popular tool for crop lodging detection because it conveniently provides features for building machine learning models and predicting lodging. However, difficulties in interpreting machine learning models and their predictions limit the confidence of using remote sensing images to detect lodging. In addition, the lodging datasets used for modeling are difficult to balance under natural conditions. Designing a robust and interpretable classification model for the detection of lodging in an imbalanced distribution dataset poses a particularly difficult challenge. In this study, visible and multi-spectral images were collected with a UAV to extract relevant features from remote sensing images. In a preliminary step, Synthetic Minority Oversampling Technique (SMOTE) and Edited Nearest Neighbors (ENN) method were used to treat imbalanced datasets. The SMOTE-ENN-XGBoost model is proposed for the efficient identification of maize lodging at the plot scale. The SMOTE-ENN-XGBoost model achieved an F1-score of 0.930 and a recall of 0.899 on a testing set, suggesting that it can be used for modeling lodging detection. Additionally, the SHapley Additive exPlanations (SHAP) approach was employed to interpret the identification and prioritization of features that determine lodging classification and activity prediction. The results showed that canopy structure and textural features are relatively stable compared with spectral features, which are susceptible to the external environment when modeling is employed to detect lodging. This work also showed that canopy structural, spectral, and textural information should be considered simultaneously rather than separately when detecting crop lodging in a crop breeding program in order to prevent differences in expression controlled by the interaction between genotype and environment obscuring the change in a single feature before and after lodging. For practical applications of machine learning models in crop lodging detection, such insights are of critical relevance. Taken together, the results of this study encourage further applications of remote sensing techniques to build interpretable machine learning models. [ABSTRACT FROM AUTHOR]

• Determined the optimal UAV wide-band vegetation indices (VIs) for LAI and LCC retrievals. • Analyzed the sensitivity of VIs used for retrieval to influencing factors. • Constructed a hybrid inversion method of LAI and chlorophyll for individual apple tree crowns. • Quantitatively characterized the effect of canopy shading on inversion. The structural and chemical characteristics of individual apple tree crowns can indicate the nutritional and growth status of the trees, making them crucial for advancing orchard management practices. In this study, we collected multispectral imagery and ground validation data from two representative apple orchards in Beijing, China. We employed a hybrid inversion method to estimate the Leaf Area Index (LAI), Leaf Chlorophyll Content (LCC), and Canopy Chlorophyll Content (CCC) of individual apple tree crowns. Furthermore, we quantitatively evaluated the impact of canopy shading on the inversion results and mapped these traits at the scale of individual tree crowns (ITCs). To determine the optimal broad-band Vegetation Indices (VIs) for estimating LAI and LCC, we empirically analyzed 22 VIs using the PROSAIL simulation dataset. We constructed two measured datasets of canopy reflectance by masking canopy shadows, one containing shaded pixels and the other consisting of sunlit-only pixels. Using the Artificial Neural Network (ANN) algorithm and PROSAIL model, we developed a hybrid inversion model to assess the performance of the filtered VIs on the two measured datasets. The results demonstrated that TCARI/OSAVI and SR3 were the most accurate VIs for estimating LAI (including shaded pixels: R2 = 0.67, RMSE = 0.31 m2/m2; sunlit-only pixels: R2 = 0.74, RMSE = 0.28 m2/m2) and LCC (including shaded pixels: R2 = 0.70, RMSE = 7.11 μg/cm2; sunlit-only pixels: R2 = 0.73, RMSE = 6.63 μg/cm2) in the two measured reflectance datasets, respectively. Removing canopy shadows significantly improved the accuracy of LAI and LCC retrieval, although there was no significant difference in CCC retrieval accuracy (including shaded pixels: R2 = 0.78, RMSE = 31.25 μg/cm2; sunlit-only pixels: R2 = 0.79, RMSE = 28.48 μg/cm2). Moreover, we utilized UAV imaging multispectral data to map the estimated variability of leaf and canopy traits. The results revealed trait variability among different apple tree canopies, highlighting the potential of UAV imaging multispectral techniques in characterizing and mapping individual apple tree crown traits while capturing variability among crowns. We recommend performing canopy shading pixel masking to enhance the accuracy of ITCs trait retrieval. [ABSTRACT FROM AUTHOR]

• The averaged dry stem mass content (C sm) is defined. • The crop leaf dry matter content (C m), height (C h), and density (C d) were considered. • C sm , C h , and C d improved vertically growing crop stem AGB (VGC-AGB) estimation. • The VGC-AGB = LAI C m + C d C h C sm did not saturate at medium-to-high crop covers. The accurate estimation of crop above-ground biomass (AGB) can assist in crop growth monitoring and grain yield prediction. Remote sensing has been widely used for AGB estimation at regional and local scales in recent years. However, optical remote sensing spectral indices (SIs) become saturated at medium-to-high crop covers. The combined use of remote sensing techniques and statistical regression models is not based on an understanding of how crop leaves and vertical organs contribute to the crop AGB. This causes difficulties in measuring the biomass stored in vertical organs (e.g., plant stem, wheat-spike, maize-tassel; abbreviated as AGB v) using optical remote sensing. This study aims to develop an unmanned aerial vehicle (UAV)-based vertically growing crop AGB (VGC-AGB) model. We defined C sm (g/m) to describe the crop stem and reproductive organs' average dry mass content. This was done to improve the estimation of AGB v. The crop leaf area index (LAI, m2/m2), leaf dry matter content (C m , g/m2), height (C h , m), and density (C d , m−2) were used in the VGC-AGB. The VGC-AGB calculated crop leaf AGB (AGB l) using LAI × C m (g/m2) and AGB v using C d × C h × C sm (g/m2). The proposed VGC-AGB (AGB = LAI × C m + C d × C h × C sm) was verified using field and UAV-based hyperspectral datasets of winter-wheat and summer-maize at three growth stages. Our results indicate that UAV-based VGC-AGB (R 2 = 0.92–0.93, RMSE = 68.82–75.15 g/m2) is superior to the statistical regression model that is based on remote sensing SIs and CSMs (R 2 = 0.77, RMSE = 134.94 g/m2). The results indicate that the UAV-based VGC-AGB supports the analysis of crop photosynthetic product transfers and high-performance UAV-based high-performance non-destructive AGB monitoring. [ABSTRACT FROM AUTHOR]

• Current soil moisture indices (SMIs) are sensitive to VWC. • Vegetation-water Resistant SMI (VRSMI) insensitive to VWC. • VRSMI incorporates a physical relation between NIR and SWIR reflectance. • A linear relationship is found between soil VRSMI and soil moisture. • VRSMI supports SMC estimation under low-moderate crop cover. Soil surface moisture content (SMC) is one of the most critical and fundamental indicators of crop drought because water is vital for vegetation growth. In recent decades, satellite high-spatial-resolution optical multi-spectral remote-sensing sensors have entered a new stage. However, current optical remote sensing SMC indices (SMIs) are sensitive to cropland SMC and vegetation water content (VWC). From dry to saturated, the soil spectra decreased quickly as the SMC increased. For the vegetation spectra, the short-wave infrared (SWIR) bands decreased as the VWC increased. The differences in the responses of near-infrared (NIR) and SWIR bands to water may help distinguish soil moisture from mixed soil-vegetation spectra. This study focuses on (a) analyzing how soil and vegetation reflectance are affected by VWC and SMC, (b) designing a new vegetation-water resistant soil moisture index (VRSMI) based on the differences in the responses of NIR bands and SWIR bands to water, and (c) evaluating the potential of using VRSMI to remotely estimate SMC in low-moderate crop-covered regions. VRSMI uses SWIR1-SWIR2 as the numerator to detect the water absorption features and NIR×(SWIR1)0.5 as the denominator to weaken the vegetation effect. VRSMI and soil-VRSMI (VRSMIs) were tested using laboratory- and field-based spectra datasets. Our study presents the following conclusions: (1) The differences in the responses of NIR and SWIR bands to water help distinguish the effects of soil moisture on mixed soil-vegetation spectra. (2) VRSMIs can provide high-performance SMC estimates under low-moderate vegetation cover (NDVI < 0.55). (3) VRSMIs can be used for cropland SMC evaluation without considering crop VWC (laboratory-based spectra dataset: RMSE = 0.07–0.11, R 2 = 0.96–0.98; field-based spectra dataset: RMSE = 0.17–0.18, R 2 = 0.63–0.79). Our study indicates that VRSMI and VRSMIs can provide high-performance relative SMC estimates under low-to-moderate vegetation cover. [ABSTRACT FROM AUTHOR]

