Method and device for monitoring dynamic of cultivated land soil desertification based on multi-source satellite remote sensing

Abstract

This disclosure provides a method and apparatus for dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing. The method, applied in the field of environmental protection technology, includes: training a machine learning model using a constructed inversion dataset to obtain a soil sand content inversion model; performing inversion using the soil sand content inversion model based on a constructed long-term time-series farmland bare soil dataset and corresponding topographic data to obtain farmland soil sand content data for the study area; based on the farmland soil sand content data, screening farmland soil desertification early warning factors from climate data, topographic data, socioeconomic data, and soil data of the study area, calculating the weights corresponding to these factors, and establishing a farmland soil desertification early warning model based on these factors and their corresponding weights; and issuing early warnings for farmland soil desertification in the study area based on this model. In this way, dynamic monitoring and risk warning of farmland soil desertification can be achieved.

Description

Technical Field

[0001] This disclosure relates to the field of environmental protection technology, and in particular to a method and device for dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing. Background Technology

[0002] Soil desertification in arable land poses a severe challenge to agricultural ecosystems, food production, and the human living environment. Currently, monitoring soil desertification in arable land mainly relies on field surveys and remote sensing technology. Although field surveys offer high accuracy, they are time-consuming and costly, making it difficult to achieve large-scale, high-frequency dynamic monitoring. Traditional remote sensing monitoring often relies on single satellite data, which has limitations in temporal and spatial resolution. For example, although the Landsat series satellites possess long-term time-series data, their temporal resolution is low, making it difficult to capture short-term dynamic changes in soil desertification. Furthermore, single data sources are susceptible to interference from factors such as cloud cover and vegetation cover in the face of complex surface features, affecting the accuracy and reliability of monitoring results.

[0003] The introduction of multi-source satellite remote sensing technology has brought new opportunities for monitoring farmland soil desertification. First, multi-source satellite data fusion can significantly improve the spatial and temporal resolution of monitoring data. For example, combining high-resolution optical satellite data with high-temporal-resolution MODIS data can not only obtain detailed surface information but also enable continuous monitoring of the soil desertification process. Second, multi-source data fusion can fully leverage the advantages of different sensors, compensating for the shortcomings of a single data source. Furthermore, multi-source satellite remote sensing technology can provide richer information on surface features, such as spectral and temperature characteristics, which helps to delve deeper into the causes and evolution patterns of soil desertification.

[0004] Conducting research on dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing has significant practical and scientific value. First, it is crucial for improving the accuracy and efficiency of farmland soil desertification monitoring. Through multi-source data fusion, refined monitoring of the soil desertification process can be achieved, allowing for timely understanding of the distribution, extent, and trends of desertified land. This provides important guidance for the scientific formulation of desertification prevention and control policies, rational land use planning, and effective implementation of desertified land management and restoration. Second, this research helps promote innovation and development of remote sensing technology in the field of soil desertification monitoring. By continuously optimizing multi-source data fusion algorithms and expanding the application scope of remote sensing monitoring, more advanced technical means and methods can be provided for soil desertification monitoring. Furthermore, this research also has significant ecological and social benefits. Accurate and timely farmland soil desertification monitoring information can provide strong support for ecological security assessments and regional sustainable development decisions, contributing to ecological environment protection, food security, and the promotion of sustainable economic and social development. Summary of the Invention

[0005] In a first aspect, embodiments of this disclosure provide a method for dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing, the method comprising:

[0006] The study obtained sand content data and laboratory spectral data of surface soil samples of typical soil types collected within the cultivated land area of ​​the study area through laboratory testing, and obtained multi-source satellite spectral data of the sample locations based on the transit images during the sample collection period.

[0007] Using multi-source satellite spectral data from sample locations, spectral features of common bands are extracted to construct a dual-band spectral index between arbitrary bands. A third band is then added to the dual-band spectral index to construct a three-band spectral index. The constructed spectral index, the corresponding topographic data, and the corresponding bands are used as input features, and the laboratory-measured sand content data is used as the target output to construct an inversion dataset. The machine learning model is trained using the inversion dataset to obtain a soil sand content inversion model.

[0008] Based on the bare soil period window time corresponding to multiple sub-regions within the study area, query all available multi-source satellite remote sensing data for all historical periods to obtain the historical remote sensing dataset of the study area; extract bare soil pixels from the remote sensing data in the historical remote sensing dataset within the selected bare soil synthesis period, and synthesize bare soil images to obtain the bare soil image synthesis results for the corresponding time period, thereby constructing a long-term bare soil dataset.

[0009] Farmland extraction processing was performed on the long-term bare soil dataset to obtain a long-term farmland bare soil dataset. Based on the long-term farmland bare soil dataset and the corresponding topographic data, a soil sand content inversion model was used to invert the farmland soil sand content data of the study area.

[0010] Based on the sand content data of arable land soil, early warning factors for arable land soil desertification were screened from climate data, topographic data, socio-economic data and soil data of the study area, and the weights corresponding to the early warning factors for arable land soil desertification were calculated. Based on the early warning factors for arable land soil desertification and their corresponding weights, an early warning model for arable land soil desertification was established.

[0011] The farmland soil desertification early warning model was used to conduct early warning of farmland soil desertification in the study area.

[0012] Among some feasible approaches to the first aspect, a third band is added to the two-band spectral index to construct a three-band spectral index, including:

[0013] Calculate the correlation coefficient between the dual-band index and the laboratory-measured sand content data, and add a third band to the dual-band spectral index where the correlation coefficient is greater than a preset threshold to construct a three-band spectral index. Then, supplement the existing spectral indexes in the constructed three-band spectral index.

[0014] In some possible implementations of the first aspect, a machine learning model is trained using an inversion dataset to obtain a soil sand content inversion model, including:

[0015] Based on the inversion dataset, the training set and validation set are divided. The cross-validation algorithm is used to select the input features that are actually used for modeling, and the training set is then used to remove features, retaining the input features that are actually used for modeling.

[0016] The updated training set was used to train the machine learning model, and hyperparameters were tuned to obtain the soil sand content inversion model.

[0017] In some possible implementations of the first aspect, bare soil pixels are extracted from remote sensing data in a historical remote sensing dataset within a selected bare soil synthesis period, and bare soil images are synthesized to obtain the bare soil image synthesis results for the corresponding time period, thereby constructing a long-term bare soil dataset, including:

[0018] Based on laboratory spectral data and the spectral response functions of various satellites, the laboratory spectral data is resampled to simulate the band characteristics of each sensor and establish a spectral response relationship model between different sensors. This ensures standardized processing and unified scale conversion of cross-sensor data. Bare soil pixels are extracted from the converted remote sensing data, and bare soil images are synthesized to obtain the synthesized bare soil images for the corresponding time period, thereby constructing a long-term bare soil dataset.

