CN109213964A - A kind of satellite AOD product bearing calibration for merging multi-source feature geographic factor - Google Patents
A kind of satellite AOD product bearing calibration for merging multi-source feature geographic factor Download PDFInfo
- Publication number
- CN109213964A CN109213964A CN201810776814.2A CN201810776814A CN109213964A CN 109213964 A CN109213964 A CN 109213964A CN 201810776814 A CN201810776814 A CN 201810776814A CN 109213964 A CN109213964 A CN 109213964A
- Authority
- CN
- China
- Prior art keywords
- aod
- aeronet
- satellite
- data
- variable
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Granted
Links
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- General Physics & Mathematics (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Theoretical Computer Science (AREA)
- Mathematical Physics (AREA)
- Computational Mathematics (AREA)
- Mathematical Analysis (AREA)
- Life Sciences & Earth Sciences (AREA)
- Operations Research (AREA)
- Probability & Statistics with Applications (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Algebra (AREA)
- Evolutionary Biology (AREA)
- Databases & Information Systems (AREA)
- Software Systems (AREA)
- General Engineering & Computer Science (AREA)
- Bioinformatics & Computational Biology (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a kind of satellite AOD product bearing calibrations for merging multi-source feature geographic factor, first, regression model is corrected based on random forests algorithm building satellite AOD product, AERONET AOD planar is obtained and simulates spatial distribution result, as the satellite AOD data after preliminary corrections;Then the comparison of done site by site point is carried out with ground AERONET AOD truthful data, obtains the deviation of the two, and carry out space interpolation estimation, deviation obtained based on estimation, the satellite AOD value of preliminary corrections is modified;Finally, modified satellite AOD data and ground AERONET AOD data are carried out the comparison of done site by site point by setting corrected threshold.This AOD data for being present satellites remote-sensing inversion and true AOD value be in the case that there is some difference, one kind of invention it is comprehensive it is high, the property of can refer to is strong, effectively accurately satellite AOD product bearing calibration, study atmosphere PM as accurate land productivity satellite AOD2.5Effective concentration spatial-temporal distribution characteristic and changing rule simultaneously assess continuous large geographical area the state of air pollution offer reference.
Description
Technical field
The present invention relates to satellite remote sensing retrieval products field, in particular to a kind of aerosol optical depth (AOD) correction side
Method.
Background technique
Atmospheric aerosol refers to that the various solids that the diameter that left floating in atmosphere is 0.001~100 μm and liquid particle are total
With the heterogeneous system of composition.Atmospheric aerosol main source include sand and dust, salt grain, volcano eruption fall-out and forest combustion
The natural sources and burning, communications and transportation and the flue dust of various industrial discharges from fossil and non-fossil fuel etc. such as flue dust
Artificial source.In recent decades, the urbanization of global range and industrialized sharply development are so that anthropogenic discharge is continuously increased, seriously
Atmosphere pollution air quality and Regional climate change caused seriously affect, become a global environmental problem.
Both at home and abroad it is most to optical properties of aerosol research and concern be aerosol optical depth (Aerosol Optical Depth,
AOD).AOD is an important parameter for describing consumption of the aerosol to light, is constructed using satellite AOD product
PM2.5- AOD relational model and then estimation ground PM2.5The simulation of concentration spatial and temporal distributions is difficult to standard to solve traditional ground observation method
Really disclose PM2.5The spatial and temporal distributions situation and evolution Feature polluted in extensive area provides new scheme.
Ground aerosol remote sensing monitoring can obtain AOD accurate information on point scale, and the current world is widely recognized that and makes
Ground AOD product is aerosol automatic Observational Network (the Aerosol Robotic that U.S. NASA is initiated and supported
Network,AERONET).AERONET observational network 1497 websites covering the whole world are existed using ground CIMEL heliograph
The AOD with regional representativeness is obtained in global range.The observation error of AERONET AOD is 0.01~0.02, can be incited somebody to action
The observation of AERONET is verified and is assessed as precision of the true value to the AOD data product that satellite remote sensing obtains.So
And compared with groundwork detection, satellite remote sensing observation range is wider, is limited by space-time smaller, can obtain large geographical area
AOD spatial and temporal distributions characteristic.But the aerosol product of satellite remote sensing inverting by ground mulching, atmospheric radiation, meteorologic factor,
The influence of many factors such as instrument itself error, inversion algorithm and product, satellite AOD data and true AOD value exist certain
Difference leads to the result error using direct satellite AOD product research air pollution concentration spatial-temporal distribution characteristic, needs invention one
The method that kind carries out efficient spatio-temporal correction for satellite AOD data product.
The variation of regional geographic characteristics element has relatively strong influence, land use, traffic route, people for the height of AOD
Mouth is equal closely related with synthetic aerosol discharge intensity, and the meteorologic factors such as wind speed, precipitation, relative humidity can influence atmospheric aerosol
Diffusion precipitin.Therefore, the spatial and temporal distributions spy of AOD can be simulated by analyzing the space distribution rule of multi-source geographical feature parameter
Property;Meanwhile the AOD product and ground observation of satellite remote sensing inverting can be explained further in the difference of regional geography element distribution
The difference of AOD product provides feasibility for correction satellite AOD data product.Therefore, satellite AOD data and ground are utilized
The deviation of the true AOD data of AERONET, based on machine learning algorithm and geo-statistic theory, fusion multi-source feature geography ginseng
Several pairs of satellite AOD products are corrected, and can obtain optimal, unbiased, accurate, reliable satellite AOD correction product.
