The expanding utilization of unmanned aerial vehicle (UAV) remote sensing (RS) technology has significantly advanced crop monitoring and detection. Despite its widespread application, the use of UAVs for examining rice grain starch accumulation (GSA) remains in its infancy. The preflowering nutritional organs' nonstructural carbohydrate transport and the postflowering plant's photosynthesis products are the primary sources of GSA. This study constructs a dynamic change curve based on the spectral index (SI) red edge re-normalized different vegetation index (RERDVI) before rice flowering. It introduces a novel indicator, the preflowering biomass accumulation dynamics (PBAD), identified through the dynamic curve's distinct shape characteristics. Results show that PBAD has a good correlation with the aboveground biomass (AGB) at different preflowering stages. After flowering, a nutrient distribution composite index (NDCI) is developed by combining SIs and color indices (CIs), providing a precise monitoring tool for the nitrogen harvest index (NHI), which is important in GSA. By comprehensively considering preflowering nonstructural carbohydrate accumulation (AGB), postflowering photosynthetic capacity (NHI), canopy temperature depression (CTD) sensitive to GSA, and meteorological factors (sunshine duration (SSD) and precipitation), a GSA estimation model based on multisource RS data fusion was constructed using a multiple linear regression (MLR), random forest regression (RFR), and extreme gradient boosting (XGBoost). This approach significantly improved the accuracy of GSA estimation, with the XGBoost model achieving a validation R(2 )of 0.76 and a root mean square error (RMSE) of 0.11 kg/m2 on a multiecological dataset, notably reducing the underestimation observed in traditional linear models.

Fractional vegetation cover (FVC) plays an important role in spectral unmixing, crop growth monitoring, crop light interception calculation, and yield estimation. However, the spectral reflectance would change with view zenith angles (VZAs), and retrieved FVC is also affected by VZAs. Therefore, in this paper, the observed multiangular (+45 degrees, degrees , +30 degrees, degrees , 0 degrees) degrees ) spectral datasets with different crops (wheat and rice), along with simulated spectral dataset, were used to explore method of correcting the influence of angle effect in FVC retrieval. Firstly, a simulated dataset of multi-angular hyperspectral data and directional FVC were constructed using the PROSAIL model, and a single-angular FVC retrieval model (Sin-FVC) was established based on the Gaussian process regression (GPR) algorithm. Then, the single-angular FVC (i.e., FVCB) B ) was converted into vertical FVC (FVCBv) B v ) based on the Beer-Lambert law. Finally, a Multi-FVC model was developed, which summed outputs of the SinFVC model (FVC0 degrees, 0 degrees , FVCBv) B v ) according to their respective weights (i.e., FVC correct ), to correct the angle effect. Using the pixel bisection model (FVCVI) VI ) as a comparison, different FVC retrieval models (Sin-FVC, Multi-FVC, and FVCVI) VI ) were compared and evaluated in terms of FVC retrieval accuracy, ability in weakening soil background, and LNC retrieval accuracy after spectral decomposition using different retrieved FVC (FVC0 degrees, 0 degrees , FVC-30 degrees v,-30 degrees v , FVC-45 degrees v,-45 degrees v , FVC correct ). The results showed that FVC correct had the highest FVC retrieval accuracy (RRMSE = 8.5 %) compared with FVC retrieved from other models (FVCVI, VI , FVC0 degrees, 0 degrees , FVC-30 degrees v and FVC-45 degrees v).-45 degrees v ). Meanwhile, when using the reflectance at NIR band in the black soil background as the baseline, after decoupling mixing spectra based on FVC correct , the relative offset (i.e., RO correct ) of each vegetation index (NDVI, EVI, SAVI, OSAVI) was minimum (least-ROcorrect- RO correct = 0.42 %). That is, the soil background was effectively suppressed by FVC correct spectral decomposition, along with the highest LNC retrieval accuracy, with an RRMSE of 14.2 %.

