General methods for the validation of terrestrial quantitative remote sensing products
Some standard content:
ICS35.240.70
National Standard of the People's Republic of China
GB/T39468—2020
General methods for the validation of terrestrial quantitative remote sensing products
Published on 2020-11-19
State Administration for Market Regulation
National Administration of Standardization
Implementation on 2021-06-01
Normative reference documents
Terms and definitions
General methods for direct verification
Steps for direct verification
Methods for analysis of spatial heterogeneity
Spatial sampling methods
Scale conversion methods
General methods for indirect verification
Steps for indirect verification
Cross-check Verification method
Test method for temporal and spatial change trend analysis
Multi-scale step-by-step verification method based on ground observation and high-resolution remote sensing data 6 Evaluation method
6.1 Evaluation of consistency between remote sensing products and relative true value/reference object 6.2 Uncertainty evaluation of remote sensing product authenticity verification process Appendix A (Informative Appendix) Surface spatial heterogeneity analysis method Appendix B (Informative Appendix) Spatial sampling model References
-rrKaeerKAca-
GB/T39468—2020
This standard was drafted in accordance with the rules given in GB/T1.1-2009. GB/T39468—2020
Please note that some contents of this document may involve patents. The issuing agency of this document does not assume the responsibility for identifying these patents. This standard was proposed by the Chinese Academy of Sciences.
This standard is under the jurisdiction of the National Remote Sensing Technology Standardization Technical Committee (SAC/TC327). The drafting units of this standard are: Institute of Geographic Sciences and Natural Resources Research, Chinese Academy of Sciences, Institute of Tibetan Plateau Research, Chinese Academy of Sciences, Institute of Space Information Innovation, Chinese Academy of Sciences, and China Resources Satellite Application Center. The main drafters of this standard are: Ge Yong, Hu Maogui, Wang Jianghao, Li Xin, Wang Jinfeng, Zhang Renhua, Wu Hua, Liu Qinhuo, Wang Xinhong, Pan Zhiqiang, and Liu Zhaoyan.
rrKacerKAca-
1 Scope
General method for authenticity verification of land quantitative remote sensing products GB/T39468—2020
This standard specifies the general direct verification method, indirect verification method and evaluation method for authenticity verification of land quantitative remote sensing products.
This standard applies to the authenticity verification of land quantitative remote sensing products. Normative reference documents
The following documents are indispensable for the application of this document. ：For any dated referenced document, only the dated version applies to this document. For any undated referenced document, the latest version (including all amendments) applies to this document. GB/T36296-2018 Remote Sensing Product Authenticity Verification Guidelines Terms and Definitions
The terms and definitions defined in GB/T36296-2018 and the following terms and definitions apply to this document. 3.1
estimator
Estimator
Statistical inference value of a population attribute through a sample. [GB/Z33451-2016, definition 3.1.18]］3.2
sample
A subset of a population consisting of one or more sampling units [GB/T3358.2-2009. definition 1.2.17]］3.3
sampling
The action of extracting or forming a sample.
[GB/T3358.2—2009. Definition 1.3.1]3.4
Sampling unit samplingunit
Each part after the population is divided
Note 1: Sampling units can be graded. The population consists of primary (sampling) units, each primary (sampling) unit consists of secondary sampling units, and so on. Note 2: Rewrite GB/T3358.2—2009, definition 1.2.14.3.5
simplerandomsampling
Simple random sampling
Select n sampling units from the population to form a sample, so that all possible combinations of n sampling units have equal probability of being drawn.
［GB/T3358.2—2009, definition 1.3.4a
rKaeerKAca-
GB/T39468—2020
Systematic samplingsystematicsampling
Arrange the sampling units in the population in a certain order, randomly select one or a group of initial units within the specified range, and then determine the sampling of other sample units according to certain rules. GB/T3358.2-2009. Definition 1.3.12
Stratified samplingstratifiedsampling
Samples are drawn from different layers of the population, and each layer has at least one sampling unit. Note 1: The units of the population are divided into several secondary populations (layers) according to certain characteristics, and then random sampling is carried out from each layer to form a sample. Note 2: Rewrite GB/T3358.2-2009. Definition 1.3.6. 3.8
Kriging model
Based on the theory of variogram and structural analysis, it considers the geometric characteristics of the sample points such as size, shape, and spatial distribution, and uses a limited number of sample points to make the best unbiased estimate of the value of the region or a local area within the region. 3.9
Mean of surface with non-homogeneity model; MSN is a model for making the best unbiased estimate of the mean of heterogeneous spatial surfaces that can be homogenized within the layers by stratification, taking into account the spatial correlation within and between layers. 4 General method for direct inspection
4.1 Direct inspection steps
The inspection process of direct inspection is shown in Figure 1.
Test site selection
Sample selection
Data measurement
Accuracy, consistency judgment and image
Accuracy evaluation of the test results
·Surface spatial heterogeneity analysis
, typical verification site selection
, sampling method based on sample point observation
·Sampling method of Biting footprint observation
·Point to pixel conversion
Footprint to image scale conversion
Accuracy evaluation
●Uncertainty evaluation
Figure 1 Direct verification process
The main steps are as follows:
Verification site selection: According to the surface spatial heterogeneity of the observation target (analysis method see 4.2), it is advisable to a
Select high, medium and low value pixels or areas of the target variable in different underlying surfaces or product coverage within the product coverage. It is recommended to select the central area of the homogeneous area as the verification field or observation field
KaeerKAca-
b) Sample selection: Select the sampling method according to 4.3 to determine the sample size and sample selection. GB/T39468—2020
c) Data measurement: According to the selected samples, select the corresponding measuring instrument at the pixel scale to continuously observe the target variable or select representative time points for observation d) Scale consistency judgment and pixel scale relative true value determination: According to the pixel size of the product to be inspected, judge whether the spatial scale of the measurement data meets the requirements. If it is satisfied, it can be directly compared and verified; otherwise, select the scale conversion method to estimate the relative true value of the pixel scale according to the requirements of 4.4.
