Rational allocation of irrigation water within large-scale, hierarchical canal networks is crucial for food security. However, this stability is increasingly compromised by hydro-climatic volatility and the inherent spatial heterogeneity of the soil-crop system. Current allocation paradigms face a limitation: centralized solvers, while theoretically optimal, become computationally intractable when scaled to mega-districts and are hindered by information silos across administrative boundaries. Meanwhile, existing distributed algorithms typically rely on idealized flow assumptions, neglecting nonlinear conveyance attenuation, asymmetric crop-water production functions, and the absence of a global state observer. To address these gaps, we propose physics-informed decentralized optimization (PI-DAO), a physics-informed, fully decentralized optimization framework. PI-DAO integrates two critical physical constraints into local objective functions: (1) a water-demand-strength-amplified quadratic yield-loss model that captures non-linear stress responses, (2) distance-dependent hydraulic transmission loss coefficients, and (3) irrigation-method-dependent effective water use coefficients. The framework uses a dynamic average-consensus protocol to reconstruct the global supply-demand ratio using only peer-to-peer communication, eliminating the need for a central coordinator. Validated within a high-fidelity digital simulation system of the Irrigation District, PI-DAO demonstrates high numerical stability with a consensus error <0.005%. Furthermore, it yields an average 9% reduction in the coefficient of variation of yield-loss inequality and recovers an average 2.1% of production under drought scenarios compared to the conventional proportional-share method. This decentralized paradigm provides a robust, scalable architecture for achieving distributive equity and drought resilience in complex hydraulic infrastructures.

Accurate and efficient estimation of aboveground biomass (AGB) is essential for precision agriculture and sustainable crop management. Traditional methods for AGB estimation rely on destructive sampling, which is time-consuming and impractical for frequent, large-scale monitoring. Although remote sensing techniques offer non-destructive alternatives, many existing models lack integration with crop physiological processes. The study introduced an optimized FAO66 approach combined time-series unmanned aerial vehicle (UAV) multispectral and thermal imagery to improve the accuracy of maize AGB model. We proposed and compared water stress coefficients derived from stomatal conductance (Ksg), canopy temperature (Ksr), and soil moisture (Ksw). The accumulated product of the basal crop coefficient (Kcb), temperature stress coefficient (Kst), and water stress coefficients was used as input variables for AGB modelling using stepwise regression, random forest regression (RFR), support vector regression, and convolutional neural networks. Across multiple irrigation treatments and growth stages, the model utilizing the accumulated KcbxKsrxKst achieved the highest accuracy among the FAO66-based AGB models (R2 = 0.80, RMSE = 254.8 g m-2 in 2018 and 2019). Machine learning models, particularly RFR (R2 = 0.84, RMSE = 242.5 g m-2 in both years), improved robustness by capturing nonlinear relationships among multi-temporal features, while the canopy temperature-based water stress coefficient offered a practical alternative to soil moisture - based metrics, reducing fieldwork without sacrificing accuracy. The proposed UAV-based framework provided a scalable, high-resolution solution for field-scale AGB mapping to support precision irrigation and water management.

Accurately linking farmland parameters to radar backscatter is vital for agricultural water conservation and growth monitoring, a process dependent on precise surface scattering models. High-resolution, near-field observations (e.g., UAV-based) reveal canopy heterogeneity that traditional models like the Water Cloud Model cannot resolve, leading to significant backscatter errors and limiting soil moisture inversion. To address this, we propose a novel Three-Phase Mixture (TPM) model for winter wheat, conceptualizing the canopy as a heterogeneous mix of water, dry matter, and air. Using dielectric homogenization, the model derives an equivalent dielectric constant for physically interpretable electromagnetic characterization. Innovatively, a phenologydriven adaptive mechanism dynamically adjusts canopy layer thickness and scatterer distribution via height and spatial factors, capturing structural evolution across growth stages. Validated with a custom low-altitude Cband radar platform across six key growth stages, the model effectively mitigates scattering errors from heterogeneity, improving backscatter simulation (R2=0.7817, R=0.8841, RMSE=0.8252 dB, MD=0.0824%). The results confirm its practical potential for winter wheat monitoring and parameter inversion.