Spectral remote sensing can effectively, rapidly and non-destructively detect the nitrogen status of crop plants. Estimation of crop leaf nitrogen concentration (LNC, %) using canopy bidirectional reflectance factor (BRF) is an effective method to diagnose nitrogen deficiency in crops. It is challenging to estimate LNC with empirical remote sensing models because the variability of the canopy structure at different growth stages affects the model accuracy. Over the years, the canopy scattering coefficient [CSC, the ratio of BRF to directional area scattering factor (DASF)] has been used for LNC estimation by suppressing the effect of the canopy structure on BRF. However, this method often regards leaves as the main factor and has less consideration for the canopy structure effects on BRF caused by other organs (e.g., panicles). Incorporating the changes with the emergence of rice panicles into the DASF algorithm may generate reliable results in LNC estimation. Herein, we propose the PROSPECT-P model, which is based on the PROSPECT model and combines the panicle spectra to quantify the structural properties of the panicles and realize the simulation of the panicle albedo at different growth stages. Utilizing the spectral invariants theory, panicle-leaf structure correction factor (DASF (LP) ) was calculated based on canopy BRF, panicle albedo, leaf albedo and canopy component fraction. CSC after correction for panicle and leaf structure (CSC (LP) ) from 400-2500 nm can be obtained by the ratio of BRF and DASF (LP) . The CSC (LP) was further subjected to continuous wavelet analysis and achieved an accurate estimation of the LNC using four machine learning models. The results showed that when combined with eight wavelet features, CSC (LP) can accurately invert rice LNC using the random forest algorithm (R (2) = 0.81, RMSE = 0.30, RE = 14.02%), which was more exact than CSC (R (2) = 0.76, RMSE = 0.33, RE = 16.13%) that corrected only for leaf structure. Moreover, the results on UAV multispectral also showed that UAV-CSC (LP) predicted LNC by XGBoost model (R (2) = 0.61, RMSE = 0.35, RE = 17.27%) more accurately than the traditional method UAV-CSC (R (2) = 0.50, RMSE = 0.40, RE = 19.52%) on the independent test set. Herein, we propose the accurate inversion of crop growth parameters by remote sensing using PROSPECT-P to correct for panicle and leaf structure effects.

Context or problem: Timely and accurately predicting grain yield before harvest is of great importance for rice production management and grain trade.Objective or research question: Numerous methods based on remote sensing (RS) technology have explored to estimate rice grain yield. However, in most cases, these methods are negatively affected by the mixed pixels of RS images due to the effect of background and panicles.Methods: To resolve such issues, the abundance information of leaves, soil and water background and rice panicles were extracted from the multispectral images of unmanned aerial vehicle (UAV) for the multiple end-member spectral mixture analysis (MESMA) method. Based on the analysis of contribution of vegetation index (VI) and abundance (ABD) to rice grain yield, a rice grain yield prediction model was developed by combining multi-stage time-series, ABD and VI.Results: Results showed that MESMA can mitigate endmember variability in the estimation of rice yield, and achieve higher accuracy than conventional spectral mixture analysis (SMA). The NDRE and the sum of leaf and panicle abundance (ABDL+P) can be well correlated to grain yield before heading stage (R2 = 0.75) and at heading stage (R2 = 0.72), respectively. In comparison to them, the multi-stage time-series model consisted of VI and ABD [ n-ary sumation (VI & ABD)] produces the highest correlation (R2 = 0.80). It could also resolve the issues about the underestimation of yield using different datasets from various multi-spectral camera in comparison to the single-parameter models n-ary sumation VI and n-ary sumation ABD, and produce the good validation accuracy with R2 = 0.73, RRMSE= 0.22 and R2 = 0.75, RRMSE= 0.15, respectively. Conclusions: This study suggests that the combination of UAV multi-temporal VI and ABD data can achieve ac-curate prediction of rice grain yield.Implications or significance: This method effectively utilizes the optical information of leaves and rice panicles and reduces background effects, which provides a new idea for accurate rice yield prediction.

Rapid and accurate estimation of plant potassium accumulation (PKA) using hyperspectral remote sensing is of significance for the precise management of crop K fertilizer. This study focused on the separation of non-negative matrix factorization (NMF) for hyperspectral reflectance from the ground and unmanned aerial vehicle (UAV) platforms and its mitigation effect on the water and soil background. Pure vegetation spectra were extracted from the canopy mixed spectra using NMF, and then a partial least-squares regression (PLSR) model was established based on the extracted vegetation spectra and rice PKA to construct an estimation model of rice PKA. The results showed that the green light and red edge bands contributed significantly to the rice PKA estimation. NMF could effectively extract pure vegetation and water and soil spectra from mixed spectra, and enhance the green peak, red valley, and red edge information of the extracted vegetation spectra. Compared with spectral indices, the PLSR performed best for ground and UAV data. Besides, the R2 of the PLSR model based on NMF-extracted vegetation spectra increased by 15.15% to 0.76%, and the verified RMSE and RE decreased by 16.93% and 16.77% to 3.19 g m−2 and 45.07%, respectively. Hyperspectral dataset testing from different years, growth stages and varieties, and UAV platforms showed that NMF could improve the estimation accuracy of rice PKA. This study showed that NMF could be applied to both ground and UAV hyperspectral platforms to improve the estimation accuracy of rice K nutrition.