Summary of the invention
In order to effectively solve the problems, such as that the AOD Product Precision of satellite remote sensing inverting is low, error is big in order to effectively solve, this hair
It is bright that a kind of method of satellite AOD product correction for merging multi-source feature geographic factor is provided.
In order to achieve the above technical purposes, the technical scheme is that,
A kind of satellite AOD product bearing calibration for merging multi-source feature geographic factor, comprising the following steps:
Step 1, sample is established using the modeling factors data set of matched ground AERONET monitoring station corresponding position to instruct
Practice collection, then using the ground AERONET AOD data of training set as dependent variable, with multi-source feature geographic factor and satellite AOD number
According to for independent variable, randomly selecting the branch node building pair of corresponding dependent variable and independent variable as decision tree from each sample
The decision tree of each sample is answered, to form random forest regression model with these decision trees, and to the conduct feature in model
The independent variable of variable carries out important sex determination, then optimizes the quantity of characteristic variable and decision tree, completes random forest and returns mould
The foundation of type, then the multi-source feature geographic factor of corrected satellite AOD data and known corresponding place will be needed to input
Into random forest regression model, AERONET AOD planar simulation spatial distribution result is obtained, as the satellite after preliminary corrections
AOD data;
Step 2, by after the preliminary corrections obtained in step 1 satellite AOD data and ground done site by site point AERONET AOD it is true
Real data is compared, the deviation both obtained, carries out ordinary kriging interpolation to deviation, obtain ground AERONET AOD with
Deviation planar spatial distribution between the satellite AOD of preliminary corrections is obtained according to the satellite AOD data and deviation after preliminary corrections
Obtain the satellite AOD data of secondary correction;
Step 3, corrected threshold is set, the satellite AOD data of secondary correction and ground done site by site point AERONET AOD are true
Data compare after with corrected threshold compared with, if more than threshold value then return step 2, until the result that currently obtains and it is preceding once
Result it is identical or be less than corrected threshold, export finally obtained satellite AOD correction product and precision evaluation result.
The method, in the step one, the matched ground AERONET monitoring station corresponding position is built
Mould factor data is concentrated, including ground AERONET AOD station data and multi-source feature geographic factor, and the multi-source is characteristically
Reason parameter includes land use/covering, road traffic, landform altitude, wind speed, precipitation, relative humidity, temperature and air pressure, is incited somebody to action
Ground AERONETAOD station data as dependent variable and the multi-source feature geographic factor as independent variable are according to monitoring station
It is matched one by one, forms modeling factors data set.
The method, in the step one, training set is established by sample the following steps are included:
To the original sample D for including p characteristic variable, taken out using part observation of the Bootstrap method to D
K training set θ is randomly generated in sample1、θ2、…θk。
The method, in the step one, composition random forest regression model the following steps are included:
For the sample in each training set, it is the variable of n as decision tree that fixed number is randomly selected from p variable
Branch node construct decision tree, wherein n < p, to generate corresponding decision tree { H (X, θ based on each training set1)}、{H
(X,θ2)}、…{H(X,θk), wherein X is independent variable, this K decision tree constitutes random forest regression model, use Geordie system
Number selection random forest regression model characteristic variable, Gini (v) calculation formula of the model at node v are as follows:
In formula,It is observation of j-th of variable in node v, XiIn the Geordie information gain Gain (X of split vertexes vi,
V) be node v and v child node between impurity level difference.
The method, Geordie information gain Gain (Xi, v) and calculation formula is as follows:
Gain(Xi, v) and=Gini (Xi,v)-wLGini(Xi,vL)-wRGini(Xi,vR)
In formula, vLWith vRIt is the left and right child node at node v, w respectivelyLWith wRIt is that characteristic variable belongs to left and right son section respectively
The ratio of point randomly selects mtry variable at each node in p variable, final acquisition maximum information gain
Characteristic variable is used for the division of node v, wherein
The method, in the step one, carrying out important sex determination to the characteristic variable in model includes following step
It is rapid:
Carry out evaluation and foreca characteristic variable X using predictive variable importance scoresiTo random forest regression model contribute, from
The big characteristic variable of middle selection contribution degree, calculation formula are as follows:
In formula,It is in the random forest set at ntree, by XiThe node set of division.
The method, in the step one, optimize the quantity of characteristic variable and decision tree the following steps are included:
By traversal setting mtry parameter from 1 to n, carries out n times test and record the error rate tested every time, select mistake
The minimum mtry value of rate is as optimal mtry value, and for the quantity ntree of decision tree, fixed optimal mtry value obtained above is not
Become, traverse ntree from 1 to n, carry out n times test and records the error rate tested every time, the ntree value for selecting error rate minimum
As optimal ntree value.
The method in the step two, carries out ordinary kriging interpolation to deviation, formula is as follows:
In formula,For point (x0,y0) estimated value, i.e. Z0=Z (x0,y0);λiFor weight coefficient, and meet point (x simultaneously0,
y0) at estimated valueWith true value Z0The smallest a set of optimal coefficient of difference;
1. unbiasedness:
2. optimality:
The method, in the step three, setting corrected threshold the following steps are included:
Corrected threshold indicates that formula is as follows using anticipation error (EE):
EE=a+b × σAERONET
Wherein, σAERONETIndicate that ground AERONET AOD, coefficient a and b are to adjust the parameter for correcting Product Precision, 0 < a <
1,0<b<1。
The method, in the step three, precision evaluation passes through following steps and realizes:
Precision evaluation index indicates that formula is as follows with relative error δ using coefficient R:
In formula, x, y respectively indicate the satellite AOD value after AERONET AOD value and correction,Respectively indicate year
The average value of AERONET AOD and the satellite AOD after correction.
