CN117611993B - Method for estimating vegetation classification based on remote sensing actual evapotranspiration - Google Patents
Method for estimating vegetation classification based on remote sensing actual evapotranspiration Download PDFInfo
- Publication number
- CN117611993B CN117611993B CN202311491601.2A CN202311491601A CN117611993B CN 117611993 B CN117611993 B CN 117611993B CN 202311491601 A CN202311491601 A CN 202311491601A CN 117611993 B CN117611993 B CN 117611993B
- Authority
- CN
- China
- Prior art keywords
- evapotranspiration
- vegetation
- land
- remote sensing
- classification
- 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.)
- Active
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000011160 research Methods 0.000 claims abstract description 18
- 238000004364 calculation method Methods 0.000 claims abstract description 16
- 238000001704 evaporation Methods 0.000 claims abstract description 12
- 230000008020 evaporation Effects 0.000 claims abstract description 12
- 238000012876 topography Methods 0.000 claims abstract description 8
- 230000000007 visual effect Effects 0.000 claims abstract description 8
- 230000003595 spectral effect Effects 0.000 claims description 16
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 claims description 16
- 238000012549 training Methods 0.000 claims description 12
- 238000012360 testing method Methods 0.000 claims description 9
- 230000006870 function Effects 0.000 claims description 7
- 238000002310 reflectometry Methods 0.000 claims description 5
- 238000004422 calculation algorithm Methods 0.000 claims description 4
- 238000012952 Resampling Methods 0.000 claims description 2
- 238000000605 extraction Methods 0.000 claims description 2
- 238000010801 machine learning Methods 0.000 claims description 2
- 230000007613 environmental effect Effects 0.000 abstract description 9
- 238000007740 vapor deposition Methods 0.000 description 17
- 238000004088 simulation Methods 0.000 description 10
- 238000001228 spectrum Methods 0.000 description 10
- 238000012795 verification Methods 0.000 description 9
- 244000025254 Cannabis sativa Species 0.000 description 7
- 230000008859 change Effects 0.000 description 7
- 230000004907 flux Effects 0.000 description 6
- 238000001556 precipitation Methods 0.000 description 6
- 230000000694 effects Effects 0.000 description 4
- 238000010586 diagram Methods 0.000 description 3
- 238000007637 random forest analysis Methods 0.000 description 3
- 230000003247 decreasing effect Effects 0.000 description 2
- 238000011161 development Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000002156 mixing Methods 0.000 description 2
- 230000008569 process Effects 0.000 description 2
- 239000002689 soil Substances 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 238000006243 chemical reaction Methods 0.000 description 1
- 238000010276 construction Methods 0.000 description 1
- 238000003066 decision tree Methods 0.000 description 1
- 238000000354 decomposition reaction Methods 0.000 description 1
- 230000007423 decrease Effects 0.000 description 1
- 238000011156 evaluation Methods 0.000 description 1
- 230000004927 fusion Effects 0.000 description 1
- 230000005484 gravity Effects 0.000 description 1
- 238000013507 mapping Methods 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 230000005855 radiation Effects 0.000 description 1
- 238000005070 sampling Methods 0.000 description 1
- 239000004576 sand Substances 0.000 description 1
- 230000001932 seasonal effect Effects 0.000 description 1
- 238000010025 steaming Methods 0.000 description 1
- 230000001052 transient effect Effects 0.000 description 1
- 230000007704 transition Effects 0.000 description 1
- 238000011144 upstream manufacturing Methods 0.000 description 1
- 238000009834 vaporization Methods 0.000 description 1
- 230000008016 vaporization Effects 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/188—Vegetation
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N20/00—Machine learning
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/764—Arrangements for image or video recognition or understanding using pattern recognition or machine learning using classification, e.g. of video objects
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V10/00—Arrangements for image or video recognition or understanding
- G06V10/70—Arrangements for image or video recognition or understanding using pattern recognition or machine learning
- G06V10/77—Processing image or video features in feature spaces; using data integration or data reduction, e.g. principal component analysis [PCA] or independent component analysis [ICA] or self-organising maps [SOM]; Blind source separation
- G06V10/774—Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Software Systems (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- Computer Vision & Pattern Recognition (AREA)
- Multimedia (AREA)
- Artificial Intelligence (AREA)
- Health & Medical Sciences (AREA)
- Computing Systems (AREA)
- Evolutionary Computation (AREA)
- Medical Informatics (AREA)
- General Health & Medical Sciences (AREA)
- Databases & Information Systems (AREA)
- Data Mining & Analysis (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Image Processing (AREA)
Abstract
The invention discloses a method for estimating vegetation classification based on remote sensing actual evapotranspiration, which comprises the steps of collecting hydrology, weather, topography, underlying surface and remote sensing image data in a research flow field; constructing an SEBAL actual evapotranspiration estimation model; calculating the actual evaporation of the research river basin on a multidimensional time scale; generating random sample points, identifying the land coverage types of the sample points by using a visual interpretation method, and editing a sample point attribute table; establishing a classification system and a feature input database; and constructing a classifier. The invention considers the evapotranspiration space-time diversity characteristics of different vegetation types, digs out the influence of vegetation distribution on environmental factors, namely vegetation distribution difference information contained in the evapotranspiration space difference, estimates continuous multi-period instantaneous evapotranspiration by constructing an actual evapotranspiration calculation model based on an energy balance principle and introduces the estimated continuous multi-period instantaneous evapotranspiration into a characteristic input database, thereby improving the classification precision of land coverage, in particular the vegetation classification precision of vegetation mixed regions.
Description
Technical Field
The invention relates to the technical field of remote sensing, in particular to a method for estimating vegetation classification based on remote sensing actual evapotranspiration.