[0019] In some possible implementations of the first aspect, arable land extraction processing is performed on the long-term bare soil dataset to obtain a long-term arable land bare soil dataset, including:

[0020] A farmland mask dataset was created using the annual land use dataset. Based on this, farmland extraction processing was performed on the long-term bare soil dataset to obtain the long-term farmland bare soil dataset.

[0021] In some feasible approaches to the first aspect, based on data on soil sand content in arable land, early warning factors for arable land desertification are screened from climate, topographic, socioeconomic, and soil data of the study area, and the weights corresponding to these early warning factors are calculated, including:

[0022] Based on the sand content data of arable land soil, the correlation analysis algorithm was used to screen early warning factors for arable land soil desertification from climate data, topographic data, socio-economic data and soil data of the study area.

[0023] The relative importance of each soil desertification early warning factor in cultivated land is calculated by comparing them pairwise, and the corresponding weights are obtained accordingly.

[0024] Among the feasible approaches to the first aspect, an early warning model for farmland soil desertification is established based on early warning factors for farmland soil desertification and their corresponding weights, including:

[0025] Based on the early warning factors of arable land soil desertification and their corresponding weights, a comprehensive evaluation algorithm is used to establish an early warning evaluation model for arable land soil desertification. This model is used to quantitatively assess the risk of arable land soil desertification, predict the risk level of arable land soil desertification, and establish an early warning model for arable land soil desertification based on this.

[0026] Secondly, embodiments of this disclosure provide a dynamic monitoring device for farmland soil desertification based on multi-source satellite remote sensing, the device comprising:

[0027] The acquisition module is used to acquire sand content data and laboratory spectral data of surface soil samples of typical soil types collected within the cultivated land area of ​​the study area through laboratory testing, and to acquire multi-source satellite spectral data of the sample locations based on the transit images during the sample collection period.

[0028] The module is used to extract spectral features of common bands from multi-source satellite spectral data of sample points, construct a dual-band spectral index between arbitrary bands, and add a third band to the dual-band spectral index to construct a three-band spectral index. The constructed spectral index, the corresponding topographic data, and the corresponding bands are used as input features, and the laboratory-measured sand content data is used as the target output to construct an inversion dataset. The machine learning model is trained using the inversion dataset to obtain the soil sand content inversion model.

[0029] The extraction module is used to query all available multi-source satellite remote sensing data for all historical periods based on the bare soil period window time corresponding to multiple sub-regions within the study area, and obtain the historical remote sensing dataset of the study area; the remote sensing data in the historical remote sensing dataset within the selected bare soil synthesis period are used to extract bare soil pixels and synthesize bare soil images to obtain the bare soil image synthesis results for the corresponding time period, thereby constructing a long-term bare soil dataset.

[0030] The inversion module is used to extract farmland from the long-term bare soil dataset to obtain a long-term farmland bare soil dataset. Based on the long-term farmland bare soil dataset and the corresponding topographic data, the soil sand content inversion model is used to perform inversion to obtain the farmland soil sand content data of the study area.

[0031] A module is established to screen early warning factors of farmland soil desertification from climate data, topographic data, socio-economic data, and soil data of the study area based on farmland soil sand content data, calculate the weights corresponding to the early warning factors of farmland soil desertification, and establish an early warning model of farmland soil desertification based on the early warning factors of farmland soil desertification and their corresponding weights.

[0032] The early warning module is used to provide early warning of farmland soil desertification in the study area using the farmland soil desertification early warning model.

[0033] Thirdly, embodiments of this disclosure provide an electronic device comprising: at least one processor; and a memory communicatively connected to the at least one processor; the memory storing instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the method described above.

[0034] Fourthly, embodiments of this disclosure provide a non-transitory computer-readable storage medium storing computer instructions for causing a computer to perform the methods described above.

[0035] In this embodiment, a machine learning model is trained using a constructed inversion dataset to obtain a soil sand content inversion model. Based on a constructed long-term time-series bare soil dataset and corresponding topographic data, the soil sand content inversion model is used to invert the soil sand content data of the study area. Based on the soil sand content data, early warning factors for soil desertification are selected from climate, topographic, socioeconomic, and soil data of the study area, and the weights corresponding to these factors are calculated. An early warning model for soil desertification is established based on these factors and their corresponding weights. This model is then used to provide early warning of soil desertification in the study area. In this way, dynamic monitoring and risk warning of soil desertification can be achieved.

[0036] It should be understood that the description in the Summary of the Invention is not intended to limit the key or essential features of the embodiments of this disclosure, nor is it intended to restrict the scope of this disclosure. Other features of this disclosure will become readily apparent from the following description. Attached Figure Description

[0037] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. The drawings are provided for a better understanding of the invention and are not intended to limit the scope of this disclosure. In the drawings, the same or similar reference numerals denote the same or similar elements, wherein:

[0038] Figure 1 A flowchart illustrating a method for dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing, provided in this embodiment of the disclosure;

[0039] Figure 2 A synthetic soil image of one phase provided in an embodiment of this disclosure is shown;

[0040] Figure 3 This diagram illustrates the mapping results of soil sand content in cultivated land in the first phase, as provided in an embodiment of this disclosure.

[0041] Figure 4 This diagram shows a structural diagram of a dynamic monitoring device for farmland soil desertification based on multi-source satellite remote sensing, provided in an embodiment of this disclosure.

[0042] Figure 5 A structural diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. Detailed Implementation

[0043] To make the objectives, technical solutions, and advantages of the embodiments of this disclosure clearer, the technical solutions of the embodiments of this disclosure will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this disclosure, and not all embodiments. Based on the embodiments of this disclosure, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this disclosure.

[0044] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article generally indicates that the preceding and following related objects have an "or" relationship.

[0045] To address the problems encountered in the background art, this disclosure provides a method, apparatus, device, and storage medium for dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing. The following, in conjunction with the accompanying drawings, provides a detailed description of the method, apparatus, device, and storage medium for dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing provided by this disclosure through specific embodiments.

[0046] Figure 1 The flowchart shown is a method for dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing provided in an embodiment of this disclosure. Figure 1 As shown, the dynamic monitoring method 100 for farmland soil desertification may include the following steps:

[0047] S110: Obtain sand content data and laboratory spectral data of surface soil samples of typical soil types collected within the cultivated land area of ​​the study area through laboratory testing, and obtain multi-source satellite spectral data of the sample locations based on transit images during the sample collection period.