The technical effects of the invention are that: (1) this method be in the case where satellite AOD error is big, precision is low, invention
A kind of New Satellite AOD product bearing calibration of coupling machine learning algorithm and Geostatistical theory and technology;(2) our simultaneously
Method is a kind of more accurate satellite AOD data correcting method, has merged multi-source feature geographic factor, is based on ground AERONET
The deviation of AOD truthful data and satellite AOD data is iterated amendment to satellite AOD data product, optimal unbiased until obtaining
It is accurately reliable to repair result for the satellite AOD product of estimation;(3) this method or a kind of efficient satellite AOD product correction of comparison
Method, required technology and data can obtain in real time, and algorithm is efficient, can quickly correct satellite AOD product.
Detailed description of the invention
Fig. 1 is the satellite AOD product bearing calibration flow chart of present invention fusion multi-source feature geographic factor, wherein (a) table
The building of AOD random forest regression model and simulation for showing fusion multi-source feature geographic factor, (b) indicate satellite AOD and ground
The Geostatistical of the deviation of AERONET AOD models, and (c) indicates that iterative model building solves the MODIS AOD of Best unbiased estimator
Correct product.
Fig. 2 is that the satellite AOD of present invention fusion multi-source feature geographic factor corrects product and not corrected satellite AOD is produced
Product accuracy comparison figure, wherein (a) is original AOD Product Precision figure, it is (b) the AOD Product Precision figure after present invention correction.
Specific embodiment
Here is to a preferred embodiment of the invention, in conjunction with the detailed description of attached drawing progress.
1, the building of AOD random forest regression model and simulation of multi-source feature geographic factor are merged
The building of AOD random forest regression model and simulation for the fusion multi-source feature geographic factor that the present invention uses, including
Following steps:
Step 1): collecting and extracts multi-source feature geographic element.This example is collected big with region by taking Beijing-tianjin-hebei Region as an example
Gas aerosol discharge and spread relevant geographic factor collection, it is main include meteorological day Value Data (temperature, air pressure, humidity, precipitation,
Wind speed), land cover pattern/use pattern (construction land, meadow, forest land, waters), road traffic, population grid and landform altitude
Data.For the geographic element of planar, such as population grid, landform altitude raster data, directly extraction ground AERONET monitoring
The corresponding grid attribute value of website is as corresponding characteristic variable;For dotted geographic element, such as dotted meteorological element, make
The spatial distribution raster data of element is obtained with spatial interpolation methods, then is extracted;For road and land element, due to it
With spatial scale effects, thus use with the area of the total length of road and all kinds of lands used in ground monitoring website a certain range
Accounting is extracted around calculating website within the scope of 500-5000m as characteristic variable by establishing buffer area respectively, all kinds of lands used
Area accounting and all road total length degree.
Step 2): satellite AOD product data processing.This example is to be mounted in the MODIS sensor inverting of TERRA satellite
For DT DB Combined AOD data product, by downloading the product in the official website NASA, is projected, spliced and adopted again
Sample generates the raster data of 10 × 10km spatial resolution and cuts out the corresponding AOD data in Beijing-tianjin-hebei Region.Last base area
The satellite AOD data of face AERONET monitoring station spatial position extraction corresponding points.
Step 3): ground AERONET AOD product data processing.
1. wave band matches
AERONET can provide 1020,870,670,500, the AOD data at the wave bands such as 440nm, and satellite remote sensing inverting produces
Product MODIS only provides 660,550, the AOD data of 3 wave bands of 470nm.Two kinds of AOD data products do not have wave corresponding to the same
Section.It thus needs to carry out wave band interpolation to AERONET AOD data, obtains the AOD data to match with MODIS product wave band.
On the wave band that no steam influences, the Spectral structure of particulate meets Junge distribution, between AOD and wavelength
In the presence ofRelational expression
σ (λ)=β λ-α
In formula, σ (λ) indicates that wavelength is the AOD value at λ;β is indicatedAtmospheric turbidity coefficient, value size by
The influence of particulate sum, volume size distribution and refraction coefficient;α is indicatedWavelength Indices, value and gas
Sol particles radius is related.Assuming that wave band λ1, λ2Place is not influenced by steam, is had according to above formula:
σ(λ1)=β λ1 -α
σ(λ2)=β λ2 -α
Simultaneous above formula can obtain:
In formula,For wavelength X1-λ2Between Wavelength Indices.In known λ1, λ2And Wavelength IndicesThe case where
Under, wavelength X1-λ2Between any af at wavelength lambda aerosol optical depth value can by following formula carry out interpolation obtain:
Based on above-mentioned formula, this example is using the AERONETAOD data at 440 and 870nm and the wave between 440-870nm
Long index α440-870, the AERONET AOD at 550nm is obtained by interpolation.