Background
The land utilization/land coverage map is a basic data support for scientific researches such as ecological environment, hydrologic simulation, geographic monitoring, socioeconomic performance and the like. The spectrum features of the land coverage types such as water, city, unused land and the like are relatively obvious, the texture features are relatively clear, and the satellite coverage type spectrum features are easy to identify on satellite images. The spectrum features of different vegetation types are similar, the spatial distribution is staggered, certain ambiguity and transition exist between different types, and the different vegetation types are difficult to accurately distinguish in remote sensing classification. Vegetation is one of the major components of earth's surface coverage, which occupies a large specific gravity in land. How to innovate or improve the existing method to improve the vegetation classification precision is still a research difficulty in the remote sensing field.
There are many methods of interpreting land coverage by remote sensing, where supervised classification has been considered a very efficient surface coverage classification method. The current method for eliminating or reducing the influence of mixed pixels so as to improve the vegetation classification precision mainly comprises the following steps: a fusion mixed pixel decomposition algorithm, a classification method based on vegetation indexes, a classification method based on multi-phase information, a classification method of additional auxiliary information and the like. As in prior document 1: zhang Xiaoyu land satellite-8 remote sensing image forest vegetation classification based on random forest model [ J ], university of northeast forestry report, 2016,44 (6), using three vegetation indexes to distinguish between woodland and non-woodland, features of woodland are relatively more obvious and easy to distinguish, but for other types, such as cultivated land and grassland, it is difficult to distinguish using only vegetation indexes. Existing document 2: li w. et al in literature Integrating Google Earth imagery with Landsat data to improve 30-mresolution land cover mapping[J].Remote Sensing of Environment,2020,237:111563.(, SCI, doi No./10.1016/j.rse.2019.111563) improves vegetation classification accuracy by using a vegetation index, but the vegetation index is more suitable for distinguishing vegetation from non-vegetation, and has poor recognition effect on different vegetation types. Existing document 3: sun W, et al, document Spatiotemporal vegetation cover variations associated with climate change and ecological restoration in the Loess Plateau[J].Agricultural and Forest Meteorology,209:87-99.(, SCI, doi 10.1016/j. Agrsurface. 2015.05.002) adds some environmental factors affecting vegetation distribution, mainly including terrain, weather and socioeconomic indexes, as additional inputs to improve classification accuracy based on spectral band information of remote sensing images, but vegetation may be the same under different environmental factors, and vegetation may be different under the same environmental factors.
Existing document 4: sun Na et al, combined with the method for classifying the coverage of vegetation, university of Beijing university, 2022,58 (6): 917-924. In the area of study, the change curve of FVC is used to extract the characteristics of vegetation, and the difference of vegetation is used as the classifying basis, and as can be seen from fig. 2 of this document, the vegetation area is small, and the vegetation coverage change curve of cultivated land and grassland is very different, as can be seen from fig. 4 of this document. Existing document 5: liu Zhen et al, 2003-2014, the analysis of seasonal vegetation coverage space-time change in the middle region of the gateway [ J ]. Geospatial information, 2018,16 (05), it can be seen from this prior art document 5 that fig. 3 (I-11 for grass, I-12 for cultivated land, I-15 for forest land) shows that the vegetation coverage changes for cultivated land and grass are relatively close, and that prior art document 5 does not have the same results as prior art document 4, so that vegetation coverage FVC is essentially a vegetation index that can be used to distinguish vegetation well from non-vegetation areas, and that it is not good to distinguish different vegetation types.
Disclosure of Invention
The invention aims to solve the problems that in the existing remote sensing image interpretation land coverage method, vegetation types relatively close to spectral features are difficult to accurately identify and the vegetation types are only judged to have errors according to environmental factors and vegetation coverage, and provides a method for estimating vegetation classification based on remote sensing actual evapotranspiration, wherein the evapotranspiration spatial difference of different vegetation types is used as feature input to improve the identification precision of different vegetation.
The invention adopts the following technical scheme:
A method for estimating vegetation classification based on remote sensing actual evapotranspiration, comprising the steps of:
Step one: collecting hydrologic, meteorological, topographic, underlying surface and remote sensing image data in a research flow field, sorting site data according to time sequence, cutting space data according to a vector range of the research flow field, and resampling to the same resolution;
step two: constructing an SEBAL actual evapotranspiration estimation model suitable for a research basin based on the data collected in the step one;
step three: calculating to obtain the actual evaporation of the research drainage basin on a multidimensional time scale by using the SEBAL actual evaporation model built in the second step;
step four: generating random sample points, identifying the land coverage types of the sample points by using a visual interpretation method, and editing a sample point attribute table; and according to 7:3, dividing the ratio into training samples and test samples;
Step five: establishing a classification system and a characteristic input database, reasonably setting land coverage type classification of a research river basin, and arranging the database taking five types of information of spectrum bands, spectrum indexes, topography factors, night lights and evapotranspiration as characteristic input;
Step six: constructing a classifier, selecting a proper machine learning algorithm as the classifier, substituting the classifier into the extraction values of the sample points in the training samples in the step four on the characteristic input, and training the optimized classifier; and (3) verifying the sample points in the test sample in the step (IV), and comparing the result calculated by the classifier with the real land coverage type on the sample points in the test sample obtained by visual interpretation in the step (IV), so that the classification precision can be obtained, and the precision is evaluated.
Further, the main calculation of the SEBAL model in the second step is based on the energy balance principle:
λE＝Rn-G-H
where λ is the latent heat of vaporization, E is the actual amount of evapotranspiration, R n is the net radiant flux, G is the soil heat flux, and H is the sensible heat flux.
