CN115310719B - Design method of farmland soil sampling plan based on three-stage k-means - Google Patents

Abstract

The invention discloses a method for designing a farmland soil sampling scheme based on three-stage k-means, including: S1, based on DEM data, using the k-means method and the geographic detector method to form farmland sub-regions with spatial differentiation, and According to the area ratio, allocate the number of soil samples in the farmland sub-region; S2. Based on the NDVI data, use the local variation coefficient CV and k-means method to subdivide the farmland sub-region obtained by k-means in the first stage to form a local space A series of sub-patches with similar variation levels, and obtain the number of soil samples for each sub-patch according to the area and local spatial variation level; S3. Based on the remote sensing yield estimation data, use the k-means method and variance statistics to determine the representative soil The spatial location of the sample. The invention comprehensively considers the spatial heterogeneity of soil and the level of local spatial variation, and can more scientifically and rationally design farmland soil sampling schemes to obtain good-quality soil spatial data.

Description

Design method of farmland soil sampling plan based on three-stage k-means

technical field

The invention belongs to the technical field of geographic information systems, and discloses a three-stage k-means-based farmland soil sampling scheme design method.

Background technique

China feeds nearly 20% of the world's population with about 7% of the world's arable land. However, in recent years, there have been many problems in the development of modern agriculture in my country (high agricultural production costs, unreasonable use of chemical fertilizers or pesticides, low degree of agricultural mechanization, and soil compaction, etc.), leading to increasingly prominent contradictions between population, environmental security and food security . Precision agriculture is a new trend of modern agriculture, a key means for modern agriculture to transform from resource-input to technology-based, and a symbol of the combination of agricultural informatization, mechanization and modernization, and will help resolve the above contradictions. The country has always attached great importance to the development of precision agriculture. In recent years, with the continuous acceleration of urbanization, many farmers have migrated to cities to work, resulting in idle rural land. Intensive management of rural land will become a new trend of the times, which also provides a new opportunity for the smooth development and implementation of precision agriculture. . The specific meaning of precision agriculture is to rationally adjust the input of agricultural materials according to the soil nutrients in the farmland, the soil nutrients required for the growth of crops, and the target yield, so as to improve the utilization rate of agricultural materials and agricultural productivity, save resources, and protect Agricultural environment and other purposes.

As an important part of the earth's surface system, soil has spatial differentiation. The spatial inconsistency between soil nutrient supply and crop growth demand is the fundamental reason affecting crop yield. Therefore, fully understanding the spatial differentiation of farmland soil nutrients will help to achieve the development goals of precision agriculture. At present, the main method to grasp the spatial distribution of soil nutrients is to collect field soil samples from farmland and perform physical and chemical analysis in the laboratory to determine soil nutrient content. However, collecting soil samples on a large scale and intensively will consume a lot of manpower, material resources and financial resources. Soil sampling is a scientific method to estimate the spatial distribution of farmland soil nutrients through a certain soil prediction model by obtaining a small amount of key soil sample point nutrient information, which can achieve a better balance between sampling costs and expected soil prediction accuracy. Different sampling schemes will obtain different spatial distribution characteristics of farmland soil nutrients. Research and selection of reasonable soil sampling methods will be a powerful guarantee for obtaining reliable spatial distribution of soil nutrients in farmland.

At present, soil sampling methods are mainly divided into design-based sampling methods and model-based sampling methods. The design-based soil sampling method is to quantitatively determine the optimal sample size based on prior knowledge, and then use random sampling, systematic sampling and stratified sampling to determine the sample layout. For example, Wang Jinfeng et al. proposed a “sandwich” spatial sampling model based on stratified sampling and considering spatial variability. In addition, according to the synergistic relationship between soil and environmental variables, some studies use environmental variables to assist in the design of sampling plans to improve soil sampling efficiency. Model-based soil sampling methods are based on geostatistical theory, and through certain criteria, the implementation can accurately fit the population. This method designs the optimal sample number and spatial distribution pattern by minimizing the model estimation variance. For example, Wang Zhenhua and others combined fuzzy set theory with sampling inspection theory and proposed a two-level sampling model based on spatial data quality inspection. In addition to the above-mentioned soil sampling methods, recent research trends also include resampling methods based on spatially extrapolated uncertainties and sampling methods considering accessibility, etc.

