Xinwei Li

Accurate and efficient estimation of leaf chlorophyll content (LCC) is essential for crop physiological monitoring and precision agriculture. Current remote sensing approaches for LCC estimation mainly rely on machine learning and deep learning methods, which exhibit strong nonlinear modelling capabilities but require sufficient and representative samples. Radiative transfer models (RTMs) can simulate canopy spectral response based on physical mechanisms and generate large-scale datasets, thereby overcoming the limitations of purely data-driven approaches. However, their applications under diverse environmental conditions remain challenging due to uncertainties caused by model parameterization and discrepancies between simulated and observed data. This study proposed a hybrid framework, PROSAIL-DNN-TL, integrating the PROSAIL RTM with full-network fine-tuned transfer learning for winter wheat LCC estimation. First, 85,680 simulated samples generated by PROSAIL were used to pre-train a deep neural network (DNN), and field observations (n = 270) were subsequently used for full-network fine-tuning. A conventional DNN model trained only with field data and machine learning models based on different feature selection strategies were established for comparison. The results showed that: (1) UVE-selected bands were distributed across visible, red-edge, and near-infrared regions; CARS retained more bands in the visible region, whereas SPA mainly selected near-infrared bands; (2) RR and SVR achieved higher accuracy than KNN, with the full-spectrum and CARS feature sets showing superior performance and SPA showing the lowest accuracy; (3) PROSAIL-DNN-TL achieved reliable LCC estimation performance, with an average test R² of 0.79 across 500 repeated experiments, and showed improved performance compared with conventional machine learning models; (4) PROSAIL-DNN-TL achieved reliable predictive performance when evaluated using datasets collected during the 2024 and 2025 growing seasons, with R² values of 0.75 and 0.92, respectively. Overall, the proposed hybrid framework integrates physical knowledge from RTMs with data-driven modelling through transfer learning, providing a potential approach for crop chlorophyll monitoring. However, the SPAD-to-LCC conversion used in this study was not specifically calibrated for the instrument and crop, and therefore, the absolute LCC values should be interpreted with caution.

Leaf chlorophyll content (LCC) is a key indicator for assessing the photosynthetic capacity and nutritional status of winter wheat. Among traditional LCC estimation methods, empirical models lack a physical basis and have poor generalisability, while physical models are widely applicable but suffer from ill-posed inversion problems. Hybrid inversion methods, which integrate radiation transfer models such as PROSAIL with machine learning, offer both the interpretability of physical models and the efficiency of machine learning; however, they are still affected by the domain shift between simulated and measured data, which limits their generalisation performance. Transfer Component Analysis (TCA), a domain adaption method, can effectively alleviate this problem. In this study, hyperspectral and LCC data were collected in the field, and simulated data were generated using the PROSAIL model; a sensitivity analysis was conducted to identify LCC-sensitive bands. A genetic algorithm was applied to the measured data for band selection and, together with the results of the sensitivity analysis, yielded an optimal set of 30 characteristic bands for subsequent modelling. Three datasets were constructed: measured data only, a direct mixture of measured and simulated data, and a TCA-fused mixture of measured and simulated data. Four models—gradient boosting regression (GBR), random forest (RF), support vector regression (SVR) and deep neural network (DNN)—were developed for each dataset. The results show that: (1) the LCC-sensitive bands are concentrated in the 450–660 nm and 680–720 nm ranges; (2) the model built on the TCA-fused data (R² = 0.722, RMSE = 6.792) outperformed those built on the measured-only data (R² = 0.682, RMSE = 7.259) and the directly mixed data (R² = 0.616, RMSE = 7.976); (3) for the TCA-fused data, the four models differed considerably in accuracy, with SVR performing best (R² = 0.723, RMSE = 5.363), followed by RF (R² = 0.630, RMSE = 6.201) and GBR (R² = 0.575, RMSE = 6.650), whereas the DNN performed worst (R² = 0.388, RMSE = 7.947), probably owing to the limited sample size.

Remote sensing using unmanned aerial vehicles (UAV) combined with machine learning (ML) has significantly advanced field-scale prediction of aboveground biomass. Although ensemble learning frameworks (ELFs) typically outperform individual ML algorithms in accuracy, systematic evaluations of meta learner selection and the effects of base learner quantity and diversity remain limited. Leveraging vegetation indices and texture features extracted from multi-temporal UAV imagery, ELFs were constructed incorporating nine ML algorithms with three meta learners, Linear model, Random forest and Bayesian model averaging (BMA), to systematically evaluate how base learner configuration affects prediction accuracy. Using the fused feature set of sensitive vegetation indices and texture features, Gaussian process regression (GPR) achieved the highest accuracy among all base learners, with R2 = 0.769 and RMSE = 1.83 t·ha–1. Also, the three meta learners outperformed the best base learner, with the BMA meta learner yielding the superior accuracy (R2 = 0.795, RMSE = 1.73 t·ha–1). However, meta learner performance depended strongly on the composition of the base learners pool, stability was optimal with five base learners, and maximum accuracy was achieved by hybrid ensembles that combined linear-kernel models with GPR. This study highlights the importance of both meta learner selection and base learner composition in ELFs for aboveground biomass prediction in rice. These findings offer methodological guidance for UAV-based high-precision monitoring of crop AGB, with practical implications for precision agriculture and crop management.