Precise carbon dioxide (CO2) and water vapor (H2O) flux measurements are critical for understanding how ecosystems work, thereby improving agricultural water use efficiency. Conventional chamber-based systems are widely used to monitor CO2 and H2O fluxes. However, their widespread adoption in precision agriculture has been significantly hindered by high costs and uncertainty in CO2 and H2O measurements. The development of a solar-powered low-cost multi-channel automated chamber system (Loco-MACS) is being analysed in this paper for real-time, high-precision, and synchronous monitoring of CO2 and H2O fluxes in maize fields. Loco-MACS integrates a cost-effective NDIR CO2 sensor and employs multiplexing technology to sequentially control ten automated chambers. System performance was evaluated through comparative testing against a commercial LI8100A system and a micro-lysimeter. Statistically, CO2 fluxes measured by Loco-MACS exhibited strong agreement with the commercial system (R2 = 0.93, RMSE = 0.32 mu mol m-2 s-1), while H2O fluxes closely matched micro-lysimeter measurements (R2 = 0.85, RMSE = 0.05 mm h-1) with a minimal bias of 0.03 mm h-1. Particularly, the H2O monitoring component maintained high precision even under high-humidity conditions by mitigating condensation errors. Long-term installation in maize field further demonstrated the system's capability to continuously capture canopy-soil CO2 and H2O flux dynamics, including diurnal and seasonal variations in Net Ecosystem Exchange (NEE), Soil Respiration (RS), Evapotranspiration (ET), and Evaporation (E). By combining long-term high accuracy, affordability, and scalability, Loco-MACS offers a cost-effective solution (& euro;2153) for monitoring of CO2 and H2O fluxes, supporting precision agriculture and ecosystem sustainability.

Precise crop mapping via remote sensing is critical for the rational utilization of cultivated land resources. Remote sensing techniques and deep learning-based semantic segmentation methods provide effective approaches to large-scale crop mapping. However, due to variations among remote sensing imaging systems, significant distributional shifts exist across different remote sensing datasets, which limit the generalizability of deep learning models. To address the issue, we propose a semantic segmentation network named PLGCA-SAM for crop mapping in remote sensing imagery. PLGCA-SAM integrates the contextual perception capability of PLGCA with the powerful generalization of SAM. Specifically, we introduce a task-adaptive encoding (TAE) module, which includes a SAM adapter to align SAM’s generic visual features with the PLGCA feature space, and a task-guided mechanism that refines the fused representation using a task-aware loss. This design enables more effective feature integration and improves adaptability across diverse remote sensing datasets. Experiments conducted on GF-2 and Sentinel-2 datasets demonstrate that PLGCA-SAM achieves 2.03% and 1.50% improvements in mIoU, respectively, over SOTA methods. Our code will be obtained by: https://github.com/Hanhlab/PLGCA-SAM.git.

[Background] Improving crop yield while mitigating nitrogen application and water-saving has become increasingly important in recent years. Excessive nitrogen and irrational irrigation lead to nitrogen loss and nitrate accumulation in the soil, and nitrate leaching varies with annual precipitation. This study aimed to evaluate effects of nitrogen rate and supplemental water on nitrate distribution, and nitrogen absorption and utilization by winter wheat in different pre-sowing stored soil water. [Methods] A split-split plot experiment was conducted for two wheat seasons using three pre-sowing stored soil waters (S1: 350 mm, S2: 450 mm, and S3: 650 mm), four nitrogen rates (N0: 0, N105: 105 kg ha- 1, N210: 210 kg ha- 1, and N315: 315 kg ha- 1), and four supplemental waters (W0: 0, W1: 56.3 mm, W2: 78.1 mm, and W3: 100 mm). [Conclusion] The results showed that higher pre-sowing stored soil water increased soil NO3--N content throughout the profile, but also enhanced crop uptake, thereby reducing residual nitrate in the deepest layer under S3 treatment. Nitrogen rate was the dominant factor influencing nitrate leaching, with N210 and N315 markedly increasing NO3--N accumulation in the 80-160 cm layer. Grain yield peaked at N105W3 under S1 treatment, and at N210W3 under S2 and S3 treatments, while excessive nitrogen (>= 315 kg ha-1) reduced both yield and nitrogen use efficiency (NUE). NUE increased with moderate nitrogen (105-210 kg ha-1) and adequate supplemental water (78.1-100 mm), but declined when nitrogen or water exceeded optimal levels; the optimal nitrogen rates derived from multi-objective regression (yield, NUE, leaching) were 105 kg ha-1 (S1), 162 kg ha-1 (S2) and 168 kg ha-1 (S3), achieving high grain yields (1000-5600 kg ha-1) with acceptable nitrate leaching. [Suggestion] The findings provide a scientific basis for water-nitrogen management in dryland wheat production, suggesting that "moderate nitrogen and appropriate water" can maximize yield while minimizing environmental risks.