• Fusion of remote sensing and meteorological data to estimate maize BBCH. • Real-time phenological stage from single time-phase imagery is possible. • Phenological differences and growth rates in breeding plots were found. • Model is resilient to fluctuations in sowing date. Real-time crop phenological information can provide crucial guidance for field management, agricultural machinery scheduling, and other agricultural activities. However, in previous research, phenological monitoring is often done post-seasonally, which typically entails a certain degree of lag. To overcome this, here we combined time-series UAV data, image-derived structural information, and cumulative temperature required for maize growth to explore the mapping relationships between multiple data and maize phenology (Bundessortenamt and CHemical Industry scale, BBCH). Then we developed a model for phenology's real-time monitoring for applications requiring only single-time-phase UAV imagery and a single environmental factor (cumulative temperature). Specifically, a composite index was built using the one-to-one multiplication of vegetation index (VI), structural features (SF), and relative growing degree-days (RGS). Finally, the real-time monitoring model of maize BBCH was constructed via the ordinary least squares (OLS) fitting method. The results reveal the DATT-RGS model performs best, showing significant advantages over other combinations (VI-PH, VI-CC, CC-RGS, and PH-RGS), with R2, RMSE, NRMSE, and RMSEd values of 0.92, 7.66, 13.57, and 38.14 d, respectively. Plant phenology is a combined response outcome to genotype, field management, and regional environment; and while a VI can indicate the maize genotype and mode of field management, cumulus temperature is indicative of the regional environment and hence more mechanistic. Moreover, fluctuations in the sowing date had less of an effect on VI-RGS. However, when meteorological data is unavailable, the VI-PH model is recommended. Further, the VI-RGS model is able to determine the phenological differences and growth rates of maize in various breeding plots. This study presents new insights for the real-time monitoring of phenology from single-time-phase UAV imagery, providing timely crop phenotypic information for enhancing maize field management and smart breeding. The findings also offer technical support for the identification and selection of maize varieties. [ABSTRACT FROM AUTHOR]

• Rapidly constructing 3D scenes of maize canopy based on key structural parameters. • Combining 3D scenes with the 3D radiative transfer model to evaluate spectral response. • Analyzing the effect of leaf area and leaf base angle on canopy spectra. Three-dimensional (3D) scenes of a maize canopy can be utilized to depict its vertical structure and serve as the foundation for a 3D radiative transfer model. Given the inherent heterogeneity of canopy structures, evaluating their spectral response is crucial for remote-sensing inversion algorithms. Yet existing methods to model 3D canopy scenes of crops lack the high efficiency required for large-scale breeding applications. Hence, this study aims to develop a method to rapidly construct 3D canopy scenes of maize based on key structural parameters—namely, leaf area, base angle, and inclination angle—across multiple cultivars and growth stages. The accuracy of this canopy scenes modeling method was validated by comparison with a fine-scale 3D model and a multi-growth stage reflectance dataset. The root mean square error (RMSE) and normalized root mean square error (NRMSE) between the fine-scale 3D model and the constructed 3D model were less than 0.021 and 6.4%, respectively. Moreover, the RMSE and NRMSE between the simulated reflectance and measured reflectance were less than 0.042 and 11.1%, respectively. Next, we integrated the 3D radiative transfer model with 3D scenes to analyze the spectral response of maize canopy structure. The findings revealed that the near-infrared band was affected more by leaf area than leaf base angle, while the opposite is observed for the red-edge band. Furthermore, the middle layer contributed more to the canopy reflectance than either the upper or lower layers. Notably, two maize scenes where two structural parameters simultaneously change could elicit the same canopy spectra. The proposed method thus offers the dual advantages of rapid modeling and high efficiency, making it highly applicable for maize and other similar crops. [ABSTRACT FROM AUTHOR]

• Elucidated the possibility of using above-ground parameter to monitor yield of potato. • HLM was used to establish connections among VIs, EVs, and potato yield. • EVs successfully corrected inter-annual and regional variability in VIs-yield relationship. Timely monitoring of potato yield status is crucial for guiding field management and ensuring the sustainability of food trade. In recent years, non-destructive remote sensing techniques have emerged as one of the most promising methods for crop yield monitoring. However, most existing potato yield estimation models have primarily been tested at specific growth stages within a single ecosystem, highlighting the urgent need for a model that can be applied across different ecological environments. In this study, we developed a semi-physical model based on hierarchical linear modeling (HLM) to automatically correct yield estimation based on remotely sensed vegetation indices using environmentally relevant variables such as net solar radiation at the surface (ssr), air temperature at 2 m above the surface (t2m), and soil water content in the root zone (swvl). The accuracy and transferability of the model were further evaluated by comparing it with aboveground destructive sampling methods, conventional empirical models, and machine learning approaches. The results demonstrated that site-specific relationships exist between aboveground agronomic traits parameters like biomass and leaf area index (LAI) with respect to yield estimation. Conventional linear statistical and machine learning methods faced challenges in transferring their constructed yield estimation models across regions and years. However, the HLM approach exhibited excellent generalization ability across all trial samples. In the middle and later stages of potato growth, the standardized LAI determining index proved to be the most effective vegetation for predicting yield, while ssr, skt max , and shallow swvl were identified as key environmental variables affecting yield. By integrating these variables using HLM modeling techniques, we achieved optimal generalization performance (R2 = 0.57, RMSE = 7.65 t/ha, NRMSE = 25.02 % during tuber growth stage; R2 = 0.60, RMSE = 7.37 t/ha, NRMSE = 24.13 % during starch accumulation stage). This study underscores the importance of considering environmental factors when constructing a reliable model for estimating potato yield at different spatial and temporal scales using remote sensing technology. [ABSTRACT FROM AUTHOR]

• Meteorological data was introduced to build the HLM. • An AGB model for different regions was proposed and validated. • HLM models provided better AGB estimation at different regions. Above-ground biomass (AGB) is an important agronomic indicator that reflects crop growth and estimates yield. The AGB estimation using remote sensing becomes a non-destructive, rapid, and alternative method to post-harvest laboratory measurements. However, most of the AGB estimation models constructed based on remote sensing data are difficult to expand regionally, which limits the applicability of the models. This study combined ground-based hyperspectral and meteorological data by using a hierarchical linear modeling (HLM) to construct an AGB estimation model that was generalized across different regions. Experimental data from both regions were acquired and validated, namely from Xiaotangshan Experimental Base, Beijing, 2019 (North China) and Keshan Farm, Qiqihar Branch, Heilongjiang General Bureau of Reclamation, 2022 (Northeast China). Compared to OLS, RFR and GRU, the HLM method was better for estimating potato AGB in different regions with R2 = 0.54, RMSE = 429.62 kg/hm2, NRMSE = 28.20 %. The results of this study demonstrated that HLM could be used as a powerful method to improve the transferability of AGB estimation at different regions. [ABSTRACT FROM AUTHOR]