e) Evaluation of test results: Compare the relative true value of the pixel scale with the value of the product to be tested, and evaluate the consistency and uncertainty between the quantitative remote sensing product and the relative true value from the two aspects of accuracy and uncertainty. 4.2 Spatial heterogeneity analysis method
According to spatial heterogeneity, the land surface can be divided into two categories: homogeneous surface and heterogeneous surface. It is recommended to use methods such as spatial autocorrelation analysis or spatial variance analysis to calculate the spatial differences of surface variables, such as spatial autocorrelation Moran's index (see A.1 in Appendix A), semivariogram (see A.2), coefficient of variation (see A.3) or geographic detector 9 index (see A.4), etc., combined with the scale of the remote sensing product to be tested, surface characteristics and expert knowledge to determine the surface spatial heterogeneity. 4.3 Spatial sampling method
4.3.1 Sampling method based on sample point observation
For sampling based on sample point observation targets, the sampling model should be selected according to the surface conditions (see Appendix B). The specific requirements are as follows: a) When the spatial distribution of the observation target does not have spatial autocorrelation and does not have spatial stratification heterogeneity, a simple random sampling model or a systematic sampling model should be used;
b) When the spatial distribution of the observation target does not have spatial autocorrelation but has spatial stratification heterogeneity, a stratified sampling model should be used. When the spatial distribution of the observation target has spatial autocorrelation but does not have spatial stratification When the spatial distribution of the observed target has both spatial autocorrelation and spatial stratified heterogeneity, it is advisable to use a spatial sampling model that considers spatial autocorrelation and stratified heterogeneity, such as the MSN model; when it is necessary to estimate the observed target at multiple scales, a two-stage sampling strategy can be used, that is, the sampling samples obtained by the sampling models a) to d) are used as primary sampling units, and secondary sampling is performed within each primary sampling unit to obtain secondary sampling units. f) When the spatial structure characteristics of the observed target change significantly over time, a systematic sampling model can be used to arrange sample points; after the observation data are collected, the estimation method is selected based on whether the data analysis shows spatial autocorrelation and spatial stratification heterogeneity.
4.3.2 Sampling method based on footprint observations
The ground measured values of land quantitative remote sensing products (such as evapotranspiration products, etc.) can be characterized based on footprint observations. The spatial heterogeneity of the product to be tested should be combined with footprint observation instruments (such as eddy correlators). According to the characteristics of the observation instrument, large aperture observation instrument, cosmic ray observation instrument, etc., select the appropriate spatial sampling method (see Appendix B), the specific requirements are as follows: a) When the pixel to be tested is a homogeneous surface, it is advisable to use a systematic sampling model for sample layout; b): When the pixel to be tested is a non-homogeneous surface and does not change much over time, spatial partitioning is performed according to the spatial characteristics of the observation target, and a stratified sampling model is advisable for sample layout: c) When the pixel to be tested is a non-homogeneous surface and changes over time, auxiliary information should be used to combine prior knowledge to determine the location of the representative observation point.
nKaeerKAca
GB/T39468—2020
4.4 Scale conversion method
4.4.1 Scale conversion from sample point observation value to pixel Combine the measured data and variable characteristics of each sample point and select a scale conversion model to estimate the relative true value of the pixel. The scale conversion requirements for different surface types are as follows:
a) Homogeneous surface test pixels: Combined with prior knowledge, if a single sample point can represent the surface covered by the entire pixel, no scale conversion is required; otherwise, a simple random sampling statistical inference model should be used to convert the sample point observation value to the pixel scale: b) Heterogeneous surface test pixels: When the observation target does not have spatial autocorrelation but has spatial stratified heterogeneity, a stratified sampling statistical inference model should be used; when the observation target has spatial autocorrelation but does not have spatial stratified heterogeneity, a Kriging statistical inference model should be used; when the observation target has both spatial autocorrelation and spatial stratified heterogeneity, an MSN statistical inference model should be used.
4.4.2 Footprint observation value to pixel scale conversion For land quantitative remote sensing products using footprint observations, the spatial variation law of the observed variable footprint should be analyzed to select the corresponding footprint observation value to pixel scale conversion method. In the scale conversion process, the scale consistency and time consistency of the ground observation value range and remote sensing product should be ensured. The specific requirements are as follows:
a) When the site to be inspected is at the pixel scale, different scale conversion methods can be selected according to the homogeneity of the surface in the pixel to be inspected:
Pixels to be inspected on homogeneous surfaces: The surface area occupied should be weighted by the footprint range to obtain the relative true value of the pixel scale.
Pixels to be inspected on non-homogeneous surfaces: The observation area should be discretized into grids of appropriate sizes based on the footprint observations and combined with auxiliary information, and then the relative true value of the pixel scale should be obtained based on the statistical model. It is recommended to use the surface-to-point Kriging method for interpolation first, and then use the point-to-surface Kriging method to convert the grid values in the area to be valued to the pixel scale. b) When the site to be inspected is at the regional scale, the physical model can be used to estimate the true value at the regional scale in combination with the physical meaning of the variable and the footprint observations.