Preferably, the land cover types described in step four and step five include cultivated land, woodland, grassland, town land, water, and unused land.
Preferably, the characteristic input database in the fifth step specifically includes 6 band spectrum data of Landsat8, that is, B2 blue band, B3 green band, B4 red band, B5 near infrared band, B6 short wave infrared band 1 and B7 short wave infrared band 2,3 spectrum indexes, that is, NDVI, MNDWI and NDBI, a night light index, elevation and slope 2 topography factors, and 21 instantaneous vapor emission total 33 characteristics.
Preferably, 3 spectral indexes of Landsat8 are calculated as follows:
NDVI＝(B5-B4)/(B5+B4)
MNDWI＝(B3-B6)/(B3+B6)
NDBI＝(B6-B5)/(B6+B5)。
Preferably, the objective function employed in step six to evaluate classification accuracy includes, but is not limited to, user accuracy UA, producer accuracy PA, F1 score, overall accuracy OA, and Kappa coefficient.
Further, the calculation formulas of the objective function user precision UA, the producer precision PA, the F1 score, the overall precision OA, and the Kappa coefficient are as follows, respectively:
PA＝nni/yyi
Kappa＝(OA-Pe)/(1-Pe)
Where nn i is the number of pixels of the whole image correctly divided into the i-th type of pixels, xx i,yyi is the total number of pixels of the whole image divided into the i-th type and the i-th type of real reference total number of pixels by the classifier, and Pe is the sum of products of the number of various real pixels and the number of predicted samples, and divided by the square of the total number of samples.
Preferably, the hydrology, weather, topography, underlying surface and remote sensing image data in the research flow field collected in the first step specifically include: runoff data of hydrologic stations in a river basin, meteorological data of surrounding meteorological stations in the river basin, a digital elevation data set, a vegetation index, a ground surface temperature, a ground surface reflectivity data set, a land water reserve data set, a remote sensing image and a night light data set are researched.
The invention has the beneficial effects that:
The existing documents 1 and 2 are difficult to accurately identify vegetation types which are relatively close to spectral features and are mixed by spatial distribution. The invention considers the evapotranspiration time-space diversity characteristics of different vegetation types, estimates continuous multi-period instantaneous evapotranspiration by constructing an actual evapotranspiration calculation model based on an energy balance principle and introduces the continuous multi-period instantaneous evapotranspiration into a characteristic input database, thereby improving the classification precision of land coverage, in particular the vegetation classification precision of vegetation mixed regions.
The existing document 3 only judges that the vegetation type has an error according to the environmental factors, because the vegetation may be the same under different environmental factors, and the vegetation may be different under the same environmental factors. The vegetation climatic features of the cultivated land and the grasslands in the prior documents 4 and 5 are overlapped to a certain extent, and are difficult to distinguish. The invention digs the influence of vegetation distribution on environmental factors, takes the effect of vegetation on hydrology (the feedback effect of vegetation growth on atmospheric moisture-evapotranspiration) as a classification basis, and the number and the change rule of evapotranspiration with time are different on different vegetation types, namely vegetation distribution difference information is contained in the evapotranspiration space difference. In general, a woodland is a region with the largest evapotranspiration, and for two types of difficult distinction, the evapotranspiration of the cultivated land is larger than that of the grass land in the crop growth and development period, the evapotranspiration of the grass land in the crop withering period is larger than that of the cultivated land, and the grass land and the cultivated area can be well distinguished through the number and the change rule of the instantaneous evapotranspiration in the whole year, so that the method is applicable to all research type areas.
The invention demonstrates that the characteristic input of the evapotranspiration is introduced to improve the vegetation classification precision through theory and embodiments, and has certain theoretical value and practical significance for realizing the multi-type large-scale rapid identification of vegetation in a mixed area or a high heterogeneous area of the perfect vegetation.
Drawings
FIG. 1 is a graph showing the results of transient evapotranspiration over the period of the period 2020 for the Hurricane basin modeled by the SEBAL model of example 1;
FIG. 2 is a simulation result of verifying SEBAL based on water balance on the sub-basin scale in example 1;
FIG. 3 is a graph showing the classification result of land coverage estimated based on actual evapotranspiration in example 1;
FIG. 4 is a graph of accuracy versus bar graph for different input schemes in example 1;
FIG. 5 is a comparison of land cover classification results estimated based on actual evapotranspiration in example 1 with existing products;
FIG. 6 shows the degree of difference and importance of feature input in embodiment 1;
fig. 7 is the difference in instantaneous evaporation under three vegetation types in example 1.
Detailed Description
The technical scheme of the invention is further described below with reference to the attached drawings and specific embodiments.
Example 1
This example takes the example of an Erdos basin, which is located upstream and downstream of the yellow river basin, with an area of about 38 ten thousand square kilometers. The Erdos basin vegetation types have a zonal distribution rule from southeast to northwest, are influenced by terrain, and the regional vegetation landscape pattern presents high spatial heterogeneity of arbor-shrub-grass mixing. The terrain breaking and vegetation site conditions under the natural attribute of the Erdos basin have high spatial heterogeneity, and the existing remote sensing inversion products are difficult to accurately reflect the vegetation spatial pattern under the arbor-shrub-grass mixed pattern in addition to human policy intervention such as ecological engineering construction.