2. time-space registration
AERONET AOD data are the continuous observations based on monitoring station location interval set time (generally 15min)
Data, and MODIS Combined AOD value indicates the instantaneous monitoring value of 10km × 10km area, while satellite transit time is
Local time at 10 points in the morning 30 minutes.Since MODIS AOD and AERONET AOD is different from time and space scale.Such as
Fruit directly takes the AERONET observation of satellite transit time to be compared with the MODIS AOD value of single pixel, and this comparison can
Reliability is lower, and singular value phenomenon easily occurs.In order to improve the comparability of two kinds of products, during this example with AERONET website is
The heart takes mean value as the satellite AOD at AERONET website after removing exceptional value using the MODISAOD of 5 × 5 pixel of website periphery
Mean value, using satellite pass by front and back 1.5h ground AERONET observation data time average as AERONET observation mean value,
Certain screening conditions are added simultaneously, need to meet around AERONET website at least 5 or more effective pictures in 5 × 5 pixels
Member, AERONET surface observation data satellite pass by front and back 1.5h at least 2 or more effective observations just can be into
Row matching comparison.
Step 4): building random forest regression model
Construct a random forest (Random Forest) regression model, using ground AERONET AOD station data as because
Variable, merge multi-source feature geographic factor (meteorological element, land use/covering, road traffic, landform altitude etc.) and
MODIS DT DB Combined AOD product is independent variable.The key step that random forest returns has: 1. sample data is random
Sampling: to the original sample D for including p characteristic variable, using Bootstrap method (having the sampling put back to) to the part of D
Observation is sampled, and K training set θ is randomly generated1、θ2、…θk, and data set (out- outside the data composition bag that do not drawn
Of-bag, OBB), it can be used as test sample collection.2. characteristic variable randomly selects: for the sample in each training set, from p
The variable that fixed number is n (n < p) is randomly selected in variable and constructs decision tree as the branch node of decision tree, thus each instruction
Practice collection and generates corresponding decision tree { H (X, θ1)}、{H(X,θ2)}、…{H(X,θk), this K decision tree just constitutes random gloomy
Woods regression model, model obtain predicted value finally by the mode being averaged.
This example selects random forest regression model characteristic variable, Gini (v) of the model at node v using Gini coefficient
Calculation formula is as follows:
In formula,It is observation of j-th of variable in node v.XiIn the Geordie information gain Gain (X of split vertexes vi,
V) be node v and v child node impurity level difference.Calculation formula is as follows:
Gain(Xi, v) and=Gini(Xi,v)-wLGini(Xi,vL)-wRGini(Xi,vR)
In formula, vLWith vRIt is the left and right child node at node v, w respectivelyLWith wRIt is that characteristic variable belongs to left and right son section respectively
The ratio of point.At each node, mtry is randomly selected in p variableA variable, the final characteristic variable for obtaining maximum information gain are used for node
The division of v.
This example carrys out evaluation and foreca characteristic variable X using predictive variable importance scoresiTo random forest regression model
Contribution, the characteristic variable for therefrom selecting contribution degree big, calculation formula are as follows:
In formula,It is in the random forest set at ntree, by XiThe node set of division.
In Random Forest model building process, mtry and ntree are two important parameters, when respectively representing construction branch
The variable number of random sampling and the quantity of decision tree.Wherein, it is randomly selected from P variable when mtry constructs decision tree
N variable, also the as branch number of nodes of decision tree, selects suitable mtry value that can reduce the pre- sniffing of Random Forest model
Accidentally rate.The present invention, from 1 to n, is carried out n times test and records the error rate tested every time by traversal setting mtry parameter, selection
The minimum mtry value of error rate is as optimal mtry value.Ntree indicates the quantity of decision tree, usually, when ntree value is too small
When to will lead to model errors rate higher, and when ntree value is excessive model complexity can be promoted, reduce model efficiency.This
Example is constant by fixed optimal mtry value obtained above, traverses ntree from 1 to n, carries out n times test and records examination every time
The error rate tested, the ntree value for selecting error rate minimum is as optimal ntree value.
Step 5): simulation AERONET AOD spatial distribution
After the completion of random forest regression model, input is constituted by multi-source feature geographic factor and MODIS AOD data
Variable is entered into the random forest regression model that building is completed, and is obtained AERONET AOD planar and is simulated spatial distribution knot
Fruit, the MODIS AOD data as preliminary corrections.
2, the Geostatistical of the deviation of MODIS AOD and ground AERONET AOD models
The Geostatistical of the deviation of MODIS AOD and ground AERONET AOD that the present invention uses models, including following
Step:
Step 1): the deviation of MODIS AOD and ground AERONET AOD are solved
The preliminary corrections MODIS AOD data obtained in above-mentioned steps 1 and ground done site by site point AERONET AOD are really counted
According to being compared, the deviation delta (t) of the two is obtained.
Step 2): Geostatistical modeling is carried out to deviation using Kriging regression method
The ground station AERONET AOD and MODIS AOD deviation delta (t) obtain to above-mentioned steps carries out Ke Lijin and inserts
Value obtains deviation delta (t) the planar spatial distribution between the ground AERONET AOD and MODIS AOD of preliminary corrections.Common gram
In golden interpolation method (Ordinary Kriging, OK), be a kind of common Kriging regression method, basic principle such as following formula
It is shown:
In formula,For point (x0,y0) estimated value, i.e. Z0=Z (x0,y0);λiFor weight coefficient, and meet point (x simultaneously0,
y0) at estimated valueWith true value Z0The smallest a set of optimal coefficient of difference.
1. unbiasedness:
2. optimality:
Step 3): amendment MODIS AOD product
The preliminary corrections MODIS AOD data obtained using above-mentioned steps 1, in addition the ground AERONET of step 2) simulation
Deviation delta (t) between AOD and the MODIS AOD of preliminary corrections, so that it may which the MODIS AOD data for obtaining secondary correction produce
Product.
