Search

Generating spatial crop yield information is of great significance for academic research and guiding agricultural policy. Existing public yield datasets have a coarse spatial resolution, spanning from 1 to 43 km. Although these datasets are useful for analyzing large-scale temporal and spatial change in yield, they cannot deal with small-scale spatial heterogeneity, which happens to be the most significant characteristic of the Chinese farmers' economy. Hence, we generated a 30 m Chinese winter wheat yield dataset (ChinaWheatYield30m) for major winter-wheat-producing provinces in China for the period 2016–2021 with a semi-mechanistic model (hierarchical linear model, HLM). The yield prediction model was built by considering the wheat growth status and climatic factors. It can estimate wheat yield with excellent accuracy and low cost using a combination of satellite observations and regional meteorological information (i.e., Landsat 8, Sentinel 2 and ERA5 data from the Google Earth Engine (GEE) platform). The results were validated using in situ measurements and census statistics and indicated a stable performance of the HLM based on calibration datasets across China, with a correlation coefficient (r) of 0.81 and a relative root mean square error (rRMSE) of 12.59 %. With regards to validation, the ChinaWheatYield30m dataset was highly consistent with in situ measurement data and statistical data (p<0.01), indicated by an r (rRMSE) of 0.72** (15.34 %) and 0.69** (19.16 %). The ChinaWheatYield30m is a sophisticated dataset with both high spatial resolution and excellent accuracy; such a dataset will provide basic knowledge of detailed wheat yield distribution, which can be applied for many purposes including crop production modeling and regional climate evaluation. The ChinaWheatYield30m dataset generated from this study can be downloaded from 10.5281/zenodo.7360753 (Zhao et al., 2022b). [ABSTRACT FROM AUTHOR]

Crop production is one of the major interactions between humans and the natural environment, in the process, carbon is translocated cyclically inside the ecosystem. Data assimilation algorithm has advantages in mechanism and robustness in yield estimation, however, the computational efficiency is still a major obstacle for widespread application. To address the issue, a novel hybrid method based on the combination of the Crop Biomass Algorithm of Wheat (CBA-Wheat) to the Simple Algorithm For Yield (SAFY) model and the transfer learning method was proposed in this paper, which enables winter wheat yield estimation with acceptable accuracy and calculation efficiency. The transfer learning techniques learn the knowledge from the SAFY model and then use the knowledge to predict wheat yield. The main results showed that: (1) The comparison using CBA-Wheat between measured AGB and predicted AGB all reveal a good correlation with R2 of 0.83 and RMSE of 1.91 t ha−1, respectively; (2) The performance of yield prediction was as follows: transfer learning method (R2 of 0.64, RMSE of 1.05 t ha−1) and data assimilation (R2 of 0.64, RMSE of 1.01 t ha−1). At the farm scale, the two yield estimation models are still similar in performance with RMSE of 1.33 t ha−1 for data assimilation and 1.13 t ha−1 for transfer learning; (3) The time consumption of transfer learning with complete simulation data set is significantly lower than that of the other two yield estimation tests. The number of pixels to be simulated was about 16,000, and the computational efficiency of the data assimilation algorithm and transfer learning without complete simulation datasets. The transfer learning model shows great potential in improving the efficiency of production estimation. [ABSTRACT FROM AUTHOR]