The present embodiment constructs and validates the SEBAL model based on the MODIS vegetation index, surface temperature, surface reflectivity dataset provided by the united states space agency, the GDEMV digital elevation dataset provided by the japan space agency JAXA, the daily meteorological data (air temperature, wind speed and precipitation) provided by the chinese meteorological network, the hydrological station runoff data provided by the chinese hydrological annual survey, and the GRACE land water reserve dataset provided by the spatial research center CSR. And classifying land coverage of the Erdos basin 2020 by utilizing the Sentinel-2 and Landsat8 remote sensing images provided by the global scale remote sensing cloud computing platform GEE, the VIIRS night light data set provided by NOAA of the national ocean and atmosphere administration, the digital elevation data and the SEBAL model computing result. And comparing the classification precision under different input schemes, and downloading the comparison classification precision of the existing land cover/land utilization products. The products used for comparison here are of 6 classes, including: CNLUCC products (1000 m) provided by the national academy of sciences and resource research, MCD12Q1 products (500 m) of MODIS, global Land products (30 m) provided by the national basic geographic information center, CLCD products (30 m) and CRLC products (10 m) provided by the university of Wuhan, and GLC_FCS products (30 m) provided by the national academy of sciences' space-sky information innovation. The vegetation classification precision improving method based on remote sensing actual evapotranspiration estimation provided by the invention has higher precision.
The specific method of the embodiment is as follows:
Step one: downloading digital elevation data of MODIS vegetation index MOD13Q1, ground surface temperature MOD11A2, ground surface reflectivity MCD43A3 and GDEMV2, GRACE land water reserve data, sentinel-2 images, landsat8 remote sensing images, VIIRS night light data sets, downloading meteorological data of the inner part of a current area and surrounding meteorological sites, including wind speed, air temperature, precipitation and the like, and extracting hydrological station runoff data provided by Chinese hydrological annual survey. The air temperature and the air speed are the meteorological input calculated by the SEBAL model, and the precipitation and the runoff are the factors needed when the water balance method is adopted to verify the evapotranspiration result calculated by the SEBAL model. The site data are sorted in time sequence, the spatial data are cut according to the Hutterdos basin vector boundary, and resampled to the same resolution.
Step two: and (3) selecting an SEBAL actual evapotranspiration estimation model, inverting earth surface characteristic parameters such as earth surface clear sky albedo, earth surface emissivity, earth surface temperature and the like by utilizing the data collected and processed in the step (I), calculating instantaneous net radiation flux and soil heat flux when a satellite passes through an environment by taking a single pixel as a unit, iteratively correcting aerodynamic impedance by Monin-Obukhov to obtain stable region sensible heat flux, and finally constructing the SEBAL actual evapotranspiration estimation model applicable to the Erdos basin based on an energy balance equation.
Step three: and (3) utilizing the actual evapotranspiration estimation model built in the second step, and firstly, simulating the instantaneous evapotranspiration of the Erdos basin.
Fig. 1 shows the results of instantaneous vapor deposition, i.e., actual vapor deposition, over the period of 2020 for the erdos basin simulated by the SEBAL model. The annual instantaneous vapor-emission generally shows a trend of increasing and then decreasing with time, and the 4-9 month instantaneous vapor-emission is relatively large. The instantaneous vapor emission is not more than 0.4mm/h in 1-3 months, the average vapor emission is 0.5-0.7mm/h in 4-9 months, the local area can reach more than 1mm/h, and the vapor emission is gradually reduced in 10-12 months. Spatially, the instantaneous evapotranspiration decreases from southeast to northwest, and the evapotranspiration in the loess region in the south is generally greater than that in the sand region in the north.
Secondly, because the calculation of the instantaneous vapor deposition lacks the verification of the actual measurement data, the instantaneous vapor deposition is subjected to scale conversion, the daily vapor deposition is calculated, and the daily vapor deposition and the annual vapor deposition are respectively counted. The statistics on month and year scales is used for verification, and the verification of whether the calculated instantaneous vapor emission is accurate or not is accurate in month and year, namely the instantaneous vapor emission is considered to be accurate.
Finally, selecting a plurality of small watercourses in a research area, constructing a water balance equation by utilizing precipitation, runoff, water reserves and the like above a control station to calculate a theoretical true value of the vapor emission, comparing the theoretical true value with the month vapor emission and year vapor emission values obtained by calculation, analyzing and verifying the accuracy of a simulation result, and calculating a correlation coefficient r and a relative error RE of an objective function, wherein the formula is as follows:
RE＝(ETsim-ETreal)/ETreal
In the method, in the process of the invention, The theoretical true value and the model simulation value of the evapotranspiration in the j-th period are respectively/> The average values of the theoretical true value and the model simulation value of the evapotranspiration are respectively.
The reason for selecting the small watershed is that no accurate data are observed at present, and the water balance method is mostly adopted, based on the watershed scale, by utilizing the actually measured data of precipitation and runoff observed by the watershed. Because the measured data for runoff is only available at each drainage basin outlet.
Fig. 2 shows the verification result of the SEBAL model on the sub-drainage basin scale, where (a) is the geographical location of the sub-drainage basin and the control station, (b) is a scatter diagram of month-scale vapor deposition theoretical truth values and simulation values, (c) is a scatter diagram of year-scale vapor deposition theoretical truth values and simulation values, and in fig. 2, (b) and (c) are scatter diagrams, which can be explained as follows: for each point in the graph, the abscissa corresponding to that point represents the theoretical true value, and the ordinate corresponding to that point represents the simulated value. In fig. 2, (d) shows annual emission and relative error distribution of each sub-basin (wherein Esim represents the simulated value, eobs represents the theoretical true value). The 44 sub-basins basically cover the range of different areas with different characteristics of the Erdos basin, and the verification result can represent the simulation precision of the whole area. On a month scale, the correlation coefficient r between the theoretical true value of the evapotranspiration and the evapotranspiration simulation value of the SEBAL model is 0.63 on average in all sub-watershed, and 0.78 on an annual scale. On each sub-basin, the correlation coefficient of the theoretical true value of the evapotranspiration and the evapotranspiration simulation value of the SEBAL model is mostly above 0.7 except that the individual basin is lower, and the evapotranspiration of each sub-basin in 2020 is basically within 350-700 mm in magnitude. In 2020, the actual vapor deposition is not different from the simulated vapor deposition on the sub-watershed, and the error is mostly within 10%, so that the verification result of the multi-scale multi-watershed shows that the simulated vapor deposition of the SEBAL model has reliability.
