Liming Zhu

High-precision, multi-tier categorical soil classification faces critical bottlenecks, including the neglect of spatial context by conventional pixel-based models, the error cascade propagation phenomenon in multi-level classification networks, and the disruption of geophysical directional anisotropy by traditional geometric data augmentation. To address these challenges, in this study, we propose a multi-level soil classification model based on MTSC-ResNet-Trans, which organically couples residual convolutional blocks with a 3-layer Transformer encoder to model long-range spatial dependencies. The framework integrates a geospatial-safe data augmentation pipeline to preserve the topological fidelity of absolute geographic coordinates alongside four decoupled parallel multi-task classification heads to substantially suppress inter-level error propagation. Evaluated in the Southern Foothills of Qilian Mountains using 18 environmental covariates, the framework achieves an Overall Accuracy of 0.8931 at the Great Group level under conventional random splitting. Under a distance-stratified spatial evaluation—which isolates the contribution of spatial autocorrelation to accuracy estimates—the framework maintains robust performance, with MTSC-ResNet-Trans consistently outperforming pixel-based baselines (Random Forest) by approximately 3.7 percentage points even at spatial separation distances exceeding 400 m. This protocol transparently decomposes predictive accuracy into a component attributable to spatial proximity and a component reflecting reduced spatial proximity performance. Across the four taxonomic levels, accuracy decay is suppressed to 5.11%. Although spatial-block cross-validation indicates a lower regional extrapolation accuracy (OA = 0.7389 ± 0.0671), the decoupled parallel framework provides an effective and robust baseline for high-resolution regional digital soil mapping.

Old urban areas are often prone to waterlogging and sewage contamination owing to their haphazard spatial arrangements, extensive impervious surfaces, and insufficient drainage infrastructure, thereby posing significant risks to both public safety and aquatic ecosystems. Sponge City retrofitting offers a viable solution. Currently, the study area is facing issues of waterlogging and pollution caused by rainfall. Conventional modeling approaches for optimizing the spatial allocation of Low-Impact Development (LID) practices typically quantify only the overall retrofit proportion. However, these methods fail to specify the optimal placement of individual facilities to balance hydrological benefits against construction costs. To bridge this gap between theoretical optimization and practical implementation, this study proposes an iterative approximation framework. First, the Non-Dominated Sorting Genetic Algorithm II (NSGA-II) was coupled with the Storm Water Management Model (SWMM) to generate a Pareto front, from which optimal solutions were selected using the Analytic Hierarchy Process (AHP). The configuration was further refined through multiple iterations of “exhaustive search combined with Euclidean distance” analysis to determine the optimal types and locations of LID facilities. The results show that: In Scenario 3, the Euclidean distance after LID retrofitting achieved a narrowing gap from 5 to 3 to 1. This indicates that the proposed progressive approximation solving process can be directly applied to specific retrofit targets, providing concrete construction guidance for LID retrofitting in older communities’ areas. Conclusions showed that (1) the specific locations for implementing LID facilities within sub-catchments become progressively clearer, ultimately defining precise retrofitting sites. (2) The proposed progressive approximation approach effectively and systematically reduces this disparity. (3) Retrofitted LID measures effectively managed stormwater and controlled pollution.