Crop yields are important for food security and people's living standards, and it is therefore very important to predict the yield in a timely manner. This study used different vegetation indices and red-edge parameters calculated based on the canopy reflectance obtained from near-surface hyperspectral data and UAV hyperspectral data and used the partial least squares regression (PLSR) and artificial neural network (ANN) methods to estimate the yield of winter wheat at different growth stages. Verification was performed based on these two types of hyperspectral remote sensing data and the yield was estimated using vegetation indices and a combination of vegetation indices and red-edge parameters as the modeling independent variables, respectively, using PLSR and ANN regression, respectively. The results showed that, for the same data source, the optimal vegetation index for estimating the yield was the same in all of the studied growth stages; however, the optimal red-edge parameters were different for different growth stages. Compared with using only the vegetation indices as the modeling factor to estimate yield, the combination of the vegetation indices and red-edge parameters obtained superior estimation results. Additionally, the accuracy of yield estimation was shown to be improved by using the PLSR and ANN methods, with the yield estimation model constructed using the PLSR method having a better prediction effect. Moreover, the yield prediction model obtained using the near-surface hyperspectral sensors had a higher fitting and accuracy than the model obtained using the UAV hyperspectral remote sensing data (the results were based on the specific growth stressors, N and water supply). This study shows that the use of a combination of vegetation indices and red-edge parameters achieved an improved yield estimation compared to the use of vegetation indices alone. In the future, the selection of suitable sensors and methods needs to be considered when constructing models to estimate crop yield. [ABSTRACT FROM AUTHOR]

Although crop-growth monitoring is important for agricultural managers, it has always been a difficult research topic. However, unmanned aerial vehicles (UAVs) equipped with RGB and hyperspectral cameras can now acquire high-resolution remote-sensing images, which facilitates and accelerates such monitoring. To explore the effect of monitoring a single crop-growth indicator and multiple indicators, this study combines six growth indicators (plant nitrogen content, above-ground biomass, plant water content, chlorophyll, leaf area index, and plant height) into the new comprehensive growth index (CGI). We investigate the performance of RGB imagery and hyperspectral data for monitoring crop growth based on multi-time estimation of the CGI. The CGI is estimated from the vegetation indices based on UAV hyperspectral data treated by linear, nonlinear, and multiple linear regression (MLR), partial least squares (PLSR), and random forest (RF). The results are as follows: (1) The RGB-imagery indices red reflectance (r), the excess-red index (EXR), the vegetation atmospherically resistant index (VARI), and the modified green-red vegetation index (MGRVI), as well as the spectral indices consisting of the linear combination index (LCI), the modified simple ratio index (MSR), the simple ratio vegetation index (SR), and the normalized difference vegetation index (NDVI), are more strongly correlated with the CGI than a single growth-monitoring indicator. (2) The CGI estimation model is constructed by comparing a single RGB-imagery index and a spectral index, and the optimal RGB-imagery index corresponding to each of the four growth stages in order is r, r, r, EXR; the optimal spectral index is LCI for all four growth stages. (3) The MLR, PLSR, and RF methods are used to estimate the CGI. The MLR method produces the best estimates. (4) Finally, the CGI is more accurately estimated using the UAV hyperspectral indices than using the RGB-image indices. [ABSTRACT FROM AUTHOR]

Rapidly developing remote sensing techniques are shedding new light on large-scale crop growth status monitoring, especially in recent applications of unmanned aerial vehicles (UAVs). Many inversion models have been built to estimate crop growth variables. However, the present methods focused on building models for each single crop stage, and the features generally used in the models are vegetation indices (VI) or joint VI with data derived from UAV-based sensors (e.g., texture, RGB color information, or canopy height). It is obvious these models are either limited to a single stage or have an unstable performance across stages. To address these issues, this study selected four key wheat growth parameters for inversion: above-ground biomass (AGB), plant nitrogen accumulation (PNA) and concentration (PNC), and the nitrogen nutrition index (NNI). Crop data and multispectral data were acquired in five wheat growth stages. Then, the band reflectance and VI were obtained from multispectral data, along with the five stages that were recorded as phenology indicators (PIs) according to the stage of Zadok's scale. These three types of data formed six combinations (C1–C6): C1 used all of the band reflectances, C2 used all VIs, C3 used bands and VIs, C4 used bands and PIs, C5 used VIs and PIs, and C6 used bands, Vis, and PIs. Some of the combinations were integrated with PIs to verify if PIs can improve the model accuracy. Random forest (RF) was used to build models with combinations of different parameters and evaluate the feature importance. The results showed that all models of different combinations have good performance in the modeling of crop parameters, such as R2 from 0.6 to 0.79 and NRMSE from 10.51 to 15.83%. Then, the model was optimized to understand the importance of PIs. The results showed that the combinations that integrated PIs showed better estimations and the potential of using PIs to minimize features while still achieving good predictions. Finally, the varied model results were evaluated to analyze their performances in different stages or fertilizer treatments. The results showed the models have good performances at different stages or treatments (R2 > 0.6). This paper provides a reference for monitoring and estimating wheat growth parameters based on UAV multispectral imagery and phenology information. [ABSTRACT FROM AUTHOR]

