For the problem of multi-dimensional feature redundancy in remote sensing detection of wheat stripe rust using reflectance spectrum and solar-induced chlorophyll fluorescence (SIF), a feature selection and disease index (DI) monitoring model combining mRMR and XGBoost algorithm was proposed in this study. Firstly, characteristic wavelengths selected by successive projections algorithm (SPA) were combined with the vegetation indices, trilateral parameters, and canopy SIF parameters to constitute the initial feature set. Then, the max-relevance and min-redundancy (mRMR) algorithm and correlation coefficient (CC) analysis were used to reduce the dimensionality of the initial feature set, respectively. Features selected by mRMR and CC were input as independent variables into the extreme gradient boosting regression (XGBoost) and gradient boosting regression tree (GBRT) to monitor the severity of stripe rust. The experimental results show that, compared with CC analysis, the monitoring accuracy of the features selected by mRMR in the XGBoost and GBRT models increased by 12% and 17% on average, respectively. Meanwhile, the mRMR-XGBoost model achieved the best monitoring accuracy (R² = 0.8894, RMSE = 0.1135). The R² between the measured DI and predicted DI of mRMR-XGBoost was improved by an average of 5%, 12%, and 22% compared with mRMR-GBRT, CC-XGBoost, and CC-GBRT models. These results suggested that XGBoost is more suitable for the remote sensing monitoring of wheat stripe rust, and mRMR has more advantages than the commonly used CC analysis in feature selection. Field survey data validation results also confirm that the mRMR-XGBoost algorithm has excellent monitoring applicability and scalability. The proposed model could provide a reference for data dimensionality reduction and crop disease index monitoring based on hyperspectral data.

Sun-induced chlorophyll fluorescence (SIF) has shown potential in quantifying plant responses to environmental changes by which abiotic drivers are dominated. However, SIF is a mixed signal influenced by factors such as leaf physiology, canopy structure, and sun-sensor geometry. Whether the physiological information contained in SIF can better quantify crop disease stresses dominated by biological drivers, and clearly explain the physiological variability of stressed crops, has not yet been sufficiently explored. On this basis, we took winter wheat naturally infected with stripe rust as the research object and conducted a study on the responses of physiological signals and reflectivity spectrum signals to crop disease stress dominated by biological drivers, based on in situ canopy-scale and leaf-scale data. Physiological signals include SIF, SIF_yield (normalized by absorbed photosynthetically active radiation), fluorescence yield (Φ_F) retrieved by NIRvP (non-physiological components of canopy SIF) and relative fluorescence yield (Φ_F-r) retrieved by near-infrared radiance of vegetation (NIRvR). Reflectance spectrum signals include normalized difference vegetation index (NDVI) and near-infrared reflectance of vegetation (NIRv). At the canopy scale, six signals reached extremely significant correlations (P < 0.001) with disease severity levels (SL) under comprehensive experimental conditions (SL without dividing the experimental samples) and light disease conditions (SL < 20%). The strongest correlation between NDVI and SL (R = 0.69) was observed under the comprehensive experimental conditions, followed by NIRv (R = 0.56), Φ_F-r (R = 0.53) and SIF (R = 0.51), and the response of Φ_F (R = 0.45) and SIF_yield (R = 0.34) to SL was weak. Under lightly diseased conditions, Φ_F-r (R = 0.62) showed the strongest response to disease, followed by SIF_yield (R = 0.60), SIF (R = 0.56) and NIRv (R = 0.54). The weakest correlation was observed between Φ_F and SL (R = 0.51), which also showed a result approximating NDVI (R = 0.52). In the case of a high level of crop disease severity, NDVI showed advantages in disease monitoring. In the early stage of crop diseases, which we pay more attention to, compared with SIF and reflectivity spectrum signals, Φ_F-r estimated by the newly proposed ‘NIRvR approach’ (which uses SIF together with NIRvR (i.e., SIF/ NIRvR) as a substitute for Φ_F) showed superior ability to monitor crop physiological stress, and was more sensitive to plant physiological variation. At the leaf scale, the response of SIF to SL was stronger than that of NDVI. These results validate the potential of Φ_F-r estimated by the NIRvR approach to monitoring disease stress dominated by biological drivers, thus providing a new research avenue for quantifying crop responses to disease stress.

