CN113807285A

CN113807285A - GEE-based rapid automatic extraction method for long-time-sequence large-scale lake water body

Info

Publication number: CN113807285A
Application number: CN202111118564.1A
Authority: CN
Inventors: 胡贵锋; 高扬; 赵宏辉
Original assignee: Xingtu Zhihua Xi'an Digital Technology Co ltd
Current assignee: Xingtu Zhihua Xi'an Digital Technology Co ltd
Priority date: 2021-09-24
Filing date: 2021-09-24
Publication date: 2021-12-17

Abstract

The invention discloses a dynamic long-time-sequence rapid automatic monitoring and drawing method for the current situation of large-scale multi-lake range in China by means of a Google Earth Engine (GEE) platform. The method comprises the following steps: obtaining massive Landsat and Sentinel image data of a lake region from a GEE cloud platform, wherein the massive Landsat and Sentinel image data comprise remote sensing reflectivity and remark information data, screening effective matching data from the data, establishing a long-time sequence automatic water body range inversion model suitable for land reflectivity by using a water body extraction algorithm, and applying the model to GEE remote sensing products to realize rapid mapping of water body ranges with different space-time scales. The method can realize the rapid drawing of the lake water body range in a large area, a long-time sequence and rapid automation, and provides data support for the long-term monitoring and comparative analysis of the large-area lake water body range.

Description

GEE-based rapid automatic extraction method for long-time-sequence large-scale lake water body

Technical Field

The invention relates to the field of satellite remote sensing technology and application thereof, in particular to a GEE-based automatic extraction method for long-time-sequence large-scale lake water body.

Background

Inland lakes are sensitive indicators of regional climate and environmental changes, and the changes in area and water level can objectively reflect the water balance process in arid inland regions, thereby maintaining the stability of regional ecosystems. In recent years, the water body area in China is more than 10km²231 of 635 lakesThe total shrinkage area of the lake reaches 18 percent, the lake protection is not slow, and long-time serial national-scale lake water body range monitoring can provide data reference for the lake protection.

The traditional lake monitoring is limited by manpower and financial resources, the basic conditions and related information of lakes across the country cannot be comprehensively mastered, the lake information is incomplete, the indexes of dynamic information of lakes are incomplete, and the dynamic development of lakes across the country and the river management conditions cannot be known in time. The remote sensing lake monitoring has the characteristics of low cost, wide range, high timeliness and strong intuition, and becomes a powerful technical means for lake monitoring in recent years. However, remote sensing monitoring has the characteristics of large data volume, complex and tedious processing flow, high requirement on computer storage space and processing performance, and once long-time and large-range lake monitoring is involved, the remote sensing lake monitoring based on local processing has the disadvantages of long processing time and many repetitive work, so that the remote sensing lake monitoring based on local processing has defects in the aspects of convenience, automation, timeliness and the like.

Disclosure of Invention

The invention aims to solve the problems of difficult data acquisition, low precision, small monitoring range, low data processing efficiency, low automation level and the like, and provides a quick automatic drawing method for the water body range of a long-time-sequence large-scale lake based on a GEE cloud computing platform so as to realize national-scale, long-time-sequence and efficient automatic water body range monitoring.

According to the invention, an API (application programming interface) provided by GEE (generic object analyzer) is utilized, a regional multi-source long-time-sequence remote sensing image is obtained through the platform, a water body extraction model algorithm is established by combining water body monitoring data, and national lake water body range extraction and automatic treatment in nearly 50 years are realized. The specific technical scheme is as follows.

A GEE-based rapid automatic extraction method for large-scale lake water bodies comprises the following steps.

(1) And acquiring a remote sensing image data set of nearly 50 years from the GEE cloud platform.

(2) And calling a GEE function method for the acquired remote sensing data to perform cloud screening quality control processing.

(3) Establishing a lake water body index method extraction model based on the processed remote sensing image data, and compiling a program algorithm in GEE to extract the water body.

(4) And (4) performing smoothing treatment spot removal aiming at the preliminary water body extraction result to obtain the final water body extraction result.

(5) And calling a GEE Python API, modifying the space-time range, and realizing automatic extraction of any lake water body range and export of extracted result data and products to the local.

As a further improvement of the invention, in the step (1), in 1972-. And acquiring a remote sensing image data set by using an ee.image method provided by the GEE platform or the ID identification of the image.

