CN108876917A - A kind of forest ground biomass remote sensing estimation universal model construction method - Google Patents

Abstract

The invention discloses a method for constructing a general model for forest aboveground biomass remote sensing estimation, using Landsat 1987-2011 remote sensing data and 6-period measured data as data sources, first preprocessing the remote sensing data to obtain normalized data Set, calculate the remote sensing factors, biomass values and change values of the sample plot, construct the static and dynamic models of forest aboveground biomass through multiple regression, linear simultaneous equations, geographically weighted regression, and linear models, and determine the optimal model through verification As a general model of forest aboveground biomass, the parametric form of the model is given. The invention fully excavates the relationship between the change of remote sensing factors and the change of stand biomass, and reduces the uncertainty caused by single-period remote sensing data estimation; the aboveground biomass model of the forest constructed by the linear mixed model greatly improves the The accuracy of simulation and prediction is improved; the general model constructed can perform real-time and time-to-place estimation of forest biomass.

Description

A general model construction method for forest aboveground biomass remote sensing estimation

technical field

The invention relates to the field of forest aboveground biomass estimation, in particular to a method for constructing a general model for forest aboveground biomass estimation based on time series remote sensing data.

Background technique

In forest biomass measurement, traditional measurement methods and remote sensing monitoring methods are included. The ever-growing remote sensing technology has the advantages of being fast, accurate, non-destructive to forests, and capable of long-term, dynamic, and continuous macroscopic monitoring, making remote sensing monitoring methods the main way to obtain forest aboveground biomass. Because the parametric model established by the existing remote sensing estimation means relies on field survey data, the fitting and prediction accuracy is low, and the calculation is more complicated. Although the non-parametric model has high accuracy, the model is poor in portability, which leads to difficulties in popularization and application. .

Landsat remote sensing data was released free of charge in 2007. It covers the whole world and has the longest archive time. Although the accuracy of Lidar data estimation of aboveground biomass in forests will be improved, the cost is relatively high. A general model refers to a method or technology that can be generally applied in different environments and conditions. The general model for biomass remote sensing estimation refers to the unified mathematical function form of the model, and the model is suitable for multi-source remote sensing data, data of different time phases, different forest types, and different geographical environments.

However, there are few reports on the general model of biomass estimation using remote sensing. The research on the general model established by Landsat includes: 1) Townsend et al. published "A general Landsat model" in Volume 119 of "Remote Sensing of Environment" in 2012. to predict canopy defoliation in broadleafdeciduous forests", based on the difference of Landsat vegetation index values during non-defoliation and defoliation periods, a general model of canopy defoliation in deciduous forests was constructed, and the final model was in the form of an exponential function including vegetation indices. 2) Carmona et al. published "Development of a general model to estimate the instantaneous, daily, and daytime net radiation with satellite data on clear-sky days" in Volume 171 of "Remote Sensing of Environment" in 2015. The study used Landsat data to establish A general model for real-time estimation of net radiation. 3) Roy et al. published "Ageneral method to normalize Landsat reflectance data to nadir BRDF adjusted reflectance" in Volume 176 of "Remote Sensing of Environment" in 2016, and developed a method to convert Landsat reflectance data into adjusted BRDF Generic method for reflectivity. None of the above general models involve the estimation of forest biomass. In addition, the general models of forest biomass that have been established are basically based on tree measurement factors, and few are based on remote sensing data. For example, West et al published "A general model for the origin of allometric scaling" in "Science" in 1997. lawsin biology" and "A general model for the structure and allometry of plant vascular systems" were published in "Nature" in 1999, deriving a general model for the relationship between tree biomass and diameter. In China, Zeng Weisheng and others published "A New Universal Relative Growth Biomass Model" in "Forestry Science" in 2012, deriving a new general model for the relationship between tree biomass and diameter. Judging from the published related invention patents, the patent "A Method for Remote Sensing Forest Biomass Retrieval (CN201510056042)" uses LiDAR point cloud data with high cost, and the more commonly used stepwise regression model is used for model construction , the present invention has compared stepwise regression and linear mixed models, and found that the simulation accuracy of the latter has been significantly improved; Multi-scale estimation method of forest biomass based on weibull distribution (CN201510170607)", "Forest biomass inversion method based on collaboration of optical reflection model and microwave scattering model (CN201410799878)", etc. are mainly related to modeling methods, multi-source data combination, etc. In terms of innovation, it has not been found that multi-period remote sensing data and multi-period field sample data are used to establish a general estimation model to avoid problems such as uncertainty caused by single-period remote sensing data estimation.