• The optimized vegetation indices (Opt-VIs) could resist spectral saturation. • Sensitive wavelengths in response to potato AGB were 734 and 742 nm. • Opt-VIs combined with texture features improved potato AGB estimation. Above-ground biomass (AGB) estimation is critical for monitoring crop growth and assessing yields. Unmanned aerial vehicle (UAV) optical remote sensing technology offers robust support for crop AGB estimation through vegetation indices (VIs). However, under conditions of high nitrogen or high AGB, most VIs lose their response to the presence of a dense plant canopy. To address the inaccuracy of estimating crop multistage AGB based on VIs formed in two bands, a UAV imaging hyperspectral experiment was conducted on potato test plots with different varieties, planting densities, and fertilizer application gradients. The ratio vegetation index (RVI) and normalized difference vegetation index (NDVI) were initially modified using a lambda-by-lambda optimization algorithm, which yielded the optimized RVI (Opt-RVI) and NDVI (Opt-NDVI). Their performances were compared with published VIs. Then, the corresponding bands from optimized VIs were used to extract textural features and their estimation performances were evaluated. Finally, based on published VIs, Opt-VIs, textures of sensitive bands, and Opt-VIs combined with the textures, the potato multistage AGB was estimated by optimized random forest (RF) regression. The RF estimation model was simplified according to variable importance scores. Results showed the spectral regions sensitive to AGB were mainly located in the red-edge range. The wavelengths of Opt-RVI and Opt-NDVI that exhibited the strongest correlation with AGB were 734 and 742 nm. Except for the normalized red-edge index formed by two red-edge bands, the published VIs (R2 = 0.07–0.28) demonstrated saturation in the context of high AGB. The optimized Opt-RVI and Opt-NDVI suppressed this phenomenon, such that R2 values for both reached 0.44 for estimating AGB. The R2 values of textural features relative to AGB for both bands ranged from 0.06 and 0.31, and CON 734 , COR 734 , DIS 742 , ENT 742 , SEC 742 , and COR 742 exhibited the strongest correlations with AGB (R2 > 0.2). The performance of estimating potato AGB by a single indicator was sensitive texture features, Opt-VIs and published VIs from high to low and the estimation capacities of these models were limited. The combination of Opt-VIs and textural features of sensitive bands demonstrated the greatest estimation accuracy (R2 = 0.62 and RMSE = 293.08 kg/hm2). Based on variable importance scores, the accuracy of the simplified RF estimation model (four variables: Opt-RVI, CON 734 , ENT 742 , COR 742) slightly decreased (R2 = 0.59 and RMSE = 301.01 kg/hm2), but model complexity was reduced and computational efficiency was improved. The simplified model accuracy was mainly affected by potato variety, fertilizer gradient, and growth stage, but not by planting density. The results of this study can be used as a reference for potato growth monitoring in the field. [ABSTRACT FROM AUTHOR]

• Deep features were extracted by combining SPA and LSTM. • The AGB estimation performance of spectral and deep features was compared. • The accuracy of the SPA-LSTM model across years and growth stages was evaluated. The accurate estimation of above ground biomass (AGB) is valuable in grasping the status of potato growth and assessing yield. Spectral analysis techniques play an important role in AGB estimation by virtue of its non-destructive and rapid advantages. However, the models constructed based on spectral features at the entire growth stages due to differences in phenological stages, such as vegetation indices (VIs) and locations at fixed wavelengths, can be not applicable to single growth stages. Therefore, the main objective of this study is to improve the applicability of the AGB estimation model to mitigate the impacts of phenological stages by mining depth features from hyperspectral data. The canopy hyperspectral and AGB data of potato tuber formation, tuber growth and starch accumulation stages were measured in 2017–2019. The acquired canopy hyperspectral reflectance was smoothed by Savitzky-Golay (SG) filter, and the green edge (GE), the red edge (RE), red edge chlorophyll index (CIred-edge), optimized soil adjusted vegetation index (OSAVI), green normalized difference vegetation index (GNDVI), and normalized difference vegetation index (NDVI) were extracted. The spectral changes were analyzed at different growth stages of each year. To obtain features in response to AGB changes, a new approach is proposed to analyze the hidden depth features between sensitive wavelengths by combining a successive projections algorithm (SPA) and a long-short-term memory network (LSTM). The AGB estimation models were constructed by partial least squares regression (PLSR) based on traditional VIs + GE + RE, full spectrum, sensitive wavelengths obtained by SPA, and depth features obtained by SPA-LSTM. The results showed that GE, RE and VIs alone explained only 24 %-36 % of the variation in AGB. The accuracy of estimating AGB at sensitive wavelengths obtained after SPA was higher (R2 = 0.70, RMSE = 406 kg/hm2, and NRMSE = 27.14 %) than that of conventional VIs + GE + RE (R2 = 0.51, RMSE = 536 kg/hm2, and NRMSE = 35.81 %) and full spectrum (R2 = 0.63, RMSE = 452 kg/hm2, and NRMSE = 30.19 %) in the three-year dataset. The depth features extracted by the SPA-LSTM method proposed in this study achieved the best AGB estimation (R2 = 0.82, RMSE = 318 kg/hm2, and NRMSE = 21.25 %). The applicability of the model was validated in different years and growth stages. The R2 of each year was in the range of 0.48–0.77, the RMSE was in the range of 290–367 kg/hm2, and the NRMSE was in the range of 16.69 %-25.81 %. The R2 of each growth stage within three year was in the range of 0.48–0.78, the RMSE was in the range of 184–386 kg/hm2, and the NRMSE was in the range of 14.91 %-29.45 %. The method proposed in this study mitigates the impacts of phenological stage and improves the robustness and accuracy of the AGB estimation model, which provides remote sensing technical support for realizing potato growth monitoring and yield assessment in the field. [ABSTRACT FROM AUTHOR]

• A hyperspectral data-processing framework was developed using Boruta and DNN. • Hyperspectral bands that can estimate the LNC of apple trees were explicitly identified. • Investigated the potential of phenological data to enhance apple LNC estimation. Rapid and accurate assessment of apple-tree leaf nitrogen content (LNC) plays an important role in smart orchards for accurate fertilization. Hyperspectral remote sensing technology provides a reliable, economical, and fast method to estimate LNC. However, high-dimensional spectral data contain redundant information that is irrelevant to the target parameters, necessitating the development of robust hyperspectral analysis techniques to estimate apple-tree LNC. Additionally, the role of phenological information in improving LNC estimates across multiple growth stages remains uncertain. The present study thus proposes a novel hyperspectral data-processing framework (Boruta-Iteration-DNN) to optimize estimates of apple-tree LNC and investigates the potential of phenological information to improve LNC estimates of multi-fertility apple trees. The proposed framework incorporates the Boruta feature-selection method for identifying important spectral bands and assessing how they affect model performance through iterative feature analysis. A deep neural network (DNN) regression model is then developed for LNC estimation. In addition, a widely used Pearson-DNN method is used for comparison to demonstrate the superiority of the proposed approach. Furthermore, the potential of incorporating phenological information (Day After Anthesis, DAA) into the estimation models is examined to evaluate its ability to better estimate LNC. The results demonstrate that the proposed framework not only enhances the accuracy of apple-tree LNC estimates with respect to the full-spectrum model (without DAA: validation R2 improved from 0.69 to 0.75, NRMSE reduced from 11.92 % to 9.97 %; with DAA: validation R2 improved from 0.72 to 0.79, NRMSE reduced from 11.09 % to 9.06 %), but also reduces the complexity and dimensionality of the models by eliminating 96 % of the redundant spectral bands (2015 bands). Moreover, the inferior results of the Pearson-Iteration-DNN framework further underscore the effectiveness of the Boruta method in selecting relevant spectral bands. Additionally, including the DAA phenological information enhances the accuracy of all models (the total R2 of training and validation increased by 0.05 ∼ 0.12, and the total NRMSE reduced by 1.49 % ∼ 1.93 %). These findings highlight the ability of DNN models integrated with hyperspectral data and phenological information to accurately assess LNC in apple trees, thereby offering valuable guidance to orchard management for developing precise fertilization strategies. [ABSTRACT FROM AUTHOR]