• 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]

Accurate determination of crop phenology is key to field management and decision making. The existing research on phenology based on remote sensing data is mainly phenology monitoring, which cannot realize the prediction of phenology. In this paper, we propose a method to predict the maturity date (MD) of winter wheat based on a combination of phenology monitoring method and accumulated temperature. The method is divided into three steps. First, 2-band Enhanced Vegetation Index (EVI2) time series data were generated using the moderate-resolution imaging spectroradiometer (MODIS) reflectance data at 8-day intervals; then, the time series were reconstructed using polynomial fitting and the heading date (HD) of winter wheat was extracted using the maximum method. Secondly, the average cumulative temperature required for winter wheat to go from HD to MD was calculated based on historical phenological data and meteorological data. Finally, the timing of winter wheat HD and the current year's Meteorological Data were combined to predict winter wheat MD. The method was used to predict the MD of winter wheat in Hebei in 2018 and was validated with data from the phenology station and the Modis Land Cover Dynamics (MCD12Q2) product. The results showed that the coefficient of determination (R2) for predicting MD using this method was 0.48 and 0.74, the root mean square error (RMSE) was 7.03 and 4.91 days, and Bias was 4.93 and −3.59 days, respectively. In summary, the method is capable of predicting winter wheat MD at the regional scale. [ABSTRACT FROM AUTHOR]

Solar-induced chlorophyll fluorescence (SIF) is closely related to the light-reaction process and has been recognized as a good indicator for tracking gross primary productivity (GPP). Nevertheless, it has not been widely examined how SIF and GPP respond to temperature. Here, we explored the linkage mechanisms between SIF and GPP in winter wheat based on continuous measurements of canopy SIF ( c S I F ), GPP, and meteorological data. To separately explore the structural and physiological mechanisms underlying the SIF–GPP relationship, we studied the temperature responses of the estimated light use efficiency ( L U E p ), canopy-level chlorophyll fluorescence yield ( c S I F y i e l d ) and photosystem-level chlorophyll fluorescence yield ( Φ F ) estimated using canopy-scale remote sensing measurements. We found that GPP, red canopy SIF ( c S I F 688 ) and far-red canopy SIF ( c S I F 760 ) all exhibited a decreasing trend during overwintering periods. However, GPP and c S I F 688 showed relatively more obvious changes in response to air temperature ( T a ) than c S I F 760 did. In addition, the L U E p responded sensitively to T a (the correlation coefficient, r = 0.83, p-value < 0.01). The c S I F y i e l d _ 688 and Φ F _ 688 ( Φ F at 688 nm) also exhibited significantly positive correlations with T a (r > 0.7, p-value < 0.05), while c S I F y i e l d _ 760 and Φ F _ 760 ( Φ F at 760 nm) were weakly correlated with T a (r < 0.3, p-value > 0.05) during overwintering periods. The results also show that L U E p was more sensitive to T a than Φ F , which caused changes in the L U E p / Φ F ratio in response to T a . By considering the influence of T a , the GPP estimation based on the total SIF emitted at the photosystem level ( t S I F ) was improved (with R2 increased by more than 0.12 for t S I F 760 and more than 0.05 for t S I F 688 ). Therefore, our results indicate that the L U E p / Φ F ratio is affected by temperature conditions and highlights that the SIF–GPP model should consider the influence of temperature. [ABSTRACT FROM AUTHOR]