Contents of the invention

Aiming at the deficiencies of the prior art, the purpose of the present invention is to provide a general model construction method for forest aboveground biomass estimation based on time series remote sensing data, so as to realize real-time, time-to-place estimation and prediction of biomass.

In order to achieve the above object, the technical solution adopted in the present invention is: a method for constructing a general model for forest aboveground biomass remote sensing estimation, comprising the following steps:

Step 1. Obtain Landsat time series images of the United States, and obtain a normalized data set of surface albedo and geometric position after radiometric calibration, atmospheric correction, geometric correction, and terrain correction;

Step 2, extracting the original wave band, vegetation factor, image enhancement factor, texture factor, terrain factor of the remote sensing image, and calculating the variation of the remote sensing factor;

Step 3, carry out field sample land investigation, carry out sampling and weighing to felled wood, take back to laboratory and dry and weigh, adopt different function forms to fit and obtain the above-ground biomass estimation model of single tree; in the present invention, single The wood biomass model can also directly refer to the single tree biomass model already established by the forestry department.

Step 4. Calculate the aboveground biomass at the level of the sample plot according to the data of the first-class forest resource survey sample plots completed by the country every five years, and calculate the variation of the aboveground biomass of the sample plot forest and the variation of remote sensing factors,

ΔAGB＝ _AGBn -AGBn _-5 (1)

_ΔRS ＝RSn-RSn _-5 (2)

Among them, ΔAGB and ΔRS represent the change of aboveground biomass and remote sensing factors respectively, and n represents the year; for example, n=1992, the change represents the change between 1992 and 1987.

Step 5, using the factors obtained in steps 2 and 4 as variables to construct a model data set;

In step 6, the general model form for constructing forest aboveground biomass is,

Among them, f() is a mathematical function of a given form, and the independent variable in the brackets is the input variable. Among the input variables, DBH is the average diameter at breast height of the trees in the plot, H is the average tree height, Age is the average tree age, and B ₁ …B _m is the original satellite band factor, C ₁ …C _n is the band combination factor, V ₁ …V _o is the vegetation index value, I ₁ …I _p is the information enhancement factor, W ₁ …W _q is the texture factor, G ₁ ...G _r is the geographical factor, R ₁ ...R _u is the reference parameter, which depends on the monitored object category and the spatial resolution of the satellite image;

General model construction is mainly based on the characteristics of operability and reliability of the model, but limited to a given area and tree species.

Step 7, screen the remote sensing factors with strong importance according to the correlation, and use the Pearson correlation coefficient and scatter plot curve method to comprehensively judge the factors with strong correlation with biomass;

Step 8: Construct forest aboveground biomass through stepwise regression, linear mixed model, linear simultaneous equations, and geographically weighted regression model; eliminate a small number of biomass values below 10 and high outliers to avoid the saturation of remote sensing data. 80% of the data was randomly selected to build a model, and 20% of the data was used for testing;

Step 9, model inspection, evaluate the simulation accuracy and inspection accuracy of the model through the following indicators;

decisive factor:

Relative root mean square error:

Relative mean absolute error:

in, is the predicted value of aboveground biomass of the i-th and point forest, y _i is the observed value, is the average value of observations.

Further: in step 1, the source of the Landsat remote sensing data is USGS, using the data of Level1, using the blue light band 0.45-0.52um, the green light band 0.52-0.60um, the red light band 0.63-0.69um, Near-infrared band 0.76-0.90um, short-wave infrared Ⅰ 1.55-1.75um, short-wave infrared Ⅱ 2.08-2.35um, the spatial resolution is 30m.

Further: in step 1, the terrain correction adopts a slope matching method.

Further: in step 2, the variation of the remote sensing factor needs to be matched with the time of the ground survey.

Further: In step 4, the sample plot data can also be continuously observed data in other ways, and the sample plot data needs to include average diameter at breast height, average tree height, number of trees, and dominant tree species.