[0048] In some embodiments, topsoil samples can be collected within the cultivated land area of ​​the study area to ensure that the samples cover typical soil types. Subsequently, the sand content data and laboratory spectral data of the samples are determined by laboratory testing, and satellite spectral data of the sample locations are obtained based on transit images during the sample collection period (multi-source satellite remote sensing data can be selected and filtered according to image quality).

[0049] At the same time, topographic data with a spatial resolution of no less than 30m, annual land use data, and auxiliary data such as historical annual climate data, socio-economic data, and soil data of the study area were queried.

[0050] S120 utilizes multi-source satellite spectral data from sample locations to extract spectral features of common bands, constructs a dual-band spectral index between arbitrary bands, and adds a third band to the dual-band spectral index to construct a three-band spectral index. Using the constructed spectral index, corresponding topographic data, and corresponding bands as input features, and laboratory-measured sand content data as the target output, an inversion dataset is constructed. The machine learning model is trained using the inversion dataset to obtain a soil sand content inversion model.

[0051] In some embodiments, multi-source satellite spectral data of sample locations can be used to extract spectral features of common bands, construct a dual-band spectral index between arbitrary bands, and then calculate the correlation coefficient between the dual-band index and laboratory-measured sand content data. A third band is added to dual-band spectral indices with correlation coefficients greater than a preset threshold to construct a tri-band spectral index, supplementing existing spectral indices within the constructed tri-band index. Next, using the constructed and supplemented spectral indices, corresponding topographic data, and corresponding bands as input features, and laboratory-measured sand content data as the target output, an inversion dataset is constructed. Based on the inversion dataset, training and validation sets are divided. Cross-validation algorithms are used to filter input features actually used for modeling, and these features are used to remove features from the training set, retaining only the input features actually used for modeling. The updated training set is used to train the machine learning model, and hyperparameter tuning is performed to obtain the soil sand content inversion model.

[0052] In other words, this method utilizes multi-source satellite spectral data from sample locations to construct dual-band spectral indices (e.g., difference index, ratio index, normalization index, root-square index, etc.) between arbitrary bands, and calculates the correlation coefficient between the dual-band spectral indices and the measured sand content data. For dual-band spectral indices with high correlation, a third band is added to construct a three-band spectral indices. Existing soil texture-related spectral indices are supplemented into the constructed three-band spectral indices, and an inversion dataset is constructed by combining topographic data and measured sand content data. The inversion dataset is stratified proportionally into training and test sets. The modeling features constructed using the training set are cross-validated with the original bands for feature selection. The selected features are used to construct a soil sand content inversion model, and hyperparameter tuning is performed to further optimize the model (the model accuracy evaluation criterion is R0). 2 RMSE, RPD).

[0053] It is worth noting that this study systematically compares all available dual-band spectral indices based on conventional spectral indices. In addition to difference, ratio, and normalized indices, a root-sum-square index is also introduced. Furthermore, based on existing spectral indices with good performance, a three-band index is constructed to further explore deeper spectral features in multispectral data.

[0054] S130. Based on the bare soil period window time corresponding to multiple sub-regions within the study area, query all available multi-source satellite remote sensing data for all historical periods to obtain the historical remote sensing dataset of the study area; extract bare soil pixels from the remote sensing data in the historical remote sensing dataset within the selected bare soil synthesis period, and synthesize bare soil images to obtain the bare soil image synthesis results for the corresponding time period, thereby constructing a long-term bare soil dataset.

[0055] In some embodiments, the influence of factors such as latitude, climate, and the main crop growth cycle on the bare soil period can be comprehensively considered to set the remote sensing image query time (i.e., the bare soil period window time) for multiple sub-regions within the study area. Based on the study area and the set bare soil period window time, all available multi-source satellite remote sensing data for all historical periods are queried to obtain the historical remote sensing dataset for the study area. Subsequently, the laboratory spectral data is resampled based on laboratory spectral data and the spectral response functions of each satellite to simulate the characteristics of each sensor band and establish a spectral response relationship model between different sensors. This ensures standardized processing and unified scale conversion of cross-sensor data. Bare soil pixels are extracted from the converted remote sensing data, and bare soil images are synthesized to obtain the synthesized bare soil images for the corresponding time period, thereby constructing a long-term bare soil dataset.

[0056] As an example, to ensure spectral consistency between laboratory spectral data and multi-source satellite remote sensing data (i.e., data from different remote sensing sensors), an adaptation conversion was performed based on the laboratory spectral data and the spectral response functions of each satellite to match the spectral response ranges of different remote sensing sensors such as Landsat 5, Landsat 8, and Landsat 9. Through this process, a regression model was established to determine the band conversion coefficients between different sensors, thereby ensuring consistency between laboratory data and remote sensing data within the same spectral range.

[0057] Furthermore, a set of bare soil pixel extraction rules was constructed, and the bare soil index threshold was adjusted based on the satellite spectral data of the sample locations to improve its applicability in the study area. By annually compiling and screening the acquired remote sensing datasets, it was ensured that the data covered the corresponding cultivated land area each year. For missing data, supplementation was achieved by expanding the query time range or adding data sources. Simultaneously, considering the differences in spectral response between different sensors, spectral response conversion coefficients were calculated to uniformly convert the time-series remote sensing image data acquired by different sensors into image data consistent with the sensors used for modeling, thereby ensuring data consistency and comparability. Bare soil pixels were extracted from each image, ultimately yielding bare soil pixel sequences for each image dataset. Based on this, a long-term time-series bare soil dataset was constructed, providing high-quality data support for the soil sand content inversion model.

[0058] It is worth noting that, based on the bare soil screening images, a complete bare soil extraction technology process was formed by utilizing multiple bare soil spectral features. The index threshold was further corrected by combining sample data, and the large-scale, long-term bare soil distribution information was extracted by integrating multi-source remote sensing data.

[0059] S140, the long-term bare soil dataset is processed to extract farmland, resulting in a long-term farmland bare soil dataset. Based on the long-term farmland bare soil dataset and the corresponding topographic data, the soil sand content inversion model is used to invert the farmland sand content data of the study area.

[0060] In some embodiments, a cultivated land mask dataset can be created using an annual land use dataset. Based on this, cultivated land extraction processing can be performed on the long-term bare soil dataset to obtain a long-term cultivated land bare soil dataset. Based on the long-term cultivated land bare soil dataset and the corresponding topographic data, a soil sand content inversion model is used to invert the soil sand content data of the cultivated land in the study area.

[0061] As an example, see S130. Here, bare soil pixel sequences from a long-term farmland bare soil dataset are synthesized year by year. The median or mean method is used to ensure numerical stability, and annual synthetic image data of bare soil in the study area are obtained. Combined with farmland data of the corresponding years, annual synthetic results of farmland bare soil in the study area are further generated, and the original spectral information is preserved to obtain a long-term farmland bare soil dataset. Using the constructed soil sand content inversion model, based on the spectral information of the farmland bare soil synthetic image data in the long-term farmland bare soil dataset and the corresponding topographic data, a spatial distribution map of farmland soil sand content in the study area is generated.