Net primary productivity (NPP) is a key indicator of ecohydrological processes and vegetation responses–including in croplands–to hydrological and carbon dynamics under environmental change. However, existing studies rarely jointly analyze seasonal, interannual, and extreme-event NPP responses or account for spatial heterogeneity and autocorrelation, limiting ecohydrological attribution. This study integrates remote sensing and climate data (2001-2020) with time-series, statistical, machine-learning, and spatial analyses to assess NPP dynamics across China's nine major river basins (SLR, HHR, HER, YWR, YER, PLR, SEB, SWB, ILB). Spatial patterns show a southeast-to-northwest gradient, with peak NPP in humid PLR/SEB and minima in arid ILB; growing seasons account for 59.5%–98.9% of annual totals. Temperate basins show significant increases (3.60–5.98 g C m-2 yr-1, p < 0.05) from restoration, while ILB declines (-0.78 g C m-2yr-1, < 0.05p) under drought stress; the 2011 drought reduced temperate NPP by 4.6%–12.0%, with humid basins showing greater resilience. SHAP analysis identifies vegetation metrics (FVC, LAI) and hydrological factors (precipitation, temperature, ET) as dominant, topography-modulated drivers. Among four candidate models, LGM performs best (R2 = 0.87, RMSE = 78.76 g C m-2yr-1), significantly outperforming XGB (R2 = 0.85), Random Forest (R2 = 0.80), and a linear regression baseline (R2 = 0.62; p < 0.05); incorporating spatial autocorrelation (Moran's I = 0.32-0.90, P < 0.001) via a spatial lag model further improves accuracy (R2 = 0.89, RMSE = 70.63 g C m-2 yr-1), and ablation experiments attribute the largest gains to spatial dependency and hydrological features. These findings support adaptive ecohydrological strategies for ecosystem resilience and inform crop monitoring in intensive-agriculture basins such as YWR and YER.

Agricultural drought is a highly complex and devastating natural hazard that poses significant threats to food security and socioeconomic stability in Pakistan. Traditional drought indices, relying on single variables and data sources, fail to capture the complex patterns of agricultural drought. To address this, we developed a Multi-Model Drought Index (MMDI) using 23 years (2000-2023) of multi-source remote sensing and reanalysis data. Ten carefully screened input indices were derived from MODIS, CHIRPS, and ERA5-Land datasets. Three modelling approaches were integrated to develop the MMDI: the Entity Embedding Deep Neural Network (EEDNN, deep learning), the Light Gradient Boosting Machine (LGBM, machine learning), and the Analytical Hierarchy Process (AHP, multi-criteria decision-making). Each MMDI variant was evaluated against the TerraClimate Palmer Drought Severity Index (PDSI) across croplands of Punjab, Sindh, and Khyber Pakhtunkhwa (KPK), Pakistan. Validation results indicated that MMDIEEDNN achieved the best performance with a ROC-AUC of 0.97, followed by MMDILGBM (ROC-AUC = 0.83) and MMDIAHP (ROC-AUC = 0.75). Sindh exhibited a statistically significant increasing drought trend (Z < -1.96, p < 0.05) based on Run Theory and the Mann-Kendall test. The highest drought frequency (>10 episodes) and intensity (0.33-0.40) were recorded in the eastern and southern parts of Sindh. Punjab experienced longer-duration droughts (>20 months) with moderate to high severity (8.25-9.45) across south-central areas. KPK showed the lowest drought occurrence, with only 3-4 episodes of shorter duration (1.2-5.8 months). MMDIEEDNN provides farmers and policymakers with actionable information to support drought mitigation and adaptation strategies.