Chlorophyll serves as a crucial indicator for crop growth monitoring and reflects the health status of crops. Hyperspectral remote sensing technology, leveraging its advantages of repeated observations and high-throughput analysis, provides an effective approach for non-destructive chlorophyll monitoring. However, determining the optimal spectral scale remains the primary bottleneck constraining the widespread application of hyperspectral remote sensing in crop chlorophyll estimation: excessively fine spectral scale readily introduces redundant information, leading to dramatically increased data dimensions and reduced computational efficiency; conversely, overly coarse spectral scale risks losing critical spectral features such as absorption peaks and reflection troughs, thereby compromising model accuracy. Therefore, establishing an appropriate spectral scale that effectively preserves spectral feature information while maintaining computational efficiency is crucial for enhancing the accuracy and practicality of chlorophyll remote sensing estimation. To address this, this study proposes a three-dimensional analytical framework integrating “spectral scale—machine learning algorithm—crop growth stage” to systematically solve the scale optimization problem. Ground-truth measurements and hyperspectral data from five growth stages of winter wheat in Fengyang County, Anhui Province, were collected. Spectral bands sensitive to chlorophyll were analyzed, and four modeling methods—Ridge Regression (RR), K-Nearest Neighbors (KNN), Random Forest (RF), and Support Vector Regression (SVR)—were employed to integrate data from different spectral scales with respective bandwidths of 2, 3, 5, 7, 10, 20, and 50 nanometers (nm). The results evaluated the response characteristics of raw band reflectance to chlorophyll values and its impact on machine learning-based chlorophyll estimation across different spectral scales. Results indicate: (1) Canopy spectra significantly correlated with winter wheat chlorophyll primarily reside in the red and red-edge bands; (2) For single-scale analysis, larger spectral scales (10, 20 nm) enhance monitoring accuracy compared to 1 nm high-resolution data, while medium and small scales (5, 7 nm) may degrade accuracy due to redundant noise introduction. (3) Integrating growth stages, spectral scales, and machine learning revealed optimal monitoring accuracy during the jointing and heading stages using 1–5 nm spectral scales combined with the KNN algorithm. For the booting, flowering, and grain filling stages, the highest accuracy was achieved using 20–50 nm spectral scales combined with either the KNN or RF algorithm. The results indicate that high-precision chlorophyll inversion for winter wheat does not rely on a single fixed model or scale, but rather on the dynamic adaptation of the “scale-model-growth stage” triad. The proposed systematic framework not only provides a theoretical basis for chlorophyll monitoring using multi-platform remote sensing data, but also offers methodological support for future crop-sensing sensor design and data processing strategy optimization.

Chlorophyll plays a vital role in wheat growth and fertilization management. Accurate and efficient estimation of chlorophyll content is crucial for providing a scientific foundation for precision agricultural management. Unmanned aerial vehicles (UAVs), characterized by high flexibility, spatial resolution, and operational efficiency, have emerged as effective tools for estimating chlorophyll content in wheat. Although multi-source data derived from UAV-based multispectral imagery have shown potential for wheat chlorophyll estimation, the importance of multi-source deep feature fusion has not been adequately addressed. Therefore, this study aims to estimate wheat chlorophyll content by integrating spectral and textural features extracted from UAV multispectral imagery, in conjunction with partial least squares regression (PLSR), random forest regression (RFR), deep neural network (DNN), and a novel multi-source deep feature neural network (MDFNN) proposed in this research. The results demonstrate the following: (1) Except for the RFR model, models based on texture features exhibit superior accuracy compared to those based on spectral features. Furthermore, the estimation accuracy achieved by fusing spectral and texture features is significantly greater than that obtained using a single type of data. (2) The MDFNN proposed in this study outperformed other models in chlorophyll content estimation, with an R² of 0.850, an RMSE of 5.602, and an RRMSE of 15.76%. Compared to the second-best model, the DNN (R² = 0.799, RMSE = 6.479, RRMSE = 18.23%), the MDFNN achieved a 6.4% increase in R², and 13.5% reductions in both RMSE and RRMSE. (3) The MDFNN exhibited strong robustness and adaptability across varying years, wheat varieties, and nitrogen application levels. The findings of this study offer important insights into UAV-based remote sensing applications for estimating wheat chlorophyll under field conditions.

