CN115797501A - Time-series forest age mapping method combining forest disturbance and recovery events - Google Patents

Abstract

The invention discloses a time series forest age mapping method combined with forest disturbance and restoration events, comprising the following steps: (1) acquisition of remote sensing satellite image data source; (2) construction of forest age estimation model; (3) random forest algorithm estimation Spatial distribution of forest age in the base year; (4) Improved LandTrendr algorithm; (5) Landsat disturbance and trend detection (LandTrendr) algorithm to detect forest disturbance and restoration information; (6) Integrating the results of different changing years; (7) Estimation of time series forest age . The invention adopts the LandTrendr algorithm to detect forest changes, can identify broad interference intensity, completes the mapping of natural forest and artificial forest forest age, combines the detected forest and restoration events into forest age time series estimation, and improves the accuracy of mapping.

Description

A Time-Series Stand Age Mapping Method Combining Forest Disturbance and Restoration Events

technical field

The invention relates to the field of forest age mapping, in particular to a time series forest age mapping method combining forest disturbance and restoration events.

Background technique

Forest age is a key parameter that determines the carbon sequestration status and potential of forest ecosystems. The lack of time-series forest age spatial pattern makes it impossible to capture forest disturbance and restoration history in sufficient space and time domains, which also increases the uncertainty of past and future forest carbon sink estimation. Therefore, it is of great scientific significance to accurately estimate forest ecosystem carbon sinks by considering the history of forest disturbance and restoration and exploring the dynamic changes of forest age.

The combination of remote sensing and ground observation records provides an effective and feasible method for obtaining the temporal and spatial distribution of forest age. At present, the forest age estimation method based on remote sensing tends to establish a regression relationship between the forest age and the characteristic variables such as the spectrum and texture of the vegetation, and use this to predict the spatial distribution of the forest age. The basic physical mechanism of this method is that the different optical properties of tree species, branches, leaves, bark, trunk, and cones in the stand will cause the light reflectance and transmittance of the canopy structure to be different. , the image texture of the forest canopy also changes, from a uniform style to a clumping pattern, which also allows the relationship between the canopy texture features and forest age to be established. Based on the above reasons, researchers at home and abroad use multiple linear regression, artificial neural network, support vector machine and other regression methods to invert the spatial distribution of forest age in combination with spectral and texture features.

However, the above methods can only obtain the spatial distribution of stand age in a specific year. How to combine the history of disturbance and recovery with the forest age in a specific year is the key to deduce the forest age in time series. Therefore, taking a specific year as the base year, combined with disturbance and restoration events, and calculating the forest age backwards, it is possible to map the spatial distribution of forest age in time series.

At present, there is only one most similar prior art invention, "A New Method for Forest Age Spatial Mapping of Planted Forests" (application number: CN201710592204.2, authorized announcement number: CN107247809B). It first uses the random forest algorithm to simulate the forest age distribution in a certain year, then uses the vegetation change tracking (VCT) algorithm to detect forest disturbance events, and finally calculates the time series plantation age spatial distribution affected by the disturbance based on the first two.

Compared with the VCT algorithm, the Landsat Disturbance and Trend Detection (LandTrendr) algorithm is widely used worldwide as an effective method for detecting forest disturbance and restoration events. The advantage of the LandTrendr algorithm is that straight-line segmentation can detect sharp disturbance events (clearfelling) as well as long-lasting slow events (forest growth, post-disturbance recovery). At the same time, the LandTrendr algorithm can set different spectral indices to obtain different detection results, unlike the VCT algorithm which can only be based on the IFZ index. Therefore, the present invention attempts to use the Google Earth Engine (GEE) cloud platform, based on the LandTrendr algorithm of different spectral indices, to more accurately obtain the year of forest disturbance and restoration, thereby coupling the random forest model to calculate the age of the time series forest, and this process It is fully programmed and automatically completes time-series forest age mapping at different scales and time ranges.

The Vegetation Change Tracking (VCT) algorithm used in the prior art technology to detect forest changes can only identify limited interference intensity and can only complete the mapping of plantation forest age; the forest restoration events detected by the VCT algorithm are not integrated into the forest. In age estimation; or only based on a single spectral index, the forest disturbance and restoration detection algorithm can be run to obtain the corresponding change information of the forest.