Further: in step 6, the geographical factors include slope, aspect, altitude and soil.

Further: in step 8, the linear mixed model uses plot number and survey year as fixed effects, and altitude grade and slope grade as random effects.

The beneficial technical effects of the present invention are: the present invention builds the method for the general model of forest aboveground biomass remote sensing based on time series data, takes the Landsat remote sensing data of the United States and the measured data once every five years as data sources, and extracts remote sensing factors and Variation factors, a general model for forest aboveground biomass estimation was constructed by stepwise regression, linear mixed model, linear simultaneous equations, and geographically weighted regression model, and the model was verified by randomly selected sample plot values. The relationship between changes in remote sensing factors and stand biomass changes has been fully explored, reducing the uncertainty caused by estimation of single-period remote sensing data; the forest aboveground biomass model constructed by a linear mixed model has improved the simulation and Forecasting accuracy, a general model of forest aboveground biomass is constructed, which can estimate and predict forest aboveground biomass without conducting field surveys.

Description of drawings

In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. For those skilled in the art, other drawings can also be obtained according to these drawings without any creative effort.

Fig. 1 is a flow chart of the inventive method;

Fig. 2 is the distribution diagram of the experimental area and sample plots of the present invention.

Detailed ways

The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only some, not all, embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by persons of ordinary skill in the art without creative efforts fall within the protection scope of the present invention.

A method for constructing a general model of forest biomass remote sensing estimation, comprising the following steps (as shown in Figure 1):

1) Overview of the study area

The study area is Shangri-La City in Northwest Yunnan, China, the hinterland of the Hengduan Mountains and the southeastern edge of the Qinghai-Tibet Plateau. Its geographical range is from 26°52′ to 28°52′ north latitude and from 99°20′ to 100°19 east longitude. big. The average annual temperature is 6.3 degrees Celsius, ranging from 1.9 to 14.4 degrees Celsius. The annual rainfall is 651 millimeters, with the most precipitation being 156 millimeters in July, followed by 146 millimeters in August. Its main vegetation type is cold-temperate coniferous forest, and the dominant tree species are spruce fir (Abiesferreana Bordères-Reyet Gaussen), alpine pine (Pinusdensata Mast), Yunnan pine (Pinusyunnanensis Franch.), alpine oak (Quercusaquifolioides Rehd. et Wils.), etc. . Its forest coverage rate is relatively high, reaching 75%. Alpine pine is distributed throughout the county, mainly at altitudes above 2,800 meters. Shangri-La City has a unique type of forest landscape and rich animal and plant resources, and is a UNESCO World Natural and Cultural Heritage (as shown in Figure 1).

2) Acquisition of remote sensing data

The remote sensing data used in this example is the Landsat TM data of the US Landsat. Landsat has been released for free and has the longest historical archive data. It has obtained six Landsat TM satellite images in 1987, 1992, 1997, 2002, 2007, and 2012. Since the TM sensor stopped operating in 2012, the TM data in December 2011 was selected instead to correspond to the year of the plot survey. All images are concentrated from October to March, the cloud cover is less than 10%, the spatial resolution is 30 meters, and 6 bands are used: band 1-blue, band 2-green, band 3-red, band 4 - near infrared, band 5 - short wave infrared, band 7 - medium wave infrared. The blue light band used is 0.45-0.52um, the green light band is 0.52-0.60um, the red light band is 0.63-0.69um, the near-infrared band is 0.76-0.90um, the short-wave infrared Ⅰ 1.55-1.75um, and the short-wave infrared Ⅱ 2.08- 2.35um, the spatial resolution is 30m

3) Preprocessing of remote sensing images

First convert the original digital value to radiation value, and perform radiation calibration in ENVI5.3. Atmospheric correction is then performed using the FLAASHS model. Geometric precision correction is performed based on the topographic map data of the study area, and the error is controlled within one pixel. Finally, the slope matching method is used for terrain correction, which effectively eliminates the shadow caused by the terrain.