• Coupling HLM method with remote sensing to estimate rice yield and quality. • Propose a modeling methed combining remote sensing with meteorological data. • A model was proposed with a better scalability at interannual and regional scales. Timely and accurate estimation of crop yield and quality can provide a practical and effective basis for the formulation of national food policies. Based on remote sensing data of rice multi-growth stages from 2021 to 2022 in Ninghe District of Tianjin, the methods coupling hierarchical linear model (HLM) with typical vegetation indices and meteorological data were used to estimate yield and quality in rice. The results showed that: (1) Compared with the multiple linear regression model, the methods of estimating rice yield and quality based on HLM had higher stability in interannual expansion; (2) For the three growth stages of rice, namely, jointing, booting and filling stage, the hierarchical linear model of yield and quality estimation at booting stage showed the highest performance. The R^2 of the GRVI rice yield estimation model, MSR rice amylose content estimation model, and RDVI rice protein content estimation model were 0.52, 0.70, and 0.88, and RMSE were 1.17 t/ha, 0.85 %, and 0.33 %, respectively. The nRMSE was 11.06 %, 4.46 %, and 3.37 %, respectively. The HLM rice yield and quality estimation models based on multi-growth period data were further established. The GCI rice yield estimation model, MSR amylose content estimation model, and DVI protein content estimation model R^2 were 0.67, 0.75, and 0.91, and RMSE were 1.01 t/ha, 0.78 %, and 0.30 %. The nRMSE was 9.48 %, 4.07 %, and 3.05 %. It indicates that the method using HLM has a good potential for accessing yield and quality in rice and shows better scalability at interannual and regional scales. [ABSTRACT FROM AUTHOR]

• The AGB estimation performance of ASD and UHD185 was compared. • HD was employed to improve spectral sensitivity to AGB. • The AGB spatial distribution map was created. Accurately estimating potato above-ground biomass (AGB), which is closely associated with the growth and yield of crops, carries significant importance for guiding field management practices. Hyperspectral techniques have emerged as a powerful and efficient tool for quickly and non-invasively acquiring information about AGB due to its capability to provide rich spectral data closely related to crop physiology and biochemistry. However, using spectral features obtained from hyperspectral data, such as spectral reflectance and vegetation indices (VIs), often leads to inaccurate estimations of crop AGB at multiple growth stages due to spectral saturation effects and dynamic changes in spectral responses. To enhance the robustness of AGB estimation models, this study proposed a harmonic decomposition (HD) method derived from Fourier series to extract energy features. The ground (referred to as ASD) and unmanned aerial vehicle hyperspectral (referred to as UHD185) remote sensing data from three growth stages of potatoes in 2018 (validation set) and 2019 (calibration set) were utilized in the study. Firstly, a comparison was made between the spectral reflectance of the potato canopy measured by the ASD and UHD185 sensors. Subsequently, the correlation between spectral reflectance, VIs, and harmonic components obtained from ASD and UHD185 sensors was analyzed in relation to AGB at both the individual and whole growth stage. Then, sensitive bands selected through CARS (competitive adaptive reweighted sampling), the entire spectral reflectance, VIs, and harmonic components, were utilized to construct AGB estimation models by partial least squares regression (PLSR). Finally, the optimal model performance was validated across different years, growth stages, and treatment conditions. The results showed there were differences in spectral reflectance acquired by ASD and UHD185 sensors across various wavelengths, but overall, there was a high level of consistency between the two. The correlation of spectral reflectance and VIs with potato AGB at individual growth stage was notably higher than that observed for entire growth stages. The accuracy of AGB estimation using VIs obtained from ASD (the R2, RMSE and NRMSE of validation sets were 0.52, 592 kg/hm2 and 26.91 %, respectively) and UHD185 (the R2, RMSE and NRMSE of validation sets were 0.46, 612 kg/hm2 and 27.82 %, respectively) sensors were low. Utilizing sensitive bands and full spectral reflectance separately improved the precision of models, although the enhancement was somewhat limited. The HD-PLSR models from ASD (the R2, RMSE and NRMSE of validation sets were 0.69, 477 kg/hm2 and 21.69 %, respectively) and UHD185 (the R2, RMSE and NRMSE of validation sets were 0.66, 481 kg/hm2 and 21.86 %, respectively) achieved the best AGB estimation results. Using the HD-PLSR model to estimate AGB for two years, the R2 values were 0.79 and 0.76 for ASD and UHD185, with RMSE values of 381 kg/hm2 and 386 kg/hm2 and NRMSE values of 22.35 % and 22.70 %, respectively. The capability of the HD-PLSR model was confirmed at various growth stages and treatments. This work offers valuable remote sensing technical support for implementing potato growth monitoring and yield assessment in the field. [ABSTRACT FROM AUTHOR]

• Late Blight in Potato, a prevalent potato disease, significantly impacts potato yield and stands as one of the principal afflictions affecting the potato crop. Timely and effective monitoring of late blight severity in potato and its spatial distribution has become an immediate priority. In the studies we know of, the research focus on late blight in potato monitoring still predominantly revolves around single-sensor approaches, such as multispectral UAV monitoring and hyperspectral UAV monitoring. However, as of now, there remains a lack of research utilizing the collaborative monitoring of potato late blight using multiple sensors on drones. By integrating the vegetation index derived from the multispectral UAV, the texture index from the RGB UAV, and the estimated crown coverage feature, we combined the Relief-mRmR technique with machine learning modeling algorithms to monitor late blight in potato. We compared and constructed severity monitoring models for late blight in potato using Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN) classification algorithms. The results indicate that the RF model exhibited the highest accuracy. For the training dataset, the overall accuracy (OA) and kappa coefficient were 92.50% and 0.90, respectively, while for the independent validation dataset, the OA and kappa coefficient reached 97.50% and 0.96, respectively. Late Blight in Potato, a prevalent potato disease, significantly impacts potato yield and stands as one of the principal afflictions affecting the potato crop. Timely and effective monitoring of late blight severity in potato and its spatial distribution has become an immediate priority. In this study, we employed a dual-drone collaborative approach, utilizing the multispectral drone from DJI as well as an ultra-high-resolution RGB drone. By integrating the vegetation index derived from the multispectral UAV, the texture index from the RGB drone, and the estimated crown coverage feature, we combined the relief-mRmR technique with machine learning modeling algorithms to monitor late blight in potato. We compared and constructed severity monitoring models for late blight in potato using Support Vector Machine (SVM), Random Forest (RF), and K-Nearest Neighbors (KNN) classification algorithms. The results indicate that the RF model exhibited the highest accuracy. For the training dataset, the overall accuracy (OA) and kappa coefficient were 92.50% and 0.90, respectively, while for the independent validation dataset, the OA and kappa coefficient reached 97.50% and 0.96, respectively. The findings also demonstrate that augmenting the vegetation index and texture index with the estimated high-resolution plant crown coverage information significantly enhances the accuracy of the late blight in potato monitoring model. [ABSTRACT FROM AUTHOR]