5 General method of indirect verification
Indirect verification steps
When the relative true value of the pixel cannot be obtained through ground observation, the authenticity of the land quantitative remote sensing products can be verified based on the indirect verification method and the reference objects. The verification process is shown in Figure 2. Products to be tested and reference objects
Methods
Cross-cross test method
Temporal and spatial change trend analysis test method
Test result evaluation
Figure 2 Indirect test process
rKaeerKAca-
Based on ground observation and high-resolution remote sensing
Multi-scale grid test method
·Accuracy evaluation
, not accurate evaluation
The main steps are as follows:
GB/T39468—2020
a) Analysis of reference object availability: According to the characteristics of the product to be tested, determine the data or products with acceptable reference values for authenticity test of remote sensing products.
b) Selection of inspection methods: According to the characteristics of the product to be inspected and the reference object, select appropriate indirect inspection methods, including cross-inspection method, spatiotemporal change trend analysis inspection method and multi-scale step-by-step inspection method based on ground observation and high-resolution remote sensing data. Inspection result evaluation: Compare the pixel scale reference object value with the product to be inspected value, and evaluate the consistency and uncertainty between the remote sensing product and the reference object from the two aspects of accuracy and uncertainty. C)
5.2 Cross-inspection method
The cross-inspection method is suitable for the simultaneous inspection of multiple quantitative remote sensing products and the direct inspection method cannot be implemented. This method uses the inspected quantitative remote sensing product or the average of multiple inspected remote sensing products as the relative true value to inspect the remote sensing product to be inspected. The cross-inspection method focuses on evaluating the relative accuracy of remote sensing products. The inspection process should comply with the provisions of 8.2.1 of GB/T36296-2018. The general methods involved in the cross-inspection include: a) Cross-inspection product selection method: It is advisable to select the inspected land quantitative remote sensing product of the same type as the product to be inspected as the reference object. b) Spatial, temporal, spectral and angular consistency conversion method: The space, time, spectrum and angle of the product to be inspected should be consistent and comparable with those of the inspected product. If there are differences, the following methods should be used for consistency conversion: Spatial consistency conversion: If there is a difference in the spatial resolution between the product to be inspected and the inspected product, there is a spatial mismatch between the pixels, or the projection coordinate system is inconsistent, the remote sensing product geometric correction method should be used to spatially align the product to be inspected with the inspected product;
Temporal consistency conversion: If there is a difference in time between the product to be inspected and the inspected product, it is advisable to perform temporal consistency conversion based on the time series characteristics of the product to be inspected and the inspected product: Spectral Consistency conversion: If the imaging spectrum of the product to be tested is inconsistent with that of the tested product, it is advisable to use methods such as spectral matching to perform spectral consistency conversion;
-Angle consistency conversion: If the imaging angle of the product to be tested is inconsistent with that of the tested product, it is advisable to use methods such as bidirectional reflectance function to perform angle consistency conversion
c) Inspection sample sampling method: It is advisable to use a simple random sampling method (see B.1) to select samples for verification and comparison 5.3 Spatiotemporal change trend analysis test method
The spatiotemporal change trend analysis test method is applicable to situations where direct inspection and cross-inspection cannot be implemented, but there is prior knowledge of the spatiotemporal change trend of the quantitative remote sensing product to be tested. This method compares the spatiotemporal change trend of the remote sensing product to be tested with the spatiotemporal change trend of the relevant influencing factors to analyze their consistency. The inspection process should comply with the provisions of 8.2.3 of GB/T36296-2018. The general methods for temporal and spatial trend analysis include: a) Temporal and spatial trend analysis method: conduct temporal or spatial trend analysis on the land quantitative remote sensing products and influencing factors to be tested, and obtain their temporal and spatial trend patterns within a certain temporal and spatial range: Temporal trend analysis: analyze the temporal trend patterns of the land quantitative remote sensing products and influencing factors to be tested within a certain time range, such as period, frequency, amplitude, etc.; Spatial trend analysis: analyze the spatial trend patterns of the land quantitative remote sensing products and influencing factors to be tested within a certain spatial range, such as the spatial distribution of correlation coefficients, etc. b) Test sample selection method: it is advisable to use systematic sampling method or simple random sampling method (see Appendix B) to select samples for verification and comparison.
5.4 Multi-scale step-by-step verification method based on ground observation and high-resolution remote sensing data The multi-scale step-by-step verification method based on ground observation and high-resolution remote sensing data is also called the comprehensive verification method. This method uses high-resolution 5
iKaeeiAca
GB/T39468—2020
satellite remote sensing data as the input of the land quantitative remote sensing product estimation model, and performs consistency verification on the model output results and ground observation data, and then converts the model output results into medium and low resolutions through scale conversion, and then verifies the medium and low resolution remote sensing products. The inspection process should comply with the provisions of 8.1 and 8.2.1 in GB/T36296-2018. The general method of multi-scale step-by-step verification based on ground observations and high-resolution remote sensing data includes: a) Generate high-resolution remote sensing products based on the estimation model of land quantitative remote sensing products: take the verified high-resolution remote sensing data with the same or similar scale as the ground observations, and consistent in time and space as the input, and use the verified land quantitative remote sensing product estimation model to generate the corresponding high-resolution remote sensing products; b) Ground sampling method: obtain ground observation data according to the sampling method in direct verification; c) Consistency verification method: verify the estimation results of remote sensing products generated based on high-resolution remote sensing data in combination with ground observation data, determine the model error and provide feedback and correction. Only when the accuracy of the generated high-resolution remote sensing product meets the established threshold requirements can it be used for the next verification; otherwise, re-prepare the high-resolution remote sensing data and verify the generated remote sensing product; d)