3, the MODIS AOD that iterative model building solves Best unbiased estimator corrects product
The MODIS AOD that the iterative model building that the present invention uses solves Best unbiased estimator corrects product, comprising the following steps:
Step 1): according to actual product accuracy requirement, corrected threshold is set.Corrected threshold uses anticipation error (EE) table
Show, formula is as follows:
EE=a+b × σAERONET
Wherein, σAERONETIndicate ground AERONET AOD.Coefficient a and b are predefined constant (0 < a < 1,0 <b < 1).
This example by setting coefficient a=0.05, b=0.2, obtain corrected threshold be EE=0.05+0.2 ×
σAERONET。
Step 2): the MODIS AOD data of the secondary correction that step 2 is obtained and ground AERONETAOD data carry out by
Whether website comparison, the difference DELTA (t+1) both judged are greater than the threshold value of setting, if if, then repeatedly step 2, until working as
The result of preceding acquisition is identical with previous result or is less than corrected threshold, exports the school finally obtained MODIS AOD
Positive product and precision evaluation result.Wherein, precision evaluation index is indicated using coefficient R with relative error (δ), formula
It is as follows:
In formula, x, y respectively indicate the MODIS AOD value after AERONET AOD value and correction, Respectively indicate year
The average value of AERONET AOD and the MODIS AOD after correction.
In this example, t=3 when circulation stops, finally obtained satellite AOD correction product and ground AERONET AOD
The coefficient R of data is 0.98, and relative error δ is 10.12%, and coefficient R rises 0.22 than original AOD data,
Relative error δ reduces 34.01%, shows that the present invention can effectively promote satellite AOD data product precision.
The above content is a further detailed description of the present invention in conjunction with specific preferred embodiments, and it cannot be said that
A specific embodiment of the invention is only limitted to this, for those of ordinary skill in the art to which the present invention belongs, is not taking off
Under the premise of from present inventive concept, several simple deduction or replace can also be made, all shall be regarded as belonging to the present invention by institute
Claims of submission determine scope of patent protection.
Claims (10)
1. a kind of satellite AOD product bearing calibration for merging multi-source feature geographic factor, which comprises the following steps:
Step 1, sample training is established using the modeling factors data set of matched ground AERONET monitoring station corresponding position
Collection, then using the ground AERONET AOD data of training set as dependent variable, with multi-source feature geographic factor and satellite AOD data
For independent variable, the branch node building that corresponding dependent variable and independent variable are randomly selected from each sample as decision tree is corresponded to
The decision tree of each sample to form random forest regression model with these decision trees, and becomes the conduct feature in model
The independent variable of amount carries out important sex determination, then optimizes the quantity of characteristic variable and decision tree, completes random forest regression model
Foundation, then the multi-source feature geographic factor of corrected satellite AOD data and known corresponding place will be needed to be input to
In random forest regression model, AERONET AOD planar simulation spatial distribution result is obtained, as the satellite after preliminary corrections
AOD data;
Step 2, the satellite AOD data after the preliminary corrections obtained in step 1 are really counted with ground done site by site point AERONET AOD
According to being compared, the deviation both obtained carries out ordinary kriging interpolation to deviation, obtains ground AERONET AOD and preliminary
Deviation planar spatial distribution between the satellite AOD of correction obtains two according to the satellite AOD data and deviation after preliminary corrections
The satellite AOD data of secondary correction;
Step 3, corrected threshold is set, by the satellite AOD data of secondary correction and ground done site by site point AERONET AOD truthful data
After comparing with corrected threshold compared with, if more than threshold value then return step 2, until the result currently obtained and previous knot
Fruit is identical or is less than corrected threshold, exports finally obtained satellite AOD correction product and precision evaluation result.
2. the method according to claim 1, wherein in the step one, the matched ground
In the modeling factors data set of AERONET monitoring station corresponding position, including ground AERONET AOD station data and multi-source spy
Geographic factor is levied, the multi-source feature geographic factor includes land use/covering, road traffic, landform altitude, wind speed, drop
Water, relative humidity, temperature and air pressure, will be as the ground AERONET AOD station data of dependent variable and as independent variable
Multi-source feature geographic factor is matched one by one according to monitoring station, forms modeling factors data set.
3. the method according to claim 1, wherein establishing training set packet by sample in the step one
Include following steps:
To the original sample D for including p characteristic variable, it is sampled using part observation of the Bootstrap method to D,
K training set θ is randomly generated1、θ2、…θk。
4. according to the method described in claim 3, it is characterized in that, forming random forest regression model in the step one
The following steps are included:
For the sample in each training set, variable dividing as decision tree for randomly selecting that fixed number is n from p variable
Minor matters point constructs decision tree, wherein n < p, to generate corresponding decision tree { H (X, θ based on each training set1)}、{H(X,
θ2)}、…{H(X,θk), wherein X is independent variable, this K decision tree constitutes random forest regression model, use Gini coefficient
Random forest regression model characteristic variable is selected, Gini (v) calculation formula of the model at node v is as follows:
In formula,It is observation of j-th of variable in node v, XiIn the Geordie information gain Gain (X of split vertexes vi, v) be
The difference of impurity level between the child node of node v and v.