Step four: sample points are randomly selected in the Erdos basin based on a 2m high-resolution Sentinel-2 image according to a layered sampling method, the land coverage type is identified by a visual interpretation method, a sample point attribute table is edited, wherein 555 sample points are selected in total, and 121, 107, 126, 63, 61 and 77 sample points are respectively selected in cultivated land, woodland, grassland, urban land, water body and unused land. And according to 7:3, dividing the ratio into training samples and test samples;
Step five: six classification systems of farmland, woodland, grassland, town land, water body and unused land are arranged in consideration of the distribution of main underlying surfaces in the Erdos basin. And constructing a characteristic input database, wherein the characteristic input database comprises five types of information including spectral bands, spectral indexes, topography factors, night light and evapotranspiration. The device specifically comprises 6 wave band spectrum data (B2 blue wave band, B3 green wave band, B4 red wave band, B5 near infrared wave band, B6 short wave infrared wave band 1, B7 short wave infrared wave band 2) of Landsat8, 3 spectrum indexes (NDVI, MNDWI, NDBI), 2 topography factors (elevation and gradient), 1 night lamplight index and 33 instantaneous evaporation and emission characteristics. Wherein for Landsat8, the calculation formula of the spectral index is as follows:
NDVI＝(B5-B4)/(B5+B4)
MNDWI＝(B3-B6)/(B3+B6)
NDBI＝(B6-B5)/(B6+B5)
In the calculation process of the evapotranspiration, an aerodynamic method is needed to carry out iterative calculation, but under the influence of weather and cloud cover, the convergence solution (4-26 and 8-28 respectively) cannot be obtained by the instantaneous evapotranspiration of two days, the results of the two days in fig. 1 are shown as interpolation results, no new information is basically entered, and in practical application, only 21 instantaneous evapotranspiration which can obtain the convergence result are used.
The feature input database constructed here is not just for 555 samples, but for the whole area. That is, these 33 feature inputs, each input having specific data at each location over the entire area. Only when training the classifier in step six, the values of these feature inputs are extracted to the sample points, and finally the classifier is applied to the whole region, so that the whole feature inputs of the region are required.
Step six: constructing a classifier, selecting a random forest algorithm as the classifier, extracting feature input values at the positions of sample points by using the sample point data generated in the step four and the feature input database established in the step five, and training the feature input values as the input of the classifier, specifically, extracting values of each sample point on 33 feature inputs, wherein for 70% of sample points, the training data quantity of 388 training sample points is (388×33=12804).
Verification and accuracy assessment were performed with the remaining 30% of the sample points (555 x 0.3 = 167 test samples), which 30% were not used to train the classifier and therefore could be used for accuracy verification. At the moment, the land coverage classification result of the remaining 30% of sample points on the area is obtained through the classifier calculation, and the classification precision can be obtained by comparing the result of the classifier calculation with the real land coverage type of the 30% of sample points obtained through visual interpretation in the step four. The accuracy is evaluated, and the adopted objective function comprises User Accuracy (UA), producer Accuracy (PA), F1 fraction, overall Accuracy (OA), kappa coefficient and the like, and the calculation formula is as follows:
UA＝nni/xxi,PA＝nni/yyi
Kappa＝(OA-Pe)/(1-Pe)
Where nn i is the number of pixels of the whole image correctly divided into the i-th type of pixels, xx i,yyi is the total number of pixels of the whole image divided into the i-th type and the i-th type of real reference total number of pixels by the classifier, and Pe is the sum of products of the number of various real pixels and the number of predicted samples, and divided by the square of the total number of samples.
After training the classifier by using 70% of sample points (555×0.7=388), applying the trained classifier and input to the whole area together for calculation, obtaining a classification result of the whole area, and then passing through 30% of sample point tests. As shown in fig. 3, the vegetation classification result by using the actual evapotranspiration is presented by the method, and the classification result of the whole area is compared with the visual interpretation result in the step four.
In the whole, OA and Kappa of classification are respectively 0.930 and 0.914, PA and UA of six types of land cover types are above 85%, F1 is above 90%, and classification accuracy is relatively high. The maximum grass percentage of the erdos basin in 2020 is about 46.1%. Secondly, the cultivated land and the grassland are 21.3 percent and 16.5 percent respectively, the unused land accounts for 11.2 percent, and the part is mainly desert. The land used in towns is about 3.9 percent, and the water body is less than 1 percent.
Example 2
Using the classifier in the step six, respectively simulating and calculating the land coverage classification results under the following 5 input schemes (S1-S5) and comparing the precision, wherein the input of the scheme 1 (S1) is only 6 spectral band information of Landsat images; the input of scheme 2 (S2) increases three spectral indices in addition to the spectral band in S1; the input of the scheme 3 (S3) is to increase the digital elevation information on the basis of the S2, wherein the digital elevation information comprises the elevation and the terrain gradient; the input of the scheme 4 (S4) is to increase night light data on the basis of S3; scheme 5 (S5) integrates all input information, i.e., increases the number of consecutive sets of instantaneous vapor-emission information over the year on the basis of S4.