scale conversion and relative truth value acquisition method: according to the scale conversion method specified in 4.4, the verified high-resolution remote sensing product is scaled to the same spatial scale as the medium and low resolution remote sensing products to be verified, and the relative truth value of the remote sensing product to be verified is obtained. 6 Evaluation methods
6.1 Evaluation of consistency between remote sensing products and relative true values/reference objects The accuracy evaluation index and uncertainty evaluation index are used to analyze and evaluate the consistency and uncertainty between the land quantitative remote sensing products to be tested and the relative true values/reference objects. For specific evaluation indicators, see Chapter 6 of GB/T36296-2018. 6.2 Evaluation of uncertainty in the authenticity test process of remote sensing products When analyzing the sources of uncertainty in the authenticity test of land quantitative remote sensing products, various factors should be considered, and the impact of various errors on the uncertainty of the results during the test process should be comprehensively evaluated. The ways to reduce and control uncertainty during the authenticity test process should be determined to make the authenticity test results of land quantitative remote sensing products more reliable. The main contents of the analysis and evaluation of uncertainty sources include: a) Uncertainty of geometric positioning: Due to the measurement error of the geometric positioning instrument and the image space registration error, in the test model based on considering the spatial position (such as the Kriging statistical inference model of direct test, the MSN statistical inference model, etc.), the sensitivity of the output results of the quantitative analysis model to the change of spatial position can be measured by indicators such as standard deviation and variance. b) Uncertainty of ground observation: mainly includes the error caused by the measurement performance and measurement method of the measuring instrument and the spatial representative error of the fixed-point observation. It is advisable to evaluate the range of this error through multiple observations. c) Uncertainty of scale conversion: When using statistical inference models for scale conversion, the variance of the estimated value output by the model and the confidence interval of the estimated result should be considered
nKaeerKAca-
A.1 Spatial autocorrelation Moran's I index
Appendix A
(Informative Appendix)
Surface spatial heterogeneity analysis method
A statistical indicator for measuring the autocorrelation of spatial objects. The calculation formula is shown in formula (A.1). n
Where:
w,(rr-r)(r-)
Z(ai-)2
-Moran'sI statistic, with values between -1 and 1; number of regional units;
The cell value of the row and column in the spatial weight matrix W: the value of the interval or ratio variable of regional unit i; the mean value of work
GB/T39468—2020
....(A.1)
The degree of spatial autocorrelation of surface variables can be judged according to the value of 1. The closer the value is to -1, the stronger the negative spatial autocorrelation; the closer the value is to 1, the stronger the positive spatial autocorrelation A.2 Semivariogram
The semivariogram quantifies the assumption that nearby things are more similar than distant things, and measures the strength of statistical correlation as a function of distance. This method groups all the point pairs according to the size and direction of the interval distance. In each group, the difference of the attribute value of each point pair is calculated, and finally the average value is taken as the difference (variance value) of the attribute value of the group. y(h)
is calculated by formula (A.2) where:
spatial interval distance;
2N(h)2
[z()-Z(+h)]
semivariogram value with a distance interval of h;
number of point pairs with a distance interval of h;
spatial position point;
attribute value at spatial position 3,;
Z(, + h) - attribute value at spatial position + h. .A.2)
When calculating, first fit the scattered points calculated by the sample points to obtain the empirical variogram; then observe the distribution map of the variogram (see Figure A.1) and find a theoretical model or a linear combination of multiple theoretical models provided by geostatistics for fitting. The sill value and nugget value in the semivariogram can measure the degree of spatial heterogeneity. The sill value indicates the maximum variation of the regionalized variable. The larger the sill value, the higher the degree of overall spatial heterogeneity: the nugget value indicates the spatial heterogeneity of the random part, and the larger the nugget value indicates the spatial heterogeneity of the random part.
-rKaeerKAca-
GB/T39468—2020
A.3 Coefficient of variation
Quick value
Schematic diagram of semivariogram
Basic value
The ratio w/p and c/u between the range w, mean u and standard deviation of the target variable in the target area can reflect the drastic degree of change in the value of the spatial object. Among them, the range calculation is shown in formula (A.3), the mean calculation is shown in formula (A.4), and the standard deviation calculation is shown in formula (A.5). W=XH-X
Where:
Full range;
The maximum value of all sub-pixel values;
The minimum value of all sub-pixel values;
Average value;
Number of sub-pixels;
Each sub-pixel value:
Standard deviation.
A.4 Geographic detector q index
....(A.3)
...(A.4)
..........(A..5)
The Geographic detector 9 index is a spatial hierarchical heterogeneity detection index. It uses 9 values to compare the consistency of spatial distribution and judge the correlation and explanatory power between two spatial distribution elements. The calculation of the Geographic detector 9 index is shown in formula (A.6). q=l
Where:
..(A.6)
Geographic detector 9 index, the value range is [0,1]. When there is no spatial differentiation of a certain attribute, 9=0; when there is perfect spatial differentiation of a certain attribute, that is, the discrete variance within the layer is zero and the discrete variance between layers is not zero, then 91; total area;
Number of layers;
rKaeerKAca
Layer number;
Area of the hth layer;
Total variance;wwW.bzxz.Net
Variance of the hth layer.
GB/T39468—2020
The degree of heterogeneity of surface variables can be judged according to the size of the q value. The closer the q value is to 1, the higher the spatial heterogeneity: the closer the q value is to 0, the lower the heterogeneity.