Further, the method for acquiring Landsat4 data in the GEE cloud platform comprises the following steps: imagecollection ('LANDSAT/LT04/C01/T1_ SR'); the method for acquiring Landsat5 data comprises the following steps: imagecollection ('LANDSAT/LT05/C01/T1_ SR'); the method for acquiring Landsat7 data comprises the following steps: imagecollection ('LANDSAT/LE07/C01/T1_ SR'); the method for acquiring Landsat8 data comprises the following steps: imagecollection ('LANDSAT/LC08/C01/T1_ RT _ TOA'); the method for acquiring the Sentinel-1 data comprises the following steps: imagecollection ('COPERNICUS/S1_ GRD').

As a further improvement of the present invention, in the step (2), since Sentinel-1 is radar data and is not affected by cloud, cloud amount screening processing is performed only on Landsat series data. By using the Landsat Simple Cloud Score algorithm method of GEE internal package, the Landsat TOA image is processed by setting the 'Cloud Threshold' parameter.

Further, cloud screening is performed on the Landsat4, the Landsat5 and the Landsat7 data in the GEE cloud platform to call a ('pixel _ qa') band in the image, and cloud screening is performed on the Landsat8 data to call a ('BQA') band in the image.

As a further improvement of the present invention, in the step (3), for Landsat data, the water body index method is to extract the water body information in the image by selecting a wave band sensitive to the water body information, establishing a mathematical model or a wave band combination to highlight the difference between the water body and the non-water body. In the method, the automatic water body extraction index (AWEI) without threshold value provided by Feyisa is adopted, so that the problem that different lake threshold values need to be considered in the process of extracting multiple lakes is avoided, and the water body identification is more convenient and faster. The AWEI calculation formula is as follows.

AWEI = 4 (Green – SWIR1) - (0.25NIR + 2.75SWIR2) 。

In the formula: green, NIR, SWIR1, and SWIR2 are the surface reflectance of Green, near-infrared, short-wave infrared 1, and short-wave infrared 2 bands, respectively.

Further, Green, NIR, SWIR1 and SWIR2 are extracted from Landsat4, Landsat5 and Landsat7 images in the GEE cloud platform, and image selection ('BX') is adopted as a method for calculating the AWEI.

For the Sentinel-1 radar data, a Dual-Polarized data SDWI (Sentinel-1 Dual-Polarized Water Index) Water body Index method is selected as a mathematical model for information extraction to highlight the difference between the Water body and the non-Water body, and the identification of the Water body is carried out, wherein the SDWI calculation formula is as follows.

SDWI = ln(10×VV×VH)−8。

In the formula: VV and VH are Sentinel-1 wave bands.

Further, the bands VV and VH used for calculating SDWI are extracted from the Sentinel-1 image in the GEE cloud platform by an image.

Further, water body extraction is carried out according to an optimal threshold value method, and 0.35 is selected as an optimal threshold value for SDWI to carry out water body extraction.

As a further improvement of the invention, in the step (4), because the acquired water body distribution binary image has more speckle breaking errors, the image is smoothed by using a linear convolution algorithm, and the high-frequency speckle information is deleted by using a function image. container () in the GEE and completing an algorithm of a smoothing kernel, so that the remote sensing inversion model of the lake water body range is optimized.

As a further improvement of the present invention, in the step (5), in a control parameter file in a local python API, a time range and a space boundary of the lake water body to be extracted are set, that is, data and a model of the GEE platform are automatically called by an ee.initialize () method, real-time automatic water body extraction of a target area is performed, and a final result is automatically downloaded to a local area for use by a user without excessive manual participation.

The technical scheme has the following advantages or beneficial effects: the method provides a rapid and automatic long-time-sequence large-scale lake water body extraction method. Firstly, a large amount of image data and high-performance cluster server cloud computing capacity are provided on line through the GEE cloud platform, a user can analyze and research the large amount of image data only by using a local Python API, the efficiency of a researcher is greatly improved, the researcher is enabled to concentrate on research result analysis, and the productivity of workers is liberated. Secondly, the water body extraction method adopts multi-source satellite data, adopts advanced cloud computing algorithm extraction, image spot smoothing processing and the like, solves the problems of short time sequence, low precision and poor quality of single satellite data, and can perform long-time sequence high-precision water body extraction. And finally, the constructed water body range recognition model is accessed to the GEE platform by using a python API program, the model can be automatically operated in a fixed-area and fixed-time mode, the model is synchronized to the local, the final result is obtained, both hands are completely liberated, and the service efficiency is greatly improved.

