Search

CN-121999358-A - Vegetation interference recovery process identification method based on cloud platform and improvement BfastMonitor

CN121999358ACN 121999358 ACN121999358 ACN 121999358ACN-121999358-A

Abstract

The invention provides a forest disturbance recovery process identification method, and belongs to the field of forest protection and monitoring. The method comprises the steps of 1) screening Landsat ground surface reflectivity products of a research area in a research period, preprocessing the ground surface reflectivity products, creating an area NBR image dataset, 2) dividing the cloud-length-free time sequence remote sensing dataset of the research area into a stable period and a monitoring period, obtaining primary multi-window vegetation interference recovery information of the research area by utilizing BfastMonitor algorithm, extracting continuous interference pixels for 6 times and removing false changes detected by Chow test false positive detection algorithm, and 3) automatically identifying different vegetation interference recovery processes based on a preset screening algorithm. The embodiment of the invention provides a remote sensing image time sequence-based vegetation interference recovery process identification method, which is based on dense terrestrial satellite time sequence images, can rapidly detect historical time sequence structural changes of a research area, effectively remove most of pseudo changes detected by an original algorithm, and further judge different interference recovery change modes so as to realize effective identification of the vegetation history interference recovery process of the research area.

Inventors

  • LIN WENSHU
  • XU KE
  • ZHANG NING

Assignees

  • 东北林业大学

Dates

Publication Date
20260508
Application Date
20241104

Claims (8)

  1. 1. The vegetation interference recovery process identification method based on the cloud platform and the improvement BfastMonitor is characterized by comprising the following steps: S1, screening Landsat surface reflectivity products of a research area in a research period, preprocessing the surface reflectivity products, and creating an area NBR image dataset; s2, dividing a cloud length-free time sequence remote sensing data set of a research area into a stable period and a monitoring period, and obtaining primary multi-window vegetation interference recovery information of the research area by utilizing BfastMonitor algorithm; And S3, automatically identifying different vegetation interference recovery processes based on a preset screening algorithm.
  2. 2. The vegetation interference recovery process identification method based on the cloud platform of claim 1, wherein in S1, screening of Landsat ground surface reflectivity products is based on Landsat Collection scene Tier 2-level reflectivity data in Gu Geyun platforms (Google EARTH ENGINE, GEE), all available images in a research area and research time are selected, preprocessing of the ground surface reflectivity products is firstly image screening, cloud pixels and cloud-free pixels are marked by CFMASK algorithm, image cutting and splicing are carried out after cloud pixels in the images are removed to obtain a remote sensing image preprocessed in the research area, and in S1, an area NBR image dataset is created by calculating vegetation index NBR by the following formula: Wherein NIR is near infrared Band (LANDSAT TM/ETM+band 4 in image, band 5 in OLI), SWIR is short wave infrared Band (Band 7 in Landsat image).
  3. 3. The vegetation interference recovery process identification method based on the cloud platform according to claim 1 is characterized in that in the step S2, a BfastMonitor algorithm is utilized to obtain primary multi-window vegetation interference recovery information of a research area, specifically, remote sensing data of a stable period in NBR image data set of the area are obtained for each pixel, a second harmonic model is obtained through data fitting, a fitting reflectivity value corresponding to each observation time point in the monitoring period is calculated according to the harmonic model, a MOSUM (MO t ) is calculated through the fitting reflectivity and actual reflectivity in the monitoring period, when a time sequence is in a stable state, MO t is close to 0 and fluctuates randomly, and when the time sequence has a large change, MO t deviates systematically by 0, and when the deviation of 0 exceeds a 95% significance boundary, a breakpoint occurs in a time sequence structure.
  4. 4. A method according to claim 3, wherein the expression of the harmonic model is: Where y t is the value observed at t, β 0 is the intercept, k is the number of harmonic terms of the established harmonic model, amplitude γ and phase δ j are unknowns, f is the known frequency, ε t is the error.
  5. 5. A method according to claim 3, wherein the MOSUM (MO t ) has the formula: wherein y s is equal to Representing the actual and predicted observations respectively, Is an estimate of variance, n is the number of historical observations, and h is the size of the moving window.
  6. 6. The method for identifying vegetation recovery process based on a cloud platform according to claim 1, wherein in the step S2, the false change detected by the algorithm is detected by extracting continuous interference pixels and Chow test false positive detection and removal algorithm for 6 times, specifically comprising the steps of judging that the pixel is interfered when MO t values calculated by continuous 6 clear observations in a time sequence deviate from a significance boundary, and if MO t values calculated by less than 6 clear observations deviate from the significance boundary, the change detected by the algorithm is regarded as the false change, and after extracting continuous interference pixels for 6 times, detecting the false change detected by a further removal algorithm by using a Chow test.
  7. 7. The method of claim 6, wherein the chow test has a formula: Wherein F (T) is a test statistic, SSR A limits the sum of squares of residuals of the model, SSR B is SSR of model 1+SSR of model 2, k is a model fitting coefficient, and n is an observed time span.
  8. 8. The vegetation interference recovery process identification method based on the cloud platform of claim 1 is characterized in that in the step S3, different vegetation interference recovery processes are automatically identified based on a preset screening algorithm, specifically comprising the steps of firstly classifying MOSUM pixels deviating from a significant stable boundary for multiple times and returning to the significant stable boundary into multiple interference classes, judging whether interference occurs before a research period through an average angle threshold value between adjacent change vectors when initial interference occurs in the continuous interference classes, performing vegetation recovery detection by utilizing dynamic changes of MOSUM values calculated by the algorithm, classifying MOSUM values continuously deviating from the significant stable boundary, classifying the pixels which do not return to the significant stable boundary until the research period is finished as an interference-free recovery class, classifying MOSUM values returning to the significant stable boundary after the time is more than or equal to 3 years, and classifying the pixels which remain within the stable boundary for one year or more as a post-interference recovery class.