CONTEXT: Water scarcity and erratic precipitation driven by climate change adversely affected peanut yields, water and nitrogen use efficiencies in the North China Plain. OBJECTIVE: This study aimed to analyze the impact of irrigation practices and long-term precipitation deviation on peanut production systems using a modelling approach. METHODS: Flood-irrigated, drip and rainfed peanut production systems were evaluated in a two-year field experiment and modelling analysis. The irrigation amounts in flood and drip were 160-180 mm and 90-110 mm, respectively. The rainfed system was completely maintained under natural precipitation conditions. The WHCNS (Water-heat-carbon-nitrogen-simulator) model was calibrated using measured plant growth, yields, soil and weather data of the first year and validated using a data set of the second year. Model simulation robustness was investigated using Root mean square error, Mean absolute error, Index of agreement, Kling-Gupta coefficient and R-2. The validated model was applied to assess the impact of long-term 29-year seasonal precipitation deviation on each cultivation system. RESULTS AND CONCLUSIONS: The results showed that the flood-irrigated peanut production system consumed 70% (70 mm ha(-1)) more average irrigation water and led to 46% (8 kg ha(-1)) greater NO3- leaching, making it inefficient compared to the drip-irrigation method. The drip-irrigated peanut production system had significantly high yields, water and nitrogen use efficiencies at P < 0.05. The water use efficiencies were 2.81 and 2.72 kg m(-3) in drip and flood systems, respectively. The rainfed peanut declined an average yield by 51% (2891 kg ha(-1)) compared to drip-irrigation (5905 kg ha(-1)). Furthermore, the long-term precipitation scenario prediction reflected obvious water and nitrogen losses in flood-irrigated peanut at increasing precipitation rates. The rainfed peanut production had consistently low yields with 17-19% yield loss in dry seasons compared to normal and wet conditions. Thus, it is found to be most vulnerable to drought. Changing the flood and rainfed peanut cultivated area to the drip-irrigation method is recommended. SIGNIFICANCE: The WHCNS model was successfully applied to explore water and nitrogen dynamics and simulate peanut yields in flood, drip and rainfed conditions. Yield fluctuations were quantified in the normal, wet and dry seasons.

The ratio of variable fluorescence to maximum fluorescence (Fv/Fm) is a critical indicator of crop photosynthetic health, yet its accurate estimation via remote sensing remains challenging. While vegetation indices (VIs) have been widely used for this purpose, integrating them with texture features (TFs) and chlorophyll content (SPAD) to enhance model accuracy remains unexplored. To address this gap, a field experiment was conducted across the 2022-2023 winter wheat growing season. Five feature fusion strategies-combining VIs derived from autonomous aerial vehicle (AAV) multispectral images, TFs extracted from AAV RGB images, and leaf chlorophyll content- were evaluated using Gaussian process (GP) and other machine learning algorithms. Results indicated significant correlations between all three VI categories and Fv/Fm. The highest correlations within each category were observed for ExGR (visible spectrum only; r = 0.80), SIPI (near infrared without red edge; r = 0.79), and both MCARI and NDVI705 (red-edge spectrum; r = 0.69). In contrast, TFs exhibited weaker correlations with Fv/Fm, peaking at r = -0.53 for B_M and B_VAR. However, feature selection based solely on maximal within-category correlations may yield suboptimal results due to interfeature collinearity. Among the five fusion strategies, integrating VIs, TFs, and physiological features (SPAD) achieved optimal performance, increasing validation R-2 from 0.70 to 0.85 and reducing root-mean-square error from 0.08 to 0.05 compared with single-feature models (GP regression). This approach demonstrates the viability of multifeature fusion for accurate Fv/Fm monitoring, providing a scalable tool for precision agriculture.