The causes of agricultural drought are complex, and its actual occurrence process is often characterized by rapid onset in terms of time and small scale in terms of space. Monitoring agricultural drought using satellite remote sensing with low spatial resolution makes it difficult to accurately capture the details of small-scale drought events. High-resolution satellite remote sensing has relatively long revisit cycles, making it difficult to capture the rapid evolution of drought conditions. Furthermore, the occurrence of agricultural drought is linked to multiple factors including precipitation, evapotranspiration, soil properties, and crop physiological characteristics. Consequently, relying on a single variable or indicator is insufficient for multidimensional monitoring of agricultural drought. This study takes Hebi City, Henan Province as the research area. It uses Sentinel-1 satellite data (HV, VV), Sentinel-2 data (NDVI, B2, B11), elevation, slope, aspect, and GPM precipitation data from 2019 to 2024 as independent variables. Three machine learning algorithms—Random Forest (RF), Random Forest-Recursive Feature Elimination (RF-RFE), and eXtreme Gradient Boosting (XGBoost)—were employed to construct a multi-dimensional agricultural drought monitoring model at the field scale. Additionally, the study verified the sensitivity of different environmental variables to agricultural drought monitoring and analyzed the accuracy performance of different machine learning algorithms in agricultural drought monitoring. The research results indicate that under the condition of full-factor input, all three models exhibit the optimal predictive performance. Among them, the XGBoost model performs the best, with the smallest Relative Root Mean Square Error (RRMSE) of 0.45 and the highest Correlation Coefficient (R) of 0.79. The absence of Digital Elevation Model (DEM) data impairs the models’ ability to capture the patterns of key features, which in turn leads to a reduction in predictive accuracy. Meanwhile, there is a significant correlation between model performance and sample size. Ultimately, the constructed XGBoost model takes the lead with an accuracy of 89%, while the accuracies of Random Forest (RF) and Random Forest-Recursive Feature Elimination (RF-RFE) are 88% and 86%, respectively. Based on these three drought monitoring models, this study further monitored a drought event that occurred in Hebi City in 2023, presented the spatiotemporal distribution of agricultural drought in Hebi City, and applied the Mann–Kendall test for time series analysis, aiming to identify the abrupt change process of agricultural drought. Meanwhile, on the basis of the research results, the feasibility of verifying drought occurrence using irrigation signals was discussed, and the potential reasons for the significantly lower drought occurrence probability in the western mountainous areas of the study region were analyzed.

Characterized by soil moisture content and plant growth, agricultural drought occurs when the soil moisture content is lower than the water requirement of plants. Microwave remote sensing observation has the advantages of all-weather application and sensitivity to soil moisture change. However, microwave remote sensing can only invert 0~5 cm of soil surface moisture, so it cannot effectively reflect the drought situation of farmland. Therefore, this study took Henan Province as the study area, used soil moisture active and passive (SMAP) satellite soil moisture data, employed NDVI, LST, and ET as the independent variables, and took the drought grade on the sample as the dependent variable. Using the 2017–2019 data as the training set and the 2020 data as the testing set, a random forest drought monitoring model with comprehensive influence of multiple factors was constructed based on the training set data. In the process of model training, the cross-validation method was employed to establish and verify the model. This involved allocating 80% of the sample data for model construction and reserving 20% for model verification. The results demonstrated an 85% accuracy on the training set and an 87% accuracy on the testing set. Additionally, two drought events occurring during the winter wheat growing period in Henan Province were monitored, and the validity of these droughts was confirmed using on-site soil moisture and the vegetation supply water index (VSWI). The findings indicated a high incidence of agricultural drought in the southwestern part of Henan Province, while the central and northern regions experienced a lower incidence during the jointing to heading and filling stages. Subsequently, leveraging the results from the random forest drought monitoring, this study conducted a time series analysis using the Mann–Kendall test and a spatial analysis employing Moran’s I index to examine the temporal and spatial distribution of agricultural drought in Henan Province. This analysis aimed to unveil trends in soil moisture changes affecting agricultural drought, as observed via the SMAP satellite (NASA). The results suggested a possible significant spatial auto-correlation in the occurrence of agricultural drought.