Grain protein content (GPC) is an important indicator of nutritional quality of rice. In this study, nitrogen fertilization experiments were conducted to monitor GPC for high-quality Indica rice varieties Meixiangzhan 2 (V1) and Wufengyou 615 (V2) in 2019 and 2020. Three types of parameters, including photosynthetic sensitive vegetation indices (VIs), canopy leaf area index (LAI), and crop plant nitrogen accumulation (PNA), obtained from UAV hyperspectral images were used to estimate rice GPC. Two-dimensional and three-dimensional GPC indices were constructed by combining any two of the three types of parameters and all three, respectively, based on the Euclidean distance method. The R² and RMSE of the two-dimensional GPC index model for variety V1 at the tillering stage were 0.81 and 0.40% for modeling and 0.95 and 0.38% for validation, and 0.91 and 0.27% for modeling and 0.83 and 0.36% for validation for variety V2. The three-dimensional GPC index model for variety V1 had R² and RMSE of 0.86 and 0.34% for modeling and 0.78 and 0.45% for validation, and 0.97 and 0.17% for modeling and 0.96 and 0.17% for validation for variety V2 at the panicle initiation stage. At the heading stage, the R² and RMSE of the three-dimensional model for variety V1 were 0.92 and 0.26% for modeling and 0.91 and 0.37% for validation, and 0.96 and 0.20% for modeling and 0.99 and 0.15% for validation for variety V2. These results demonstrate that the GPC monitoring models incorporating multiple crop growth parameters based on Euclidean distance can improve GPC estimation accuracy and have the potential for field-scale GPC monitoring.

Alpine meadow ecosystems are extremely vulnerable to climate change and serve an essential function in terrestrial carbon sinks. Accurately estimating their gross primary productivity (GPP) is essential for understanding the global carbon cycle. Solar-induced chlorophyll fluorescence (SIF), as a companion product directly related to plant photosynthesis process, has become an attractive pathway for estimating GPP accurately. To date, the quantitative SIF-GPP relationship in terrestrial ecosystems is not yet clear. Especially, red SIF and far-red SIF present differences in their ability to track GPP under different environmental conditions. In this study, we investigated the performance of SIF at both red and far-red band in monitoring the GPP of an alpine meadow ecosystem based on continuous tower-based observations in 2019 and 2020. The results show that the canopy red SIF (${\mathrm{SIF}}_{\mathrm{Red}}$) and far-red SIF (${\mathrm{SIF}}_{\text{Far-red}}$) were both strongly correlated with GPP. ${\mathrm{SIF}}_{\mathrm{Red}}$ was comparable to ${\mathrm{SIF}}_{\text{Far-red}}$ for monitoring GPP based on comparisons of both half-hourly averaged and daily averaged datasets. Moreover, the relationship between ${\mathrm{SIF}}_{\mathrm{Red}}$ and GPP was linearly correlated, while the relationship between ${\mathrm{SIF}}_{\text{Far-red}}$ and GPP tended to be nonlinear. At a diurnal scale, dramatic changes in photosynthetically active radiation (PAR), air temperature (Ta), and vapor pressure deficit (VPD) all had effects on the slope of the linear fitted line with zero intercept for ${\mathrm{SIF}}_{\mathrm{Red}}$-GPP and ${\mathrm{SIF}}_{\text{Far-red}}$-GPP, and the effect on the slope of the linear fitted line with zero intercept for ${\mathrm{SIF}}_{\text{Far-red}}$-GPP was obviously stronger than that for ${\mathrm{SIF}}_{\mathrm{Red}}$-GPP. PAR was the dominant factor among the three environmental factors in determining the diurnal variation of the slope of SIF-GPP. At a seasonal scale, the ${\mathrm{SIF}}_{\text{Far-red}}$/GPP was susceptible to PAR, Ta, and VPD, while the ${\mathrm{SIF}}_{\mathrm{Red}}$/GPP remained relatively stable at different levels of Ta and VPD, and it was only weakly affected by PAR, suggesting that ${\mathrm{SIF}}_{\mathrm{Red}}$ was more consistent than ${\mathrm{SIF}}_{\text{Far-red}}$ with GPP in response to seasonal variations in environmental factors. These results indicate that ${\mathrm{SIF}}_{\mathrm{Red}}$ has more potential than ${\mathrm{SIF}}_{\text{Far-red}}$ for monitoring the GPP of alpine meadow ecosystems and can also assist researchers in gaining a more comprehensive understanding of the diversity of SIF-GPP relationships in different ecosystems.