5. according to the method described in claim 4, it is characterized in that, Geordie information gain Gain (Xi, v) and calculation formula is as follows:
Gain(Xi, v) and=Gini (Xi,v)-wLGini(Xi,vL)-wRGini(Xi,vR)
In formula, vLWith vRIt is the left and right child node at node v, w respectivelyLWith wRIt is the ratio that characteristic variable belongs to left and right child node respectively
Example, at each node, randomly selects mtry variable in p variable, and the final feature for obtaining maximum information gain becomes
Amount is used for the division of node v, wherein
6. according to the method described in claim 4, it is characterized in that, in the step one, to the characteristic variable in model into
The important sex determination of row the following steps are included:
Carry out evaluation and foreca characteristic variable X using predictive variable importance scoresiTo random forest regression model contribute, Cong Zhongxuan
The big characteristic variable of contribution degree is selected, calculation formula is as follows:
In formula,It is in the random forest set at ntree, by XiThe node set of division.
7. according to the method described in claim 4, it is characterized in that, optimizing characteristic variable and decision tree in the step one
Quantity the following steps are included:
By traversal setting mtry parameter from 1 to n, carries out n times test and record the error rate tested every time, select error rate most
Low mtry value is as optimal mtry value, and for the quantity ntree of decision tree, fixed optimal mtry value obtained above is constant,
Ntree is traversed from 1 to n, n times test is carried out and simultaneously records the error rate tested every time, the ntree value for selecting error rate minimum as
Optimal ntree value.
8. according to the method described in claim 4, it is characterized in that, carrying out common Ke Lijin to deviation in the step two
Interpolation, formula are as follows:
In formula,For point (x0,y0) estimated value, i.e. Z0=Z (x0,y0);λiFor weight coefficient, and meet point (x simultaneously0,y0) at
Estimated valueWith true value Z0The smallest a set of optimal coefficient of difference;
1. unbiasedness:
2. optimality:
9. according to the method described in claim 4, it is characterized in that, setting corrected threshold includes following in the step three
Step:
Corrected threshold indicates that formula is as follows using anticipation error (EE):
EE=a+b × σAERONET
Wherein, σAERONETIndicate that ground AERONET AOD, coefficient a and b are to adjust the parameter for correcting Product Precision, 0 < a < 1,0 <b <
1。
10. according to the method described in claim 4, it is characterized in that, precision evaluation passes through following steps in the step three
It realizes:
Precision evaluation index indicates that formula is as follows with relative error δ using coefficient R:
In formula, x, y respectively indicate the satellite AOD value after AERONET AOD value and correction,Respectively indicate annual AERONET
The average value of AOD and the satellite AOD after correction.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810776814.2A CN109213964B (en) | 2018-07-13 | 2018-07-13 | Satellite AOD product correction method fusing multi-source characteristic geographic parameters |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201810776814.2A CN109213964B (en) | 2018-07-13 | 2018-07-13 | Satellite AOD product correction method fusing multi-source characteristic geographic parameters |
Publications (2)
Publication Number | Publication Date |
---|---|
CN109213964A true CN109213964A (en) | 2019-01-15 |
CN109213964B CN109213964B (en) | 2021-08-17 |
Family
ID=64990105
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201810776814.2A Active CN109213964B (en) | 2018-07-13 | 2018-07-13 | Satellite AOD product correction method fusing multi-source characteristic geographic parameters |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN109213964B (en) |
Cited By (20)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109785569A (en) * | 2019-01-28 | 2019-05-21 | 中科光启空间信息技术有限公司 | A kind of forest fire monitoring method based on 3S technology |
CN110398446A (en) * | 2019-08-01 | 2019-11-01 | 成都信息工程大学 | A kind of PM of combination AOD and meteorologic parameter2.5Concentration approximating method |
CN110610258A (en) * | 2019-08-20 | 2019-12-24 | 中国地质大学(武汉) | Urban air quality refined estimation method and device fusing multi-source space-time data |
CN111078678A (en) * | 2019-12-18 | 2020-04-28 | 中国气象局乌鲁木齐沙漠气象研究所 | Satellite precipitation data correction method based on multi-source information fusion and scale reduction |
CN111104639A (en) * | 2019-12-24 | 2020-05-05 | 福州大学 | Point-surface fused time sequence PM2.5 spatial distribution estimation method |
CN111323352A (en) * | 2020-04-09 | 2020-06-23 | 中南大学 | Regional PM2.5 remote sensing inversion model fusing fine particulate matter concentration data |
CN111737850A (en) * | 2020-05-15 | 2020-10-02 | 中国科学院空天信息创新研究院 | Multisource satellite AOD fusion method based on uncertainty on pixel scale |
CN112016696A (en) * | 2020-08-14 | 2020-12-01 | 武汉大学 | PM integrating satellite observation and ground observation1Concentration inversion method and system |
CN112131789A (en) * | 2020-09-18 | 2020-12-25 | 中国人民解放军国防科技大学 | Multispectral rainfall detection system and method based on random forest algorithm |