Soil moisture (SM) is a crucial environmental variable, and it plays an important role in energy and water cycles. SM data retrieval based on microwave satellite remote sensing has garnered significant attention due to its spatial continuity, wide observational coverage, and relatively low cost. Validating the accuracy of satellite remote sensing SM products is a critical step in enhancing data credibility, which plays a vital role in ensuring the effective application of satellite remote sensing data across various fields. Firstly, this study focused on Henan Province and evaluated the accuracy of the SMAP Enhanced L3 Radiometer Global and Polar Grid Daily 9 km EASE-Grid Soil Moisture (SPL3SMP_E) product along with its application in agriculture. The evaluation was based on in situ SM data from 55 stations in Henan Province. The assessment metrics used in this study include mean difference (MD), root mean square error (RMSE), unbiased root mean square error (ubRMSE), and the Pearson correlation coefficient (R). The time span of this study is from 2017 to 2020. The evaluation results indicated that the SPL3SMP_E soil moisture product performs well, as reflected by an ubRMSE value of 0.045 (m³/m³), which was relatively close to the product’s design accuracy of 0.04 (m³/m³). Moreover, the accuracy of the product was unaffected by temporal factors, but the product exhibited strong spatial aggregation, which was closely related to land use types. Then, this study explored the response of the SPL3SMP_E product to irrigation signals. The precipitation and irrigation data from Henan Province were employed to investigate the response of the SPL3SMP_E soil moisture product to irrigation. Our findings revealed that the SPL3SMP_E soil moisture product was capable of capturing over 70% of irrigation events in the study area, indicating its high sensitivity to irrigation signals in this region. In this study, the SPL3SMP_E product was also employed for monitoring agricultural drought in Henan Province. The findings revealed that the collaborative use of the SPL3SMP_E soil moisture product and machine learning algorithms proves highly effective in monitoring significant drought events. Furthermore, the integration of multiple indices demonstrated a notable enhancement in the accuracy of drought monitoring. Such an evaluation holds significant implications for the effective application of satellite remote sensing SM data in agriculture and other domains.

Accurate irrigation water-use data are essential to agricultural water resources management and optimal allocation. The obscuration presented by ground cover in farmland and the subjectivity of irrigation-related decision-making processes mean that effectively identifying regional irrigation water use remains a critical problem to be solved. In view of the advantages of satellite microwave remote sensing in monitoring soil moisture, previous studies have proposed a method for estimating irrigation water use using the satellite microwave remote sensing of soil moisture. However, the method is affected by false irrigation signals from soil moisture increases caused by non-irrigation factors, causing irrigation water use to be overestimated. Therefore, the purpose of this study is to improve the estimation of irrigation water use in drylands by using irrigation signals from SMAP soil moisture data. In this paper, the irrigation water use in Henan Province is estimated by using the irrigation signals from SMAP (soil moisture active and passive) soil moisture data. Firstly, a method for recognizing irrigation signals in soil moisture data obtained by microwave satellite remote sensing was used. Then, an estimation model of the amount of irrigation water (SM2Rainfall model) was built on each data pixel of the satellite microwave remote sensing of soil moisture. Finally, the amount of irrigation water utilized in Henan Province was estimated by combining the irrigation signals and irrigation water-use estimation model, and the results were evaluated. According to the findings, this study improved the estimation accuracy of irrigation water use by using the irrigation signals in Henan Province. The result of this study is of great importance to accurately obtain irrigation water use in the region.

This study addresses the problem of restricted ability for large-scale monitoring due to the limited cruising time of unmanned aerial vehicles (UAV) by identifying an optimal leaf ChlorophyII content (LCC) inversion machine learning model at different scales and under different parameterization schemes based on simultaneous observations of ground sampling, UAV flight, and satellite imagery. The following results emerged: (1) The correlation coefficient between most remote sensing features (RSFs) and LCC increased as the remote scale expanded; thus, the scale error caused by the random position difference between GPS and measuring equipment should be considered in field sampling observations. (2) The LCC simulation accuracy of the UAV multi-spectral camera using four machine learning algorithms was ExtraTree > GradientBoost > AdaBoost > RandomForest, and the 20- and 30-pixel scales had better accuracy than the 10-pixel scale, while the accuracy for three feature combination schemes ranked combination of extremely significantly correlated RSFs > combination of significantly correlated and above RSFs > combination of all features. ExtraTree was confirmed as the optimal model with the feature combination of scheme 2 at the 20-pixel scale. (3) Of the Sentinel-2 RSFs, 27 of 28 were extremely significantly correlated with LCC, while original band reflectance was negatively correlated, and VIs were positively correlated. (4) The LCC simulation accuracy of the four machine learning algorithms ranked as ExtraTree > GradientBoost > RandomForest > AdaBoost. In a comparison of two parameterization schemes, scheme 1 had better accuracy, while ExtraTree was the best algorithm, with 11 band reflectance as input RSFs; the RMSE values for the training and testing data sets of 0.7213 and 1.7198, respectively.