Fig. 4 is a graph of accuracy versus bar graph for different input schemes in example 1, 4a is an overall accuracy versus for different input schemes, and 4b is a Kappa coefficient versus for different input schemes. For the overall accuracy OA, S1 (0.676) is close to S2 (0.665), S3 (0.730) is close to S4 (0.733), S5 is highest (0.807), and Kappa is similar to OA. Indicating that increasing the spectral index on the basis of S1 or increasing the night light on the basis of S3 has little effect on the result. After DEM addition, the accuracy of S3 is greatly improved, which demonstrates that elevation and slope play a great role in vegetation classification as described above. In particular, after adding the instant steaming, the accuracy of S5 is obviously improved, and S5 combines all the characteristics as the input of the classifier. S5 improves OA by 7% -14% and Kappa by 9% -17%. For different land cover types, relative to S1-S4, S5 increases F1 by 5% -15%,0% -16%,3% -20%,10% -30%,10% -13%,1% -10% respectively, and the precision improvement degree on the different land cover types is respectively referred to as for 6 land cover types: the classification precision of the cultivated land, the woodland, the grassland, the urban land, the water body and the unused land, such as S5, is improved by 5-15 percent compared with the classification precision of the cultivated land by the former four schemes (S1-S4). The results prove that the S5 not only remarkably improves the classification precision of cultivated land, woodland and grassland, but also proves that the instant evaporation is an effective input characteristic.
Example 3
The classification result of the method is compared with the existing Land cover products (CNLUCC products, MCD12Q1, global Land products, CLCD products, CRLC products and GLC_FCS products) with various resolutions, and the classification accuracy is represented to be higher as the objective function is closer to 1 through the calculation result of various objective functions in the step six.
Fig. 5 is a comparison of land cover classification results estimated based on actual evapotranspiration in example 1 with existing products. 5a is CNLUCC product, 5b is MCD12Q1 product, 5c is Globe Land product, 5d is CLCD product, 5e is glc_fcs product, 5f is CRLC product, 5g is the result AETLULC of the proposed method of the invention; from (a) to (f), the spatial resolutions of the six data were 1km,500m,30m,10 m, respectively. The resolution of the product obtained by the method proposed by the present invention is 30m. The land cover of CNLUCC products is more dispersed, the land cover product of the MCD12Q1 is concentrated in the overall land feature type, particularly the grasslands and the cultivated lands are gathered in a large range, and the actual situation of vegetation mixing cannot be represented. The CLCD product, CRLC product, glc_fcs product are visually closer to the classification result (AETLULC) of the present invention, while there is a slight difference in unused land, mainly cultivated land, grassland and desert. Compared with six existing products, AETLULC has 32.7%,43.5%,21.0%,15.8%,10.9% and 4.9% improvement in F1 on grasslands; the F1 on the cultivated land is increased by 28.1 percent, 18.1 percent, 10.8 percent, 2.4 percent, 5.2 percent and 2.1 percent; the F1 on the forest land is improved by 29.2%,24.1%,8.2%,12.0%,11.2% and 1.0%.
Example 4
In order to verify the rationality and the advancement of the method, the KL divergence and JS divergence are adopted to evaluate the difference degree of the feature input under different vegetation types, the larger the difference degree is, the more effective the feature input is in classification, the importance of the feature input is evaluated by adopting the importance of FI, and the larger the FI is, the higher the importance of the input feature in classification is. The calculation formula is as follows:
Wherein F (X) and P (X) represent probability distribution over a discrete space X, KL (F||P), JS (F|P) are respectively the KL divergence and JS divergence of F (X) to P (X), ntree is the number of decision trees in a random forest, errOOB t1 represents the out-of-bag error after the characteristic input value in the t-th tree changes, and errOOB t2 represents the out-of-bag error of the normal characteristic input value in the t-th tree.
Fig. 6 is a histogram of feature input variability and importance in embodiment 1, where a is KL and JS divergences of various feature inputs under different land coverage types, and b is the importance ranking of various feature inputs. For different land cover types, the input features show a large difference: b6, B7, B4, NDVI, NDBI, B2, B3, ET inst21,ETinst3,ETinst10, etc., the input feature importance ranking is as follows: NDVI, B5, MNDWI, B3, B2, B6, ELEVATION, B7, B4, ET inst12,ETinst21. The reflectivity information of the spectrum band can represent the difference of vegetation types to a great extent, and the additional information also displays the difference of vegetation types to a certain extent, so that the input of additional information such as instantaneous evaporation and the like can assist in improving the classification precision.
Fig. 7 shows the difference of instantaneous vapor deposition under three vegetation types in example 1, a is the variation of the instantaneous vapor deposition in the years, and b is the variation of the instantaneous vapor deposition in different regions of the precipitation. For the instantaneous vapor-deposition of different dates throughout the year 2020, the vapor-deposition of three vegetation types of cultivated land, woodland and grassland shows the change rule of increasing firstly and then decreasing. In number, the evapotranspiration of woodlands is generally larger than that of grasslands and cultivated lands, and the evapotranspiration of grasslands and cultivated lands is relatively close. The average value of the instantaneous evaporation in the forest land, cultivated land and grassland is 0.56mm,0.42mm and 0.41mm respectively, the evaporation in the autumn and winter is slightly larger than that in the cultivated land, and the cultivated land is larger than that in the grass land in summer and in the last spring, which is related to the regional agricultural planting system. The different periods of evaporation exhibit different characteristics over different vegetation types. The method is refined on each dewatering area, the evapotranspiration of the forest land is maximum and can be obviously distinguished from cultivated lands and grasslands, and the evapotranspiration of the cultivated lands and the grasslands is similar, but the distinction degree is increased along with the increase of dewatering. Through verification and evaluation by various methods, the accuracy and feasibility of the method for improving the classification accuracy by introducing instantaneous evapotranspiration can be proved.