Compared with the design-based soil sampling method, the model-based soil sampling method can fully excavate the structural information of the soil in the study area, so that the soil information of key points can be obtained better. However, most model-based soil sampling methods are based on geostatistical theory. The design of sampling schemes using this type of method needs to rely on the variation function, and the variation function is usually known only after sampling, or before adopting the study area. Variation function modeling of the soil data. Due to the strong intervention of human activities in agricultural areas, there may be large differences in the spatial distribution of soil between different years. In this case, it will not be possible to model the coefficient of variation using previous soil data for that agricultural area. Therefore, design-based soil sampling methods may be more suitable for application in precision agriculture. Different crop types absorb different soil nutrients. The complex planting structure and artificial fertilization management in agricultural areas will lead to certain local spatial differences in the soil in this area. It is urgent to develop a new soil sampling method based on design. By considering the local spatial variability of soil in agricultural areas to identify and obtain soil information at key points, the accuracy of digital soil mapping for precision agriculture can be improved.

Contents of the invention

In view of the above technical problems, the present invention provides a method for designing a farmland soil sampling scheme based on three-stage k-means, and tries to infer the local spatial variability of soil by mining the environmental information closely related to the soil and using the three-stage k-means method. Determine the spatial location of representative samples to form a farmland soil sampling plan. Among them, the first stage of k-means processing is to construct farmland sub-regions with spatial differentiation and reasonably allocate the number of soil samples; the second stage of k-means processing is to form sub-patches with similar soil local spatial variation levels and obtain corresponding The sampling number of sub-patches; the third stage of k-means processing determines the spatial location of representative samples.

The design method of farmland soil sampling scheme based on three-stage k-means includes the following steps:

S1. Based on the DEM data, use the k-means method and the geographic detector method to form farmland sub-regions with spatial differentiation, and allocate the number of soil samples in the farmland sub-regions according to the area ratio to realize the first stage of k-means processing ;

S1 specifically includes the following sub-steps:

S11. Using the k-means method to construct farmland sub-regions with spatial differentiation:

Based on the DEM data, the k-means method is used to divide the farmland to form farmland sub-regions with similar topographic conditions within the region and different topographic conditions between regions, that is, spatial differentiation;

S12. Using geographic detectors to determine the optimal number of farmland sub-regions:

Based on the DEM data, use the Q value in the geographic detector to detect the spatial differentiation of farmland sub-regions under different numbers, obtain the change curve of Q value, and select the number corresponding to the inflection point as the optimal number of farmland sub-regions;

S13. Assign the number of soil samples for each farmland sub-region:

The design of the soil sampling plan is to determine the sample size first, and then determine the sample location; based on the results of the division of farmland sub-regions, calculate its area, and use this as a weight to allocate the number of soil samples for each farmland sub-region.

S2. Based on the NDVI data, use the local variation coefficient CV and k-means method to subdivide the farmland sub-region obtained by k-means in the first stage to form a series of sub-patches with similar levels of local spatial variation. and the level of local spatial variation, obtain the number of soil samples of each sub-patch, and realize the second stage of k-means processing;

S21. Infer the local spatial variation level of farmland soil:

Select the NDVI data that is closely related to crop growth, use the local variation coefficient CV to infer the local spatial variation level of farmland soil, and provide data support for the further allocation of soil samples;

S22. Generate sub-patches with similar soil local spatial variation levels

Based on the calculation results of the local spatial variation level of farmland soil, using the k-means method, sub-patches with similar soil local spatial variation levels are formed within the farmland sub-region; the number of clusters in each farmland sub-region is determined according to the Half of the number of soil samples in the area shall be determined;

S23. Obtain the sampling quantity of the corresponding sub-patch:

Based on the sub-patch division results, calculate its area and local variation coefficient CV, and use this as a weight to further allocate soil samples, so as to obtain the sampling number of the corresponding sub-patch, and provide quantitative support for determining the spatial location of soil sampling.

S3. Based on the remote sensing yield estimation data, use the k-means method and variance statistics to determine the spatial location of representative soil samples, realize the third stage of k-means processing, and complete the design of the farmland soil sampling plan.

S3 specifically includes the following sub-steps:

S31, forming subsets with similar crop yield levels inside the subpatch:

Based on crop remote sensing yield estimation data, using the k-means method, a subset with similar crop yield levels is formed within each sub-patch; the clustering number of each sub-patch is equal to the sampling number of the corresponding sub-patch;

S32. Determine the spatial position of the representative sample:

After going through the above process, each subset has similar terrain conditions, local soil variation levels, and crop yield levels; in order to determine the spatial location of representative samples, the expected variance is calculated; then, a sampling point is randomly selected within each subset to form a sample The sample point set is searched for the sample point set closest to the expected variance as the sampling point, so as to determine the spatial position of the representative sample and form the farmland soil sampling plan.