The leaf area index (LAI) played a crucial role in ecological, hydrological, and climate models. The normalized difference vegetation index (NDVI) has been a widely used tool for LAI estimation. However, the NDVI quickly saturates in dense vegetation and is susceptible to soil background interference in sparse vegetation. We proposed a multi-angular NDVI (MAVI) to enhance LAI estimation using tower-based multi-angular observations, aiming to minimize the interference of soil background and saturation effects. Our methodology involved collecting continuous tower-based multi-angular reflectance and the LAI over a three-year period in maize cropland. Then we proposed the MAVI based on an analysis of how canopy reflectance varies with solar zenith angle (SZA). Finally, we quantitatively evaluated the MAVI’s performance in LAI retrieval by comparing it to eight other vegetation indices (VIs). Statistical tests revealed that the MAVI exhibited an improved curvilinear relationship with the LAI when the NDVI is corrected using multi-angular observations (R² = 0.945, RMSE = 0.345, rRMSE = 0.147). Furthermore, the MAVI-based model effectively mitigated soil background effects in sparse vegetation (R² = 0.934, RMSE = 0.155, rRMSE = 0.157). Our findings demonstrated the utility of tower-based multi-angular spectral observations in LAI retrieval, having the potential to provide continuous data for validating space-borne LAI products. This research significantly expanded the potential applications of multi-angular observations.

Solar-induced chlorophyll fluorescence (SIF) has great advantages in the remote sensing detection of crop stress. However, under stripe rust stress, the effects of canopy structure and leaf physiology on the variations in canopy SIF are unclear, and these influencing factors are entangled during the development of disease, resulting in an unclear coupling relationship between SIF_canopy and the severity level (SL) of disease, which affects the remote sensing detection accuracy of wheat stripe rust. In this study, the observed canopy SIF was decomposed into NIR_VP, which can characterize the canopy structure, and SIF_tot, which can sensitively reflect the physiological status of crops. Additionally, the main factors driving the variations in canopy SIF under different disease severities were analyzed, and the response characteristics of SIF_canopy, NIR_VP, and SIF_tot to SL under stripe rust stress were studied. The results showed that when the severity level (SL) of disease was lower than 20%, NIR_VP was more sensitive to variation in SIF_canopy than SIF_tot, and the correlation between SIF_tot and SL was 6.6% higher than that of SIF_canopy. Using the decomposed SIF_tot component allows one to detect the stress state of plants before variations in vegetation canopy structure and leaf area index and can realize the early diagnosis of crop diseases. When the severity level (SL) of disease was in the state of moderate incidence (20% < SL ≤ 45%), the variation in SIF_canopy was affected by both NIR_VP and SIF_tot, and the detection accuracy of SIF_canopy for wheat stripe rust was better than that of the NIR_VP and SIF_tot components. When the severity level (SL) of disease reached a severe level (SL > 45%), SIF_tot was more sensitive to the variation in SIF_canopy, and NIR_VP reached a highly significant level with SL, which could better realize the remote sensing detection of wheat stripe rust disease severity. The research results showed that analyzing variations in SIF_canopy by using the decomposed canopy structure and physiological response signals can effectively capture additional information about plant physiology, detect crop pathological variations caused by disease stress earlier and more accurately, and promote crop disease monitoring and research progress.