CN108230310B

CN108230310B - A method for extracting non-fire spatiotemporal data based on semivariogram

Info

Publication number: CN108230310B
Application number: CN201810003693.8A
Authority: CN
Inventors: 何彬彬; 文崇波; 全兴文
Original assignee: University of Electronic Science and Technology of China
Current assignee: University of Electronic Science and Technology of China
Priority date: 2018-01-03
Filing date: 2018-01-03
Publication date: 2021-12-17
Anticipated expiration: 2038-01-03
Also published as: CN108230310A

Abstract

A method for extracting non-fire spatiotemporal data based on semivariogram. The present invention first determines the positions and occurrence times of all fire pixels in the burning area data, judges the vegetation types of all fire pixels, and then extracts the data of all fire pixels. And the surrounding non-fire pixel data with the same vegetation type as the fire pixel, use the semivariogram to analyze the spatial correlation, and use the range as the buffer distance to establish a corresponding fire buffer, and then extract the correlation time of the fire pixel. The process of establishing fire buffers is repeated, and finally the data of pixels that do not fall into all buffers are selected as non-fire data. The invention can not only eliminate the subjective influence caused by human factors, but also eliminate the differences caused by factors such as geographical location, vegetation type, etc., and also eliminate the accidental factors of fire occurrence, and improve the contrast between non-fire data and fire data. It is beneficial to improve the accuracy of the simulation results of the training model.

Description

Method for extracting non-fire spatio-temporal data based on semi-variogram

Technical Field

The invention particularly relates to a method for extracting non-fire spatio-temporal data based on a half-variogram.

Background

The forest fire is a very common and destructive natural disaster, the forest fire happens more than 20 ten thousand times every year in the world on average, and the burning area of the forest occupies more than 1 per thousand of the total area of the forest in the world. The forest fire happens more than 1 ten thousand times per year in China, hundreds of thousands of forests to millions of hectares are burnt, and the forest area occupies 5-8 per thousand of the whole country. Forest fires not only burn trees to directly reduce the forest area, but also seriously destroy the forest structure and the forest environment, so that the forest ecological system is out of balance, the forest biomass is reduced, and even people and livestock are killed.

The forest fire danger level prediction method is very meaningful and necessary work, can realize the forest fire danger level prediction with large range and high space-time resolution by a remote sensing technology, and has certain scientific guiding significance for the allocation of forest fire protection resources and the personnel movement. At present, the research on fire risk level prediction models is relatively deep, and the training models generally need to input climate parameters, terrain parameters and vegetation parameters, wherein the vegetation parameters can reflect the physical characteristics (such as type, coverage and the like) of vegetation. The training model requires comparison data, i.e. the input data to the model includes not only the data of a fire (fire data) but also the data of a non-fire (non-fire data). The quality of the input data has a significant impact on the final simulation effect of the training model.

At present, a method for acquiring fire data and non-fire data based on a remote sensing technology mainly comprises the following steps: firstly, carrying out fire burning area mapping according to a fire point extraction algorithm or directly utilizing related fire products, and then extracting data of fire pixel according to a mapping result or the fire products, thereby acquiring fire data; the obtained non-fire data usually takes a fire pixel as a circle center, the buffer radius is determined according to experience to further establish a buffer area, and data of the non-fire pixel is randomly extracted from an area outside the buffer area to serve as comparison data of the fire pixel. Although the method for extracting the non-fire data is simple in operation, the method has the following problems: firstly, the determination of the fire pixel buffer radius is divided according to experience, subjective influence caused by human factors exists, meanwhile, the size of the buffer radius changes along with the difference of a research area, a vegetation type and the like, that is, the buffer radius of each fire pixel is different, if the buffer radius is too small, space correlation exists between extracted non-fire data and corresponding fire data, so that the contrast is lacked, the simulation effect of a training model is influenced, and the prediction purpose cannot be well achieved; if the buffer radius is set too large, the area for non-fire data extraction will be too small. Secondly, the extraction method of non-fire data only considers spatial correlation and does not consider temporal correlation, such as: a certain picture element has no fire in a certain time phase, but a fire in the next time phase occurs, which indicates that the data of the picture element is close to the fire data, so that the non-fire data extracted by the method still lacks the contrast. In summary, how to accurately, conveniently and efficiently extract non-fire data becomes a technical problem to be solved urgently in the field.