Under the background of a changing environment, the method provided by the invention can provide technical support for rapid and accurate identification of vegetation in a high heterogeneous area, and provide theoretical basis in application of future regional homeland resource planning, ecological environment protection and the like.
The results obtained in the prior document 4 and the prior document 5 are different, and it is not good to distinguish different vegetation types. The vegetation features of the cultivated land and the grasslands are overlapped to a certain extent, and are difficult to distinguish, for example, crops which are mature for one year are quite close to the growth and development period of the grasslands, the crops start to develop in spring, grow rapidly in summer, and are withered in autumn and winter, and are difficult to distinguish by virtue of vegetation coverage only, namely, the same vegetation coverage or similar weather rules can be of different land coverage types.
It should be noted that, the above specific embodiments and examples are preferred embodiments of the present invention, and the scope of the present invention is not limited thereto, and any equivalent changes or equivalent modifications made on the basis of the technical solution according to the technical idea of the present invention still fall within the scope of the technical solution of the present invention.
Claims (5)
1. A method for estimating vegetation classification based on remote sensing actual evapotranspiration, which is characterized by comprising the following steps:
Step one: collecting hydrologic, meteorological, topographic, underlying surface and remote sensing image data in a research flow field, sorting site data according to time sequence, cutting space data according to a vector range of the research flow field, and resampling to the same resolution;
step two: constructing an SEBAL actual evapotranspiration estimation model suitable for a research basin based on the data collected in the step one;
step three: calculating to obtain the actual evaporation of the research drainage basin on a multidimensional time scale by using the SEBAL actual evaporation model built in the second step;
step four: generating random sample points, identifying the land coverage types of the sample points by using a visual interpretation method, and editing a sample point attribute table; and according to 7:3, dividing the ratio into training samples and test samples;
Step five: establishing a classification system and a characteristic input database, reasonably setting land coverage type classification, and arranging a database which takes five types of information of spectral bands, spectral indexes, topography factors, night lights and evapotranspiration as characteristic input;
Step six: constructing a classifier, selecting a proper machine learning algorithm as the classifier, substituting the classifier into the extraction value of the training sample point in the fourth step on each characteristic input, and training the optimized classifier; verifying the sample points in the test sample in the step four, comparing the result calculated by the classifier with the real land coverage type on the test sample points obtained by visual interpretation in the step four, so as to obtain classification precision and evaluate the precision;
The objective function adopted for evaluating the classification accuracy comprises a user accuracy UA, a producer accuracy PA, an F1 score, an overall accuracy OA and a Kappa coefficient; the calculation formulas of the user precision UA, the producer precision PA, the F1 fraction, the overall precision OA and the Kappa coefficient are respectively as follows:
PA＝nni/yyi
Kappa＝(OA-Pe)/(1-Pe)
Where nn i is the number of pixels of the whole image correctly divided into the i-th type of pixels, xx i,yyi is the total number of pixels of the whole image divided into the i-th type and the i-th type of real reference total number of pixels by the classifier, and Pe is the sum of products of the number of various real pixels and the number of predicted samples, and divided by the square of the total number of samples.
2. The method of estimating vegetation classification based on remote sensing actual evapotranspiration as claimed in claim 1, wherein the characteristic input database in the fifth step specifically includes 6 band spectral data of Landsat8, namely, 2,3, 4, 5, 1 and 7 short wave infrared band, 3 spectral indexes of NDVI, MNDWI and NDBI, a night light index, 2 topography factors of elevation and slope, and 33 total characteristics of 21 instantaneous evapotranspiration.
3. A method for estimating vegetation classification based on remote sensing actual evapotranspiration as claimed in claim 2 wherein 3 spectral indexes of Landsat8 are calculated as follows:
NDVI＝(B5-B4)/(B5+B4)
MNDWI＝(B3-B6)/(B3+B6)
NDBI＝(B6-B5)/(B6+B5)。
4. The method of estimating vegetation classification based on remote sensing actual evapotranspiration as claimed in claim 1 wherein the land cover types in step four and step five include cultivated land, woodland, grassland, town land, water and unutilized land.