Contents of the invention

The purpose of the present invention is to address the deficiencies of the prior art, to propose a time-series forest age mapping method combining forest disturbance and restoration events, to solve the problem that only small and medium-scale plantation forest age mapping can be realized before, and to realize large-scale plantation Mapping with natural forest age can obtain more accurate forest disturbance and restoration information, improve the accuracy of time series forest age spatial mapping, and users can set a variety of spectral indices at will, so as to improve the accuracy of forest age spatial mapping by combining the results of each index precision.

The purpose of the present invention is achieved by the following technical solutions: a time-series forest age mapping method in conjunction with forest disturbance and recovery events, the technical scheme of time-series forest age spatial distribution mapping comprises the steps:

(1) Remote sensing satellite image data source acquisition:

According to the mapping range, year and resolution required by the user, obtain the remote sensing satellite data of the surface albedo image of the corresponding location, corresponding year and corresponding resolution; and obtain the corresponding forest age ground survey data;

(2) Build a forest age estimation model:

Based on the original spectral band of the surface reflectance data, calculate the principal component analysis variables, spectral index, tasseled cap transformation and texture feature variables with the gray level co-occurrence matrix; combine with the forest age ground survey data to construct a training set, and use the training set to train Random forest regression model.

(3) The random forest algorithm estimates the spatial distribution of forest age in the base year:

After the training of the random forest regression model is completed, the non-forest areas within the mapping range are masked out, so that the forest age in the base year within the mapping range can be obtained by simulating and estimating based on the random forest regression model;

(4) Improved LandTrendr algorithm:

Expand the input data of the LandTrenr algorithm, realize that it can call the Sentinel (Sentinel) and the medium resolution imaging spectrometer (MODIS) surface reflectance image data, and construct the corresponding time series stack data, that is, the image collection (ImageCollection) object in GEE , and then calculate and segment the spectral trajectory through this object, and obtain the change year according to the trajectory segmentation algorithm;

(5) Landsat interference and trend detection (LandTrendr) algorithm detects forest interference and restoration information:

Set different spectral indices to obtain different forest disturbance and restoration occurrence years based on the improved LandTrendr algorithm;

(6) Integrate the results of different changing years:

(6.1) Integrate the change year results of different spectral indices into one image through the majority voting method;

(6.2) The last 5 band variables in the results of different spectral indices are used as independent variables of the random forest classification model, and the interference and restoration information judged by the user's visual interpretation is used as a label to make a forest interference and restoration training set, and use the training set To train the random forest classification model, simulating the boundary area of forest interference and restoration, the output of the model is two binary raster images;

(6.3) Finally, use the binary raster image to mask the integration results of the year of change, so as to obtain the spatial distribution map of the year of forest disturbance and restoration respectively;

(7) Calculation of time series forest age:

First, the forest age in the next year is calculated based on the forest age in the base year, and the corresponding pixel is retrieved from the detection results of the year when the forest disturbance and restoration occurred in LandTrendr. The value of 0 is assigned to the value of 0, and the value of 1 is assigned to the pixel value of the forest restoration in the next year, and 1 is added to the age value of the previous year for the pixel of the forest that has not changed in the next year. Create the spatial distribution of forest age in the next year; then, iteratively use the forest age in the previous year to obtain the forest age in the next year; finally, complete the mapping of the time series forest age required by the user.

Furthermore, the detection result of the spectral index includes 6 bands: the year of change, the magnitude of change, the duration of change, the spectral value before the change, the rate of spectral change, and the signal-to-noise ratio.

Further, the majority voting method is specifically: when the number of majority elements is greater than n/2, assign the value of the majority elements to the pixel. Conversely, when there are no majority elements, or the number of majority elements is less than n/2, assign the value with the highest detection accuracy in the time series to the pixel;

Further, in step (6.2), the labels are respectively: "forest loss/forest unchanged", "forest restoration/forest unchanged".

Beneficial effects of the present invention:

(1) The LandTrendr algorithm is used to detect forest changes, which can identify a wide range of interference intensities, and complete the mapping of natural forests and plantation forest ages.

(2) Combining the forest and restoration events detected by the LandTrendr algorithm into forest age time series estimation to improve the accuracy of mapping.

(3) Use a variety of remote sensing characteristic variables (such as: original band, spectral index) to detect forest changes based on the LandTrendr algorithm, and combine the majority voting method to obtain more accurate forest disturbance and restoration information.

Description of drawings

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings that need to be used in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only These are some embodiments of the present invention. Those skilled in the art can also obtain other drawings based on these drawings without creative work.

Fig. 1 is a flowchart of a time-series forest age mapping method combining forest disturbance and restoration events provided by the present invention.

