Leaf nitrogen content (LNC) is a vital agronomic parameter in rice, commonly used to evaluate photosynthetic capacity and diagnose crop nutrient levels. Nitrogen deficiency can significantly reduce yield, underscoring the importance of accurate LNC estimation for practical applications. This study utilizes hyperspectral UAV imagery to acquire rice canopy data, applying various machine learning regression algorithms (MLR) to develop an LNC estimation model and create a nitrogen concentration distribution map, offering valuable guidance for subsequent field nitrogen management. The analysis incorporates four types of spectral data extracted throughout the rice growth cycle: original reflectance bands (OR bands), vegetation indices (VIs), first-derivative spectral bands (FD bands), and hyperspectral variable parameters (HSPs) as model inputs, while measured nitrogen concentration serves as the output. Results demonstrate that the random forest regression (RFR) and gradient boosting decision tree (GBDT) algorithms performed effectively, with the GBDT achieving the highest average R² of 0.76 across different nitrogen treatments. Among the nitrogen estimation models for various rice varieties, RFR exhibited superior accuracy, achieving an R² of 0.95 for the SuXiangJing100 variety, while the GBDT reached 0.93. Meanwhile, the support vector machine regression (SVMR) showed slightly lower accuracy, and partial least-squares regression (PLSR) was the least effective. This study developed an LNC estimation method applicable to the whole growth stage of common rice varieties. The method is suitable for estimating rice LNC across different growth stages, varieties, and nitrogen treatments, and it also provides a reference for nitrogen estimation and fertilization planning at flight altitudes other than the 120 m used in this study.

Surface water quality is critical to ecosystem health and sustainable development, yet conventional monitoring falls short of spatiotemporally continuous assessment. Remote sensing coupled with machine learning has become a powerful paradigm for large-scale quantitative retrieval of water quality parameters (WQPs). This review examines the progress and prospects of machine-learning-driven multi-source remote sensing for surface WQP retrieval. A systematic literature review following PRISMA 2020 guidelines, covering 437 Web of Science Core Collection publications (2000–2025), reveals exponential growth, with China and the United States contributing 70.3% of total output. A critical synthesis covers four dimensions: (1) characteristics and fusion strategies of satellite, airborne, and ground-based remote sensing data; (2) modeling features of traditional machine learning (SVR, RF, GBDT), deep learning (CNN, RNN, Transformer), and hybrid approaches; and (3) retrieval advances for optically active versus non-optically active parameters—the former approaches operational readiness while the latter remains constrained by weak indirect spectral correlations; and (4) uncertainty sources and mitigation strategies across the data–model–parameter chain. Five key challenges are identified: limited model generalizability, insufficient physical interpretability, optical heterogeneity and parameter coupling, scarce in situ data, and multi-source fusion bottlenecks. Five future directions are proposed—transfer learning, physically informed explainable machine learning, non-optically active parameter retrieval, benchmark dataset development, and intelligent multi-source fusion—offering a roadmap toward operational surface water quality monitoring.

Eutrophication has become a major environmental concern in Lake Taihu, with total phosphorus (TP) and total nitrogen (TN) serving as key indicative parameters. However, existing retrievals of TP and TN in Lake Taihu mostly rely on conventional empirical algorithms over limited periods, and the month-to-month spatial patterns of the two nutrients across the whole lake remain poorly characterized. To address this gap, the present study introduces Maximal Information Coefficient (MIC)-based feature selection into machine learning retrieval of TP and TN from Sentinel-2 imagery and produces month-to-month lake-wide maps over nine months. This study utilized Sentinel-2 satellite imagery and water quality data obtained from 15 automatic monitoring stations in Lake Taihu between January 2024 and December 2024 to develop retrieval models for TP and TN. Feature selection was first performed using the MIC. Subsequently, four machine learning algorithms—Random Forest (RF), Support Vector Machine (SVM), Backpropagation Neural Network (BP), and eXtreme Gradient Boosting (XGBoost)—were employed to establish TP and TN retrieval models. The final models were then selected to map the spatial distributions of the two nutrients. Results showed that the SVM model incorporating band combinations achieved the best TP retrieval performance (R² = 0.65, RMSE = 0.03 mg/L), while RF yielded the highest accuracy for TN estimation (R² = 0.46, RMSE = 0.36 mg/L), explaining only about 46% of the variance and indicating limited predictive power for TN retrieval. Temporally, TP concentrations across Lake Taihu remained relatively low from December to May, increased gradually from August to October, and reached the annual peak in October. Spatially, western lake zones exhibited higher TP concentrations than the eastern area. TN shared identical spatial distribution patterns with TP across the lake; temporally, TN concentrations reached the annual maximum in March and the annual minimum in August. These findings demonstrate the feasibility of satellite-based TP and TN retrieval for lake water quality monitoring and provide scientific support for water environment management in Lake Taihu.