rrkaeerkAca1. The general method of multi-scale step-by-step verification based on ground observations and high-resolution remote sensing data includes: a) Generate high-resolution remote sensing products based on the estimation model of land quantitative remote sensing products: take the verified high-resolution remote sensing data with the same or similar scale as the ground observations and consistent in time and space as input, and use the verified land quantitative remote sensing product estimation model to generate the corresponding high-resolution remote sensing products; b) Ground sampling method: obtain ground observation data according to the sampling method in direct verification; c) Consistency verification method: verify the estimation results of remote sensing products generated based on high-resolution remote sensing data in combination with ground observation data, determine the model error and provide feedback and correction. Only when the accuracy of the generated high-resolution remote sensing product meets the established threshold requirements can it be used for the next verification; otherwise, re-prepare the high-resolution remote sensing data and verify the generated remote sensing product; d)
scale conversion and relative truth value acquisition method: according to the scale conversion method specified in 4.4, the verified high-resolution remote sensing product is scaled to the same spatial scale as the medium and low resolution remote sensing products to be verified, and the relative truth value of the remote sensing product to be verified is obtained. 6 Evaluation methods
6.1 Evaluation of consistency between remote sensing products and relative true values/reference objects The accuracy evaluation index and uncertainty evaluation index are used to analyze and evaluate the consistency and uncertainty between the land quantitative remote sensing products to be tested and the relative true values/reference objects. For specific evaluation indicators, see Chapter 6 of GB/T36296-2018. 6.2 Evaluation of uncertainty in the authenticity test process of remote sensing products When analyzing the sources of uncertainty in the authenticity test of land quantitative remote sensing products, various factors should be considered, and the impact of various errors on the uncertainty of the results during the test process should be comprehensively evaluated. The ways to reduce and control uncertainty during the authenticity test process should be determined to make the authenticity test results of land quantitative remote sensing products more reliable. The main contents of the analysis and evaluation of uncertainty sources include: a) Uncertainty of geometric positioning: Due to the measurement error of the geometric positioning instrument and the image space registration error, in the test model based on considering the spatial position (such as the Kriging statistical inference model of direct test, the MSN statistical inference model, etc.), the sensitivity of the output results of the quantitative analysis model to the change of spatial position can be measured by indicators such as standard deviation and variance. b) Uncertainty of ground observation: mainly includes the error caused by the measurement performance and measurement method of the measuring instrument and the spatial representative error of the fixed-point observation. It is advisable to evaluate the range of this error through multiple observations. c) Uncertainty of scale conversion: When using statistical inference models for scale conversion, the variance of the estimated value output by the model and the confidence interval of the estimated result should be considered
nKaeerKAca-
A.1 Spatial autocorrelation Moran's I index
Appendix A
(Informative Appendix)
Surface spatial heterogeneity analysis method
A statistical indicator for measuring the autocorrelation of spatial objects. The calculation formula is shown in formula (A.1). n
Where:
w,(rr-r)(r-)
Z(ai-)2
-Moran'sI statistic, with values between -1 and 1; number of regional units;
The cell value of the row and column in the spatial weight matrix W: the value of the interval or ratio variable of regional unit i; the mean value of work
GB/T39468—2020
....(A.1)
The degree of spatial autocorrelation of surface variables can be judged according to the value of 1. The closer the value is to -1, the stronger the negative spatial autocorrelation; the closer the value is to 1, the stronger the positive spatial autocorrelation A.2 Semivariogram
The semivariogram quantifies the assumption that nearby things are more similar than distant things, and measures the strength of statistical correlation as a function of distance. This method groups all the point pairs according to the size and direction of the interval distance. In each group, the difference of the attribute value of each point pair is calculated, and finally the average value is taken as the difference (variance value) of the attribute value of the group. y(h)
is calculated by formula (A.2) where:
spatial interval distance;
2N(h)2
[z()-Z(+h)]
semivariogram value with a distance interval of h;
number of point pairs with a distance interval of h;
spatial position point;
attribute value at spatial position 3,;
Z(, + h) - attribute value at spatial position + h. .A.2)
When calculating, first fit the scattered points calculated by the sample points to obtain the empirical variogram; then observe the distribution map of the variogram (see Figure A.1) and find a theoretical model or a linear combination of multiple theoretical models provided by geostatistics for fitting. The sill value and nugget value in the semivariogram can measure the degree of spatial heterogeneity. The sill value represents the maximum variation of the regionalized variable. The larger the sill value, the higher the degree of overall spatial heterogeneity: the nugget value represents the spatial heterogeneity of the random part, and the larger the nugget value represents the spatial heterogeneity of the random part.
-rKaeerKAca-
GB/T39468—2020
A.3 Coefficient of variation
Quick value
Schematic diagram of semivariogram
Basic value
The ratio w/p and c/u between the range w, mean u and standard deviation of the target variable in the target area can reflect the drastic degree of change in the value of the spatial object. Among them, the range calculation is shown in formula (A.3), the mean calculation is shown in formula (A.4), and the standard deviation calculation is shown in formula (A.5). W=XH-X
Where:
Full range;
The maximum value of all sub-pixel values;
The minimum value of all sub-pixel values;
Average value;
Number of sub-pixels;
Each sub-pixel value:
Standard deviation.
A.4 Geographic detector q index
....(A.3)
...(A.4)
..........(A..5)
The Geographic detector 9 index is a spatial hierarchical heterogeneity detection index. It uses 9 values to compare the consistency of spatial distribution and judge the correlation and explanatory power between two spatial distribution elements. The calculation of the Geographic detector 9 index is shown in formula (A.6). q=l
Where:
..(A.6)
Geographic detector 9 index, the value range is [0,1]. When there is no spatial differentiation of a certain attribute, 9=0; when there is perfect spatial differentiation of a certain attribute, that is, the discrete variance within the layer is zero and the discrete variance between layers is not zero, then 91; total area;
Number of layers;
rKaeerKAca
Layer number;
Area of the hth layer;
Total variance;
Variance of the hth layer.
GB/T39468—2020
The degree of heterogeneity of surface variables can be judged according to the size of the q value. The closer the q value is to 1, the higher the spatial heterogeneity: the closer the q value is to 0, the lower the heterogeneity.