Drawings

FIG. 1 is a technical route of a GEE-based rapid automatic extraction method for large-scale lake water bodies.

FIG. 2 is a water body binary distribution diagram of dense cloud reservoir in 8 months in 1985.

FIG. 3 is the binary distribution diagram of 8-month dense cloud reservoir water in 1995.

Fig. 4 is a binary distribution diagram of a dense cloud reservoir water body in 8 months in 2005.

Fig. 5 is a water body binary distribution diagram of dense cloud reservoir in month 8 in 2015.

Fig. 6 is a water body binary distribution diagram of 8-month dense cloud reservoir in 2020.

Detailed Description

The specific implementation of the invention is further explained by taking a long-time sequence rapid extraction and mapping process of the water body range of the dense cloud reservoir as an example and combining the description and implementation of the attached drawings.

(1) And extracting the land remote sensing reflectivity of the dense cloud reservoir area. The GEE computing cloud platform integrates various remote sensing reflectivity data. Aiming at the water body range remote sensing inversion of the region 1984-so far, Landsat5 images used in 1984-2013, Landsat8 images used in 2013-2015 and Sentinel-1 images used so far are recommended to be synthesized and fused one pair of required dense cloud reservoir water body images month by month, and the method has unique advantages for researching long-time sequence change of the lake water body range. According to the month-by-month water body monitoring date of the dense cloud reservoir water body, the Landsat corresponding to the water body of the nested lake, the Sentinil-1 remote sensing reflectivity, the quality evaluation and other remark information are obtained in a programming mode by utilizing the GEE background database and the powerful geographical computing cloud. And analyzing the remote sensing reflectance values observed by Landsat and Sentinel-1, and specifically operating the steps as follows.

a) And independently programming and acquiring remote sensing reflectivity and quality evaluation information observed by Landsat and Sentinel-1 corresponding to the water body of the nested lake on the GEE platform. The method for obtaining the quality evaluation information in the GEE Python API is as follows.

Further, the method for acquiring the data quality evaluation information of Landsat5 in the GEE Python API comprises the following steps:

def mask_L457(image):

qa7 = image.select('pixel_qa')

cloud = qa7.bitwiseAnd(1<<5).And(qa7.bitwiseAnd(1 << 7)).Or(qa7.bitwiseAnd(1<<3))

mask2 =image.mask().reduce(ee.Reducer.min())

return image.updateMask(cloud.Not()).updateMask(mask2)。

further, the method for acquiring the data quality evaluation information of Landsat8 in the GEE Python API comprises the following steps:

def maskL8(image):

qa = image.select('BQA')

mask_cloud = qa.bitwiseAnd(1 << 4).Or(qa.bitwiseAnd(1 << 8))

return image.updateMask(mask_cloud.Not())。

b) the quality evaluation of Landsat comprises cloud coverage information of the extracted pixels, and matched non-cloud-coverage remote sensing reflectivity images can be screened out according to the information. The method for obtaining the quality evaluation information in the GEE Python API is as follows.

Further, the method for acquiring Landsat5 data in the GEE cloud platform comprises the following steps:

Landsat5 = ee.ImageCollection('LANDSAT/LT05/C01/T1_SR')/

.filterDate(start_date, start_date+relativedelta(months=1))/

.filterBounds(geometry)/

.map(mask_L57)/

.map(selected5)/

.median()。

further, the method for acquiring Landsat8 data in the GEE cloud platform comprises the following steps:

Landsat8 = ee.ImageCollection('LANDSAT/LC08/C01/T1_RT_TOA ')/

.filterDate(start_date, start_date+relativedelta(months=1))/

.filterBounds(geometry)/

.map(mask_L8)/

.map(selected8)/

.median()。

further, the method for acquiring the Sentinel-1 data in the GEE cloud platform comprises the following steps:

Sentinel-1 = ee.ImageCollection('COPERNICUS/S1_GRD')/

.filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VH'))/

.filter(ee.Filter.listContains('transmitterReceiverPolarisation', 'VV'))/

.filter(ee.Filter.eq('instrumentMode', 'IW'))/

.filterDate(start_date,start_date + relativedelta(months=1))/

.filterBounds(geometry)/

.mean()。

(2) and (5) constructing a lake water body remote sensing model. The reflection capability of the water body to electromagnetic radiation with different wavelengths is different, and the change rule of the reflectivity of the water body along with the incident wavelength is called as the spectral characteristic of the water body due to the combined action of the composition of the water body substances and the physicochemical properties of the water body. The radiation transmission process of light in the water body includes reflection, refraction, absorption, volume scattering and the like. The water body radiation transmission equation takes the inherent optical quantity of the water body and the environmental parameters as input parameters, obtains the distribution condition of radiation energy in water by simulating the water body radiation transmission process, and further calculates and obtains the water-leaving radiation, the water surface reflected light and the sky scattered light which can be received by the spectrometer. The pure water is excited by different vibration modes of water molecules, so that the pure water has larger reflectivity in a green light and blue light range (380-580 nm), the reflectivity is gradually reduced in 580-690 nm, and two small reflection valleys are formed at 610nm and 670 nm. Starting at 690nm, water absorption increases rapidly, forming a very low value of reflectance at 740 nm. The spectral characteristics are the basis for the water body to be extracted through the spectrum, and the specific operation steps are as follows.

a) Green, NIR, SWIR1 and SWIR2 bands with better quality corresponding to Landsat data and VV and VH bands with better quality corresponding to Sentiniel-1 data are selected.

Furthermore, Green, NIR, SWIR1 and SWIR2 wave bands for calculating the AWEI are extracted from the Landsat5 image in the GEE cloud platform, and the method comprises the following steps:

def selected5(image):

image = ee.Image(image)

Green=image.select('B2')

NIR=image.select('B4')

SWIR1= image.select('B5')

SWIR2=image.select('B7')。

furthermore, Green, NIR, SWIR1 and SWIR2 wave bands for calculating the AWEI are extracted from the Landsat8 image in the GEE cloud platform, and the method comprises the following steps:

def selected8(image):

image = ee.Image(image)

Green=image.select('B3')

NIR=image.select('B5')

SWIR1= image.select('B6')

SWIR2=image.select('B7')。

further, extracting VV and VH wave bands for calculating the AWEI from the Sentinel-1 image in the GEE cloud platform by the following method:

def selected1(image):

image = ee.Image(image)

VV=image.select('VV')

VH=image.select('VH')。

b) different combinations (ratio, difference, addition and the like) of corresponding wave bands are tested through autonomous programming, a model with the best correlation is determined, a data set parameterization is established through the model, a remote sensing inversion model suitable for the area of the water body of the nested lake is determined, Landsat data is applied to a formula (1), and Sentinel-1 data is applied to a formula (2).

AWEI = 4 (Green-SWIR 1) - (0.25NIR + 2.75SWIR2) formula (1).

SDWI = ln (10 × VV × VH) -8 formula (2).

In the formula: VV and VH are Sentinel-1 wave bands.

Further, the method for calculating the AWEI by utilizing the bands extracted by the Landsat5 and Landsat8 images in the GEE cloud platform comprises the following steps:

AWEI = (Green.subtract(SWIR1)).multiply(4).subtract((NIR.multiply(0.25))

.add(SWIR2.multiply(2.75)))。

further, the method for calculating the SDWI by using the band extracted by the Sentinel-1 image in the GEE cloud platform comprises the following steps:

SDWI = ((image.select("VV").multiply(image.select("VH")).multiply(10)).log()).subtract(8)。

d) and (4) spot elimination, namely eliminating data by using a smoothing algorithm special for GEE and using a function of image. The implementation method for performing the spot elimination in the GEE Python API is as follows:

boxcar = ee.Kernel.square(radius = 7, units='pixels', normalize = True)。

(3) and rapidly deriving the drawing in the water body ranges with different time scales. And uploading the result product to Google Drive by self-help programming and utilizing a batch. The method for extracting the long-time sequence water product data of the dense cloud reservoir in the GEE Python API comprises the following steps:

exporttask = batch.Export.image.toDrive(

image=Sentinel,

description=str(start_date.strftime('%Y%m')) + "-Sentinel1-" + "10m",

region=poly.getInfo()['coordinates'],

folder='MiYunShuiKu',

scale=10,

crs='EPSG:4326',

maxPixels=1E+13

)

exporttask.start()