• We developed an RRCNet that fuses the deep and shallow spectral features to map RRC. • RRCNet was trained based on an RTM-based "soil-crop-crop residue" spectra dataset. • Use transfer learning (TL) to recalibrate the fully connected layer of RRCNet. • RRCNet + TL can provide more accurate RRC estimates than SI + statistical regression. Accurate determination of rice residue cover (RRC) can improve the monitoring of tillage information. Currently, the accurate determination of RRC using optical remote sensing is hindered by variations in cropland moisture and cover of following crops. The fractional cover (FC) of the soil (f S), crop (f C), and crop residue (f CR) changes (f S + f C + f CR = 1) after the following crop is planted, which increases the difficulty of remote-sensing RRC estimation. Cropland soil moisture and crop residue moisture affect the values of cropland and crop residue spectral indices (CRSIs), thereby reducing the accuracy of remote-sensing RRC estimation. Deep learning techniques (e.g., convolutional neural networks [CNN] and transfer learning [TL]) have been proven to extract the deep features of input images with distortion invariance, such as displacement and scaling, which are similar to moisture and the following crop effects on remote-sensing CRSIs. This study aimed to evaluate the combined use of deep features of cropland spectra extracted by deep learning techniques to estimate the cropland RRC under the effects of variations in cropland moisture and cover of the following crops. This study proposes an RRCNet CNN that fuses deep and shallow features to improve RRC estimation. A PROSAIL radiative transfer model was employed to simulate a cropland "soil–crop–crop residue" mixed spectra dataset (n = 103,068), considering the variations in cropland moisture and the cover of the following crop. The RRCNet was first pre-trained using the simulated dataset, and then the knowledge from the pre-trained RRCNet was updated based on field experimental FCs, RRCs, and Sentinel-2 image spectra using the TL technique. Our study indicates that RRCNet can incorporate shallow and deep spectral features of cropland "soil–crop–crop residue" mixed spectra, providing high-performance FCs and RRC estimation. The FCs and RRC estimates from RRCNet + TL (FCs: R 2 = 0.939, root mean squared error (RMSE) = 0.071; RRC: R 2 = 0.891, RMSE = 0.083) were more accurate than those from CRSI + multiple linear regression, CRSI + random forest, and CRSI + support vector machine (FCs: R 2 = 0.877–907, RMSE = 0.086–0.101; RRC: R 2 = 0.378–0.714, RMSE = 0.133–0.229). We mapped the multistage RRC based on Sentinel-2 multispectral instrument (MSI) images and RRCNet. Tillage information can be inferred from RRC and RRC difference maps changes. [ABSTRACT FROM AUTHOR]

• AGB was estimated by combining UAV RGB and hyperspectral images. • The ability of different wavelet basis functions to extract features was compared. • The high-frequency information and wavelet coefficients were used to estimate AGB. • The problem of inaccurate estimation of potato AGB using VIs was solved. Crops' above-ground biomass (AGB) is a crucial indicator that reflects crop health and predicts crop yield. However, using only optical vegetation indices (VIs) can produce inaccurate AGB estimates due to differences in crop varieties, growth stages, and measurement environments. Given the advantages of unmanned aerial vehicle (UAV) RGB and hyperspectral image fusion, this study evaluated the performance of multi-source remote sensing data for estimating potato AGB at multiple growth stages. In 2019, this study conducted potato trials with different varieties, fertilization levels, and planting densities at the Xiaotangshan Experiment Base (Beijing). UAV image and AGB data of potato three main stages were obtained from ground survey work. High-frequency information of the potato canopy was extracted from RGB images using discrete wavelet transform (DWT). VIs and wavelet energy coefficients were extracted from hyperspectral images using continuous wavelet transform (CWT). The linear relationships between potato AGB with VIs, high-frequency information, and wavelet coefficients were analyzed. Potato AGB estimation models were constructed based on single and multiple types of variables using multiple stepwise regression (MSR) and random forest (RF) models, respectively. This work showed the following results: (i) High-frequency information and wavelet coefficients were more sensitive to potato multi-growth stage AGB than VIs, and the latter were the most sensitive. (ii) Using VIs, high-frequency information, or wavelet coefficients separately to estimate the potato multi-growth stage AGB resulted in higher error and lower model accuracy. (iii) Combining VIs with either high-frequency information or wavelet coefficients improved the accuracy of AGB estimation, which was further improved by combining high-frequency information with wavelet coefficients. (iv) Combining VIs with both high-frequency information and wavelet coefficients provided the highest estimation accuracy using the MSR method. This combined AGB estimation model reduced the RMSE by 27%, 21%, and 16%, respectively, relative to VIs, high-frequency information, or wavelet coefficients alone. This result shows that the complementary advantages of multi-source UAV data can solve the challenge of insufficient AGB estimation by optical remote sensing. The work in this study provides remote sensing technology support to achieve potato crop growth monitoring and improve yield predictions. [ABSTRACT FROM AUTHOR]

• The ability of AGB estimation based on RGB-VIs and H-VIs was compared. • The potential of RGB-VIs, SFs and TFs to estimate AGB was evaluated. • The potato AGB estimation model was built based on multiple types of features. • UAV RGB images could be used to monitor potato crop growth. Rapid and non-destructive potato above ground biomass (AGB) monitoring is a crucial step in the development of smart agriculture because AGB is closely related to crop growth, yield, and quality. Compared to time-consuming and laborious field surveys, unmanned aerial vehicle (UAV) remote sensing provides a new direction for large-scale AGB monitoring. However, estimating AGB using an optical remote sensing technique usually does not work well because of spectral saturation, but multi-source remote sensing feature fusion (e.g., fusing spectral and structural features) can mitigate that problem. Due to potato crop canopy structure and AGB change greatly during growth, the potential of fusing optical, textural (TFs), and structural features (SFs) for calculating potato AGB at multiple growth stages was unknown. In addition, the ability of optical features, TFs, and SFs and their combinations to estimate potato AGB had not been examined. Vegetation indices (RGB-VIs), TFs, and SFs were extracted from ultra-high spatial resolution RGB images and compared their performances for estimating potato AGB with those of hyperspectral vegetation indices (H-VIs) obtained from UAV hyperspectral images. The results revealed that each type of feature had its own advantages and limitations for potato AGB estimation. Except for canopy volume (CV) in SFs, the best H-VI, RGB-VI, and TF for estimating AGB in both single growth stages and the entire growth period were inconsistent. When estimating AGB with only a single type of feature, the model accuracy in descending order was SFs, TFs, H-VIs, and RGB-VIs. The fusion of any two types of remote sensing features improved AGB estimation model accuracy. Among them, TFs combined with SFs provided the best estimation performance. The fusion of RGB-VIs, TFs, and SFs produced the best AGB estimates precision (R2 = 0.81, RMSE = 207 kg/hm2, NRMSE = 17.40%). Since AGB was effectively estimated under different treatments in the field, the model applicability was confirmed. Using different types of remote sensing features, the Gaussian process regression method produced better estimation results than the partial least squares regression method did. This study provides an economic and effective method for monitoring the potato growth in the field, and thus helps improve farmland production and guide fertilization management. [ABSTRACT FROM AUTHOR]

• A-BiGRU enables more accurate modeling of crop growth to achiever high accuracy. • The Attention mechanism can focus on the critical growth period of the crop. • The bidirectional GRU structure is better than GRU. • Image reconstruction can effectively fill the data gap. Timely and accurate access to regional scale crop plant area and spatial distribution is essential for regional agricultural production and food security, especially in the context of global population growth and climate change. Deep learning has become prevalent in crop mapping under complex conditions due to its powerful feature extraction and nonlinear ability. This study proposes a time-series image classification framework using Attention-based Bidirectional Gated Recurrent Unit (A-BiGRU) to map rice, maize, and soybean in Fujin, China, from reconstructed Sentinel-2 time-series images. Firstly, the reconstructed Sentinel-2 time-series images with 10 spectral dimensions and 22 temporal dimensions were obtained by linear interpolation and Savitzky-Golay (SG) filter. Then, a neural network, the A-BiGRU was developed to identify different crops by taking advantage of their unique growth patterns. The attention mechanism enables temporal neural networks to focus on the critical growth periods of crops. Additionally, the structure of GRU is simpler than that of long short-term memory (LSTM) and simple recurrent neural network (SRNN), which reduces the number of parameters and alleviates overfitting. Compared to GRU, BiGRU can fully uses the time-series information of the entire crop growth cycle. To assess the effectiveness of the proposed method, we compared two deep learning methods (LSTM and SRNN) and three widely used non-deep learning classifiers (Spectral Angle Mapping (SAM), Support Vector Machine (SVM)) and eXtreme Gradient Boosting (XGBoost). The results demonstrate that A-BiGRU achieved the highest accuracy, with an overall accuracy of 0.9804, a macro F1 score of 0.9788 and a kappa score of 0.9714. We also selected four typical regions and compared the classification results with optical images, which showed that the proposed method has a good recognition effect. Therefore, the A-BiGRU method is capable of achieving high-precision crop mapping. [ABSTRACT FROM AUTHOR]