[0062] It is worth noting that the correspondence between soil sand content and spectral data has been extended to the time dimension. The regression model has been applied to the long-term bare soil synthesis results to achieve cross-year soil sand content inversion, providing a dynamic monitoring method for the long-term changes in soil properties in the study area.

[0063] S150, based on the sand content data of cultivated land soil, screens early warning factors for cultivated land soil desertification from climate data, topographic data, socio-economic data and soil data of the study area, calculates the weights corresponding to the early warning factors for cultivated land soil desertification, and establishes an early warning model for cultivated land soil desertification based on the early warning factors for cultivated land soil desertification and their corresponding weights.

[0064] In some embodiments, based on farmland soil sand content data, correlation analysis algorithms can be used to screen farmland soil desertification early warning factors from climate, topographic, socioeconomic, and soil data of the study area. Each farmland soil desertification early warning factor is compared pairwise to calculate its relative importance, and its corresponding weight is obtained. Based on the farmland soil desertification early warning factors and their corresponding weights, a comprehensive evaluation algorithm is used to establish a farmland soil desertification early warning evaluation model. This model quantitatively assesses the risk of farmland soil desertification, predicts the risk level of farmland soil desertification, and establishes a farmland soil desertification early warning model based on this. For example, the above process can be elaborated as follows:

[0065] (1) Selection of early warning factors for soil desertification in arable land: Based on the sample data, obtain the potential influencing factors of soil desertification such as climate data, socio-economic data, and soil data within the corresponding time period, resample, and obtain the correlation value between each factor and the sand content data of arable land soil through correlation analysis and other methods. Based on the values, early warning factors for soil desertification in arable land are selected.

[0066] (2) Determination of the weights of early warning factors for farmland soil desertification: Each early warning factor is compared pairwise to calculate its relative importance and obtain its weight. A consistency test is performed to ensure that the judgment results are reasonable. Through this process, the impact of each early warning factor on soil desertification is quantified, providing a basis for the early warning model.

[0067] (3) Establishment of an early warning and evaluation model for farmland soil desertification: Based on the selected early warning factors and their weights, and combined with relevant mathematical models and algorithms, the risk of farmland soil desertification is quantitatively assessed. The model is used to predict the risk level of soil desertification (mild, moderate, and severe).

[0068] (4) Construction of an early warning model for farmland soil desertification: Construct an early warning model based on early warning factors, combine data on climate, topography, socio-economic factors, and soil, dynamically monitor the risk level of soil desertification, and provide data support for prevention and control measures.

[0069] It is worth noting that this provides a new approach for early warning and precise prevention of farmland soil desertification. Based on a high-precision, long-term dataset of farmland soil sand content, it enables long-term evaluation of farmland soil desertification. The new dataset can be used to assess the future farmland soil desertification situation in the study area.

[0070] S160 uses a farmland soil desertification early warning model to provide early warning of farmland soil desertification in the study area.

[0071] In summary, according to the embodiments of this disclosure, at least the following technical effects are achieved:

[0072] (1) Data integration and high-precision inversion: The high-precision soil sand content inversion model constructed in this disclosure provides accurate and robust soil sand content estimation capabilities. Compared with the prior art, this disclosure, by collecting and processing multi-source satellite remote sensing data, laboratory spectral data, and long-term meteorological and topographic auxiliary data, can better adapt to the soil texture information acquisition needs of different regions and time periods, and provide more reliable soil desertification monitoring results.

[0073] (2) Long-term time series data and cross-year monitoring: By constructing a long-term time series dataset of bare soil in cultivated land, this disclosure realizes the function of historical data synthesis and continuous monitoring based on satellite imagery, overcoming the limitations of traditional methods in terms of data gaps and regional coverage. This function provides cross-year dynamic monitoring capabilities for soil desertification, enabling this disclosure to have greater depth and breadth in the analysis of soil characteristic changes in different years.

[0074] (3) Automated early warning and assessment: This disclosure achieves refined classification and real-time monitoring of farmland soil desertification risk through a screening and weighted evaluation model of early warning factors. Compared with existing technologies, this disclosure, based on a multi-factor risk level prediction model, can provide more timely and regionally targeted desertification prevention and control information, helping users to formulate reasonable soil management measures in advance.

[0075] (4) Easy-to-use multi-dimensional data visualization support: The spatiotemporal distribution map of soil sand content and the desertification risk level map generated in this disclosure provide the product with rich visualization information output capabilities. Combined with dynamic change maps of remote sensing images, soil desertification trend maps of geographical locations, and other diverse visualization methods, soil changes can be presented intuitively, providing strong data support for user decision-making.

[0076] To facilitate further understanding, the method 100 for dynamic monitoring of farmland soil desertification provided in this disclosure will be described in detail below with reference to a specific embodiment, as shown in the following figure:

[0077] (1) Data collection and processing

[0078] 1.1) Soil Data Collection

[0079] In April 2021, 162 surface soil samples (0-15 cm deep) were collected from farmland in the black soil region using random sampling. At each sampling point, five soil sub-samples were collected within a 10-meter diameter circular area using a five-point method, and these sub-samples were mixed to form a composite sample. One kilogram of composite sample was collected from each point, air-dried indoors, and then passed through a 2 mm sieve to remove gravel and plant residues. An appropriate amount of sample was taken, a dispersant and deionized water were added, and the particles were dispersed by ultrasonic vibration, followed by washing and the addition of a dispersion medium for background measurement. Scattered light signals were collected using a laser particle size analyzer, with measurements repeated 2-3 times for calibration and analysis to determine particle size distribution. Based on soil texture classification principles, the sand content (53-2000 μm) in the samples was quantitatively analyzed. In 2022 and 2023, soil data and corresponding laboratory spectral data collected from cultivated land in the study area were obtained using an ASD FieldSpec4 Hi-Res spectrometer in the soil measurement laboratory, yielding 243 spectral data points. The laboratory spectral data of soil samples had a spectral resolution of 3 nm in the 400–1000 nm band and 8 nm in the 1001–2500 nm band. These laboratory spectral data served as the primary data source for spectral conversion. Simultaneously, Landsat 8 data covering the sampling locations and matching the sampling time were selected, and spectral data for the corresponding pixels (30 m resolution) were extracted through resampling.