rrkaeerkAca1. The general method of multi-scale step-by-step verification based on ground observations and high-resolution remote sensing data includes: a) Generate high-resolution remote sensing products based on the estimation model of land quantitative remote sensing products: take the verified high-resolution remote sensing data with the same or similar scale as the ground observations and consistent in time and space as input, and use the verified land quantitative remote sensing product estimation model to generate the corresponding high-resolution remote sensing products; b) Ground sampling method: obtain ground observation data according to the sampling method in direct verification; c) Consistency verification method: verify the estimation results of remote sensing products generated based on high-resolution remote sensing data in combination with ground observation data, determine the model error and provide feedback and correction. Only when the accuracy of the generated high-resolution remote sensing product meets the established threshold requirements can it be used for the next verification; otherwise, re-prepare the high-resolution remote sensing data and verify the generated remote sensing product; d)
scale conversion and relative truth value acquisition method: according to the scale conversion method specified in 4.4, the verified high-resolution remote sensing product is scaled to the same spatial scale as the medium and low resolution remote sensing products to be verified, and the relative truth value of the remote sensing product to be verified is obtained. 6 Evaluation methods
6.1 Evaluation of consistency between remote sensing products and relative true values/reference objects The accuracy evaluation index and uncertainty evaluation index are used to analyze and evaluate the consistency and uncertainty between the land quantitative remote sensing products to be tested and the relative true values/reference objects. For specific evaluation indicators, see Chapter 6 of GB/T36296-2018. 6.2 Evaluation of uncertainty in the authenticity test process of remote sensing products When analyzing the sources of uncertainty in the authenticity test of land quantitative remote sensing products, various factors should be considered, and the impact of various errors on the uncertainty of the results during the test process should be comprehensively evaluated. The ways to reduce and control uncertainty during the authenticity test process should be determined to make the authenticity test results of land quantitative remote sensing products more reliable. The main contents of the analysis and evaluation of uncertainty sources include: a) Uncertainty of geometric positioning: Due to the measurement error of the geometric positioning instrument and the image space registration error, in the test model based on considering the spatial position (such as the Kriging statistical inference model of direct test, the MSN statistical inference model, etc.), the sensitivity of the output results of the quantitative analysis model to the change of spatial position can be measured by indicators such as standard deviation and variance. b) Uncertainty of ground observation: mainly includes the error caused by the measurement performance and measurement method of the measuring instrument and the spatial representative error of the fixed-point observation. It is advisable to evaluate the range of this error through multiple observations. c) Uncertainty of scale conversion: When using statistical inference models for scale conversion, the variance of the estimated value output by the model and the confidence interval of the estimated result should be considered
nKaeerKAca-
A.1 Spatial autocorrelation Moran's I index
Appendix A
(Informative Appendix)
Surface spatial heterogeneity analysis method
A statistical indicator for measuring the autocorrelation of spatial objects. The calculation formula is shown in formula (A.1). n
Where:
w,(rr-r)(r-)
Z(ai-)2
-Moran'sI statistic, with values between -1 and 1; number of regional units;
The cell value of the row and column in the spatial weight matrix W: the value of the interval or ratio variable of regional unit i; the mean value of work
GB/T39468—2020
....(A.1)
The degree of spatial autocorrelation of surface variables can be judged according to the value of 1. The closer the value is to -1, the stronger the negative spatial autocorrelation; the closer the value is to 1, the stronger the positive spatial autocorrelation A.2 Semivariogram
The semivariogram quantifies the assumption that nearby things are more similar than distant things, and measures the strength of statistical correlation as a function of distance. This method groups all the point pairs according to the size and direction of the interval distance. In each group, the difference of the attribute value of each point pair is calculated, and finally the average value is taken as the difference (variance value) of the attribute value of the group. y(h)
is calculated by formula (A.2) where:
spatial interval distance;
2N(h)2
[z()-Z(+h)]
semivariogram value with a distance interval of h;
number of point pairs with a distance interval of h;
spatial position point;
attribute value at spatial position 3,;
Z(, + h) - attribute value at spatial position + h. .A.2)
When calculating, first fit the scattered points calculated by the sample points to obtain the empirical variogram; then observe the distribution map of the variogram (see Figure A.1) and find a theoretical model or a linear combination of multiple theoretical models provided by geostatistics for fitting. The sill value and nugget value in the semivariogram can measure the degree of spatial heterogeneity. The sill value represents the maximum variation of the regionalized variable. The larger the sill value, the higher the degree of overall spatial heterogeneity: the nugget value represents the spatial heterogeneity of the random part, and the larger the nugget value represents the spatial heterogeneity of the random part.
-rKaeerKAca-
GB/T39468—2020
A.3 Coefficient of variation
Quick value
Schematic diagram of semivariogram
Basic value
The ratio w/p and c/u between the range w, mean u and standard deviation of the target variable in the target area can reflect the drastic degree of change in the value of the spatial object. Among them, the range calculation is shown in formula (A.3), the mean calculation is shown in formula (A.4), and the standard deviation calculation is shown in formula (A.5). W=XH-X
Where:
Full range;
The maximum value of all sub-pixel values;
The minimum value of all sub-pixel values;
Average value;
Number of sub-pixels;
Each sub-pixel value:
Standard deviation.
A.4 Geographic detector q index
....(A.3)
...(A.4)
..........(A..5)
The Geographic detector 9 index is a spatial hierarchical heterogeneity detection index. It uses 9 values to compare the consistency of spatial distribution and judge the correlation and explanatory power between two spatial distribution elements. The calculation of the Geographic detector 9 index is shown in formula (A.6). q=l
Where:
..(A.6)
Geographic detector 9 index, the value range is [0,1]. When there is no spatial differentiation of a certain attribute, 9=0; when there is perfect spatial differentiation of a certain attribute, that is, the discrete variance within the layer is zero and the discrete variance between layers is not zero, then 91; total area;
Number of layers;
rKaeerKAca
Layer number;
Area of the hth layer;
Total variance;
Variance of the hth layer.