• Using digital photos can reconstruct high-quality 3D point clouds of maize tassels. • The TreeQSM is effective in extracting phenotypes from maize tassels. • Low light has a greater impact on the reconstruction of maize point clouds. • Spatial calibration is necessary for each tassel. • The quality of point clouds obtained using image-based reconstruction methods is superior. Maize tassel is a crucial pollen producing organ that plays an important role in maize production. It is challenging to investigate the morphological and structural traits of maize tassels for breeding programs, since traditional manual measurements of organ-scale phenotypic traits are labor-intensive and prone to human errors. It is, therefore, urgent to design and develop new phenotyping systems for maize tassels to improve throughput and measurement accuracy. This study first introduced the TreeQSM in crop phenotyping at organ scale and developed TIPS (TreeQSM based Image Phenotyping System for maize tassels). The system mainly consists of three digital cameras for acquiring multi-view images of individual maize tassel. These cameras were vertically arranged with different angles of view. The acquired images were used to reconstruct the 3D point cloud data of individual maize tassel, which were fed into the TreeQSM to extract four tassel phenotypic parameters including trunk length, branch number, branch length, and branch angle. The performance analyses of the developed system were conducted on 52 tassel samples from 37 maize materials with different canopy geometric structures. The experimental results showed that the TIPS could achieve accurate tassel parameters estimation with the R2 of 0.964, 0.973, 0.935, and 0.857, and the RMSE of 1.72, 1.28, 19.94, and 2.73 for the trunk length, branch number, total branch length, and first node branch angle, respectively. This study also investigated the advantages of the data acquisition with the TIPS and the effects of different shooting angles, number of images, lighting conditions and tassel types on the four tassel parameters estimation accuracy. The comparison results indicated that the four influencing factors reduced the estimation accuracy to varying degree. Compared with the other three parameters, the estimation accuracy of branch angles less than 20 were largely affected. The higher the degree of compactness, the worse the estimation accuracy. Compared with shooting angles, the reduction of the number of images in a certain range had less impact on the quality of 3D point cloud reconstruction. The influence of low light was obviously greater than that of strong light. This study provides a valuable guide to the collection and quantitative analysis of high-throughput phenotype information of maize tassels in breeding program. [ABSTRACT FROM AUTHOR]

• Our method focuses on exploring the effect of texture indices to estimate potato PNC. • An texture index was developed for PNC estimation throughout the potato growth cycle. • O-TTI3 can further improve the estimation accuracy of potato PNC. Plant nitrogen content (PNC) is vital for evaluating crop nitrogen nutrient status and for net primary productivity. Therefore, rapid and accurate acquisition of crop PNC information can help reduce fertilizer and increase efficiency in modern agriculture. This study investigated whether an optimized texture index constructed from hyperspectral characteristic wavelength images acquired from a unmanned aerial vehicle (UAV) may be used to estimate PNC in potatoes over multiple growth stages. A potato field trial conducted in 2019 in Beijing, China, included different planting densities, nitrogen fertilizer gradients, and potato varieties with three replicates. A UAV remote sensing platform served to acquire hyperspectral images during three critical N demand periods for potatoes. In addition, we simultaneously conducted field sampling to obtain ground-truth PNC measurements. Following the classical form of vegetation indices, 12 texture indices were constructed based on hyperspectral texture features, and an arbitrary variable combination optimization algorithm was used to optimize these indices. Finally, the texture index, which had the highest correlation with potato PNC, was used as the best indicator for estimating the PNC status of potato at multiple growth stages and whether this indicator, in combination with spectral information, could further improve the accuracy of potato PNC estimation was subsequently explored. The results showed that (i) the optimal texture index TTI3 (R 494 -Cor, R 578 -Hom, R 514 -Sem) maintained a good linear relationship with PNC at the late stage of potato growth, and the accuracy and stability of the PNC estimation models constructed based on it was significantly better than that of a single texture metric. (ii) Compared with spectral information alone, the texture index combined with spectral features improved the accuracy of potato PNC estimation. More specifically, TTI3 (R 494 -Cor, R 578 -Hom, R 514 -Sem) combined with the three-band spectral index TBI 5 (530, 734, 514) achieved the best estimation accuracy with calibrated R2 , root mean square error (RMSE), and normalized RMSE of 0.77%, 0.28%, and 9.88%, respectively. The results of this study showed that the texture index constructed by combining multiple texture metrics enhanced the association between texture features and potato PNC over multiple growth stages, thus improving the monitoring accuracy of potato nitrogen nutrition status. This study can provide a reference for accurately managing crop nitrogen nutrition. [ABSTRACT FROM AUTHOR]

• Proposed a tea plantation phenological feature index (TPI). • Constructed a mapping algorithm based on multitemporal spectral features. • Tea plantations were accurately distinguished from evergreen vegetation. • The tea plantation mapping algorithm is temporally reusable. Tea is a highly demanded and valuable commodity in both domestic and international markets, with China being the world's largest producer and exporter, resulting in the expansion of tea tree cultivation in China over the past few decades. Accurate mapping of tea plantations is vital for tea markets, food security, poverty alleviation, and ecosystem value assessment. However, research on vegetation plantation monitoring has focused primarily on food crops, and the spatial information for the perennial tea tree is often neglected. Moreover, monitoring tea plantations poses significant challenges in terms of the diverse cultivation management measures, the heterogeneity of tea garden landscapes, and the limited availability of clear imagery due to cloud cover. In this study, we constructed a novel tea plantation mapping algorithm based on multitemporal spectral features. In addition, a Tea Plantation Phenological Feature Index (TPI) is proposed to enable easy and reliable monitoring of tea plantations. The algorithm harmonized the Landsat7/8 and Sentinel-2 satellite images on the Google Earth Engine (GEE) platform, utilizing a semi-automatic decision tree model to identify tea gardens in Yingde City, Guangdong Province. Our results show that the tea plantation phenological feature index (TPI), green band in October (Green Oct) and cumulative band in April (CB Apr) are the most effective multitemporal spectral features for identifying tea gardens. The F1-score, OA, and kappa coefficients (Kappa) of the 2020 tea plantation map generated by the tea plantation mapping algorithm based on multitemporal spectral features are 90.75%, 92.71%, and 0.844, respectively. Furthermore, the algorithm is time-reusable, as evidenced by the F1-score, OA, and kappa coefficients of the generated 2021 tea plantation map, which are 85.99%, 88.86%, and 0.7675, respectively. Our results underscore the exceptional ability of the multitemporal spectral features to separate tea gardens from evergreen forests. The tea plantation mapping algorithm based on multitemporal spectral features is a robust, user-friendly, and effective tool suitable for regional or national annual tea plantation mapping, enabling sustainable production and implementing. [ABSTRACT FROM AUTHOR]