Crop residue is an important component of farmland ecosystems, which is of great significance for increasing soil organic carbon, mitigating wind erosion and water erosion and conserving soil and water. Crop residue coverage (CRC) is an important parameter to characterize the number and distribution of crop residues, and also a key indicator of conservation tillage. In this study, the CRC of wheat was taken as the research object. Based on the high-resolution GF-1 satellite remote sensing imagery from China, decision tree (DT), gradient boosting decision tree (GBDT), random forest (RF), least absolute shrinkage and selection operator (LASSO), extreme gradient boosting regression (XGBR) and other machine learning algorithms were used to carry out the estimation of wheat CRC by remote sensing. In addition, the comparisons with sentinel-2 imagery data were also utilized to assess the potential of GF satellite data for CRC estimates. The results show the following: (1) Among the spectral indexes using shortwave infrared characteristic bands from sentinel-2 imagery, the dead fuel index (DFI) was the best for estimating wheat CRC, with an R2 of 0.54 and an RMSE of 10.26%. The ratio vegetation index (RVI) extracted from visible and near-infrared characteristic bands from GF-1 data performed the best, with an R2 of 0.46 and an RMSE of 11.39%. The spectral index extracted from GF-1 and sentinel-2 images had a significant response relationship with wheat residue coverage. (2) When only the characteristic bands from the visible and near-infrared spectral ranges were applied, the effects of the spatial resolution differences of different images on wheat CRC had to be taken into account. The estimations of wheat CRC with the high-resolution GF-1 data were significantly better than those with the Sentinel-2 data, and among multiple machine learning algorithms adopted to estimate wheat CRC, LASSO had the most stable capability, with an R2 of 0.46 and an RMSE of 11.4%. This indicates that GF-1 high-resolution satellite imagery without shortwave infrared bands has a good potential in applications of monitoring crop residue coverage for wheat, and the relevant technology and method can also provide a useful reference for CRC estimates of other crops. [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]

A better understanding of wheat nitrogen status is important for improving N fertilizer management in precision farming. In this study, four different sensors were evaluated for their ability to estimate winter wheat nitrogen. A Gaussian process regression (GPR) method with the sequential backward feature removal (SBBR) routine was used to identify the best combinations of vegetation indices (VIs) sensitive to wheat N indicators for different sensors. Wheat leaf N concentration (LNC), plant N concentration (PNC), and the nutrition index (NNI) were estimated by the VIs through parametric regression (PR), multivariable linear regression (MLR), and Gaussian process regression (GPR). The study results reveal that the optical fluorescence sensor provides more accurate estimates of winter wheat N status at a low-canopy coverage condition. The Dualex Nitrogen Balance Index (NBI) is the best leaf-level indicator for wheat LNC, PNC and NNI at the early wheat growth stage. At the early growth stage, Multiplex indices are the best canopy-level indicators for LNC, PNC, and NNI. At the late growth stage, ASD VIs provide accurate estimates for wheat N indicators. This study also reveals that the GPR with SBBR analysis method provides more accurate estimates of winter wheat LNC, PNC, and NNI, with the best VI combinations for these sensors across the different winter wheat growth stages, compared with the MLR and PR methods. [ABSTRACT FROM AUTHOR]