[0080] 1.2) Time-series remote sensing image collection

[0081] For the black soil region, four bare soil period intervals were defined based on latitude, as shown in Table 1. Surface reflectance (SR) data from Landsat sensors 5, 8, and 9 were selected and archived in the C02 / T1_L2 dataset in GEE. Clouds and snow were removed from the images using the QA_PIXEL band to ensure data accuracy and usability.

[0082] Table 1

[0083]

[0084] 1.3) Supplementary data

[0085] We used NASA DEM 30-meter resolution digital elevation data to calculate slope and aspect data. We also collected annual cultivated land datasets at 30-meter resolution for the black soil region. Annual average temperature and precipitation data, at a resolution of 1 km, were sourced from the National Earth System Science Data Center of China. Additionally, the ERA5 land monthly average dataset from the European Centre for Medium-Range Weather Forecasts (ECMWF) was used as another meteorological data source, including monthly average air pressure and wind speed data during the bare soil period, and was uniformly resampled to 1 km resolution in Google Earth Engine (GEE). Gross Domestic Product (GDP) and Population Density (POP) data were spatially resolved to 1 km. Spatial distribution data for soil types were obtained from the "1:1,000,000 Soil Map of the People's Republic of China" compiled and published by the National Soil Census Office in 1995, which adopted China's traditional "National Soil System Classification".

[0086] (2) Construction of soil sand content inversion model

[0087] 2.1) Constructing the optimal spectral index

[0088] Spectral information of the sampling points was obtained by selecting remote sensing imagery data that passed through the area during the sampling period. This information is correlated with soil surface properties and can mitigate the influence of soil moisture, surface roughness, and atmospheric conditions. Four dual-band indices—ratio index, difference index, normalized index, and square root index—were selected to construct spectral indices related to soil sand content. The mathematical expressions for these spectral indices are as follows:

[0089] RI(RBi,RBj)=RBi / RBj(1)

[0090] DI(RBi,RBj)=RBi-RBj (2)

[0091] NDI(RBi,RBj)=(RBi-RBj) / (RBi+RBj) (3)

[0092]

[0093] Adding a third band for a specific sensitive region to a dual-band spectral index can significantly improve estimation accuracy and enhance anti-interference capabilities. Therefore, several three-band spectral indices have been selected here:

[0094]

[0095]

[0096] TBI3(RBi,RBj,RBz)=(RBi-RBj)-(RBj-RBz) (7)

[0097]

[0098] 2.2) Random Forest Regression Model

[0099] Random forest regression is a nonlinear machine learning model based on decision trees. This algorithm is widely used to predict the spatial distribution of soil parameters. When environmental factors have a significant impact and there is no obvious linear relationship between variables, random forest algorithms typically achieve higher prediction accuracy than traditional statistical and geostatistical models. It offers advantages such as handling high-dimensional data, outputting feature importance, improving accuracy through result averaging, and supporting fast parallel computation.

[0100] The features used included six original spectral bands from Landsat data, 300 constructed spectral indices, and geographic variables such as altitude, elevation, and aspect. A forward feature selection method was employed, first selecting the variable with the highest random forest score as the initial input. Then, features were sorted by importance, and variables contributing to model accuracy were added sequentially until further additions no longer improved model performance. The final feature set was then selected using this method, choosing the top ten features for subsequent modeling.

[0101] After feature selection, the hyperparameters of the random forest regression model are tuned to improve model accuracy. After identifying the most important features, GridSearchCV is used to search for the optimal hyperparameter combination to ensure maximum model performance. The optimized model parameters are uploaded to the cloud via the Python API GeeMap, allowing for direct use of the final model in subsequent analysis and applications.

[0102] 2.3) Accuracy Analysis

[0103] The 162 samples were divided into a calibration set and a validation set at a 3:1 ratio, with 121 samples serving as training samples and 41 as testing samples. The coefficient of determination (R²) was calculated. 2 The root mean square error (RMSE) and relative prediction deviation (RPD) are used as metrics to evaluate the performance of different models.

[0104]

[0105]

[0106]

[0107]

[0108] Where n is the number of samples, It is the average of the measured values, y i These are measured values. These are the predicted values, and SD is the standard deviation. Typically, a well-developed model usually has a high R-value. 2 And RPD value, and low RMSE. Here, RPD value is divided into five levels to interpret model performance: RPD < 1.4 (unreliable), 1.4 ≤ RPD < 1.8 (average), 1.8 ≤ RPD < 2.0 (good), 2.0 ≤ RPD < 2.5 (very good), RPD ≥ 2.5 (excellent).

[0109] 2.4) Model Accuracy Comparison

[0110] Table 2

[0111]

[0112] (3) Construction of long-term time-series bare soil dataset

[0113] 3.1) Production of satellite image time series

[0114] The study area is vast, and the duration of bare soil varies across different geographical locations due to climatic factors. Therefore, the time range for image data selection was set based on latitude to minimize the impact of factors such as snow cover and vegetation cover, and to acquire as much usable remote sensing image data as possible. To ensure comprehensive coverage of the study area, bare soil image datasets were queried and selected every five years, achieving good coverage in the region. For areas with data gaps (such as high-altitude, cloudy, or snowy areas), a time-series expansion method was used to supplement the dataset with nearby dates. After filtering the image datasets, they were integrated and cropped to obtain a long-term image series for the study area. Finally, bare soil image data for the study area at different time periods was collected using GEE.

[0115] 3.2) Spectral Transformation

[0116] Numerical regression was used to compare the OLI bands with the TM and ETM+OLI-2 bands to achieve the conversion of multi-source data to OLI time-series data. Using the corresponding spectral response function, the soil laboratory spectral data was resampled into TM, ETM+, OLI, and OLI-2 bands. Then, multinomial regression was used to obtain the conversion coefficients between the TM, ETM+, OLI-2, and OLI bands. Using these conversion coefficients, the TM, ETM+, and OLI-2 image data in the dataset can be converted into OLI time-series image data, obtaining the TM and OLI-2 reflectance fitting parameters, R0. OLI =a·R TM +b, R OLI =c·R OLI2 +d. See Table 3 for details.

[0117] Table 3

[0118]

[0119]

[0120] 3.3) Extraction of bare soil pixels

[0121] By combining the spectral characteristics of bare soil pixels and applying NDVI, NBR2, and VNSIR indices in combination, bare soil pixels can be identified more effectively. The index thresholds are then appropriately adjusted based on sample points. Ultimately:

[0122]

[0123]

[0124] VNSIR=1-[(2*Red-Green-Blue)+3*(Swir2-Nir)] (15)

[0125] DI1 = Green-Blue (16)

[0126] DI2 = Red-Green (17)

[0127]

[0128] 3.4) Calculate the temporal soil spectral reflectance of each bare soil pixel.