The present invention comprehensively considers the spatial heterogeneity of soil and the level of local spatial variation, and can more scientifically and rationally design farmland soil sampling schemes to obtain good-quality soil spatial data, and further produce accurate results of farmland soil spatial distribution, which are variables Provide technical and data support for precision agricultural decisions such as fertilization and planting structure adjustment.

Description of drawings

Fig. 1 is overall technical flow chart of the present invention

Fig. 2 is the research area of embodiment

Fig. 3 is the first stage k-means processing result of embodiment

Fig. 4 is Fig. 4 local variation coefficient CV calculation of embodiment and second stage k-means processing result

Fig. 5 is the third stage k-means processing and soil sampling program result of embodiment;

Figure 6 is a comparison of different sampling methods: (a) three-stage k-means sampling of the present invention, (b) stratified random sampling, (c) k-means sampling and (d) regular grid sampling;

Fig. 7 is the soil SOM attribute mapping and error distribution of different sampling methods: (a) three-stage k-means sampling of the present invention, (b) stratified random sampling, (c) k-means sampling and (d) regular grid sampling.

Specific implementation method

The specific technical solutions of the present invention are described in conjunction with the examples.

As shown in FIG. 1 , the design method of farmland soil sampling scheme based on three-stage k-means, the research area of this embodiment is shown in FIG. 2 . The method comprises the steps of:

S1. Construct farmland sub-regions with spatial differentiation and rationally distribute the number of soil samples (k-means in the first stage)

S11 Use the k-means method to construct farmland sub-regions with spatial differentiation

The ultimate goal of soil sampling is to obtain and express the spatial distribution of soil in the study area through a small amount of soil samples. As a data type of geography, soil has spatial differentiation. Spatial heterogeneity, the full name of which is spatial stratified heterogeneity, refers to the difference in the value of an attribute between different types or regions, such as land use maps, climate zoning, ecological zoning, geographical zoning, etc. Some scholars will use environmental variables closely related to soil to divide the research area into sub-areas with spatial differentiation to assist in the design of soil sampling schemes. Among them, terrain is one of the most commonly used environment variables. In view of this, in this embodiment, based on the DEM data, the k-means method is used to divide the farmland to form farmland sub-regions with similar terrain conditions within the region and different terrain conditions between regions (that is, with spatial differentiation).

The k-means method is a clustering analysis algorithm for iterative solution. Its steps are to pre-divide the data into K groups, then randomly select K objects as the initial cluster centers, and then calculate the relationship between each object and each seed cluster. The distance between centers, assigning each object to the cluster center closest to it. The cluster centers and the objects assigned to them represent a cluster. Each time a sample is assigned, the cluster center of the cluster is recalculated based on the existing objects in the cluster. This process will be repeated until a certain termination condition is met.

S12 Determining the Optimal Number of Farmland Subregions Using Geographic Detectors

Before using the k-means method to obtain spatially differentiated farmland sub-regions, it is necessary to specify a specific number of clusters. Different numbers of clusters will lead to different spatial differentiation of farmland sub-regions. Spatial differentiation can be identified, tested, found, and attributed using geodetector Q statistics.

Geodetectors are a way to probe spatially stratified heterogeneity and reveal its inner driving forces. Its core idea is that if an independent variable has an important influence on the dependent variable, then the spatial distribution of the independent variable should be similar to that of the dependent variable, and it is a tool for detecting and utilizing the spatial differentiation of geographical phenomena. Geographic detectors include differentiation and factor detection, interaction detection, ecological detection and risk area detection. The goal of differentiation and factor detection is to detect the spatial variation of the dependent variable and to detect the extent to which an independent variable explains the spatial variation of the dependent variable. The goal of interaction detection is to identify the interaction between different dependent variables and to assess whether the interaction of the interaction factors will enhance or weaken their explanation of the dependent variable.

In this method, the differentiation and factor detection (Q statistics) in geographic detectors are selected to analyze the spatial differentiation of farmland sub-regions. The model is as follows:

(1)

In the formula, for the classification or zoning of croplands (cropland subregions), and Respectively, the classification of farmland and the area of the whole area; and are the variance of the farmland classification and the whole area of the DEM variable, respectively. , The larger the value, the more obvious the spatial differentiation.

In this embodiment, based on the DEM data, the Q value in the geographical detector is used to detect the spatial differentiation of different numbers of farmland sub-regions, and the change curve of the Q value is drawn, and the number corresponding to the inflection point is selected as the optimal The number of farmland subregions.