The ability to obtain an accurate measure of irrigation water use is urgently needed in order to provide further scientific guidance for irrigation practices. This investigation took soil moisture and precipitation as the study objects and quantitatively analyzed their relationship by establishing four models: a linear model, a logarithmic model, a soil water balance model, and a similarity model. The results from building models on every site clearly revealed the relationship between soil moisture and precipitation and confirmed the feasibility of estimating irrigation water use when soil moisture data are known. Four models combined with soil moisture data were used to estimate irrigation water use. First, the 16 sites which monitor soil moisture conditions in Hebi City were identified as study objects, from which everyday meteorological data (temperature, precipitation, atmospheric pressure, wind speed, sunshine duration) and soil moisture data from 2015 to 2020 (totaling six years) were collected. Second, the eligible data from the first four years in the date range were used to create four kinds of models (linear model, logarithmic model, soil water balance model, and similarity model) to estimate the amount of water input to the soil surface based on soil moisture. Third, the eligible data from the last two years in the established date range were used to verify the established models on every site and then judge the accuracy of the models. For example, for site 53990, the RMSE of the linear model, logarithmic model, soil water balance model, and similarity model was 10,547, 10,302, 8619, and 7524, respectively. The results demonstrate that the similarity model proposed in this study can express the quantitative relationship between soil moisture and precipitation more accurately than the other three models. Based on this conclusion, the eligible soil moisture data known in the specific site were ultimately used to estimate the irrigation water use in the field by the relationship expressed in the similarity model. Compared with the amount of irrigation water data recorded, the estimated irrigation water use yielded by the similarity model in this study was 18.11% smaller. In a future study, microwave satellite remote sensing of soil moisture data, such as SMAP and SMOS soil moisture data, will be used to evaluate the performance of estimated regional irrigation water use.

The hidden leakage of enclosure is a major risk source of deep excavation projects, and there is no effective method to detect the leakage at present. This paper introduces a non-destructive detection technology based on electric field for hidden leakage of enclosure in deep excavation projects by analyzing the distribution of electric field inside the foundation pit. The later excavation of a deep excavation project in Shanghai verifies that the detection results is convincing, which confirms that this method can accurately identify the hidden leakage in the foundation pit before excavation. And this method is applicable to various types of in-situ walls used to support excavations, including diaphragm walls, deep mixed walls, reinforced concrete retaining walls, etc. It is able to rapidly, accurately pre-detect the hidden leakage of the retaining structure, and provides early warning of leakage in enclosure. Therefore, the strong technical support for reinforcing the foundation pit is also proposed after detection.

To allow extraction of irrigation signals from satellite-derived data on soil moisture, this study describes the development of an irrigation signal extraction method that takes into account multiple environmental factors in irrigation. Firstly, the fuzzy membership functions of irrigation relating to multiple environmental factors are constructed. Then, a model is built based on the fuzzy membership functions by using operation rules of fuzzy sets, which is used to infer the relevant degree of irrigation to nonirrigation. Finally, the irrigation signals in satellite-based soil moisture data are recognized according to the relevant degree. Taking Henan Province in the North China Plain as the study area, the proposed method is used to extract irrigation signals from the SMAP Level 3 Passive Soil Moisture Product. Extracted irrigation signals from two SMAP grids are validated using daily in situ soil moisture and precipitation data, with the results showing correct identification of most of the irrigation signals. By grading the membership degree of the extracted irrigation signals, irrigation frequency maps for the 2016–2017 winter crop growth season and the 2017 summer crop growth season are obtained for Henan Province. Compared to the irrigation frequency maps with data on the annual precipitation and the annual potential evapotranspiration, the irrigation frequency maps show a spatial pattern opposite that of the annual precipitation and a spatial pattern similar to that of the annual potential evapotranspiration. It is common sense that areas with low precipitation and high evapotranspiration need more irrigation frequency and irrigation water. Thus, the spatial patterns of irrigation frequency maps are reasonable in a sense. However, it should be noted that the observed irrigation data used in the qualitative assessments are rendered less convincing by the SMAP product’s coarse resolution. Quantitative validation of extracted irrigation signals remains a significant challenge, and small-scale irrigation cannot be captured by coarse-resolution satellite-based soil moisture products. Thus, a high-resolution soil moisture product should be used to extract irrigation signals in future.