[0129] After extracting bare soil pixels from image datasets at different time periods, the temporal soil spectral reflectance at the corresponding time is obtained. The obtained data stack can be used to calculate the pixel-by-pixel statistical results of bare soil in the study area, and the percentage of times bare soil appears in each pixel can also be obtained.

[0130] 3.5) Composite soil image

[0131] To reduce errors caused by extreme values, the median statistical method was used to calculate the synthetic reflectance of each pixel. Data from the past forty years was systematically processed to generate synthetic soil images of the study area, accurately preserving the spectral bands of the original data. One phase of the synthetic soil image can be seen as follows: Figure 2 As shown.

[0132] 3.6) Farmland Extraction and Processing

[0133] Using annual land use datasets with a resolution of 30 meters from the past 40 years, cultivated land area data were acquired year by year to create a cultivated land mask dataset, thereby identifying stable cultivated land areas in the soil synthesis results. Non-cultivated land areas were removed by the cultivated land mask, ultimately yielding a composite soil image dataset of cultivated land from the past 40 years for the study area.

[0134] (4) Numerical mapping of soil sand content in arable land

[0135] For the spectral data of each composite image of cultivated land soil, the indices required for modeling are calculated and topographic information is supplemented. This information is then input into a pre-constructed soil sand content inversion model, ultimately yielding soil sand content mapping results for the study area over the past forty years. The first phase of cultivated land soil sand content mapping results can be seen as follows: Figure 3 As shown.

[0136] (5) Construction of an early warning model for soil desertification in arable land

[0137] 5.1) Selection of early warning factors for soil desertification

[0138] 1. First, collect the climate and socioeconomic data required for the farmland soil desertification early warning and evaluation model. This data includes climate factors such as temperature, precipitation, and wind speed, as well as socioeconomic factors such as land use intensity. Use farmland soil sand content as the parent sequence set and the other variables as the child sequence set.

[0139] 2. Determine the optimal set of indicators: Select the optimal value for each indicator to form the optimal set of indicators.

[0140] 3. Normalization: Since different indicators have different units, normalization is needed to convert the values ​​of each indicator into dimensionless values. Common normalization methods include linear normalization and Z-score standardization.

[0141] 4. Calculate the correlation coefficient: Calculate the correlation coefficient between each indicator of each evaluation object and the optimal indicator. Taking grey relational analysis as an example, assess the similarity between each indicator of each object and the optimal indicator. The higher the correlation coefficient, the closer the evaluation object is to the optimal indicator: For the k-th indicator of the i-th evaluation object, its correlation coefficient with the optimal indicator can be calculated using the following formula:

[0142]

[0143] Where, Δ 0k Δ represents the difference between the reference sequence (optimal index) and the reference sequence. ik Let ρ be the difference of the k-th indicator for the i-th evaluation object, and let ρ be the resolution coefficient.

[0144] 5.2) Determination of the weights of early warning factors for farmland soil desertification

[0145] In the stage of determining the weights of early warning factors, various weighting methods such as the Analytic Hierarchy Process (AHP) and the entropy method can be used to weight each early warning factor according to its importance, thereby quantifying its influence in the model. Taking the Analytic Hierarchy Process as an example, the main process includes:

[0146] 1. Construct a judgment matrix: Construct a judgment matrix by comparing the importance of different early warning factors pairwise.

[0147] 2. Weight Calculation: Calculate the weights using the eigenvalue method, or determine the weights based on the information entropy of each factor using the entropy method.

[0148] 3. Consistency check: Use the consistency check to verify the rationality of the judgment matrix. If it does not meet the consistency requirements, the judgment matrix needs to be adjusted and the weights recalculated.

[0149] 4. Weight Adjustment: Based on expert experience and actual data, the calculated weights are adjusted as necessary to ensure the practicality and scientific validity of the model.

[0150] 5.3) Establishment of an early warning and evaluation model for farmland soil desertification

[0151] After determining the factor weights, an early warning and evaluation model for farmland soil desertification can be established based on a comprehensive evaluation method.

[0152] The dimensionless values ​​of each early warning factor are multiplied by their weights and summed to obtain a comprehensive score. Based on the score range, desertification risk levels (e.g., mild, moderate, severe) are then classified. The accuracy of the model is verified using field data, and the model parameters and weights are optimized and adjusted to ensure its accuracy and reliability in practical applications.

[0153] 5.4) Establishment of an early warning model for farmland soil desertification

[0154] The establishment of an early warning model for farmland soil desertification is based on an early warning and evaluation model for farmland soil desertification. It uses time series analysis and machine learning methods to predict future desertification trends, thus achieving early warning. Models such as ARIMA and LSTM neural networks can be used to process time series data and predict future trends in farmland soil desertification. Existing soil desertification monitoring data is used to train the early warning model, enabling it to identify the impacts of climate factors and land use change on soil desertification. The model's prediction results are compared with historical data to evaluate its accuracy and stability, and adjustments are made based on errors. Based on the trained model, the future trend of soil desertification is output, and the desertification risk for a future period is classified according to the early warning level, providing targeted remediation suggestions.

[0155] It is worth noting that some of the above steps can be replaced with the following:

[0156] Regarding remote sensing datasets: In addition to Landsat data, higher spatial resolution Sentinel data, high-resolution data, and higher spectral resolution hyperspectral data can also be used as data sources to improve the spatiotemporal resolution of the dataset through data fusion.

[0157] Eliminating the influence of differences between different remote sensing sensors: Overlapping observations of images taken by different sensors at the same time at the same geographical location can be performed. By analyzing the radiometric response relationship of the image data of these overlapping areas, the calibration factor between each sensor can be calculated.

[0158] Dynamic adjustment of index formula parameters: For different geographical regions with varying spectral characteristics, the parameters of the index formula can be dynamically adjusted based on actual observations to address differences in spectral response caused by environmental changes. In addition to constructing two-band and three-band indices, new spectral indices can also be constructed using methods such as first-order and second-order differentials between two bands.

[0159] Regression model construction: In addition to the random forest algorithm, more advanced deep learning models such as DNN, CNN, and LSTM can be considered for training, combined with the spatiotemporal characteristics of remote sensing data, to improve the prediction accuracy of the model.

[0160] Bare soil extraction methods: In addition to screening bare soil based on a single image index threshold, all available data can be obtained. By constructing a time-series spectral index set, bare soil can be screened using its temporal variation characteristics, or the maximum and minimum values ​​of its time-series indexes can be used to extract bare soil from other land features.

[0161] These alternatives offer greater flexibility and adaptability to meet the soil desertification monitoring needs in different scenarios.