[Objective]Maize is one of the most important staple crops in the world and serves as a cornerstone of food security and agri-cultural sustainability.Accurate and timely prediction of maize yield is essential for optimizing agricultural management practices,supporting market regulation,and guiding policy decisions related to food supply and climate adaptation.In recent years,data-driven yield prediction methods based on machine learning and deep learning have achieved notable improvements in predictive accuracy.However,most existing approaches primarily rely on statistical correlations among variables and often treat influencing factors as in-dependent predictors,without explicitly addressing the complex causal mechanisms and time-lagged interactions that govern crop growth processes.This limitation may lead to reduced model interpretability and compromised robustness under changing environ-mental conditions.To address these challenges,a novel maize yield prediction framework that integrates causal inference with a hy-brid deep learning model was proposed,aiming to improve both predictive performance and mechanistic understanding.[Methods]Multi-source heterogeneous datasets collected across the maize growing season were utilized,including remote sensing-derived vege-tation indices,meteorological variables(such as temperature and precipitation),soil profile moisture measurements at multiple depths,and crop observation data corresponding to key phenological stages.First,the Peter-Clark and momentary conditional independence(PCMCI)causal discovery algorithm was applied to systematically identify causal relationships between maize yield and its potential driving factors.The PCMCI method enables the detection of both contemporaneous and time-lagged causal links while effectively controlling for confounding effects in high-dimensional time series data.Through this process,the causal structure of yield formation was explicitly characterized,and key variables with statistically significant causal impacts were selected as inputs for the prediction model.Subsequently,a hybrid moving average,convolutional neural network-long short-term memory(MA-CNN-LSTM)model was constructed to capture the complex spatiotemporal patterns in the causally screened input variables.Specifically,a moving average module was employed as a preprocessing step to suppress high-frequency noise and enhance signal stability.A CNN was then used to extract latent correlation features among multiple variables,reflecting their joint influence on yield formation.Finally,an LSTM net-work was adopted to model temporal dependencies and cumulative effects across the growing season,enabling effective representa-tion of dynamic yield responses.[Results and Discussions]The causal analysis revealed that soil moisture at depths of 10 cm and 50 cm exerted a significant positive influence on maize yield(P＜0.01),with deeper soil moisture showing a stronger and more persistent time-lagged effect.This finding highlighted the critical role of subsurface water availability in sustaining crop growth during later de-velopmental stages.In addition,vegetation indiced such as the modified chlorophyll absorption ratio index and the normalized differ-ence vegetation index exhibited significant short-term causal relationships with yield during the mid-growth stage of maize,indicating their sensitivity to canopy structure and photosynthetic activity during this period.Comparative experiments conducted against tradi-tional statistical models and conventional machine learning approaches demonstrated that the proposed PCMCI-MA-CNN-LSTM framework consistently achieved superior predictive performance.On the test dataset,the coefficient of determination(R2)reached 0.955,while the mean absolute error(MAE)and root mean square error(RMSE)were reduced to 1.201 kg/mu and 1.474 kg/mu(1 hm2=15 mu).These results indicated that incorporating causal variable selection effectively enhances model accuracy and stability by reducing redundant and spurious correlations.[Conclusions]The results confirm that incorporating causal analysis into yield model-ing provides a robust basis for identifying key driving variables and effectively enhances the accuracy and interpretability of maize yield prediction.The proposed framework offers a promising approach for precision agriculture and decision support in crop yield forecasting,particularly under complex and dynamic agro-environmental conditions.

Spectral remote sensing enables efficient acquisition of large-scale land surface information and is a key approach for monitoring soil salinity content (SSC). However, surface mulching significantly alters the spectral reflectance responses of croplands, increasing the uncertainty of SSC retrieval using remote sensing. This study aimed to systematically identify SSC-sensitive spectral features under different mulching conditions and to evaluate the effects of spatial resolution on SSC–spectral relationships. Multi-resolution datasets were constructed based on plastic mulch geometric parameters, and SSC–spectral relationships were analyzed using correlation methods and recursive feature elimination (RFE). Results indicate that under near-ground ultra-high-resolution conditions, the correlation between inter-mulch bare soil spectral features and SSC was weakly influenced by mulch type, and distinguishing mulch types provides limited improvement in inter-variable relationships. Pearson’s r exceeded 0.40 for both white- and black-mulched samples, and distinguishing mulch types provided only marginal gains in model accuracy (RFR–RFE R2 = 0.9524 for white-mulched and 0.9252 without distinguishing; R2 = 0.9387 for black-mulched). In contrast, under multi-resolution settings at the field scale, separating black-mulched, white-mulched, and non-mulched fields significantly enhanced the correlation between spectral indices (SIs) and SSC, with the coefficient of determination (R2) based on the recursive feature elimination (RFE) algorithm increasing by up to 0.28. The highly sensitive SIs of non-mulched farmland are generally consistent with those of white-mulched farmland but differ markedly from those of black-mulched farmland. Scale optimization analysis further indicated that the optimal spatial resolution was 1.35 m for white-mulched and non-mulched farmland. Black-mulched farmland performed best at 5.4 m, likely because stronger spectral masking by black mulch increases mixed-pixel dominance and benefits from spatial aggregation. These findings provide methodological guidance and practical approaches to accurately retrieve SSC in plastic-mulched croplands and to determine the optimal image spatial resolution.