The normalized difference vegetation index (NDVI) is an important agricultural parameter that is closely correlated with crop growth. In this study, a novel method combining the dynamic time warping (DTW) model and the long short-term memory (LSTM) deep recurrent neural network model was developed to predict the short and medium-term winter wheat NDVI. LSTM is well-suited for modelling long-term dependencies, but this method may be susceptible to overfitting. In contrast, DTW possesses good predictive ability and is less susceptible to overfitting. Therefore, by utilizing the combination of these two models, the prediction error caused by overfitting is reduced, thus improving the final prediction accuracy. The combined method proposed here utilizes the historical MODIS time series data with an 8-day time resolution from 2015 to 2020. First, fast Fourier transform (FFT) is used to decompose the time series into two parts. The first part reflects the inter-annual and seasonal variation characteristics of winter wheat NDVI, and the DTW model is applied for prediction. The second part reflects the short-term change characteristics of winter wheat NDVI, and the LSTM model is applied for prediction. Next, the results from both models are combined to produce a final prediction. A case study in Hebei Province that predicts the NDVI of winter wheat at five prediction horizons in the future indicates that the DTW–LSTM model proposed here outperforms the LSTM model according to multiple evaluation indicators. The results of this study suggest that the DTW–LSTM model is highly promising for short and medium-term NDVI prediction. [ABSTRACT FROM AUTHOR]

Crop failure detection using UAV images is helpful for precision agriculture, enabling the precision management of failure areas to reduce crop loss. For wheat failure area detection at the seedling stage using UAV images, the commonly used methods are not sufficiently accurate. Thus, herein, a new tool for precision wheat management at the seedling stage is designed. For this purpose, field experiments with two wheat cultivars and four nitrogen (N) treatments were conducted to create different scenarios for the failure area, and multispectral UAV images were acquired at the seedling growth stage. Based on the above data, a new failure detection method was designed by assimilating prior knowledge and a filter analysis strategy and compared with classical filter-based methods and Hough transform-based methods for wheat failure area detection. The results showed that the newly proposed assimilation method had a detection accuracy between 83.86% and 97.67% for different N levels and cultivars. In contrast, the filter-based methods and Hough transform-based methods had detection accuracies between 53.73% and 83.95% and between 20.71% and 75.79%, respectively. Thus, the assimilation method demonstrated the best failure detection performance. • An assimilation method was proposed for wheat failure area detection. • Different methods were compared for wheat failure area detection. • Wheat failure area can be detected using unmanned aerial vehicles imagery. [ABSTRACT FROM AUTHOR]

Timely monitoring of above-ground biomass (AGB) is essential for indicating the crop growth status and predicting grain yield and carbon dynamics. Non-destructive remote sensing techniques with a large spatial coverage have become a promising method for crop biomass monitoring. However, most existing crop biomass models have only been tested at a single growth stage or only at a small number of growth stages at a single location. This has limited the ability of these AGB models to be transferred spatially, to other fields or regions, to predict AGB at any growth stage during the season, or to be potentially used with data from other sensing systems. Here, a new crop biomass algorithm (CBA-Wheat) was developed to estimate AGB over the entire growing season. It uses information on the crop growth stage, based on phenological scale observations (Zadoks scale or ZS), the day of the year or thermal indices (growing degree days), to correct AGB estimations from remotely sensed vegetation indices. The model transferability was evaluated across multiple regional test sites and different data sources (UAV and hand-held spectroscopic data). Results showed that the coefficient values [slope (k) and intercept (b) ] of ordinary least squares regression (OLSR) of AGB with vegetation indices had a strong relationship with ZS. These k and b relationships were used to correct the OLSR model parameters based on the observed phenological stage (ZS value). The two-band enhanced vegetation index (EVI2) was the best vegetation index for predicting AGB with the new CBA-Wheat ZS model, with R2 and RMSE values of 0.83 and 2.07 t/ha for an experimental trial site, 0.78 and 2.05 t/ha for multiple independent regional test sites, and 0.69 and 1.87 t/ha when transferred to EVI2 derived from UAV. Model performance was lower with the day of the year and thermal index corrections; however, the use of relative growing degree-days (RGS; CBA-Wheat RGS), instead of ZS information, to adjust the model parameters showed a high consistency with the CBA-Wheat ZS model, and a good potential for estimation of AGB at regional scales without the need for local phenological observations. The CBA-Wheat RGS had validated R2 and RMSE values of 0.82 and 2.01 t/ha for the experimental trial site, 0.76 and 2.39 t/ha for multiple independent regional test sites, and 0.66 and 2.14 t/ha for UAV hyperspectral imagery. These results demonstrated a good potential to estimate biomass from remotely sensed imagery at varying spatio-temporal scales in winter wheat. • A model for wheat crop biomass based on a single evolving relationship with phenological period was developed. • Biomass prediction can be achieved by combining phenological information and remote sensing data. • Model transferability at varying spatio-temporal scales and to different remote sensing data sources was tested successfully. [ABSTRACT FROM AUTHOR]