Disclosure of Invention

In view of the above, the present invention aims to: aiming at the problems of poor contrast and the like of the existing non-fire data extraction method, the invention provides the method for extracting the non-fire spatio-temporal data by combining the fitting half-variation function value of the spherical model and the multi-temporal characteristics of the remote sensing image.

The technical scheme provided by the invention is as follows:

a method for extracting non-fire data based on a semi-variogram is characterized by comprising the following steps:

step 1: determining the position and the occurrence moment of a fire pixel in the combustion area data;

step 2: determining the vegetation type of the fire pixel;

and step 3: extracting data of fire pixels and surrounding non-fire pixel data with the same type as the vegetation type of the fire pixels;

and 4, step 4: performing spatial correlation analysis on the fire pixels by using a half-variation function, taking a variation range as a buffer distance of the fire pixels and establishing a corresponding fire buffer area;

and 5: extracting data of a relevant time phase of a fire pixel;

step 6: repeating the step 4 to establish a fire buffer zone of a time phase related to the fire pixel;

and 7: and (4) selecting pixels which do not fall into the fire buffer areas obtained in the step (4) and the step (6) and extracting data of the pixels to serve as non-fire data.

Furthermore, the variable range is obtained by fitting a spherical model to the half-variation function value.

Furthermore, the buffer area is specifically established by taking the fire pixel as the center of a circle and taking the variable range as the radius to form the buffer area.

Furthermore, the relevant time phases of the fire pixel in the invention are specifically as follows: and taking a single fire pixel as a research object, and taking the moment of fire occurrence of the pixel as a target time phase, wherein the adjacent time phases of the target time phase and the time phases corresponding to the target time phase in the rest years are related time phases.

Further, the combustion area data is specifically remote sensing image data.

The inventive concept is explained in detail below: in order to consider the spatial correlation between the fire pixel and the non-fire pixel, a buffer area is often required to be established for the fire pixel. And a plurality of fire pixels are often found in the fire distribution remote sensing image, so that buffer areas of all the fire pixels need to be obtained, and the comparison data of the fire pixels cannot fall into any one buffer area. Because the vegetation types have difference, the invention introduces the vegetation type factor in the process of establishing the buffer zone, and utilizes the spherical model to fit the variation trend of the half variation function value, takes the variation as the threshold distance of the spatial correlation, and establishes the buffer zone of the fire pixel; meanwhile, the time correlation of the fire pixel at different time phases is considered, and the extraction of non-fire pixel data close to the fire pixel data is avoided, so that the difference between the fire data and the non-fire data is improved, and the aim of improving the simulation precision of the fire risk grade prediction model is fulfilled.

Compared with the prior art, the invention has the beneficial effects that:

according to the invention, vegetation type factors are introduced, and the variation range of a half variation function is used as a buffer radius to establish a buffer area of a fire pixel, so that not only are subjective influences caused by human factors eliminated, but also differences caused by different geographic positions, vegetation types and the like are eliminated; meanwhile, the time relativity of the fire pixel at different time phases is considered, so that non-fire data with insufficient difference with fire data are eliminated; therefore, the method and the device can enhance the contrast between fire data and non-fire data, and further improve the simulation precision of the fire risk grade prediction model.

Drawings

FIG. 1 is a schematic view of a fire distribution image and fire pixel buffer radius in Michelle county, Yunnan province.

FIG. 2 is a graph showing the variation of the half-value of the variation function of a Normalized Difference Vegetation Index (NDVI).

FIG. 3 is a graph showing the variation of the half-value of the variation function of the water Content (FMC) of combustible.