Lightweight UAV-borne microwave radar systems exhibit significant potential for coordinated, multidimensional observation in precision agriculture. However, limitations in platform payload and instability during low-altitude flight make it challenging for existing systems to simultaneously achieve high-resolution farmland mapping and retrieval of microscopic physical parameters within a single radar platform. To address this challenge, this study develops a dual-mode microwave radar cooperative observation system-featuring both side-looking and downward-looking configurations-mounted on an unmanned aerial vehicle (UAV). A systematic data processing framework is also established for high-resolution farmland imaging and soil moisture retrieval. In the side-looking synthetic aperture radar (SAR) imaging mode, an improved range Doppler algorithm (RDA), integrating RTK-assisted two-step compensation and minimum entropy autofocus (MEA), is proposed to effectively mitigate trajectory distortions. In addition, the scale-invariant feature transform (SIFT) algorithm is employed to achieve geometrically consistent mosaicking of large-area farmland radar images. In the downward-looking quantitative retrieval mode, a height-adaptive correction and time-window truncation strategy based on edge detection is introduced to suppress interference caused by vertical height fluctuations and to extract canopy- and soil-related backscattering features. A random forest (RF) model is then constructed for soil moisture retrieval. Experimental results under complex farmland conditions demonstrate robust performance in both modes. In the side-looking mode, the structural similarity index (SSIM) of SAR image mosaics reaches 0.889, with a root mean square error (RMSE) of 0.443 pixels. In the downward-looking mode, the heightcorrected RF retrieval model was evaluated using four-fold cross-validation, yielding RMSE values of 0.0213, 0.0110, and 0.0101 cm3 /cm3 in Zones I-III, respectively, with corresponding correlation coefficients (R) of 0.6898, 0.7270, and 0.6308. The height-correction ablation further confirmed that interface alignment improved retrieval stability and reduced the influence of UAV vertical motion on radar feature extraction. The proposed framework can support field-scale soil moisture mapping, irrigation nonuniformity diagnosis, and precision irrigation decision-making in agricultural water management.

The spatial resolution of imagery is a critical factor influencing the accuracy of crop type mapping at the plot level. The lack of synchronous multi-scale imagery has led to uncertainty regarding the optimal spatial resolution for crop type mapping. This study develops a systematic framework integrating UAV remote sensing, resampling techniques, and multi-algorithm validation to determine context-specific spatial resolutions for fragmented agricultural regions, using China's Hetao Irrigation District (HID) as a case study. High-resolution UAV imagery was resampled to generate multi-scale datasets, enabling quantitative analysis of resolution impacts on plot boundary deformation and intra-plot texture representation. Bilinear interpolation emerged as the most effective resampling method, achieving PSNR values of 26.91 dB (1 m) and 23.19 dB (10 m). Resolution thresholds for preserving crop-specific textures were established: wheat required >= 2.5 m, whereas maize, sunflower, and squash necessitated >= 5 m. Plot deformation analysis revealed resolutions of 2 m, 5 m, and 10 m as optimal for plots with shortest dimensions of 10 m, 25 m, and 50 m, respectively. Algorithm compatibility tests demonstrated deep learning models (e.g. DeepLabV3+ with mIoU: 88.51%) significantly outperformed traditional methods like Random Forest (mIoU: 66.17%), particularly for crops with high intra-plot heterogeneity. This framework provides actionable guidelines for balancing texture fidelity, geometric accuracy, and algorithmic performance in fragmented landscapes, addressing a critical gap in precision agriculture.