[0162] It should be noted that, for the sake of simplicity, the foregoing method embodiments are all described as a series of actions. However, those skilled in the art should understand that this disclosure is not limited to the described order of actions, because according to this disclosure, some steps can be performed in other orders or simultaneously. Furthermore, those skilled in the art should also understand that the embodiments described in the specification are all optional embodiments, and the actions and modules involved are not necessarily essential to this disclosure.

[0163] The above is an introduction to the method embodiments. The following describes the solution described in this disclosure further through device embodiments.

[0164] Figure 4 The diagram shows a structural diagram of a dynamic monitoring device for farmland soil desertification based on multi-source satellite remote sensing, as provided in an embodiment of this disclosure. Figure 4 As shown, the dynamic monitoring device for farmland soil desertification 400 may include:

[0165] The acquisition module 410 is used to acquire sand content data and laboratory spectral data of surface soil samples of typical soil types collected within the cultivated land area of ​​the study area through laboratory testing, and to acquire multi-source satellite spectral data of the sample locations based on transit images during the sample collection period.

[0166] Module 420 is used to extract spectral features of common bands from multi-source satellite spectral data of sample points, construct a dual-band spectral index between arbitrary bands, and add a third band to the dual-band spectral index to construct a three-band spectral index. The constructed spectral index, the corresponding topographic data, and the corresponding band are used as input features, and the laboratory-measured sand content data is used as the target output to construct an inversion dataset. The machine learning model is trained using the inversion dataset to obtain the soil sand content inversion model.

[0167] The extraction module 430 is used to query all available multi-source satellite remote sensing data for all historical periods based on the bare soil period window time corresponding to multiple sub-regions within the study area, and obtain the historical remote sensing dataset of the study area; extract bare soil pixels from the remote sensing data in the historical remote sensing dataset within the selected bare soil synthesis period, and perform bare soil image synthesis to obtain the bare soil image synthesis result for the corresponding time period, thereby constructing a long-term bare soil dataset.

[0168] The inversion module 440 is used to extract farmland from the long-term bare soil dataset to obtain a long-term farmland bare soil dataset. Based on the long-term farmland bare soil dataset and the corresponding topographic data, the soil sand content inversion model is used to perform inversion to obtain the farmland soil sand content data of the study area.

[0169] Module 450 is established to screen early warning factors for soil desertification of cultivated land from climate data, topographic data, socio-economic data, and soil data of the study area based on soil sand content data, calculate the weights corresponding to the early warning factors for soil desertification of cultivated land, and establish an early warning model for soil desertification of cultivated land based on the early warning factors for soil desertification of cultivated land and their corresponding weights.

[0170] The early warning module 460 is used to provide early warning of farmland soil desertification in the study area using the farmland soil desertification early warning model.

[0171] Understandable, Figure 4 Each module / unit in the dynamic monitoring device 400 for farmland soil desertification shown has the ability to realize Figure 1 The functions of each step in the dynamic monitoring method 100 for farmland soil desertification shown are explained, and their corresponding technical effects are achieved. For the sake of brevity, these will not be elaborated here.

[0172] Figure 5 A structural diagram of an exemplary electronic device capable of implementing embodiments of the present disclosure is shown. Electronic device 500 is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. Electronic device 500 may also represent various forms of mobile devices, such as personal digital processors, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0173] like Figure 5 As shown, the electronic device 500 may include a computing unit 501, which can perform various appropriate actions and processes according to a computer program stored in a read-only memory (ROM) 502 or a computer program loaded from a storage unit 508 into a random access memory (RAM) 503. The RAM 503 may also store various programs and data required for the operation of the electronic device 500. The computing unit 501, ROM 502, and RAM 503 are interconnected via a bus 504. An input / output (I / O) interface 505 is also connected to the bus 504.

[0174] Multiple components in electronic device 500 are connected to I / O interface 505, including: input unit 506, such as keyboard, mouse, etc.; output unit 507, such as various types of monitors, speakers, etc.; storage unit 508, such as disk, optical disk, etc.; and communication unit 509, such as network card, modem, wireless transceiver, etc. Communication unit 509 allows electronic device 500 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0175] The computing unit 501 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 501 include, but are not limited to, a central processing unit (CPU), a graphics processing unit (GPU), various special-purpose artificial intelligence (AI) computing chips, various computing units running machine learning model algorithms, a digital signal processor (DSP), and any suitable processor, controller, microcontroller, etc. The computing unit 501 performs the various methods and processes described above, such as method 100. For example, in some embodiments, method 100 may be implemented as a computer program product, including a computer program tangibly contained in a computer-readable medium, such as storage unit 508. In some embodiments, part or all of the computer program may be loaded and / or installed on device 500 via ROM 502 and / or communication unit 509. When the computer program is loaded into RAM 503 and executed by the computing unit 501, one or more steps of method 100 described above may be performed. Alternatively, in other embodiments, the computing unit 501 may be configured to perform method 100 by any other suitable means (e.g., by means of firmware).

[0176] The various embodiments described above can be implemented in digital electronic circuit systems, integrated circuit systems, field-programmable gate arrays (FPGAs), application-specific integrated circuits (ASICs), application-specific standard products (ASSPs), system-on-a-chip (SoCs), payload programmable logic devices (CPLDs), computer hardware, firmware, software, and / or combinations thereof. These various embodiments may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0177] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0178] In the context of this disclosure, a computer-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A computer-readable medium can be a computer-readable signal medium or a computer-readable storage medium. A computer-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of computer-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0179] It should be noted that this disclosure also provides a non-transitory computer-readable storage medium storing computer instructions, wherein the computer instructions are used to cause a computer to execute method 100 and achieve the corresponding technical effects achieved by executing the method in the embodiments of this disclosure. For the sake of brevity, they will not be described in detail here.

[0180] In addition, this disclosure also provides a computer program product including a computer program that implements method 100 when executed by a processor.

[0181] To provide interaction with a user, the embodiments described above can be implemented on a computer having: a display device (e.g., a CRT (cathode ray tube) or LCD (liquid crystal display) monitor for displaying information to the user; and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, the feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0182] The embodiments described above can be implemented in computing systems that include backend components (e.g., as a data server), or computing systems that include middleware components (e.g., an application server), or computing systems that include frontend components (e.g., a user computer with a graphical user interface or web browser through which a user can interact with the implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication (e.g., a communication network) of any form or medium. Examples of communication networks include local area networks (LANs), wide area networks (WANs), and the Internet.

[0183] Computer systems can include clients and servers. Clients and servers are generally located far apart and typically interact via communication networks. Client-server relationships are created by computer programs running on the respective computers and having a client-server relationship with each other. Servers can be cloud servers, servers in distributed systems, or servers incorporating blockchain technology.