Detailed ways

The specific embodiments of the present invention will be described in further detail below in conjunction with the accompanying drawings.

As shown in Figure 1, a kind of time-series forest age mapping method combined with forest disturbance and recovery events provided by the present invention, the specific steps are as follows:

(1) Selection of remote sensing satellite image data sources:

Use Google Earth Engine (GEE) to call Landsat, MODIS, Sentinel and other surface reflectance data (https://developers.google.com/earth-engine/datasets/). According to the mapping range, year and resolution required by the user, select the corresponding location, corresponding year, and corresponding resolution surface albedo image data, and obtain the corresponding forest age ground survey data; in order to realize large, medium and small scale forest age mapping; The acquisition of remote sensing satellite image data can also use the PIE-Engine cloud platform to schedule the corresponding data and run the same Javascript script (including the inversion of the forest age base year mentioned below and the LandTrendr algorithm).

(2) Build a forest age estimation model:

Based on the original spectral bands of the surface reflectance data, the characteristic variables such as principal component analysis variables, spectral indices, tasseled cap transformation, and gray-level co-occurrence matrix texture are calculated. Combined with the ground survey data of forest age collected by users, a sample set is constructed, and 70% is divided into training set and 30% as test set. Use the training set to train the random forest regression model (or multiple linear regression, artificial neural network, support vector machine, etc.), and determine the quantitative relationship between the forest age and each characteristic variable through modeling.

(3) The random forest algorithm estimates the spatial distribution of forest age in the base year:

After the random forest regression model optimization and training are completed, the test set is used to verify the accuracy of the model, so that the model can be applied to the entire mapping range. At the same time, the land cover (Land cover) dataset (https://developers.google.com/earth-engine/datasets/tags/landcover) on GEE is called to mask out the non-forest areas within the mapping range, thus based on The random forest regression model simulated and estimated the forest age in the base year within the mapping range.

(4) Improved LandTrendr algorithm:

Expand the input data of the LandTrenr algorithm: In the original LandTrendr algorithm, the time series stack data is first constructed based on the Landsat image data provided by GEE, and then the spectral trajectory of the original band or spectral index is extracted from it, and then the year of change is detected according to the trajectory segmentation algorithm. After improving the LandTrendr algorithm, the data source at the input end is expanded, and the originally called Landsat image data is replaced by other surface reflectance image data (such as the sentinel Sentinel and the medium-resolution imaging spectrometer MODIS through the ee.ImageCollection() function on GEE). ) to construct the corresponding time series stack data (ImageCollection image collection object in GEE), and then use this object to calculate and obtain the corresponding segmented spectral trajectory according to the spectral trajectory segmentation algorithm of LandTrendr, so as to obtain the corresponding change year. In this way, the interference and restoration detection of large, medium and small forests can be realized.

(5) Landsat interference and trend detection (LandTrendr) algorithm detects forest interference and restoration information:

By setting different spectral indices, based on the improved LandTrendr algorithm, we can obtain different forest disturbance and restoration occurrence years. The detection result of a certain spectral index includes 6 bands: change year, change range, change duration, spectral value before change, spectral change rate, signal-to-noise ratio.

(6) Integrate the results of different changing years, the specific steps are as follows:

(6.1) The change year results of different spectral indices are integrated into an image through the majority voting method. pixels. Conversely, when there are no majority elements, or the number of majority elements is less than n/2, assign the value with the highest detection accuracy in the time series to the pixel;

(6.2) In the results of different spectral indices, except for "change year", the other 5 band variables are used as independent variables of the random forest classification model, and the interference and restoration information judged by the user's visual interpretation is used as a label to create forest interference and restoration 70% of the sample set is used as the training set and 30% as the test set. Use the training set to train the random forest classification model, and the test set to verify the model. After building the model, simulate the boundary area of forest disturbance and restoration, and the result is two binary raster images, with labels: "forest loss/forest unchanged" and "forest restoration/forest unchanged";

(6.3) Finally, the binary raster image was used to mask the integration results of the year of change, so as to obtain the spatial distribution map of the year of forest disturbance and restoration respectively. This result accurately represents which cells experienced forest loss or restoration and more accurately detects the years of change. In this way, the forest disturbance and restoration information obtained through multiple remote sensing features is more accurate, which solves the limitation that the existing technology can only use a single index.