Claims

1. a GEE-based rapid automatic extraction method for long-time-sequence large-scale lake water bodies is characterized by comprising the following steps:

(1) accessing a Google Earth Engine through a local GEE Python API, and calling the data, algorithm and computing processing capacity of the GEE;

(2) screening and acquiring a remote sensing image data set of specified time and space by modifying the time range and the parameters of the research area in the configuration file;

(3) selecting Landsat series data as a data source in the time range of 1972-2014, and selecting Sentinle-1 data as a data source in the time range of 2015 and later;

(4) carrying out cloud removal processing on Landsat data by adopting quality control information provided by GEE, and eliminating the interference of cloud on water body extraction;

(5) calculating an AWEI index by utilizing Landsat series data, and calculating an SDWI index by utilizing Sentinel-1 data so as to distinguish a water body from a non-water body;

(6) carrying out normalization segmentation on the calculated image, and extracting a water body part as a preliminary water body extraction result;

(7) in order to eliminate abnormal spot pixels in the preliminary result, smoothing the image to obtain a final water body extraction result;

(8) and (4) rapidly synchronizing the extraction result to the local in real time through a Google cloud hard disk tool for visualization or analysis processing.

2. The GEE-based rapid automatic extraction method for long-time-sequence large-scale lake water bodies according to claim 1, wherein the GEE is accessed to a Google Earth Engine through a local GEE Python API, and data, algorithm and computing processing capacity of the GEE are called, and the method comprises the following steps:

locally configuring a GEE Python API environment;

and accessing the GEE cloud terminal by adopting a command locally.

3. The GEE-based long-time-series large-scale lake water body rapid automatic extraction method as claimed in claim 1, wherein the step of screening and obtaining the remote sensing image data set of the designated time and space by modifying the time range and the parameters of the study area in the configuration file comprises the following steps:

defining a preliminary space range for water body extraction for GEE to screen data;

and limiting the time range of water body extraction for GEE to screen data.

4. The GEE-based long-time-series large-scale lake water body rapid automatic extraction method of claim 1, wherein the quality control information provided by the GEE is adopted to carry out cloud removal processing on Landsat data and eliminate the interference of cloud on water body extraction, and the method comprises the following steps:

acquiring a cloud removing method algorithm provided by the GEE;

and (4) calling a cloud removing algorithm based on the GEE quality control information waveband data to remove the cloud area.

5. The GEE-based long-time-series large-scale lake water body rapid automatic extraction method as claimed in claim 1, wherein the AWEI index is calculated by using Landsat series data, and the Sentinel-1 data is used for calculating the SDWI index so as to distinguish a water body from a non-water body, comprising the following steps:

calculating an AWEI index to distinguish between a water body and a non-water body based on Landsat series data subjected to the cloud removal processing of claim 4;

the SDWI index is calculated based on the Sentinel-1 data to distinguish between bodies of water and non-bodies of water.

6. The GEE-based long-time-series large-scale lake water body rapid automatic extraction method of claim 1, wherein the normalized segmentation is performed on the calculated image to extract a water body part, and the extraction as a preliminary water body extraction result comprises:

normalizing the water body index calculated based on the claim 5;

and dividing the acquired normalized water body index into a water body and a non-water body according to a 0 threshold value, and carrying out binarization treatment on the water body and the non-water body.

7. The GEE-based long-time-series large-scale lake water body rapid automatic extraction method of claim 1, wherein in order to eliminate abnormal spot pixels existing in the preliminary result, smoothing is performed on the image to obtain a final water body extraction result, and the method comprises the following steps:

and calling a GEE method to carry out smoothing treatment based on the preliminary water body extraction result obtained in the claim 6.

8. The GEE-based long-time-sequence large-scale lake water body rapid automatic extraction method of claim 1, wherein the extraction result is rapidly synchronized to local in real time through a Google cloud hard disk tool for visualization or analysis processing, and the method comprises the following steps:

calling a GEE batch export method for the final water body extraction result obtained in the claim 7 to export the final water body extraction result to a Google cloud hard disk;

and (4) synchronizing the extraction result to the local automatically in real time by utilizing a local Google cloud hard disk synchronization tool.