Forestry enterprises play a pivotal role in economic development, ecological civilization construction, and sustainable development. This study employs GIS-based spatial analysis to examine the distribution patterns and interrelationships of forestry enterprises, investigating their key determinants and spatial heterogeneity. The findings provide valuable insights for policymakers aiming to optimize industrial structures and enhance national ecological security. This research develops a comprehensive evaluation index system to assess the factors influencing forestry industry development in China. Nine factors are considered: human resources, economic development, industrial structure, technological support, trade development, financial environment, natural conditions, urbanization, and transportation. Using panel data from 367 cities in 2020, the Multiscale Geographically Weighted Regression (MGWR) method quantifies the influence of these factors and their spatial variations. The results show the following. (1) Forestry enterprises in China exhibit persistent spatial clustering. The eastern regions have a notably higher concentration than the western regions, and new enterprises are increasingly concentrated in a few hotspot cities in the east. (2) The spatial center of forestry enterprises has steadily moved southeast. Initially, the distribution was balanced in the eastern regions, but it has become highly concentrated in the southeastern coastal areas. (3) Regarding spatial autocorrelation, regions within the northwest cold spot cluster have been disappearing entirely. The northeast and southwest hotspot clusters have shrunk significantly, while the southeast hotspot cluster has remained large. (4) Permanent population size and green land area are the most strongly positively correlated with forestry enterprise distribution. Patent authorizations, orchard area, and forest land area also show positive effects. In contrast, road density and total import/export volume are negatively correlated with the number of forestry enterprises. This aligns with the structure of China’s forestry industry, which relies more on natural resources and market demand than on economic development level or financial environment. (5) The factors influencing forestry enterprise distribution show significant spatial variation, driven by regional factors such as resources, economy, and population. These factors ultimately determine the spatiotemporal distribution of forestry enterprises. This study provides data-driven insights to optimize the distribution of forestry industries and formulate more effective ecological protection policies.

The quality of the image data and the potential to invert crop growth parameters are essential for effectively using unmanned aerial vehicle (UAV)-based sensor systems in precision agriculture (PA). However, the existing research falls short in providing a comprehensive examination of sensor data quality and the inversion potential of crop growth parameters, and there is still ambiguity regarding how the quality of data affects the inversion potential. Therefore, this study explored the application potential of RGB and multispectral (MS) images acquired from three lightweight UAV platforms in the realm of PA: the DJI Mavic 2 Pro (M2P), Phantom 4 Multispectral (P4M), and Mavic 3 Multispectral (M3M). The reliability of pixel-scale data quality was evaluated based on image quality assessment metrics, and three winter wheat growth parameters, above-ground biomass (AGB), plant nitrogen content (PNC) and soil and plant analysis development (SPAD), were inverted using machine learning models based on multi-source image features at the plot scale. The results indicated that the RGB image quality from the M3M outperformed that of the M2P, while the MS image quality was marginally superior to that of the P4M. Nevertheless, these advantages in pixel-scale data quality did not improve inversion accuracy for crop parameters at the plot scale. Spectral features (SFs) derived from the P4M-based MS sensor demonstrated significant advantages in AGB inversion (R² = 0.86, rRMSE = 27.47%), while SFs derived from the M2P-based RGB camera exhibited the best performance in SPAD inversion (R² = 0.60, rRMSE = 7.67%). Additionally, combining spectral and textural features derived from the P4M-based MS sensor yielded the highest accuracy in PNC inversion (R² = 0.82, rRMSE = 14.62%). This study clarified the data quality of three prevalent UAV mounted sensor systems in PA and their influence on parameter inversion potential, offering guidance for selecting appropriate sensors and monitoring key crop growth parameters.

Timely and accurate yield estimation is essential for effective crop management and the grain trade. Remote sensing has emerged as a valuable tool for monitoring rice yields; however, many studies concentrate on a single period or simply aggregate multiple periods, neglecting the complexities underlying yield formation. The study enhances yield estimation by integrating cumulative time series vegetation indices (VIs) from multispectral (MS) and RGB (Red, Green, Blue) sensors to identify optimal combinations of growth periods. We utilized two unmanned aerial vehicle to capture spectral information from rice canopies through MS and RGB sensors. By analyzing the correlations between vegetation indices from different sensors and rice yields, the optimal MS-VIs and RGB-VIs for each period were identified. Following this, the relationship between the cumulative time series of MS-VIs, RGB-VIs, and rice yields was further examined. The results demonstrate that the booting stage is a crucial growth period influencing rice yield, with VIs exhibiting increased correlation with yield, peaking during this stage before declining. For the MS sensor, the rice yield model, based on the cumulative time series of MS-VIs from the tillering stage to the panicle initiation stage, achieves optimal accuracy (R² = 0.722, RRMSE = 0.555). For the RGB sensor, the rice yield model, based on the cumulative time series of RGB-VIs from the tillering stage to the grain-filling stage, yields the highest accuracy (R² = 0.727, RRMSE = 0.526). In comparison, the multi-sensor rice yield model, which combines the cumulative time series of MS-VIs from the tillering stage and RGB-VIs from the panicle initiation to grain-filling stages, achieves the highest accuracy with R² = 0.759 and RRMSE = 0.513. These findings suggest that cumulative time series VIs and the integration of multiple sensors enhance yield prediction accuracy, providing a comprehensive approach for estimating rice yield dynamics and supporting precision agriculture and informed crop management.