CN112395808A (en) * | 2020-11-18 | 2021-02-23 | 南京林业大学 | Biomass remote sensing mapping method combining random forest and collaborative kriging |
CN112884342A (en) * | 2021-03-10 | 2021-06-01 | 陕西九州遥感信息技术有限公司 | Water color satellite atmospheric layer top radiation product quality evaluation and cross calibration method |
CN112926625A (en) * | 2020-11-25 | 2021-06-08 | 北方工业大学 | Method for analyzing deviation influence factors of satellite radiation data |
CN113065098A (en) * | 2021-03-23 | 2021-07-02 | 北京航天创智科技有限公司 | PM2.5 concentration detection method and device based on satellite remote sensing and ground gas system data |
CN113077083A (en) * | 2021-03-26 | 2021-07-06 | 云南电网有限责任公司电力科学研究院 | Method for monitoring pollution condition based on satellite remote sensing technology |
CN113297527A (en) * | 2021-06-09 | 2021-08-24 | 四川大学 | PM based on multisource city big data2.5Overall domain space-time calculation inference method |
CN113534203A (en) * | 2021-07-12 | 2021-10-22 | 长光卫星技术有限公司 | On-orbit cross radiation calibration method based on AERONET aerosol data |
CN113609940A (en) * | 2021-07-26 | 2021-11-05 | 南通智能感知研究院 | Wide-coverage soil heavy metal remote sensing mixed mapping method |
CN114357885A (en) * | 2022-01-06 | 2022-04-15 | 河南大学 | Photosynthetic effective radiation scattering proportion prediction method fusing multi-source data |
CN115753687A (en) * | 2022-12-23 | 2023-03-07 | 西安工程大学 | Atmospheric pollution remote sensing monitoring method, system and computer readable medium |
CN117150894A (en) * | 2023-08-25 | 2023-12-01 | 中国气象科学研究院 | Correction method, system and device for satellite observation data |
Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103034910A (en) * | 2012-12-03 | 2013-04-10 | 北京农业信息技术研究中心 | Regional scale plant disease and insect pest prediction method based on multi-source information |
CN106933776A (en) * | 2017-03-02 | 2017-07-07 | 宁波大学 | A kind of method that MODIS AOD products missing data is repaired |
-
2018
- 2018-07-13 CN CN201810776814.2A patent/CN109213964B/en active Active
Patent Citations (2)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN103034910A (en) * | 2012-12-03 | 2013-04-10 | 北京农业信息技术研究中心 | Regional scale plant disease and insect pest prediction method based on multi-source information |
CN106933776A (en) * | 2017-03-02 | 2017-07-07 | 宁波大学 | A kind of method that MODIS AOD products missing data is repaired |
Non-Patent Citations (2)
Title |
---|
JUN-JIE HUANG ET AL.: "Fast Image Interpolation via Random Forests", 《IEEE TRANSACTIONS ON IMAGE PROCESSING》 * |
邵琦 等: "基于卫星遥感和气象再分析资料的北京市PM2.5浓度反演研究", 《地理与地理信息科学》 * |
Cited By (32)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN109785569A (en) * | 2019-01-28 | 2019-05-21 | 中科光启空间信息技术有限公司 | A kind of forest fire monitoring method based on 3S technology |
CN110398446A (en) * | 2019-08-01 | 2019-11-01 | 成都信息工程大学 | A kind of PM of combination AOD and meteorologic parameter2.5Concentration approximating method |
CN110610258A (en) * | 2019-08-20 | 2019-12-24 | 中国地质大学(武汉) | Urban air quality refined estimation method and device fusing multi-source space-time data |
CN110610258B (en) * | 2019-08-20 | 2022-05-10 | 中国地质大学(武汉) | Urban air quality refined estimation method and device fusing multi-source space-time data |
CN111078678B (en) * | 2019-12-18 | 2021-03-23 | 中国气象局乌鲁木齐沙漠气象研究所 | Satellite precipitation data correction method based on multi-source information fusion and scale reduction |
CN111078678A (en) * | 2019-12-18 | 2020-04-28 | 中国气象局乌鲁木齐沙漠气象研究所 | Satellite precipitation data correction method based on multi-source information fusion and scale reduction |
CN111104639A (en) * | 2019-12-24 | 2020-05-05 | 福州大学 | Point-surface fused time sequence PM2.5 spatial distribution estimation method |
CN111104639B (en) * | 2019-12-24 | 2022-06-10 | 福州大学 | Point-surface fused time sequence PM2.5 spatial distribution estimation method |
CN111323352A (en) * | 2020-04-09 | 2020-06-23 | 中南大学 | Regional PM2.5 remote sensing inversion model fusing fine particulate matter concentration data |
CN111323352B (en) * | 2020-04-09 | 2021-08-24 | 中南大学 | Regional PM2.5 remote sensing inversion model fusing fine particulate matter concentration data |
CN111737850B (en) * | 2020-05-15 | 2024-04-19 | 中国科学院空天信息创新研究院 | Multi-source satellite AOD fusion method based on uncertainty on pixel scale |
CN111737850A (en) * | 2020-05-15 | 2020-10-02 | 中国科学院空天信息创新研究院 | Multisource satellite AOD fusion method based on uncertainty on pixel scale |
CN112016696A (en) * | 2020-08-14 | 2020-12-01 | 武汉大学 | PM integrating satellite observation and ground observation1Concentration inversion method and system |
CN112131789A (en) * | 2020-09-18 | 2020-12-25 | 中国人民解放军国防科技大学 | Multispectral rainfall detection system and method based on random forest algorithm |
CN112131789B (en) * | 2020-09-18 | 2023-11-21 | 中国人民解放军国防科技大学 | Multispectral precipitation detection system and method based on random forest algorithm |
CN112395808A (en) * | 2020-11-18 | 2021-02-23 | 南京林业大学 | Biomass remote sensing mapping method combining random forest and collaborative kriging |
CN112926625A (en) * | 2020-11-25 | 2021-06-08 | 北方工业大学 | Method for analyzing deviation influence factors of satellite radiation data |