GB/T39468—2020
The degree of heterogeneity of surface variables can be judged according to the size of the q value. The closer the q value is to 1, the higher the spatial heterogeneity: the closer the q value is to 0, the lower the heterogeneity.
rrkaeerkAca2 Uncertainty evaluation of the authenticity verification process of remote sensing products When analyzing the sources of uncertainty in the authenticity verification of land quantitative remote sensing products, various factors should be considered, the impact of various errors on the uncertainty of the results during the verification process should be comprehensively evaluated, and the ways to reduce and control uncertainty during the authenticity verification process should be determined to make the authenticity verification results of land quantitative remote sensing products more reliable. The main contents of the analysis and evaluation of uncertainty sources include: a) Uncertainty of geometric positioning: Due to the measurement error of the geometric positioning instrument and the image space registration error, in the verification model based on considering the spatial position (such as the Kriging statistical inference model of direct verification, the MSN statistical inference model, etc.), the sensitivity of the output results of the quantitative analysis model to the change of spatial position can be measured by indicators such as standard deviation and variance. b) Uncertainty of ground observation: It mainly includes the errors caused by the measurement performance and measurement methods of the measuring instrument and the spatial representativeness error of the fixed-point observation. It is advisable to evaluate the range of this error through multiple observations. c) Uncertainty of scale conversion: When using statistical inference models for scale conversion, the variance of the estimated value output by the model and the confidence interval of the estimated result should be considered
nKaeerKAca-
A.1 Spatial autocorrelation Moran's I index
Appendix A
(Informative Appendix)
Surface spatial heterogeneity analysis method
A statistical indicator for measuring the autocorrelation of spatial objects. The calculation formula is shown in formula (A.1). n
Where:
w,(rr-r)(r-)
Z(ai-)2
-Moran'sI statistic, with values between -1 and 1; number of regional units;
The cell value of the row and column in the spatial weight matrix W: the value of the interval or ratio variable of regional unit i; the mean value of work
GB/T39468—2020
....(A.1)
The degree of spatial autocorrelation of surface variables can be judged according to the value of 1. The closer the value is to -1, the stronger the negative spatial autocorrelation; the closer the value is to 1, the stronger the positive spatial autocorrelation A.2 Semivariogram
The semivariogram quantifies the assumption that nearby things are more similar than distant things, and measures the strength of statistical correlation as a function of distance. This method groups all the point pairs according to the size and direction of the interval distance. In each group, the difference of the attribute value of each point pair is calculated, and finally the average value is taken as the difference (variance value) of the attribute value of the group. y(h)
is calculated by formula (A.2) where:
spatial interval distance;
2N(h)2
[z()-Z(+h)]
semivariogram value with a distance interval of h;
number of point pairs with a distance interval of h;
spatial position point;
attribute value at spatial position 3,;
Z(, + h) - attribute value at spatial position + h. .A.2)
When calculating, first fit the scattered points calculated by the sample points to obtain the empirical variogram; then observe the distribution map of the variogram (see Figure A.1) and find a theoretical model or a linear combination of multiple theoretical models provided by geostatistics for fitting. The sill value and nugget value in the semivariogram can measure the degree of spatial heterogeneity. The sill value represents the maximum variation of the regionalized variable. The larger the sill value, the higher the degree of overall spatial heterogeneity: the nugget value represents the spatial heterogeneity of the random part, and the larger the nugget value represents the spatial heterogeneity of the random part.
-rKaeerKAca-
GB/T39468—2020
A.3 Coefficient of variation
Quick value
Schematic diagram of semivariogram
Basic value
The ratio w/p and c/u between the range w, mean u and standard deviation of the target variable in the target area can reflect the drastic degree of change in the value of the spatial object. Among them, the range calculation is shown in formula (A.3), the mean calculation is shown in formula (A.4), and the standard deviation calculation is shown in formula (A.5). W=XH-X
Where:
Full range;
The maximum value of all sub-pixel values;
The minimum value of all sub-pixel values;
Average value;
Number of sub-pixels;
Each sub-pixel value:
Standard deviation.
A.4 Geographic detector q index
....(A.3)
...(A.4)
..........(A..5)
The Geographic detector 9 index is a spatial hierarchical heterogeneity detection index. It uses 9 values to compare the consistency of spatial distribution and judge the correlation and explanatory power between two spatial distribution elements. The calculation of the Geographic detector 9 index is shown in formula (A.6). q=l
Where:
..(A.6)
Geographic detector 9 index, the value range is [0,1]. When there is no spatial differentiation of a certain attribute, 9=0; when there is perfect spatial differentiation of a certain attribute, that is, the discrete variance within the layer is zero and the discrete variance between layers is not zero, then 91; total area;
Number of layers;
rKaeerKAca
Layer number;
Area of the hth layer;
Total variance;
Variance of the hth layer.
GB/T39468—2020
The degree of heterogeneity of surface variables can be judged according to the size of the q value. The closer the q value is to 1, the higher the spatial heterogeneity: the closer the q value is to 0, the lower the heterogeneity.
rrkaeerkAca2 Uncertainty evaluation of the authenticity verification process of remote sensing products When analyzing the sources of uncertainty in the authenticity verification of land quantitative remote sensing products, various factors should be considered, the impact of various errors on the uncertainty of the results during the verification process should be comprehensively evaluated, and the ways to reduce and control uncertainty during the authenticity verification process should be determined to make the authenticity verification results of land quantitative remote sensing products more reliable. The main contents of the analysis and evaluation of uncertainty sources include: a) Uncertainty of geometric positioning: Due to the measurement error of the geometric positioning instrument and the image space registration error, in the verification model based on considering the spatial position (such as the Kriging statistical inference model of direct verification, the MSN statistical inference model, etc.), the sensitivity of the output results of the quantitative analysis model to the change of spatial position can be measured by indicators such as standard deviation and variance. b) Uncertainty of ground observation: It mainly includes the errors caused by the measurement performance and measurement methods of the measuring instrument and the spatial representativeness error of the fixed-point observation. It is advisable to evaluate the range of this error through multiple observations. c) Uncertainty of scale conversion: When using statistical inference models for scale conversion, the variance of the estimated value output by the model and the confidence interval of the estimated result should be considered
nKaeerKAca-
A.1 Spatial autocorrelation Moran's I index
Appendix A
(Informative Appendix)
Surface spatial heterogeneity analysis method
A statistical indicator for measuring the autocorrelation of spatial objects. The calculation formula is shown in formula (A.1). n
Where:
w,(rr-r)(r-)
Z(ai-)2
-Moran'sI statistic, with values between -1 and 1; number of regional units;
The cell value of the row and column in the spatial weight matrix W: the value of the interval or ratio variable of regional unit i; the mean value of work
GB/T39468—2020
....(A.1)
The degree of spatial autocorrelation of surface variables can be judged according to the value of 1. The closer the value is to -1, the stronger the negative spatial autocorrelation; the closer the value is to 1, the stronger the positive spatial autocorrelation A.2 Semivariogram
The semivariogram quantifies the assumption that nearby things are more similar than distant things, and measures the strength of statistical correlation as a function of distance. This method groups all the point pairs according to the size and direction of the interval distance. In each group, the difference of the attribute value of each point pair is calculated, and finally the average value is taken as the difference (variance value) of the attribute value of the group. y(h)
is calculated by formula (A.2) where:
spatial interval distance;
2N(h)2
[z()-Z(+h)]
semivariogram value with a distance interval of h;
number of point pairs with a distance interval of h;
spatial position point;
attribute value at spatial position 3,;
Z(, + h) - attribute value at spatial position + h. .A.2)
When calculating, first fit the scattered points calculated by the sample points to obtain the empirical variogram; then observe the distribution map of the variogram (see Figure A.1) and find a theoretical model or a linear combination of multiple theoretical models provided by geostatistics for fitting. The sill value and nugget value in the semivariogram can measure the degree of spatial heterogeneity. The sill value indicates the maximum variation of the regionalized variable. The larger the sill value, the higher the degree of overall spatial heterogeneity: the nugget value indicates the spatial heterogeneity of the random part, and the larger the nugget value indicates the spatial heterogeneity of the random part.