Chlorophyll is a crucial indicator for monitoring crop growth and assessing nutritional status. Hyperspectral remote sensing plays an important role in precision agriculture, offering a non-destructive approach to predicting leaf chlorophyll. However, crop canopy spectra often face background noise and data redundancy challenges. To tackle these issues, this study develops an integrated processing strategy incorporating multiple preprocessing techniques, sequential module fusion, and feature mining methods. Initially, the original spectrum (OS) from 2021, 2022, and the fusion year underwent preprocessing through Fast Fourier Transform (FFT) smoothing, multiple scattering correction (MSC), the first derivative (FD), and the second derivative (SD). Secondly, feature mining was conducted using Competitive Adaptive Reweighted Sampling (CARS), Iterative Retention of Information Variables (IRIV), and Principal Component Analysis (PCA) based on the optimal preprocessing order module fusion data. Finally, Partial Least Squares Regression (PLSR) was used to construct a prediction model for winter wheat SPAD to compare the prediction effects in different years and growth stages. The findings show that the preprocessing sequential module fusion of FFT-MSC (firstly pre-processing using FFT, and secondly secondary processing of FFT spectral data using MSC) effectively reduced issues such as noisy signals and baseline drift. The FFT-MSC-IRIV-PLSR model (based on the combined FFT-MSC preprocessed spectral data, feature screening using IRIV, and then combining with PLSR to construct a prediction model) predicts SPAD with the highest overall accuracy, with an R² of 0.79–0.89, RMSE of 4.51–5.61, and MAE of 4.01–4.43. The model performed best in 2022, with an R² of 0.84–0.89 and RMSE of 4.51–6.74. The best prediction during different growth stages occurred in the early filling stage, with an R² of 0.75 and RMSE of 0.58. On the basis of this research, future work will focus on optimizing the data processing process and incorporating richer environmental data, so as to further enhance the predictive capability and applicability of the model.

Plant nitrogen concentration (PNC) is a key indicator reflecting the growth and development status of plants. The timely and accurate monitoring of plant PNC is of great significance for the refined management of crop nutrition in the field. The rapidly developing sensor technology provides a powerful means for monitoring crop PNC. Although RGB images have rich spatial information, they lack the spectral information of the red edge and near infrared bands, which are more sensitive to vegetation. Conversely, multispectral images offer superior spectral resolution but typically lag in spatial detail compared to RGB images. Therefore, the purpose of this study is to improve the accuracy and efficiency of crop PNC monitoring by combining the advantages of RGB images and multispectral images through image-fusion technology. This study was based on the booting, heading, and early-filling stages of winter wheat, synchronously acquiring UAV RGB and MS data, using Gram–Schmidt (GS) and principal component (PC) image-fusion methods to generate fused images and evaluate them with multiple image-quality indicators. Subsequently, models for predicting wheat PNC were constructed using machine-selection algorithms such as RF, GPR, and XGB. The results show that the RGB_B1 image contains richer image information and more image details compared to other bands. The GS image-fusion method is superior to the PC method, and the performance of fusing high-resolution RGB_B1 band images with MS images using the GS method is optimal. After image fusion, the correlation between vegetation indices (VIs) and wheat PNC has been enhanced to varying degrees in different growth periods, significantly enhancing the response ability of spectral information to wheat PNC. To comprehensively assess the potential of fused images in estimating wheat PNC, this study fully compared the performance of PNC models before and after fusion using machine learning algorithms such as Random Forest (RF), Gaussian Process Regression (GPR), and eXtreme Gradient Boosting (XGB). The results show that the model established by the fusion image has high stability and accuracy in a single growth period, multiple growth periods, different varieties, and different nitrogen treatments, making it significantly better than the MS image. The most significant enhancements were during the booting to early-filling stages, particularly with the RF algorithm, which achieved an 18.8% increase in R², a 26.5% increase in RPD, and a 19.7% decrease in RMSE. This study provides an effective technical means for the dynamic monitoring of crop nutritional status and provides strong technical support for the precise management of crop nutrition.