FIG. 4 is a diagram illustrating the superposition of buffers for different index parameters.

FIG. 5 is a schematic diagram of the superposition of buffers for different fire pixels.

Fig. 6 is a diagram of a multi-temporal buffer overlay.

Detailed Description

The principles and features of the present invention will be further explained with reference to the drawings and specific embodiments of the specification:

the occurrence and development of fire are very complex processes, and many influencing factors are involved, so that different representative fire risk evaluation indexes are selected according to different time, space scales and research purposes in fire risk grade evaluation. At present, the fire risk indexes commonly used can be roughly divided into the aspects of terrain conditions, weather environment, vegetation factors and the like, wherein the vegetation factors are one of the main influence factors influencing the fire maturity of vegetation.

The invention selects two important factors, namely combustible Moisture Content (FMC) and Normalized Difference Vegetation Index (NDVI), as training data of a fire risk grade prediction model. The FMC is a ratio obtained by dividing the difference between the wet weight and the dry weight of the plants in the sampling unit by the dry weight, reflects the water content of the leaves of the unit vegetation, directly influences the fire risk grade, is an important fire induction factor, and can be obtained through inversion of a vegetation radiation transmission model in the prior art. The NDVI is a common vegetation index, can reflect the growth condition and the vegetation coverage of vegetation, and can be obtained by the first and second wave band calculation of a remote sensing satellite reflectivity product (MOD09A 1). Embodiments of the present invention will be described in detail below with reference to these two vegetation parameters as examples:

example (b):

a method for extracting non-fire data based on a semi-variogram comprises the following steps:

the method comprises the following steps: preparing data;

this example is detailed in 1 month 2016 in Yunnan province, Maitreya:

the present embodiment provides a fire distribution image using a remote sensing satellite product (MCD64a1) that records the time and place of each month the fire occurred; because of the difference between the vegetation types, different vegetation types should be treated separately, and the vegetation type of each fire pixel is judged by using a land cover product (MCD12Q1) provided by a remote sensing satellite.

Step two: spatial decorrelation;

statistics on MCD64A1 fire distribution products shows that 46 pixels in Mueller county 1 month in 2016 have fires, and in the embodiment, a fire buffer area is established by calculating the buffer radius of each pixel according to FMC and NDVI. Referring to fig. 1, the left image is a fire distribution image in the mile county of Yunnan province, and the right image is a schematic buffer radius diagram of a certain fire pixel therein. Dark squares in the circles indicate fire pixels, light squares in the circles and light squares outside the circles indicate pixels without fire, and areas occupied by the circles indicate buffer areas. The pixels within the circle are within the buffer area and therefore cannot be selected as non-fire pixels, but only pixels outside the buffer area are selected as non-fire pixels. The operation of creating a buffer for only one of the fire pixels will be described in detail below, and the processing for the remaining fire pixels is the same.

The invention mainly carries out spatial decorrelation processing based on a half-mutation function:

the semivariation function is also called semivariance function, is a key function for researching soil variability in geostatistics, and is a continuous function for describing spatial continuous variation of soil properties, and can reflect the variation between different distance observed values of soil properties, and the basic formula of the semivariance function is shown as the following formula (1).

Wherein R is_(h)Is a half-variation function, N is the number of pixels at h from the fire point, Z_(x)Is the pixel value, Z, of the fire pixel_(xi+h)Is the pixel value at the pixel h away from the fire point.

The present embodiment uses a spherical model to fit half-value variation function values. The spherical model is also referred to as a Materon model. It is known in the art that the distance at which a model first assumes a horizontal state is a variable distance, and a sample position separated by a distance closer than the variable distance is spatially autocorrelation, while a sample position farther than the variable distance is not spatially autocorrelation, and the basic formula is shown in equation (2).

Wherein, Y_hIs an independent variable, h is a dependent variable, C₀C, a are model parameters, and parameter a represents a variation.