-rKaeerKAca-
GB/T39468—2020
A.3 Coefficient of variation
Quick value
Schematic diagram of semivariogram
Basic value
The ratio w/p and c/u between the range w, mean u and standard deviation of the target variable in the target area can reflect the drastic degree of change in the value of the spatial object. Among them, the range calculation is shown in formula (A.3), the mean calculation is shown in formula (A.4), and the standard deviation calculation is shown in formula (A.5). W=XH-X
Where:
Full range;
The maximum value of all sub-pixel values;
The minimum value of all sub-pixel values;
Average value;
Number of sub-pixels;
Each sub-pixel value:
Standard deviation.
A.4 Geographic detector q index
....(A.3)
...(A.4)
..........(A..5)
The Geographic detector 9 index is a spatial hierarchical heterogeneity detection index. It uses 9 values to compare the consistency of spatial distribution and judge the correlation and explanatory power between two spatial distribution elements. The calculation of the Geographic detector 9 index is shown in formula (A.6). q=l
Where:
..(A.6)
Geographic detector 9 index, the value range is [0,1]. When there is no spatial differentiation of a certain attribute, 9=0; when there is perfect spatial differentiation of a certain attribute, that is, the discrete variance within the layer is zero and the discrete variance between layers is not zero, then 91; total area;
Number of layers;
rKaeerKAca
Layer number;
Area of the hth layer;
Total variance;
Variance of the hth layer.
GB/T39468—2020
The degree of heterogeneity of surface variables can be judged according to the size of the q value. The closer the q value is to 1, the higher the spatial heterogeneity: the closer the q value is to 0, the lower the heterogeneity.
rrkaeerkAca3), the average value is calculated as shown in formula (A.4). The standard deviation is calculated as shown in formula (A.5). W=XH-X
Where:
Full range;
The maximum value of all sub-pixel values;
The minimum value of all sub-pixel values;
Average value;
Number of sub-pixels;
Each sub-pixel value:
Standard deviation.
A.4 Geographic detector q index
....(A.3)
...(A.4)
..........(A..5)
The Geographic detector 9 index is a spatial hierarchical heterogeneity detection index, which uses 9 values to compare the consistency of spatial distribution and judge the correlation and explanatory power between two spatial distribution elements. The calculation of the Geographic detector 9 index is shown in formula (A.6). q=l
Where:
..(A.6)
Geographic detector 9 index, value range [0,1]. When there is no spatial differentiation of a certain attribute, 9=0; when there is perfect spatial differentiation of a certain attribute, that is, the discrete variance within the layer is zero and the discrete variance between layers is not zero, then 91; overall area;
Number of layers;
rKaeerKAca
Layer number;
Area of the hth layer;
Overall variance;
Variance of the hth layer.
GB/T39468—2020
The degree of heterogeneity of surface variables can be judged according to the size of the q value. The closer the q value is to 1, the higher the spatial heterogeneity is: the closer the q value is to 0, the lower the heterogeneity is.
rrkaeerkAca3), the average value is calculated as shown in formula (A.4). The standard deviation is calculated as shown in formula (A.5). W=XH-X
Where:
Full range;
The maximum value of all sub-pixel values;
The minimum value of all sub-pixel values;
Average value;
Number of sub-pixels;
Each sub-pixel value:
Standard deviation.
A.4 Geographic detector q index
....(A.3)
...(A.4)
..........(A..5)
The Geographic detector 9 index is a spatial hierarchical heterogeneity detection index, which uses 9 values to compare the consistency of spatial distribution and judge the correlation and explanatory power between two spatial distribution elements. The calculation of the Geographic detector 9 index is shown in formula (A.6). q=l
Where:
..(A.6)
Geographic detector 9 index, value range [0,1]. When there is no spatial differentiation of a certain attribute, 9=0; when there is perfect spatial differentiation of a certain attribute, that is, the discrete variance within the layer is zero and the discrete variance between layers is not zero, then 91; overall area;
Number of layers;
rKaeerKAca
Layer number;
Area of the hth layer;
Overall variance;
Variance of the hth layer.
GB/T39468—2020
The degree of heterogeneity of surface variables can be judged according to the size of the q value. The closer the q value is to 1, the higher the spatial heterogeneity is: the closer the q value is to 0, the lower the heterogeneity is.
rrkaeerkAca
Tip: This standard content only shows part of the intercepted content of the complete standard. If you need the complete standard, please go to the top to download the complete standard document for free.