4) Extraction of remote sensing factors

①Original single-band factors: B1-B7;

② Band combination factors: B7/B4, B5*B4/B7, B2+B3+B4, etc.;

③ Vegetation index factors: RVI, etc. (ratio index, without considering the interaction between soil and vegetation), DVI, PVI, SAVI (based on physical knowledge, considering electromagnetic radiation, atmosphere, vegetation coverage and soil interaction), etc.;

④Information enhancement factor: principal component transformation, tasseled cap transformation;

⑤ Texture information factors: homogeneity (HO), dissimilarity (DI), entropy (EN), etc.

⑥Geographic factors: slope, aspect, altitude, abscissa, ordinate.

5) Calculation and allocation of sample plot biomass.

Field surveys were carried out from 2014 to 2018, and representative alpine pine trees were selected under different site conditions (no less than 2 trees per diameter step), and a total of 116 trees were investigated. Each sample tree is divided into trunk, branches, and needles, and each trunk is cut into 2cm discs every 2m. For each disc, 20 grams of sample wood is taken; the branches are divided into upper, middle and lower parts, which are randomly selected. The samples were taken back to the laboratory, dried and weighed, and the optimal aboveground biomass model of alpine pine single tree was established according to the coefficient of determination (R ² ) and relative median error (RMSE) (Formula 1).

W＝0.073×DBH ^1.739 ×H ^0.880 (1)

At the same time, the first-class national forest resources inventory data collected every five years from 1987 to 2012 were collected, and the above-ground biomass (t/hm ² ) of the sample plot was calculated according to the above formula.

There are 113 valid plot data from 1987 to 2012, 7 of which are less than 2, and the rest are greater than 7, which are considered too small, and 7 small ones are deleted; then, among the remaining 106 data, three times The standard deviation removed 8, leaving 98 at the end. Take 80% of the sample data (78) as the training set for model fitting, and the remaining 20% of the sample data (20) as the test set to test the model.

6) Calculation and distribution of forest aboveground biomass change and remote sensing factor change

The change amount of the sample plot is calculated by subtracting the value of the previous five years from the value of the current year, such as the value of 1992 minus the value of 1987, the value of 1997 minus the value of 1992, and so on. Among them, the change of forest aboveground biomass is calculated by formula 2, and the change of remote sensing factors is calculated by formula 3. Every five years from 1987 to 2012, there were 86 groups of fixed sample plots, 17 groups of sample plots with negative changes were discarded, 69 groups of data were screened to remove outliers, and the triple standard deviation method was used After screening, 5 outliers were finally judged, and the remaining 64 data were used to build the model after screening. Take 80% of the sample data (51) as the training set for model fitting, and the remaining 20% of the sample data (13) as the test set to test the model.

ΔAGB＝ _AGBn -AGBn _-5 (2)

_ΔRS ＝RSn-RSn _-5 (3)

Among them, ΔAGB and ΔRS represent the change of aboveground biomass and remote sensing factors respectively, and n represents the year, for example, n=1992, the change represents the change between 1992 and 1987.

7) Construction of the model: Screen the important factors according to the correlation, see Table 1.

Note: * indicates a significant level of 0.05; ** indicates a significant level of 0.01. Mean (ME), variance (VA), homogeneity (HO), contrast (CO), dissimilarity (DI), entropy (EN), angular second moment (SM), correlation (CC). R5 and R9 represent 5×5 and 9×9 windows respectively, and B1, B2, ... B7 represent bands respectively.

① Multiple stepwise regression

In this study, the significance level of variable selection was set at p ≤ 0.05, and the level of variable elimination was set at p ≥ 0.1. At the same time, in order to overcome the collinearity problem among variables, the variance inflation factor (VIF) is used to evaluate; when the VIF value is greater than 10, the corresponding independent variable will be discarded. Finally, when the variable is R5T4CC, ND32, ND54, the coefficient of determination of the model is the highest.