The glare from highly reflective metal surfaces might lead to erroneous detection in typical machine vision systems. A polarization vision-based system is proposed to suppress the highly reflected light. Existing You Only Look Once (YOLO) series defect detection methods focus on designing deeper or wider networks to detect images with moderate illumination, uniform information distribution and balanced samples. These methods can achieve impressive detection performance while showing limitations in highly reflective metals, small targets, and imbalanced samples. To cope with these problems, we propose a method named WaveBF-YOLO, which is based on YOLOv8, for the effective detection of defects in highly reflective metals. In particular, we utilize the theory of polarization vision to capture and process images to obtain images with suppressed reflections and uniform information as input to the network. The WaveletPool module, equipped with multiscale feature extraction capabilities, is used to preserve the spatial details of the image. Next, BiFPN, the module with a bidirectional flow mechanism, is employed to enhance the performance for small target detection. The improved loss function, Focaler-IoU, can effectively solve the problems of sample imbalance and inaccurate localization. Finally, the modified network is tested on a self-established polarization vision-based defect detection dataset for metal industrial parts, and the detection performance for all types of defects is significantly improved, effectively increasing the accuracy and efficiency of detection. The experimental results indicate that WaveBF-YOLO-n mAP and AP50 reach 82. 2% and 93. 4%, respectively, surpassing YOLOv8-n by 6.6% and 3.1%. The mAP and AP of WaveBF-YOLO-l are 83.2% and 94.2%, respectively, exceeding YOLOv8-l by 0.6% and 3%. The modified method addresses the prevalent challenge of detecting defects on highly reflective metal surfaces within the industry from both theoretical and technological points of view.

Timely and precise identification of the extent and intensity of field soil salinization is crucial for effective prevention and treatment. It also supports decision-making for rational irrigation planning, crop yield prediction, and precision field management. This study explored the potential of environmental covariates and spectral variables derived from unmanned aerial vehicle (UAV) multispectral images for estimating field soil salt content (SSC). Aerial and field campaigns were conducted in 18 study areas in October 2021 and April 2022 on bare farmland, simultaneously capturing ground truth data for SSC, soil water content (SWC), and soil surface roughness (SSR). The sensitivity of eleven salinity indices (SIs), ten spectral indices (VIs), and two environmental covariates to SSC in different periods were analyzed using the Pearson’s correlation coefficient method and the recursive feature elimination algorithm (RFE). The optimal parameter combination was selected as input variables, and SSC estimation was performed using linear regression model, random forest regression (RFR), artificial neural network (ANN) and support vector regression (SVR) algorithms. Results showed that SIs related to blue and red bands exhibited a strong correlation with SSC, while environmental covariate SSR showed an indirect correlation. The spectral characteristics of the soil in pre-seeding and post-harvest periods had different sensitivity responses to SSC. Among the machine learning algorithms tested, all outperformed the linear regression model in multi-parameter SSC estimation, with the SVR_SSC model demonstrating the highest accuracy (R2 < 0.72, RMSE < 0.15 %, RPD > 1.73, LCCC > 0.77). This study introduced a comprehensive method for SSC estimation that integrates environmental covariates and provides a valuable reference for the accurate assessment of field soil salinization and precision agriculture management.

In the context of global land degradation and increasing salinization of cultivated land, accurately and sustainably estimating soil salinization is essential for effective land management. This study explores a novel approach to improve soil salt content (SSC) monitoring by minimizing the influence of long-term environmental variability and incorporating relevant environmental factors. Focusing on exposed cultivated land in arid regions, we analyzed the sensitivity of multispectral indices and environmental factors to SSC during pre-seeding and post-harvest bare soil periods. Our findings indicate that soil mechanical composition and salinity indices exhibited strong correlations with SSC, which further enhanced when the periods were analyzed separately. Dividing the bare soil periods improved the model performance, increasing R2 by 10.2%-55.7%, and the support vector regression model performed better, with an R2 of 0.77, RMSE of 0.11%. This approach offers a robust framework for precision agriculture and sustainable land management in salinity-affected regions.