Potassium (K) is one of three main crop nutrients, and the high rate of potash fertilizer utilization (second only to nitrogen) leads to high prices. Therefore, efficient application, as well as rapid and time monitoring of K in crops is essential. Several turnover box and field experiments were conducted across multiple years and cultivation factors (i.e., potassium levels and plant varieties) yielding 340 groups of leaf samples with different K contents; these samples were used to examine the relationship between reflectance spectra (350–2500 nm) and leaf K content (LKC). The correlation between LKC and the two-band spectral indices computed with random two bands from 350 to 2500 nm were determined for the published K vegetation indices in rice. Results showed that the spectral reflectance, R, of the shortwave infrared (1300–2000 nm) region was sensitive to the K levels and significantly correlated with rice LKC. New shortwave infrared two-band spectral indices, Normalized difference spectral index [NDSI (R1705, R1385)], Ratio spectral index [RSI (R1385, R1705)], and Difference spectral index [DSI (R1705, R1385)], showed good correlations with LKC (R2 up to 0.68). Moreover, the three-band spectral indices (R1705 − R700)/(R1385 − R700) and (R1705 − R1385)/(R1705 + R1385 − 2 × R704) were developed by adding red edge bands to improve accuracy. Three-band spectral indices had an improved prediction accuracy for rice LKC (R2 up to 0.74). However, several previously published K-sensitive vegetation indices did not yield good results in this study. Validation with independent samples showed that the indices (R1705 − R700)/(R1385 − R700) and (R1705 − R1385)/(R1705 + R1385 − 2 × R704) had higher accuracies and stabilities than two-band indices and are suitable for quantitatively estimating rice LKC. The widescale application of these proposed vegetation indices in this paper still needs to be verified in different environmental conditions. This study provides a technical basis for LKC monitoring using spectral remote sensing in rice.

Photosynthesis is the basis of crop yield and quality. Real-time, quantitative monitoring of crop photosynthetic parameters is important to assess crop growth status, and to predict yield and quality. In the present study, we conducted two field experiments using two rice cultivars (Japonica and Indica), and nitrogen levels and light response curves (LRCs) of different leaf positions at different growth stages were determined. The leaf maximum net photosynthesis (Pn-max) and initial quantum efficiency (alpha) were estimated using LRCs and then the leaf layer maximum net photosynthesis (Pnl-max) and initial quantum efficiency (alpha(1)) were estimated using the Gaussian integration method. The results showed that the dynamic change characteristics of Pnl-max and alpha(1) at the rice leaf layer under the different growth stages presented the same trend: Increasing first and then decreasing. The relationship between the photosynthetic parameters of the leaf layer and multi-spectral vegetation indices obtained from an unmanned aerial vehicle (UAV) multi-spectral reflectance showed that the modified structure-insensitive pigment index (SIPIm(R-720-R-550)/(R-800-R-680)) correlated with an R-2 of 0.72 and 0.61 for Pnl-max and alpha(1),, respectively. Therefore, Pnl-max and alpha(1) of the rice leaf layer could be obtained quickly by UAV. In addition, the leaf layer light response curve (LRC1) model could be estimated by combining the canopy respiration (R-d) ob- tained by accumulating different leaf layers' respiration rates with Pnl-max and alpha(1). Daily photosynthetically active radiation (PAR) variation, measured using a QSO-S PAR sensor, was used as the input parameter of an LRC1 model. This allowed the prediction of daily variation of rice canopy photosynthesis based on UAV and the LRC1 model.

The emergence of rice panicle substantially changes the spectral reflectance of rice canopy and, as a result, decreases the accuracy of leaf area index (LAI) that was derived from vegetation indices (VIs). From a four-year field experiment with using rice varieties, nitrogen (N) rates, and planting densities, the spectral reflectance characteristics of panicles and the changes in canopy reflectance after panicle removal were investigated. A rice “panicle line”—graphical relationship between red-edge and near-infrared bands was constructed by using the near-infrared and red-edge spectral reflectance of rice panicles. Subsequently, a panicle-adjusted renormalized difference vegetation index (PRDVI) that was based on the “panicle line” and the renormalized difference vegetation index (RDVI) was developed to reduce the effects of rice panicles and background. The results showed that the effects of rice panicles on canopy reflectance were concentrated in the visible region and the near-infrared region. The red band (670 nm) was the most affected by panicles, while the red-edge bands (720–740 nm) were less affected. In addition, a combination of near-infrared and red-edge bands was for the one that best predicted LAI, and the difference vegetation index (DI) (976, 733) performed the best, although it had relatively low estimation accuracy (R2 = 0.60, RMSE = 1.41 m2/m2). From these findings, correcting the near-infrared band in the RDVI by the panicle adjustment factor (θ) developed the PRDVI, which was obtained while using the “panicle line”, and the less-affected red-edge band replaced the red band. Verification data from an unmanned aerial vehicle (UAV) showed that the PRDVI could minimize the panicle and background influence and was more sensitive to LAI (R2 = 0.77; RMSE = 1.01 m2/m2) than other VIs during the post-heading stage. Moreover, of all the assessed VIs, the PRDVI yielded the highest R2 (0.71) over the entire growth period, with an RMSE of 1.31 (m2/m2). These results suggest that the PRDVI is an efficient and suitable LAI estimation index.