Calculating half variation function values of FMC and NDVI with distance according to formula (1), wherein the half variation function values of NDVI with distance are shown in FIG. 2, and the half variation function values of FMC with distance are shown in FIG. 3; then, the model parameter a (namely the variation) is obtained by fitting the formula (2), so that the variation of the NDVI is 33 pixel distances, and the variation of the FMC is 35 pixel distances. At this time, the fire pixel has two buffer radiuses, and a larger buffer radius is selected to establish a fire buffer area of the fire pixel, so that a superposition schematic diagram of the fire buffer area shown in fig. 4 is obtained, wherein for the same fire pixel, the inner circle is an NDVI buffer area, and the outer circle is an FMC buffer area. Similarly, the above processing procedure is repeated for the rest fire pixels, and then the corresponding fire buffer areas are established, so that the fire buffer area superposition diagram shown in fig. 5 can be obtained, and the selection of the non-fire pixels cannot fall into any one of the buffer areas.

Step three: time decorrelation;

in practical situations, there are strong contingencies in the occurrence or non-occurrence of a fire. For a picture element, neither a fire has occurred in that phase nor is it within the buffer of any fire picture element in that phase, but its data indicates that the fire risk level for that picture element is high, most likely because a fire has occurred in its adjacent phase or the corresponding phase of another year. If the fire occurrence condition of the picture element is not considered in the adjacent time phase and the corresponding time phase of other years, the picture element is possibly selected as a non-fire picture element, so that the difference between fire and non-fire data is reduced.

Based on the multi-temporal characteristics of remote sensing images, with months as a time scale, 2016 (year 1) as a target time phase, 2015 (year 12), 2016 (year 2) and 2001-2017 (MCD64A1 begins at year 11) and yearly 1 as related time phases, extracting fire distribution images of the related time phases, and then repeating the operation of the step two to establish a fire buffer area of a fire pixel in each related time phase. As shown in fig. 6, the two-step fire buffer and the three-step fire buffer are superimposed to obtain an image, where (a) is a schematic diagram of the buffer at the target time phase, (b) is a schematic diagram of the buffer at the previous time phase (i.e., 12 months 2015), fig. (c) is a schematic diagram of the buffer at the next time phase (i.e., 2016 months 2), fig. (d) is a schematic diagram of the buffer at month 2001, fig. (e) is a schematic diagram of the buffer at month 2002, fig. (f) is a schematic diagram of the buffer at month n, and fig. (g) is a schematic diagram of the superimposition of the target time phase and all the buffers at the time phases related to the target time phase.

Step four: extracting non-fire data;

in fig. 6, 46 pixels which do not fall into the fire buffer are randomly selected as the comparison data of the fire data to ensure the extraction quality of the non-fire data.

While the present invention has been described with reference to the embodiments illustrated in the drawings, the present invention is not limited to the embodiments, which are illustrative rather than restrictive, and it will be apparent to those skilled in the art that many more modifications and variations can be made without departing from the spirit of the invention and the scope of the appended claims.

Claims

1. A method for extracting non-fire data based on a semi-variogram is characterized by comprising the following steps:

step 2: determining the vegetation type of the fire pixel;

and 4, step 4: performing spatial correlation analysis on the fire pixels by using a half-variation function, taking a variation range as a buffer distance of the fire pixels and establishing a corresponding fire buffer area; the specific method comprises the following steps: acquiring a variable range by adopting a spherical model to fit a half-variation function value, and calculating a radius of each fire pixel according to FMC and NDVI by taking the fire pixel as a circle center to establish a fire buffer area;

and 5: extracting data of a relevant time phase of a fire pixel;

2. The method for extracting non-fire data based on the semi-variogram according to claim 1, wherein the combustion region data in step 1 is specifically remote sensing image data.

3. The method for extracting non-fire data based on the semi-variogram according to claim 1 or 2, wherein the relevant time phases of the fire pixels in step 6 are: a single fire pixel is used as a research object, the moment of fire of the pixel is used as a target time phase, and the adjacent time phases of the target time phase and the time phases corresponding to the target time phase in the rest years are related time phases.