Suitable remote sensing images are crucial for the accurate monitoring of land surface information. Multi-source remote sensing image collaboration is a viable method for obtaining suitable images. Image super-resolution (SR) reconstruction can transcend the mutual limitations of image monitoring range and spatial resolution and provide a solution for multi-source remote sensing image collaboration. However, image SR reconstruction experiences the challenges of low applicability of the modelling dataset, network model, and reconstruction methods. To solve these problems, in this study, two datasets were created: one containing only Gaofen-2 (GF-2) images and the other containing both GF-2 and Sentinel-2 images. Different SR models were established by combining the lightweight-enhanced SR convolutional neural network, enhanced SR generative adversarial network, and dual regression network (DRN). The applicability of the identified SR model was evaluated by applying it to the reconstruction of Sentinel-2 images from different spatiotemporal images. The results indicated that the model formed using the dataset containing both the GF-2 and Sentinel-2 images was highly accurate, with the peak signal-to-noise ratio (PSNR) and structural similarity (SSIM) values of 23.2082 dB and 0.6408, respectively. The edges of the plots were refined, and the deformations of the plots, roads, and channels were restored. The SR model formed by the DRN was highly accurate for both single- and multi-source datasets, with PSNR and SSIM values of 24.1548 dB and 0.6912, respectively. Therefore, a method for forming an SR model using a multi-source image dataset combined with the DRN was proposed for multi-source remote sensing collaboration. The method proposed in this study has good applicability to different spatiotemporal images and can provide a reference for subsequent multi-source remote sensing research.

Analyzing the driving factors of agricultural drought is important for irrigation management. This study established a new causal framework integrating causal covariation and Structural Equation Modeling (SEM) to reveal the agricultural drought mechanisms in the Hetao Irrigation District of China. Using multi-source data (2001-2020), we quantified the drought patterns by Temperature Vegetation Dryness Index (TVDI) and identified the driving factors in 23 sub-regions using the causal covariation method. The findings are as follows: 1) Over the past two decades in the Hetao Irrigation District, 47.8 % of the region maintained a stable drought condition, 17.4 % experienced aggravated drought, and 34.8 % saw alleviated drought. The study area exhibited significant spatial heterogeneity in drought intensity: elevation explained 81 % of spatial variability (r = 0.904), with higher-elevation zones (>1035 m) facing more severe drought severity. Drainage density significantly reduced drought pressure (r = -0.76). 2) Among all sub-regions, temperature factors (LST and TEMP) consistently influenced the severity of drought, while PET, SM, and runoff exhibited significant spatial heterogeneity in their driving strength for agricultural drought in different sub-regions. The SEM, constrained by the causal covariation results, demonstrated excellent model fit (P > 0.05, CFI>0.95, GFI>0.95, RMSEA<0.05), confirming the reliability of the causal covariation results. 3) The conclusions were verified by ESI, and the non-stationarity analysis using TVDI revealed that some driving factors (such as SM and runoff) changed over time due to human interventions like water-saving techniques, but still affirmed the dominate role and spatial pattern of temperature. This study provides a replicable model for analyzing the drought mechanisms in irrigation districts, which is helpful for improving the ability of precise prediction and sustainable management of water resource.

To further evaluate the effect of water stress on soil respiration (RS), reveal the influencing factors of daily and seasonal RS, and systematically evaluate and compare the sensibility of different machine learning algorithms (multiple nonlinear regression [MNR], support vector machine regression [SVR], backpropagation artificial neural network [BPNN]) to estimate RS from a maize field under water stress condition, the field experiments were conducted within a maize field in Inner Mongolia, China, during the entire 2019 growing season. Various levels of deficit irrigation were conducted in the vegetative, reproductive, and mature stages. Our research indicated that soil CO2 fluxes from 100% evapotranspiration treatment (Tr1) were significantly greater than various deficit irrigation treatments (Tr2, Tr3, Tr4) during each growth stage of summer maize. The cumulative soil CO2 fluxes of Tr2, Tr3, and Tr4 decreased 24.8%, 30.3%, and 43.7% compared with Tr1, respectively. We determined that the drivers affecting the daily RS were soil temperature at 5 cm depth (TS,5) and soil surface temperature (TSF), followed by water-filled porosity (WFPS) at 5 cm depth, but no significant correlations were observed at 25 cm depths. TS,5 and TSF also performed similar correlation with seasonal RS with R greater than 0.753 among all water treatments, followed by chlorophyll content with R greater than 0.726. During the whole growing season, the BPNN model exhibited the best predicting result, and could explain the 60%-80% and 87.8% of the variations of RS at the daily and seasonal scales, with root mean square error of 48.7-100.9 mg m-2 h-1 and 91.5 mg m-2 h-1, respectively. The SVR and MNR models could estimate the 47.9%-57% and 39.9%-52.1% of the daily RS and 81.4% and 78.6% of the seasonal RS, respectively. Overall, our study indicated the machine learning algorithms could be successfully applied to estimate RS at daily and seasonal scales from a maize field under water stress condition.