Timely and accurate forecasting of crop grain protein content (GPC) is helpful in planning to acquire the desired target protein levels. A geographically weighted regression (GWR) model was estimated based on meteorological factors to predict the winter wheat GPC at the county level. In the Huang-Huai-Hai region, the grain protein content of winter wheat increased by 0.29% for every 1° increase in latitude. GPC prediction with this model was more precise than that of the multiple linear regressions (MLR) model. The correlation coefficient (R) and Akaike information criterion (AIC) value ranges were 0.26 ~ 0.66 and 1573.86 ~ 1710.70 for the GWR, and 0.06 ~ 0.46 and 1670.18 ~ 1939.76 for the MLR, respectively. Except for radiation in March (RAD03), radiation in April (RAD04) and radiation in May (RAD05), the sensitivity index of other monthly weather indicators to GPC had a high correlation with latitude. With 36° north latitude (L) as the limit, the correlation between RAD03 (R L<36 ° = 0.36, R L>36 ° = −0.29), RAD04 (R L<36 ° = 0.31, R L>36 ° = −0.35) and RAD05 (R L<36 ° = 0.20, R L>36 ° = −0.20) with latitude all showed an opposite trend. We highlight that spatial information needs to be considered when predicting county-level winter wheat GPC. • The grain protein content (GPC) is positively correlated with latitude. • The effect of meteorological factors on winter GPC showed spatial heterogeneity. • Geographically weighted model has advantages in winter GPC in heterogeneous region. [ABSTRACT FROM AUTHOR]

• Develop a within-growing season yield forecast system with random forest model. • Random forest model performs well in predicting grain yield in China. • We identified the most important stage-specific predictors determining crop yield. • The most important variable influencing yields varied with crop types. Accurate and timely crop yield forecasts can provide essential information to make conclusive agricultural policies and to conduct investments. Recent studies have used different machine learning techniques to develop such yield forecast systems for single crops at regional scales. However, no study has used multiple sources of environmental predictors (climate, soil, and vegetation) to forecast yields for three major crops in China. In this study, we adopted 7-year observed crop yield data (2013–2019) for three major grain crops (wheat, maize, and rice) across China, and three major data sets including climate, vegetation indices, and soil properties were used to develop a dynamic yield forecasting system based on the random forest (RF) model. The RF model showed good performance for estimating yields of all three crops with correlation coefficient (r) higher than 0.75 and normalized root means square errors (nRMSE) lower than 18.0%. Our results also showed that crop yields can be satisfactorily forecasted at one to three months prior to harvest. The optimum lead time for yield forecasting depended on crop types. In addition, we found the major predictors influencing crop yield varied between crops. In general, solar radiation and vegetation indices (especially during jointing to milk development stages) were identified as the main predictor for winter wheat; vegetation indices (throughout the growing season) and drought (especially during emergence to tasseling stages) were the most important predictors for spring maize; soil moisture (throughout the growing season) was the dominant predictor for summer maize, late rice, and mid rice; precipitation (especially during booting to heading stages) was the main predictor for early rice. Our study provides insights into practical crop yield forecasting and the understanding of yield response to environmental conditions at a large scale across China. The methods undertaken in this research can be easily implemented in other countries with available information on climate, soil, and vegetation conditions. [ABSTRACT FROM AUTHOR]