5. The method of estimating vegetation classification based on remote sensing actual evapotranspiration as claimed in claim 1, wherein the hydrologic, meteorological, topographic, underlying surface and remote sensing image data in the research flow area collected in the step one specifically comprises: runoff data of hydrologic stations in a river basin, meteorological data of surrounding meteorological stations in the river basin, a digital elevation data set, a vegetation index, a ground surface temperature, a ground surface reflectivity data set, a land water reserve data set, a remote sensing image and a night light data set are researched.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311491601.2A CN117611993B (en) | 2023-11-09 | 2023-11-09 | Method for estimating vegetation classification based on remote sensing actual evapotranspiration |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202311491601.2A CN117611993B (en) | 2023-11-09 | 2023-11-09 | Method for estimating vegetation classification based on remote sensing actual evapotranspiration |
Publications (2)
| Publication Number | Publication Date |
|---|---|
| CN117611993A CN117611993A (en) | 2024-02-27 |
| CN117611993B true CN117611993B (en) | 2024-05-14 |
Family
ID=89957044
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202311491601.2A Active CN117611993B (en) | 2023-11-09 | 2023-11-09 | Method for estimating vegetation classification based on remote sensing actual evapotranspiration |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN117611993B (en) |
Families Citing this family (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117973231A (en) * | 2024-03-29 | 2024-05-03 | 中国科学院、水利部成都山地灾害与环境研究所 | Mountain vegetation coverage field sampling point automatic optimization layout method and system |
Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551459A (en) * | 2008-10-15 | 2009-10-07 | 北京天宏金睛信息技术有限公司 | Method for monitoring regional evapotranspiration on the basis of remote sensing |
| CN106771073A (en) * | 2016-12-28 | 2017-05-31 | 中国科学院地理科学与资源研究所 | A kind of method that soil and vegetation evapotranspiration are estimated based on end member information model |
| CN110990988A (en) * | 2018-09-28 | 2020-04-10 | 华北水利水电大学 | Method for analyzing evapotranspiration influence factors and establishing multiple regression model |
| CN115376016A (en) * | 2022-08-16 | 2022-11-22 | 水利部交通运输部国家能源局南京水利科学研究院 | Actual rice field irrigation area identification method based on combination of vegetation water index and evapotranspiration |
-
2023
- 2023-11-09 CN CN202311491601.2A patent/CN117611993B/en active Active
Patent Citations (4)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN101551459A (en) * | 2008-10-15 | 2009-10-07 | 北京天宏金睛信息技术有限公司 | Method for monitoring regional evapotranspiration on the basis of remote sensing |
| CN106771073A (en) * | 2016-12-28 | 2017-05-31 | 中国科学院地理科学与资源研究所 | A kind of method that soil and vegetation evapotranspiration are estimated based on end member information model |
| CN110990988A (en) * | 2018-09-28 | 2020-04-10 | 华北水利水电大学 | Method for analyzing evapotranspiration influence factors and establishing multiple regression model |
| CN115376016A (en) * | 2022-08-16 | 2022-11-22 | 水利部交通运输部国家能源局南京水利科学研究院 | Actual rice field irrigation area identification method based on combination of vegetation water index and evapotranspiration |
Non-Patent Citations (2)
| Title |
|---|
| 基于HJ-1B数据和SEBAL模型的陆面蒸散发遥感估算;张楠楠;王文;王胤;;地理空间信息;20131028(第05期);第84-88页 * |
| 长江流域实际蒸散发的遥感估算及时空分布研究;尹剑;邱远宏;欧照凡;;北京师范大学学报(自然科学版);20200215(第01期);第90-99页 * |
Also Published As
| Publication number | Publication date |
|---|---|
| CN117611993A (en) | 2024-02-27 |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Batelaan et al. | GIS-based recharge estimation by coupling surface–subsurface water balances | |
| Kiptala et al. | Land use and land cover classification using phenological variability from MODIS vegetation in the Upper Pangani River Basin, Eastern Africa | |
| Wang et al. | Summer maize growth under different precipitation years in the Huang-Huai-Hai Plain of China | |
| CN107194160A (en) | Recover analysis method in a kind of basin Marsh Wetland space | |
| CN117611993B (en) | Method for estimating vegetation classification based on remote sensing actual evapotranspiration | |
| Ejikeme et al. | Enhancing food security in Anambra state, Nigeria using remote sensing data | |
| Zawawi et al. | Landform classification for site evaluation and forest planning: Integration between scientific approach and traditional concept | |
| Caocao et al. | Analysis on Jinghe watershed vegetation dynamics and evaluation on its relation with precipitation | |
| Girma | Potential impact of climate and land use changes on the water resources of the Upper Blue Nile Basin | |
| Jaw et al. | Evaluation for Moroccan dynamically downscaled precipitation from GCM CHAM5 and its regional hydrologic response | |
| Mücher et al. | Development of a consistent methodology to derive land cover information on a European scale from remote sensing for environmental monitoring; the PELCOM report | |
| Yimer et al. | Seasonal effect on the accuracy of Land use/Land cover classification in the Bilate Sub-basin, Abaya-Chamo Basin, Rift valley Lakes Basin of Ethiopia. | |
| Compaoré | The impact of savannah vegetation on the spatial and temporal variation of the actual evapotranspiration in the Volta Basin, Navrongo, Upper East Ghana | |
| Ueyama | Classification of recent studies by method type for surface air temperature map development and estimation of daily temperature using a radiative cooling scale | |
| Zhao et al. | Risk assessment for comprehensive meteorological disaster of winter wheat based on multisource gridded data: A case study in the upper Huaihe River basin, China | |
| Zhang et al. | Analysis of influence mechanism of spatial distribution of incoming solar radiation based on DEM | |
| Sailaja et al. | Regional rice yield estimation by integration of spatial technologies and crop model | |
| Newton | Remote sensing based estimates of reference evapotranspiration for the southwest region of Bangladesh | |
| Paiboonvorachat | Using remote sensing and GIS techniques to assess land use/land cover changes in the Nan Watershed, Thailand | |
| Jande et al. | Prediction of Land Use Change in Katsina-Ala through a Geospatial Approach | |
| Poméon | Evaluating the Contribution of Remote Sensing Data Products for Regional Simulations of Hydrological Processes in West Africa using a Multi-Model Ensemble | |
| Zushi | Regional estimation of Japanese cedar (Cryptomeria japonica D. Don) productivity by use of digital terrain analysis | |
| Baron Ruiz | Calibration and Validation of a SWAT model for the quantification of water provision ecosystem service for the Conservation Area of Guanacaste | |
| Bangelesa | Impacts of climate variability and change on Maize (Zea mays) production in tropical Africa | |
| Ndikubwimana | The analysis of the agricultural crop water productivity behaviours in semi-arid and humid regions for irrigated and rainfed areas, case studies Sudan and Rwanda |
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 |