(7) Calculation of time series forest age:

First, the forest age in the next year is calculated based on the forest age in the base year, and the corresponding pixel is retrieved from the detection results of the LandTrendr forest disturbance and restoration occurrence year. For the pixel disturbed by the forest in the next year, the value Assign a value of 0, and assign a value of 1 to the pixel of the forest restoration in the next year, and add 1 to the age value of the previous year for the pixel of the forest that has not changed in the next year, so as to make The spatial distribution of forest age in the next year; then, iteratively use the forest age in the previous year to obtain the forest age in the next year; finally, complete the mapping of the time series forest age required by the user. In this way, both forest disturbance and forest restoration information are integrated to achieve a more accurate estimation of time series forest age.

The above-mentioned embodiments are used to illustrate the present invention, rather than to limit the present invention. Within the spirit of the present invention and the protection scope of the claims, any modification and change made to the present invention will fall into the protection scope of the present invention.

Claims (4)

1. A time series forest age mapping method combining forest disturbance and recovery events is characterized in that the technical scheme of time series forest age space distribution mapping comprises the following steps:

(1) Acquiring a remote sensing satellite image data source:

acquiring earth surface reflectivity image remote sensing satellite data of corresponding positions, corresponding years and corresponding resolutions according to a drawing range, the years and the resolutions required by a user; acquiring corresponding forest age ground survey data;

(2) Constructing a forest age estimation model:

calculating principal component analysis variables, spectral indexes, tassel-cap transformation and texture characteristic variables of a gray level co-occurrence matrix based on the original spectral band of the earth surface reflectivity data; and combining with forest age ground survey data to construct a training set, and training a random forest regression model by using the training set.

(3) Estimating the spatial distribution of forest ages in the reference year by using a random forest algorithm:

after training of the random forest regression model is completed, masking the non-forest region in the drawing range, and therefore simulating and estimating to obtain the forest age of the reference year in the drawing range based on the random forest regression model;

(4) Improving the LandTrendr algorithm:

expanding input data of a LandTrenr algorithm, realizing that the LandTrenr algorithm can be used for acquiring surface reflectivity image data of sentinels (Sentinel) and medium resolution imaging spectrometers (MODIS), constructing corresponding time series stack data, namely an image collection (ImageCollection) object in GEE, calculating and segmenting a spectrum track through the object, and segmenting the algorithm according to the track to obtain the change year;

(5) Detecting forest interference and recovery information by using a Landsat interference and trend detection (LandTrendrr) algorithm:

setting different spectral indexes, and acquiring different forest disturbance and recovery years based on an improved LandTrendrr algorithm;

(6) Integrating different year-of-change results:

(6.1) integrating the changing year results of different spectral indexes into an image by a majority voting method;

(6.2) taking the last 5 wave band variables in the results of different spectral indexes as independent variables of the random forest classification model, taking interference and recovery information which is visually interpreted and judged by a user as a label, making a forest interference and recovery training set, training the random forest classification model by using the training set, simulating a forest interference and recovery boundary area, and outputting two binary grid images as a model output result;

(6.3) finally, respectively masking the change year integration result by using the binary raster image so as to respectively obtain forest disturbance and recovery year spatial distribution maps;

(7) Calculating the forest age of the time series:

firstly, calculating the forest age of the next year according to the forest age of a reference year, retrieving a corresponding pixel from the detection result of the LandTrendr forest interference and recovery occurrence year, endowing the pixel value with a value of 0 for the pixel subjected to forest interference of the next year, endowing the pixel value with a value of 1 for the pixel subjected to forest recovery of the next year, and adding 1 to the forest pixel not changed in the next year so as to make the forest age spatial distribution of the next year; then, sequentially and iteratively utilizing the forest age of the previous year to obtain the forest age of the next year; and finally, completing the mapping of the forest ages of the time series required by the user.

2. The method for time-series forest age mapping by combining forest disturbance and recovery events as claimed in claim 1, wherein the detection result of the spectral index comprises 6 bands: year of change, amplitude of change, duration of change, spectral value before change, rate of change of spectrum, signal-to-noise ratio.

3. The time-series forest age mapping method combining forest disturbance and recovery events according to claim 1, wherein the majority voting method specifically comprises the following steps: and when the number of the majority elements is larger than n/2, assigning the value of the majority elements to the pixel. Otherwise, when the majority elements do not exist or the number of the majority elements is less than n/2, the value with the highest detection precision in the time sequence is given to the pixel;

4. a method for time series forest age mapping in combination with forest disturbance and recovery events as claimed in claim 1, wherein in step (6.2), the labels are: "forest loss/forest unchanged", "forest recovery/forest unchanged".

Images