• The MP-Net model aggregates context information. • The MP-Net model reduces the feature losses. • The MP-Net model efficiently and precisely classifies crops. Accurate crop classification map is of great significance in various fields such as the survey of agricultural resource, the analysis of existing circumstance on land application, the yield estimation of crop and the disaster warning. The methods based on machine learning and deep learning are popularly used in crop classification and recognition of remote sensing images. However, the crop classification task based on neural networks still faces significant challenges due to the spatial and temporal distribution of crops and the inherent characteristics of remote sensing images. Therefore, this study proposes the multi-layer pyramid crop classification network (MP-Net) to solve the above problems. To reduce the feature loss during the crop extraction, the proposed model uses the pyramid pooling module to improve the ability of global information acquisition, and the information concatenation module to retain the upper features. Using the GF-6 and Sentinel-2 satellite data, the proposed model was tested in Erhai Lake Basin and Beian City. Compared with other five deep learning models, such as FCN, SegNet, U-Net, PSPNet and DeepLabv3+, the experimental results indicate that the proposed model achieves the highest accuracy in both study areas. Meanwhile, the proposed model has the advantages of short training time and high efficiency under the same running conditions. Overall, this study is beneficial to improve the efficiency and accuracy of crop classification task in the unbalanced temporal and spatial distribution. It also brings a feasible scheme for crop classification tasks in complex growing areas. The code has been publicly available at https://github.com/Xu-Chang-Hong/MP-Net. [ABSTRACT FROM AUTHOR]

• Hyperspectral-to-image transform technique converts hyperspectral reflectance into images. • A 2D CNN architecture was proposed for leaf chlorophyll content (LCC) estimation. • An LCCNet was pre-trained based on a PROSAIL RTM dataset. • Reusing CNN layer information of LCCNet improved the soybean LCC estimation. Leaf chlorophyll content (LCC) is a distinct indicator of crop health status used to estimate nutritional stress, diseases, and pests. Thus, accurate LCC information can assist in the monitoring of crop growth. The combined use of hyperspectral and deep learning techniques (e.g., convolutional neural network [CNN] and transfer learning [TL]) can improve the performance of crop LCC estimation. We propose a hyperspectral-to-image transform (HIT) technique for converting canopy hyperspectral reflectance into 2D images. We designed a CNN architecture called LCCNet that fuses the deep and shallow features of CNNs to improve soybean LCC estimation. This study evaluated the combined use of hyperspectral remote sensing (RS), HIT, CNN, and TL techniques to estimate soybean LCC for multiple growth stages. The LCCNet was pre-trained based on a simulated dataset (n = 114,048) from the PROSAIL radiative transfer model (RTM) and used as prior knowledge for this work. The soybean canopy hyperspectral RS dataset (n = 910) was obtained using a FieldSpec 3 spectrometer. The knowledge gained while learning to estimate LCC from PROSAIL RTM was applied when estimating field soybean LCC (Dualex readings). TL was used to enhance the soybean estimation model, called the Soybean-LCCNet (RTM + HIT + CNN + TL) model. We tested the LCC (Dualex readings) estimation performance using (a) HIT + CNN, (b) LCCNet (RTM + HIT + CNN), (c) Soybean-LCCNet (RTM + HIT + CNN + TL), and (d) widely used LCC spectral features + partial least squares regression (PLSR). Four methods were ranked based on their LCC estimation performance: Soybean-LCCNet (R2 = 0.78, RMSE = 4.13 (Dualex readings)) > HIT + CNN (R2 = 0.75, RMSE = 4.41 (Dualex readings)) > PLSR-based method (R2 = 0.61, RMSE = 5.39 (Dualex readings)) > LCCNet (R2 = 0.53, RMSE = 7.11 (Dualex readings)). The main conclusions of this work are as follows: (1) HIT + CNN can provide a more robust LCC estimation performance than the widely used LCC SIs; (2) Fusing the deep and shallow features of CNNs can improve the performance of RS soybean LCC (Dualex readings) estimation; and (3) Soybean-LCCNet can reuse the CNN layer information of a pre-trained LCCNet based on a PROSAIL RTM dataset and improve the soybean LCC estimation performance. [ABSTRACT FROM AUTHOR]

• Canopy-scale chlorophyll content (CCC) estimation conducted on open-canopy orchards. • A simplified 3D radiative transfer model is coupled with PROSPECT-D model. • Vegetation indices and model inversion methods are assessed with UAV data. • Look-up table including vegetation indices can estimate CCC of ITCs from image spectroscopy. • The inversion method based on the prior knowledge provides accurate estimation of CCC. Chlorophyll content is a key trait for understanding the functioning of agroforestry ecosystems and has important implications for leaf and canopy photosynthesis. However, fine-scale monitoring of canopy chlorophyll content (CCC) of individual fruit trees is rather challenging. This study aims to use a 3D radiative transfer model (RTM) and proposes a joint inversion model based on prior knowledge to estimate the CCC of individual tree crowns (ITCs) in apple orchards. The widely recognized 3D RTM LESS (large-scale remote sensing data and image simulation framework over heterogeneous 3D scenes) was adopted for large-scale apple orchard 3D scenes radiative transfer computing and image simulation. LESS was first evaluated with unmanned aerial vehicle (UAV) multispectral imagery and the results showed that it reasonably characterized the reflectance of apple tree canopies (RMSE = 0.02). An original look-up table (LUT) with reflectance was then produced using LESS, and the final vegetation indices LUT (VI LUT) including Normalized Difference Vegetation Index (NDVI), Green Chlorophyll Index (CIgreen), Red edge Chlorophyll Index (CIred edge) and Green NDVI (GNDVI) was generated from the original LUT form VI interpolation. A physically-based joint inversion model coupling prior knowledge of leaf pigments and leaf area index (LAI) was developed to estimate the CCC of ITCs from high-resolution UAV images. The solution first used linear interpolation to produce a weighted VI LUT corresponding to the sample based on estimated LAI. Linear interpolation was then adopted to screen multiple combinations of leaf chlorophyll a+b (C ab) and leaf carotenoids (C xc) contents from the VI LUT. A prior relationship between C ab and C xc was finally used to regularize the constraints on multiple VI combinations and determine the estimation of C ab and CCC. The joint inversion model demonstrated an accurate estimation of CCC of ITCs. The model driven by GNDVI yielded the highest result for CCC estimation (R2 = 0.84, RMSE = 24.12 μg/cm2). In addition, CIgreen (R2 = 0.82, RMSE = 32.22 μg/cm2) and CIred edge (R2 = 0.81, RMSE = 34.05 μg/cm2) also achieved satisfactory results. The proposed model facilitates CCC estimation of ITCs from high-resolution imagery in heterogeneous orchard canopies, which is important for advancing the precise nutrition management of fruit trees. [ABSTRACT FROM AUTHOR]

• Potential of land surface temperature to predict flowering phenology was explored. • A spatio-temporal reconstruction approach is developed to compenstate the missing LST. • The predicted first flowering date are validated with field measurements at multiple sites. • Feasibility for generating spatially continuous information on flowering phenology. The first flowering date (FFD) is a critical phenological parameter closely related to the apple yield, so the accurate prediction of the FFD is important for precise orchard production management. Existing methods to predict the FFD are mostly based on air temperature (Ta) measured at meteorological stations, but to great differences in meteorological variations and the ecological conditions, these methods cannot present the differences of FFD under complex meteorological conditions and provide spatially continuous FFD information at the level of a region. Therefore, we propose a method to predict spatially continuous apple FFD from remote sensing land surface temperature (LST) based on flowering prediction model. Firstly, the missing LST data were reconstructed by using spatio-temporal reconstruction (STR) approach developed. Next, new air temperature (NAT) data were generated by using the daily Ta estimation (DTE) model and the reconstructed LST. Finally, apple FFD was predicted by the NAT data and the apple flowering prediction model established based on random forest (RF) algorithm and the phenology sequential model, and the prediction accuracy was verified by comparison with the independently measured apple FFD. The LST reconstructed by using the STR approach has mean absolute error (MAE) ranging from 0.51 to 0.68 °C, and root mean square error (RMSE) ranging from 1.07 to 1.21 °C. The MAE between the NAT data and the High-Resolution Land Surface Data Assimilation System (HR-CLDAS) meteorological data ranges from 2.15 to 3.23 °C, and the RMSE ranges from 2.81 to 4.27 °C. In addition, the determination coefficient (R2) and RMSE between the predicted and measured FFD is 0.72 and 2.96 days, respectively. These results demonstrate that the developed method maximizes the potential of MODIS LST in predicting spatially continuous apple FFD, which is valuable for flower and fruit thinning, to defend against frost disasters, and in general for refined orchard production management. [ABSTRACT FROM AUTHOR]