②Linear simultaneous equations

The three variables R5T4CC, ND32, and ND54 were used as the modeling factors of the aboveground biomass of Alpine pine (Formula 4), and the correlation between canopy density (YBD) and all factors was analyzed at the same time, and the three factors of Elevation, GPS_Y, and GPS_X were most correlated Strong, as a canopy closure modeling factor (Equation 5). It is considered that there is a linear relationship between the aboveground biomass AGB of the forest and the canopy density (YBD) of the sample plot, but there are large errors in the observed values of the two, and an unbiased estimate can be obtained by using the linear measurement error method. The model form is as follows:

AGB＝a ₁ ×YBD+a ₂ ×R5T4CC+a ₃ ×ND32+a ₄ ×ND54+e ₁ (4)

YBD＝b ₁ ×AGB+b ₂ ×Elevation+b ₃ ×GPS_X+b ₄ ×GPS_Y+e ₂ (5)

Among them, a _i and b _i (i=1,2,3,4) are the coefficients of the model, and e ₁ and e ₂ are the measurement errors of forest aboveground biomass and canopy density, respectively.

③ Geographically weighted regression

The three variables of R5T4CC, ND32, and ND54 were used as modeling factors, and the horizontal and vertical coordinates of the sample plot were used as geographic factors. Geographically weighted regression (Geographically weighted regression 4) software was used for modeling.

④Linear Mixed Model

First, grade the slope and altitude. Elevation Grade (Elevation Grade) standard: 2000-2500m is "A", 2501-3000m is "B", 3001-3500m is "C", 3501-4000m is "D"; slope classification (Slope Grade) standard: 0-8° is "one", 8.1-15° is "two", 15.1-25° is "three", 25.1-35° is "four", and greater than 35° is "five". The basic form of the model is shown in Equations 6 and 7.

AGB=r5t4cc+nd32+nd54 (6)

ΔAGB=Δr5t4cc+Δnd32+Δnd54 (7)

Among them, Δ represents the amount of change. The survey year was used as a fixed effect, and the altitude grade, slope grade, and altitude grade + slope grade were used as random effects; the plot number was used as a fixed effect, and the altitude grade, slope grade, and altitude grade + slope grade were used as random effects for modeling.

8) Model evaluation results

Table 2 Model evaluation result table

Note: Linear mixing 1-year is a fixed effect, altitude grade is a random effect; linear mixing 2-year is a fixed effect, slope grade is a random effect; linear mixing 3-year is a fixed effect, altitude grade and slope grade are random effects; Linear Mixing 4—Plot No. is a fixed effect, altitude grade is a random effect; Linear Mixing 5—Plot No. is a fixed effect, slope grade is a random effect; Linear Mixing 4—Plot No. is a fixed effect, altitude grade and slope grade is a random effect. "/" indicates that the error is too large to be calculated.

From the fitting results, whether it is a static or dynamic model, the coefficient of determination of the linear mixture 4 using the plot number as a fixed effect and the altitude level as a random effect is the highest. The fitting coefficient of determination of the static model is 0.713, rRMSE is 17.69, rMAE=18.92; the coefficient of determination of the dynamic model constructed by using the interval variable of every 5 years is 0.957, rRMSE=8.32, rMAE=12.15, the model has the highest accuracy and can be used as the final estimation A general model for measuring the aboveground biomass of alpine pine.

9) General model form of alpine pine aboveground biomass in Shangri-La City

The final general model of alpine pine aboveground biomass adopts linear mixed model, plot number (ID) is a fixed effect, and altitude grade is a random effect.

① The basic form of the static model is shown in formula 6. The parameters of the fixed effects of the model are as follows, and the parameters of the random effects are 0, which are omitted here.

② If the change every 5 years is used as the input variable, the biomass in the base year plus the biomass estimated based on the change is used. The basic form of the variable model is shown in Equation 7. The parameters of the fixed effects of the model are as follows, and the parameters of the random effects are 0, which are omitted here.

Those skilled in the art can understand that, unless otherwise defined, all terms (including technical and scientific terms) used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs. It should also be understood that terms such as those defined in commonly used dictionaries should be understood to have a meaning consistent with the meaning in the context of the prior art, and will not be interpreted in an idealized or overly formal sense unless defined as herein explain.

Finally, it should be noted that the above embodiments are only used to illustrate and not limit the technical solutions of the present invention, although the present invention has been described in detail with reference to the above embodiments, those of ordinary skill in the art should understand that: the present invention can still be modified Or an equivalent replacement, any modification or partial replacement without departing from the spirit and scope of the present invention shall fall within the scope of the claims of the present invention.