[0184] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0185] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A method for dynamic monitoring of farmland soil desertification based on multi-source satellite remote sensing, characterized in that, The method includes: The study obtained sand content data and laboratory spectral data of surface soil samples of typical soil types collected within the cultivated land area of ​​the study area through laboratory testing, and obtained multi-source satellite spectral data of the sample locations based on the transit images during the sample collection period. Using multi-source satellite spectral data from sample locations, spectral features of common bands are extracted to construct a dual-band spectral index between arbitrary bands. A third band is then added to the dual-band spectral index to construct a three-band spectral index. The constructed spectral index, the corresponding topographic data, and the corresponding bands are used as input features, and the laboratory-measured sand content data is used as the target output to construct an inversion dataset. The machine learning model is trained using the inversion dataset to obtain a soil sand content inversion model. Based on the bare soil period window time corresponding to multiple sub-regions within the study area, query all available multi-source satellite remote sensing data for all historical periods to obtain the historical remote sensing dataset of the study area; extract bare soil pixels from the remote sensing data in the historical remote sensing dataset within the selected bare soil synthesis period, and synthesize bare soil images to obtain the bare soil image synthesis results for the corresponding time period, thereby constructing a long-term bare soil dataset. Farmland extraction processing was performed on the long-term bare soil dataset to obtain a long-term farmland bare soil dataset. Based on the long-term farmland bare soil dataset and the corresponding topographic data, a soil sand content inversion model was used to invert the farmland soil sand content data of the study area. Based on the sand content data of arable land soil, early warning factors for arable land soil desertification were screened from climate data, topographic data, socio-economic data and soil data of the study area, and the weights corresponding to the early warning factors for arable land soil desertification were calculated. Based on the early warning factors for arable land soil desertification and their corresponding weights, an early warning model for arable land soil desertification was established. The farmland soil desertification early warning model was used to provide early warning of farmland soil desertification in the study area.

2. The method according to claim 1, characterized in that, The process of adding a third band to a dual-band spectral index to construct a three-band spectral index includes: Calculate the correlation coefficient between the dual-band index and the laboratory-measured sand content data, and add a third band to the dual-band spectral index where the correlation coefficient is greater than a preset threshold to construct a three-band spectral index. Then, supplement the existing spectral indexes in the constructed three-band spectral index.

3. The method according to claim 1, characterized in that, The process of training a machine learning model using an inversion dataset to obtain a soil sand content inversion model includes: Based on the inversion dataset, the training set and validation set are divided. The cross-validation algorithm is used to select the input features that are actually used for modeling, and the training set is then used to remove features, retaining the input features that are actually used for modeling. The updated training set was used to train the machine learning model, and hyperparameters were tuned to obtain the soil sand content inversion model.

4. The method according to claim 1, characterized in that, The process involves extracting bare soil pixels from the remote sensing data in the historical remote sensing dataset within the selected bare soil synthesis period, and then synthesizing bare soil images to obtain the synthesized bare soil images for the corresponding time period. This process is used to construct a long-term bare soil dataset, including: Based on laboratory spectral data and the spectral response functions of various satellites, the laboratory spectral data is resampled to simulate the band characteristics of each sensor and establish a spectral response relationship model between different sensors. This ensures standardized processing and unified scale conversion of cross-sensor data. Bare soil pixels are extracted from the converted remote sensing data, and bare soil images are synthesized to obtain the synthesized bare soil images for the corresponding time period, thereby constructing a long-term bare soil dataset.

5. The method according to claim 1, characterized in that, The process of extracting farmland from the long-term bare soil dataset yields a long-term farmland bare soil dataset, including: A farmland mask dataset was created using the annual land use dataset. Based on this, farmland extraction processing was performed on the long-term bare soil dataset to obtain the long-term farmland bare soil dataset.

6. The method according to claim 1, characterized in that, Based on the soil sand content data of cultivated land, early warning factors for cultivated land desertification are screened from climate data, topographic data, socioeconomic data, and soil data of the study area, and the weights corresponding to these early warning factors are calculated, including: Based on the sand content data of arable land soil, the correlation analysis algorithm was used to screen early warning factors for arable land soil desertification from climate data, topographic data, socio-economic data and soil data of the study area. The relative importance of each soil desertification early warning factor in cultivated land is calculated by comparing them pairwise, and the corresponding weights are obtained accordingly.

7. The method according to claim 1, characterized in that, The establishment of an early warning model for farmland soil desertification based on early warning factors and their corresponding weights includes: Based on the early warning factors of arable land soil desertification and their corresponding weights, a comprehensive evaluation algorithm is used to establish an early warning evaluation model for arable land soil desertification. This model is used to quantitatively assess the risk of arable land soil desertification, predict the risk level of arable land soil desertification, and establish an early warning model for arable land soil desertification based on this.

8. A dynamic monitoring device for farmland soil desertification based on multi-source satellite remote sensing, characterized in that, The device includes: The acquisition module is used to acquire sand content data and laboratory spectral data of surface soil samples of typical soil types collected within the cultivated land area of ​​the study area through laboratory testing, and to acquire multi-source satellite spectral data of the sample locations based on the transit images during the sample collection period. The module is used to extract spectral features of common bands from multi-source satellite spectral data of sample points, construct a dual-band spectral index between arbitrary bands, and add a third band to the dual-band spectral index to construct a three-band spectral index. The constructed spectral index, the corresponding topographic data, and the corresponding bands are used as input features, and the laboratory-measured sand content data is used as the target output to construct an inversion dataset. The machine learning model is trained using the inversion dataset to obtain the soil sand content inversion model. The extraction module is used to query all available multi-source satellite remote sensing data for all historical periods based on the bare soil period window time corresponding to multiple sub-regions within the study area, and obtain the historical remote sensing dataset of the study area; the remote sensing data in the historical remote sensing dataset within the selected bare soil synthesis period are used to extract bare soil pixels and synthesize bare soil images to obtain the bare soil image synthesis results for the corresponding time period, thereby constructing a long-term bare soil dataset. The inversion module is used to extract farmland from the long-term bare soil dataset to obtain a long-term farmland bare soil dataset. Based on the long-term farmland bare soil dataset and the corresponding topographic data, the soil sand content inversion model is used to perform inversion to obtain the farmland soil sand content data of the study area. A module is established to screen early warning factors of farmland soil desertification from climate data, topographic data, socio-economic data, and soil data of the study area based on farmland soil sand content data, calculate the weights corresponding to the early warning factors of farmland soil desertification, and establish an early warning model of farmland soil desertification based on the early warning factors of farmland soil desertification and their corresponding weights. The early warning module is used to provide early warning of farmland soil desertification in the study area using the farmland soil desertification early warning model.

9. An electronic device, characterized in that, The electronic device includes: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-7.

10. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-7.

Images