• An apple tree yield prediction model was developed by using ensemble learning. • The contribution of spectral and structural features to orchard yield was discussed. • An automatic channel was developed to extract growth characteristics. As one of the world's most popular fruit, apple tree yield prediction before harvest plays an important role in optimizing orchard nutrition management, especially at the individual tree level. However, few studies focus on fruit-tree yield prediction with remote-sensing technology whereas most of them aim at field crops. Current fruits identifying and counting methods often fail to produce the expected result due to light and occlusion in complex orchard conditions. Since both the spectral and morphological characteristics of tree canopy can reflect the growth and development of fruit trees and are directly related to its potential yield. In this study, we develop a channel for automatic extraction of spectral and morphological features of apple trees using light detection and ranging (LiDAR) and multispectral imagery data from unmanned aerial vehicles. The contribution of spectral and morphological characteristics to the yield prediction of individual apple trees is discussed. With the combination of spectral and morphological features, an ensemble machine learning yield prediction model was developed by combining two widely used basic learners: support vector regression (SVR) and K-nearest neighbor (KNN). Then through extrapolating the ensemble model, the yield map was produced at the orchard level and individual tree level, respectively. The results show that the data processing channels developed in this study can accurately extract the morphological and spectral features of individual apple trees. Three features (Crown Volume 1, Ratio Vegetation Index, and CPA1) contribute most in apple tree yield prediction. The ensemble learning model outperforms all base learners with R2 = 0.813 for the validation and 0.758 for the test when using the selected three features. This study thus provides a practical example of predicting the yield of individual apple trees based on multi-source remote-sensing data and ensemble learning. [ABSTRACT FROM AUTHOR]

• The changes of potato textures and plant heights were analyzed. • The estimation models of potato above ground biomass were constructed. • The underestimation of above ground biomass when using VIs alone was solved. • The potato growth monitoring at the field scale was realized. Above-ground biomass (AGB) is a significant phenotypic index for evaluating photosynthesis capacity, healthy growth, and estimating crop yield. Accurately monitoring the AGB helps improve agricultural fertilization management and optimize planting patterns. Numerous studies have confirmed that canopy spectrum saturation causes optical vegetation indices (VIs) to underestimate the AGB of crops at multiple growth periods. To solve this problem, the present research used a remote sensing method to obtain RGB images of potato tuber formation-, tuber growth-, and starch accumulation- periods by a high-definition digital camera sensor on the unmanned aerial vehicle (UAV). From the ultrahigh spatial resolution RGB images, we then extracted RGB-VIs, textures (based on the gray level co-occurrence matrix, GLCM), and crop height (Hdsm) and analyzed the correlation between the three image features and the potato AGB for single and multiple growth periods. Finally, we estimated potato AGB at multiple growth periods based on (1) RGB-VIs, (2) RGB-VIs + GLCM-based textures, (3) RGB-VIs + Hdsm, and (4) RGB-VIs + GLCM-based textures + Hdsm by applying multiple stepwise regression (MSR) and extreme learning machine (ELM). The results showed that (i) unlike the texture features of wheat and maize that increased with growth period, the texture features and crop height of the potato canopy both increased first and then decreased with the growth period. (ii) The potato AGB was poorly estimated when using RGB-VIs, CLCM-based textures, or Hdsm individually; (iii) combining GLCM-based textures, Hdsm, and RGB-VIs solved the problem of underestimating the high AGB values of potato samples by the RGB-VIs model alone. Therefore, combining GLCM-based textures, Hdsm, and RGB-VIs obtained from UAV digital images could enhance the accuracy of potato AGB estimation under high coverage. [ABSTRACT FROM AUTHOR]

• Hyperspectral imaging monitored the quality of apple directly in orchards. • Relative spectral processing could reduce the unevenness of ambient light. • The accuracy of the prediction models met the requirements of orchard management. • The dynamic process of apple quality and maturity was visualized. Rapid and in-situ estimation of apple quality and maturity is of great significance to guarantee the quality of apples and the profit of the industry. This study explored the utility of hyperspectral imaging (HSI) in assessing the quality and maturity of apples. Hyperspectral images, in the range of 500–900 nm, were collected directly in the orchard. A total of 100 apples contained in the visual field of the camera were picked for destructive testing of firmness, soluble solids content (SSC) and starch pattern index (SPI). To reduce the complicated outdoor light interference, we adopted the PTEE ball as the white reference. Normalized reflectance spectrum (NR) and normalization different spectral index (NDSI) involving all possible two-band combinations were also extracted for apple quality estimation. Subsequently, partial least squares regression (PLSR) analyses were conducted using the characteristic wavelengths extracted by stability competitive adaptive reweighted sampling (SCARS). The PLSR model established by NDSI combined with SCARS achieved the best results. The correlation coefficient of validation set (R v) of firmness, SSC and SPI were 0.783, 0.901 and 0.834, respectively. The root-mean-square errors of the validation set (RMSEv) of firmness, SSC, SPI were 0.993 kg/cm2, 0.535% and 0.427, respectively. The Streif index was used to evaluate the maturity of apples, which was calculated from the predicted value of firmness, SSC and SPI. The RMSEv of the PLS model for Streif index was 0.0120. The results indicated the potential of the NDSI-SCARS-PLSR model based on hyperspectral imaging to observe the dynamic changes in the quality and maturity of the apples in site. This model could provide the spatial distribution of apple quality and maturity, which would be valuable for exploring the maturation pattern, grasping the growth of the area, and planning harvest time. [ABSTRACT FROM AUTHOR]

[Display omitted] • Multi-source environmental data were used to predict wheat yield in China. • We developed a robust machine learning model to predict wheat yield. • The random forest model using NIRv had better performance than models using NDVI and EVI. • NIRv from jointing to heading was the most important predictor in determining yield. Crop yield is controlled by different environmental factors. Multi-source data for site-specific soils, climates, and remotely sensed vegetation indices are essential for yield prediction. Algorithms of data-model fusion for crop growth monitoring and yield prediction are complicated and need to be optimized to deal with model uncertainty. This study integrated multi-source environmental variables (e.g., satellite-based vegetation indices, climate data, and soil properties) into random forest (RF) and support vector machine (SVM) models for wheat yield prediction in China. The performance of both RF and SVM models was investigated using different types of vegetation indices associated with other predictors. Relative importance and partial dependence analyses were used to identify the main predictors and their relationships with wheat yield. We found that using remotely sensed vegetation indices improved our model precision, and that near-infrared reflectance of terrestrial vegetation (NIRv) was slightly better than normalized difference vegetation index (NDVI) and enhanced vegetation index (EVI) in predicting yield. NIRv was better in detecting climate stress on crops, and could capture more information regarding crop growth and yield formation. Compared with the SVM model, the RF model with NIRv and other covariates had better performance in wheat yield prediction, with R2 and RMSE being 0.74 and 758 kg/ha respectively. We also found that NIRv from jointing to heading was the most important predictor in determining yield, followed by solar radiation (especially during tillering–heading), relative humidity (during planting–tillering), soil organic carbon, and wind speed (throughout the growing season). In addition, wheat yield exhibited threshold-like responses to most factors based on our RF model. These threshold values can help to better understand how different environmental factors limit wheat yield, which will provide useful information for climate-adaptive crop management. Our findings demonstrated the potential of using NIRv for yield prediction. This approach is broadly applicable to other regions globally using publicly available data. [ABSTRACT FROM AUTHOR]