S13 Rationally allocate the number of soil samples in each farmland sub-region

Spatially differentiated farmland subregions are the most fundamental basis for rationally allocating the number of soil samples. First, determine the overall sample size of the farmland soil sampling plan according to the accuracy of soil mapping or budget constraints; then, calculate the area of each farmland sub-region, and use this as a weight to reasonably allocate the number of soil samples in each farmland sub-region. The formula looks like this:

(2)

In the formula, SN is the overall sample size of the farmland soil sampling plan, is the area of farmland subregion h, is the number of soil samples in farmland subregion h.

S2. Form sub-patches with similar soil local spatial variation levels, and obtain the sampling number of corresponding sub-patches (the second stage k-means)

S21 Inferring the level of local spatial variation of farmland soils

Soil areas with strong local spatial variation need to arrange more soil samples to obtain accurate soil spatial distribution results. However, farmland soils in different regions have different levels of local soil spatial variation. Local soil conditions will affect the growth of farmland crops. Therefore, the local conditions of the soil can be inferred from the growth conditions of farmland crops. In this study, the NDVI data closely related to crop growth were selected, and the local variation coefficient CV was used to infer the local spatial variation level of farmland soil, which provided data support for the further allocation of soil samples.

The local coefficient of variation CV is a variation index expressed in relative number form. It is obtained by comparing the range, mean deviation or standard deviation in the variation index with the mean, and the standard deviation coefficient is commonly used. The application condition of the coefficient of variation is that when the levels of the two series to be compared are different, the range, mean difference or standard deviation cannot be used for comparative analysis of hundreds of lines, because they are all absolute indicators. The CV formula looks like this:

(3)

In the formula, is one of the NDVI values within the calculation window, Computes the mean of DNVI over a window. Among them, the size of the calculation window needs to be determined according to the size of the research area.

S22 generates subpatches with similar levels of soil local spatial variation

Within each farmland sub-region, the local spatial variation level of soil will have certain differences. Soil areas with high levels of local spatial variation need to set more soil sampling points to obtain accurate soil spatial distribution results. To achieve this, sub-regions of farmland need to be further divided to generate sub-patches with similar levels of local spatial variation in soil.

In this embodiment, based on the calculation results of the local spatial variation level of farmland soil, the k-means method is used to form sub-patches with similar local soil spatial variation levels within the farmland sub-region. In order to ensure that there are sampling points in each sub-patch, the number of clusters in each farmland sub-region is determined according to half of the number of soil samples in each farmland sub-region.

S23 Obtain the sampling number of the corresponding sub-patch

The larger the sub-patch area, the higher the level of local soil spatial variation, and the more sampling points are required. In this embodiment, based on the sub-patch division results, calculate its area and local coefficient of variation CV, and use this as a weight to further allocate soil sample points, so as to obtain the sampling number of the corresponding sub-patch, in order to determine the soil sampling space position, providing quantitative support. The sampling number of sub-patch is determined as follows:

(4)

In the formula, and is the area and CV of the sub-patch l inside the farmland sub-region h .

Figure 4 shows the results of local variation coefficient CV calculation and the second stage k-means processing.

S3. Determine the spatial location of representative samples (the third stage k-means)

S31 Form sub-groups with similar crop yield levels within the sub-patch

Within each sub-patch, the local spatial variation level of the soil was similar, but the spatial distribution of the local soil was not consistent. Climatic conditions, temperature conditions, and disaster conditions in the same field are similar, so crop yield can explain soil conditions to a large extent.

With the development of remote sensing technology, it is possible to quickly estimate the yield of large-scale crops. Among them, the WOFOST model is the most commonly used crop remote sensing yield estimation method. The WOFOST (WOrldFOodSTudies) model is jointly developed by Wageningen Agricultural University in the Netherlands and the Center for World Food Research (CWFS) to simulate the dynamic explanatory model of annual crop growth under specific soil and climate conditions. The model places a strong emphasis on quantitative land evaluation, regional yield forecasting, risk analysis, and quantification of interannual yield variability and climate change impacts. The model is based on the simulation of crop physiological and ecological processes such as assimilation, respiration, transpiration and dry matter distribution, and mainly includes the simulation of crop growth under potential growth conditions, water limitation conditions and nutrient limitation conditions.

In this example, based on the GF-1 remote sensing satellite data, the WOFOST model is used to quickly estimate the large-scale yield of crops; then, based on the remote sensing yield estimation data of crops, the k-means method is used to form similar A subset of crop yield levels. The number of clusters in each sub-patch is equal to the number of samples in the corresponding sub-patch.