CN112926625B (en) * | 2020-11-25 | 2023-12-22 | 北方工业大学 | Deviation influence factor analysis method for satellite radiation data |
CN112884342A (en) * | 2021-03-10 | 2021-06-01 | 陕西九州遥感信息技术有限公司 | Water color satellite atmospheric layer top radiation product quality evaluation and cross calibration method |
CN112884342B (en) * | 2021-03-10 | 2024-03-12 | 陕西九州遥感信息技术有限公司 | Quality evaluation and cross calibration method for water-color satellite atmospheric roof radiation product |
CN113065098A (en) * | 2021-03-23 | 2021-07-02 | 北京航天创智科技有限公司 | PM2.5 concentration detection method and device based on satellite remote sensing and ground gas system data |
CN113077083B (en) * | 2021-03-26 | 2023-01-20 | 云南电网有限责任公司电力科学研究院 | Method for monitoring pollution condition based on satellite remote sensing technology |
CN113077083A (en) * | 2021-03-26 | 2021-07-06 | 云南电网有限责任公司电力科学研究院 | Method for monitoring pollution condition based on satellite remote sensing technology |
CN113297527B (en) * | 2021-06-09 | 2022-07-26 | 四川大学 | PM based on multisource city big data 2.5 Overall domain space-time calculation inference method |
CN113297527A (en) * | 2021-06-09 | 2021-08-24 | 四川大学 | PM based on multisource city big data2.5Overall domain space-time calculation inference method |
CN113534203B (en) * | 2021-07-12 | 2022-08-23 | 长光卫星技术股份有限公司 | On-orbit cross-radiation calibration method based on AERONET aerosol data |
CN113534203A (en) * | 2021-07-12 | 2021-10-22 | 长光卫星技术有限公司 | On-orbit cross radiation calibration method based on AERONET aerosol data |
CN113609940A (en) * | 2021-07-26 | 2021-11-05 | 南通智能感知研究院 | Wide-coverage soil heavy metal remote sensing mixed mapping method |
CN114357885A (en) * | 2022-01-06 | 2022-04-15 | 河南大学 | Photosynthetic effective radiation scattering proportion prediction method fusing multi-source data |
CN114357885B (en) * | 2022-01-06 | 2024-03-01 | 河南大学 | Photosynthetic effective radiation scattering proportion prediction method fusing multisource data |
CN115753687A (en) * | 2022-12-23 | 2023-03-07 | 西安工程大学 | Atmospheric pollution remote sensing monitoring method, system and computer readable medium |
CN117150894A (en) * | 2023-08-25 | 2023-12-01 | 中国气象科学研究院 | Correction method, system and device for satellite observation data |
Also Published As
Publication number | Publication date |
---|---|
CN109213964B (en) | 2021-08-17 |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN109213964A (en) | A kind of satellite AOD product bearing calibration for merging multi-source feature geographic factor | |
CN102314546B (en) | Method for estimating plant growth biomass liveweight variation based on virtual plants | |
CN114757807B (en) | Multi-mode fused online accounting method for actual emission of atmospheric pollutants | |
CN108872964B (en) | Ginkgo artificial forest canopy closure degree extraction method based on unmanned aerial vehicle LiDAR data | |
CN103424405B (en) | Drought monitoring method based on HJ-1A/1B CCD data | |
CN112016696B (en) | PM integrating satellite observation and ground observation 1 Concentration inversion method and system | |
CN110595968B (en) | PM2.5 concentration estimation method based on geostationary orbit satellite | |
CN110389087B (en) | PM2.5 concentration satellite remote sensing estimation method in polluted weather | |
Chen et al. | Global datasets of leaf photosynthetic capacity for ecological and earth system research | |
CN110414738A (en) | A kind of crop yield prediction technique and system | |
CN107479065A (en) | A kind of three-dimensional structure of forest gap method for measurement based on laser radar | |
CN115170341A (en) | Forestry resource carbon sink accurate measurement and calculation method for correcting small samples by utilizing laser radar | |
CN106022470A (en) | Troposphere delay correction method based on BP-EGNOS fusion model | |
CN115952702A (en) | Forest NEP calculation method based on FORCCHN model and remote sensing data | |
CN115015486B (en) | Carbon emission measurement and calculation method based on regression tree model | |
CN115186437B (en) | Carbon isotope combined assimilation model and construction method of assimilation system for discriminating artificial carbon emission and natural carbon flux areas | |
CN115236005A (en) | Soil heavy metal content inversion method integrating spectral and spatial characteristics | |
CN116337701A (en) | Urban high-resolution aerosol optical thickness inversion method based on double-star networking | |
CN115420690A (en) | Near-surface trace gas concentration inversion model and inversion method | |
CN117219183A (en) | High coverage near ground NO in cloudy rain areas 2 Concentration estimation method and system | |
CN109543911B (en) | Sunlight radiation prediction method and system | |
CN111879915A (en) | High-resolution monthly soil salinity monitoring method and system for coastal wetland | |
CN114974453B (en) | High-coverage reconstruction method for atmospheric carbon dioxide column concentration | |
CN114880933A (en) | Atmospheric temperature and humidity profile inversion method and system for non-exploration-site foundation microwave radiometer based on reanalysis data | |
CN116030350A (en) | Urban green land vegetation carbon sink calculation method and system based on point cloud technology |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
GR01 | Patent grant | ||
GR01 | Patent grant |