Description

Vegetation interference recovery process identification method based on cloud platform and improvement BfastMonitor Technical Field The embodiment of the invention relates to the technical field of remote sensing image processing, in particular to a vegetation interference recovery process identification method based on a cloud platform and an improved BfastMonitor Background Forest as the largest carbon reservoir on land is critical to global ecological function. In recent years, extreme climate events are frequent, and simultaneously human activities such as agricultural expansion, urbanization, artificial harvest and the like affect the coverage area and quality of forests, and loss and degradation of forests affect not only land carbon circulation and climate pattern changes, but also productivity, biodiversity and landscape function loss. The forest has various application values, so that the dynamic disturbance recovery process can be better known by monitoring the forest, and the dynamic state of the forest ecological system and the development trend of forest resources can be better known, and meanwhile, a basis is provided for evaluating the health degree of the regional ecological system. Traditional plot-based forest surveys, while able to obtain more detailed and rich information, are difficult to meet the needs of long-time-series large-area monitoring. The satellite remote sensing technology, particularly hyperspectral imaging spectroscopy expands a forest observation database, the storage cost of a computer is greatly reduced due to the occurrence of a GEE cloud platform in recent years, the computing performance is rapidly improved, and meanwhile, a forest disturbance change detection algorithm based on Landsat time series data is widely developed due to the open archiving of Landsat data. Compared with ：LandTrendr spectral-time segmentation algorithm(LandTrendr)、Continuous Change Detection and Classification(CCDC)、Breaks For Additive Season andTrendMonitor(BFAST Monitor)、Vegetation Change Tracker(VCT)., LANDTRENDR and VCT, the common time sequence change detection method has the advantages that the CCDC and the BFAT Monitor can detect disturbance events in the year, specific disturbance dates can be obtained, and LANDTRENDR and VCT are mainly used for detecting the annual change information. CCDC utilizes high-frequency and multivariable image data to perform change detection while fitting harmonic modes on line, and fitting coefficients related to land utilization coverage can be obtained to further classify land utilization, but large calculation amount is required, so that the CCDC is difficult to apply on a large scale, BFAT Monitor utilizes high-frequency and univariate remote sensing images as input, and the CCDC has the main defects of high false alarm rate and more false change due to simple configuration, high operation speed and relatively good performance in detecting seasonal change. The algorithm only focuses on the detection of forest interference, does not consider the subsequent vegetation regeneration process, and has less knowledge of the forest interference recovery process. Therefore, a vegetation interference recovery process identification algorithm is needed, which can effectively remove the pseudo-change detected by the existing vegetation interference algorithm, and can identify the vegetation interference recovery process based on pixel-level time sequence information. Disclosure of Invention The invention aims to solve the problems and provides a vegetation interference recovery process identification method based on a cloud platform, so that the problem of high false alarm rate of a BFAT Monitor algorithm is solved, and vegetation interference recovery process information is provided for areas with frequent interference of a forest ecosystem. In order to achieve the aim, the invention provides the technical scheme that the vegetation interference recovery process identification method based on the cloud platform and the improvement BfastMonitor comprises the following steps: Step 1, screening Landsat surface reflectivity products of a research area in a research period, preprocessing the surface reflectivity products, including image screening, cloud and shadow masking, image cutting and splicing and NBR calculation, creating an area NBR image dataset; Step 2, dividing a cloud length-free time sequence remote sensing data set of a research area into a stable period and a monitoring period, and obtaining primary multi-window vegetation interference recovery information of the research area by utilizing BfastMonitor algorithm, wherein the false change is detected by extracting continuous interference pixels for 6 times and a Chow test false positive detection removal algorithm; Step 3, automatically identifying different vegetation interference recovery processes based on a preset screening algorithm; Further, in step S1, the data includes Collection2 scene Tier2 l