S32 determine the spatial location of the representative sample

After going through the above process, each subset has similar topographic conditions, local soil variation levels, and crop yield levels. In order to reasonably allocate soil sample resources and determine the spatial location of representative samples, the expected variance was calculated first. The formula looks like this:

(5)

In the formula, is the variance of crop remote sensing yield estimation data in farmland sub-region h . Then, a sample point is randomly selected in each subset to form a sample point set, and the sample point set closest to the expected variance is searched for as a sampling point, so as to determine the spatial position of the representative sample and form a farmland soil sampling plan. Fig. 5 is the result of the third stage k-means processing and soil sampling scheme of the embodiment.

Use different methods to compare. The quantitative evaluation results of soil SOM mapping with different sampling methods are shown in Table 1.

Table 1 Quantitative evaluation results of soil SOM mapping with different sampling methods

As shown in Figure 6, in Figure 6 are (a) three-stage k-means sampling, (b) stratified random sampling, (c) k-means sampling and (d) regular grid sampling.

Fig. 7 shows soil SOM attribute mapping and error distribution for different sampling methods: (a) three-stage k-means sampling, (b) stratified random sampling, (c) k-means sampling and (d) regular grid sampling.

Claims (1)

1. The farmland soil sampling scheme design method based on three-stage k-means, is characterized in that, comprises the steps: S1. Based on the DEM data, use the k-means method and the geographic detector method to form farmland sub-regions with spatial differentiation, and allocate the number of soil samples in the farmland sub-regions according to the area ratio to realize the first stage of k-means processing ; S1 specifically includes the following sub-steps: S11. Using the k-means method to construct farmland sub-regions with spatial differentiation: Based on the DEM data, the k-means method is used to divide the farmland to form farmland sub-regions with similar topographic conditions within the region and different topographic conditions between regions, that is, spatial differentiation; S12. Using geographic detectors to determine the optimal number of farmland sub-regions: Based on the DEM data, use the Q value in the geographic detector to detect the spatial differentiation of farmland sub-regions under different numbers, obtain the change curve of Q value, and select the number corresponding to the inflection point as the optimal number of farmland sub-regions; S13. Assign the number of soil samples for each farmland sub-region: The design of the soil sampling plan is to determine the sample size first, and then determine the sample location; based on the results of the division of farmland sub-regions, calculate the area of each farmland sub-region, and use this as a weight to reasonably allocate the number of soil samples in each farmland sub-region; The formula looks like this:

In the formula, SN is the overall sample size of the farmland soil sampling scheme, A _h is the area of the farmland sub-region h, and SN1 _h is the number of soil samples in the farmland sub-region h; S2. Based on the NDVI data, use the local variation coefficient CV and k-means method to subdivide the farmland sub-region obtained by k-means in the first stage to form a series of sub-patches with similar levels of local spatial variation. and the level of local spatial variation, obtain the number of soil samples of each sub-patch, and realize the second stage of k-means processing; S2 specifically includes the following sub-steps: S21. Infer the local spatial variation level of farmland soil: Select the NDVI data that is closely related to crop growth, use the local variation coefficient CV to infer the local spatial variation level of farmland soil, and provide data support for the further allocation of soil samples; S22. Generate sub-patches with similar soil local spatial variation levels Based on the calculation results of the local spatial variation level of farmland soil, using the k-means method, sub-patches with similar soil local spatial variation levels are formed within the farmland sub-region; the number of clusters in each farmland sub-region is determined according to the Half of the number of soil samples in the area shall be determined; S23. Obtain the sampling quantity of the corresponding sub-patch: Based on the sub-patch division results, calculate its area and local coefficient of variation CV, and use this as a weight to further allocate soil samples, so as to obtain the sampling number of the corresponding sub-patch, and provide quantitative support for determining the spatial location of soil sampling; The sampling number of sub-patch is determined as follows:

In the formula, A _hl and CV _hl are the area and CV of the sub-patch l inside the farmland sub-region h; S3. Based on the remote sensing yield estimation data, use the k-means method and variance statistics to determine the spatial location of representative soil samples, realize the third stage of k-means processing, and complete the design of the farmland soil sampling plan; S3 specifically includes the following sub-steps: S31, forming subsets with similar crop yield levels inside the sub-patch: Based on crop remote sensing yield estimation data, using the k-means method, a subset with similar crop yield levels is formed within each sub-patch; the clustering number of each sub-patch is equal to the sampling number of the corresponding sub-patch; S32. Determine the spatial position of the representative sample: After going through the above process, each subset has similar terrain conditions, local soil variation levels, and crop yield levels; in order to determine the spatial location of representative samples, the expected variance is calculated; then, a sampling point is randomly selected within each subset to form a sample The sample point set is searched for the sample point set closest to the expected variance as the sampling point, so as to determine the spatial position of the representative sample and form the farmland soil sampling plan.

Images