Claims (7)

1. A forest ground biomass remote sensing estimation general model construction method is characterized by comprising the following steps:

step 1, acquiring a Landsat time sequence image of the United states land satellite, and obtaining a data set with normalized earth surface reflectivity and geometric position through radiometric calibration, atmospheric correction, geometric correction and terrain correction;

step 2, extracting an original waveband, a vegetation factor, an image enhancement factor, a texture factor and a terrain factor of the remote sensing image, and calculating the variable quantity of the remote sensing factor;

step 3, carrying out field sample plot investigation, sampling and weighing felled trees, bringing the felled trees back to a laboratory for drying and weighing, and fitting by adopting different function forms to obtain an aboveground biomass estimation model of the single trees;

step 4, according to the forest resource survey sample plot data completed by the countries every five years, calculating the aboveground biomass at the sample plot level, simultaneously calculating the variation of the aboveground biomass of the forest sample plot and the variation of the remote sensing factor,

ΔAGB＝AGB_n-AGB_n-5(1)

ΔRS＝RS_n-RS_n-5(2)

wherein, the delta AGB and the delta RS respectively represent the variation of aboveground biomass and remote sensing factors, and n represents the year;

step 5, constructing a model data set by taking the factors obtained in the step 2 and the step 4 as variables;

step 6, constructing a general model form of the forest aboveground biomass,

wherein f () is a mathematical function of a given form, an independent variable is included in parentheses, and the input variable is input, wherein DBH is the average breast height of the tree in the sample plot, H is the average tree height, Age is the average tree Age, B₁…B_mAs a factor of the satellite's original band, C₁…C_nAs a band combination factor, V₁…V_oIs a vegetation index value, I₁…I_pAs an information enhancement factor, W₁…W_qIs a texture factor, G₁…G_rIs a geographic factor, R₁…R_uAs reference parameters, depending on the object class monitored and the spatial resolution of the satellite image;

step 7, screening the remote sensing factors with strong importance according to the correlation, and comprehensively judging the factors with strong correlation with the biomass by adopting a Pearson correlation coefficient and scatter diagram curve method;

step 8, constructing forest aboveground biomass through stepwise regression, a linear mixed model, a linear simultaneous equation set and a geographical weighted regression model; removing a small amount of biomass values below 10 and higher outliers to avoid biomass estimation errors caused by remote sensing data saturation, randomly selecting 80% of data to construct a model, and using 20% of data for inspection;

step 9, model inspection, namely evaluating the simulation precision and the inspection precision of the model according to the following indexes;

determining a coefficient:

relative root mean square error:

relative mean absolute error:

wherein,is a prediction of the biomass on the ith and forest plots, y_iIn order to be able to take the value of the observation,the average of the observations.

2. The method for remote sensing estimation of biomass on forest ground general model construction according to claim 1, characterized in that: in the step 1, the Landsat remote sensing data source is USGS, Level1 data is adopted, and blue light wave band 0.45-0.52um, green light wave band 0.52-0.60um, red light wave band 0.63-0.69um, near infrared wave band 0.76-0.90um, short wave infrared I1.55-1.75 um, short wave infrared II 2.08-2.35um and spatial resolution of 30m are used.

3. The method for remote sensing estimation of biomass on forest ground general model construction according to claim 1, characterized in that: in step 1, the terrain correction adopts a gradient matching method.

4. The method for remote sensing estimation of biomass on forest ground general model construction according to claim 1, characterized in that: in step 2, the variation of the remote sensing factor needs to be matched with the time of ground investigation.

5. The method for remote sensing estimation of biomass on forest land universal model construction according to claim 1, characterized by comprising the following steps: in step 4, the sample plot data may also be continuously observed in other manners, and the sample plot data includes an average chest diameter, an average tree height, the number of plants, and dominant tree species.

6. The method for remote sensing estimation of biomass on forest land universal model construction according to claim 1, characterized by comprising the following steps: in step 6, the geographic factors include slope, incline, elevation and soil.

7. The method for remote sensing estimation of biomass on forest land universal model construction according to claim 1, characterized by comprising the following steps: in step 8, the linear mixed model respectively selects a sample plot number and a survey year as fixed effects, and an altitude grade and a gradient grade as random effects.