Forest spatiotemporal distribution reconstruction method based on the fusion of historical archives and remote sensing big data

Abstract

The present invention discloses a method for reconstructing the spatiotemporal distribution of forests based on the fusion of historical archives and remote sensing big data, which belongs to the field of forestry monitoring technology, including: unifying forest survey data and remote sensing monitoring products in the historical period into the same standard system, and reconstructing the spatial distribution data set of forest coverage; combining the aforementioned historical archive tracing and reverse spatial modeling to reconstruct the spatiotemporal distribution data set of forest coverage in the historical period; collecting ancient tree information to verify the accuracy of the spatial distribution data set of forest coverage in the historical period; and establishing a spatiotemporal dynamic monitoring system for forest coverage changes in the historical period. The present invention unifies the forest definition in the historical forest survey process into one standard, eliminating the statistical area error caused by different forest definitions; innovatively uses ancient tree data as verification, establishes the forest distribution pattern of China's historical period at a century-long spatiotemporal scale, and has the high spatiotemporal resolution required for the "digital twin" of the earth.

Description

Technical Field

[0001] The present invention mainly relates to the field of forestry monitoring technology, and specifically to a forest spatiotemporal distribution reconstruction method based on the fusion of historical archives and remote sensing big data. Background Art

[0002] In recent years, a series of progress has been made in the reconstruction of historical land cover in the world and China. Related studies include the establishment of a global cultivated land and forest change dataset from 1700 to 1992 by the Center for Global Environment and Sustainable Development at the University of Wisconsin, USA, through remote sensing data and statistical data. The Netherlands Institute for Public Health and the Environment combined historical population estimates and allocation algorithms with time-related land use weight maps to establish the Global Environmental History Database (HYDE). And the combination of multiple types of remote sensing data with FAO land use and demographic data to obtain the 1899-2019 Historical Land Cover Change Dataset (HILDA+). However, such datasets cannot reflect the details of China's rapid forest change process since 1900 due to excessive reliance on FAO inventory data, large uncertainty in land cover allocation models, and coarse resolution. Domestic scholars use a top-down model to reconstruct the historical distribution of forests in China from 1900 to 2019 by integrating multiple inventory data and satellite images. However, such datasets still cannot completely unify the definition of forests in China's historical forest survey process into one standard, making it difficult to reflect the differences in forest statistical area caused by changes in China's definition of forests since 1900, and have a low resolution of 0.5 degrees. Summary of the invention

[0003] The technical solution of the present invention aims at the technical problem that the existing technical solutions are too single, and provides a solution that is significantly different from the existing technologies. It mainly provides a forest spatiotemporal distribution reconstruction method based on the fusion of historical archives and remote sensing big data. It is used to integrate historical archive tracing with remote sensing big data across domains and dimensions in the absence of historical forest spatiotemporal information records, and combines spatial reverse modeling technology to achieve reconstruction of forest spatiotemporal distribution in historical periods and realize spatiotemporal dynamic monitoring of forest cover changes.

[0004] The technical solution adopted by the present invention to solve the above technical problems is:

[0005] The method for reconstructing the spatiotemporal distribution of forests based on the fusion of historical archives and remote sensing big data includes the following steps:

[0006] S1. Unify historical forest survey data and remote sensing monitoring products into the same standard system, and reconstruct the historical forest cover spatial distribution dataset;

[0007] S2, combining the data set obtained in step S1 with the inverse spatial modeling to reconstruct the spatiotemporal distribution data set of forest cover in the historical period;

[0008] S3. Collect information on ancient trees to verify the accuracy of the forest cover spatial distribution dataset in the historical period; establish a spatiotemporal dynamic monitoring system for forest cover changes in the historical period.

[0009] Further, step S1 includes:

[0010] S11. Based on the definition of forest canopy density, the historical forest survey data should be unified into the same standard system in terms of time scale;

[0011] S12. Use a variety of remote sensing data voting strategies assisted by the normalized difference vegetation index to unify the land cover products monitored by remote sensing from the spatial scale to the same standard system.

[0012] Furthermore, step S11 specifically includes:

[0013] S111. Comprehensively search national forest resource inventory data, local chronicles, ancient maps, site data, statistical reports and works to extract direct information on changes in forest area during historical periods;

[0014] S112. Calculate the forest area change rate during the canopy density adjustment period for each province, as well as the forest area change rate before and after the two periods for each province;

[0015] S113. Check the forest area change rate of each province one by one, and calculate the forest cover area during the canopy density adjustment period based on the maximum forest area change rate of the two-period inventory data before and after the canopy density adjustment;

[0016] S114. Unify the definition of forests in the inventory data of the historical period to a standard system consisting of tree species with a canopy density of 0.2 or above.

[0017] Furthermore, step S12 specifically includes:

[0018] S121, automatically screen multiple sets of remote sensing monitoring land cover datasets, uniformly reclassify, resample, and reproject to Albers orthogonal equal-area double standard parallel conic projection;

[0019] S122, using a voting strategy involving multiple types of remote sensing monitoring land cover products to generate a potential forest cover probability map, wherein each pixel of the probability map is composed of a number of votes;

[0020] S123, based on the reconstructed forest resource inventory data records of each province, the annual maximum normalized difference vegetation index is used to assist the decision-making, and when the votes are the same, the pixel with the largest NDVI is defined as the forest;

[0021] S124. Reconstruct the forest cover map composed of tree species with a canopy density of 0.2 or above, and generate a dataset of the spatial distribution of forest cover in the historical period.

[0022] Furthermore, step S2 is specifically as follows:

[0023] S21, extracting the areas where changes have occurred in the forest cover map based on the reconstructed forest cover spatial distribution dataset;

[0024] S22, select 16 driving factors from four categories: meteorology, geography, economic society, and spatial relations;

[0025] S23. Use random forest machine learning methods to explore the relationship between driving factors and forest cover change, and construct a spatial probability distribution model for forest cover change;

[0026] S24. Based on the spatial probability distribution model of forest cover change, combined with the hybrid geographic cellular automaton model, the historical forest statistical data of geographical divisions are used as constraints to reconstruct the spatiotemporal distribution dataset of forest cover in the historical period.

[0027] Furthermore, the ancient tree information collected in step S3 is used to verify the accuracy of the forest cover spatial distribution dataset in the historical period, including:

[0028] S311, Kappa coefficient, overall accuracy, user accuracy, producer accuracy, relative entropy, and quality factor were used to test the accuracy of the reverse spatial modeling model;

[0029] S312. Collect a large number of ancient trees with detailed coordinate records and a tree age of more than 100 years for accuracy verification of the forest spatial distribution dataset in the historical period.

[0030] Furthermore, in step S3, a spatiotemporal dynamic monitoring system for forest cover changes in historical periods is established, specifically: based on the spatiotemporal distribution dataset of forest cover in historical periods reconstructed in step S2, the distribution and change patterns of forest cover in historical periods are analyzed from the time scale and spatial scale, and an evaluation standard for the spatiotemporal distribution of forests in historical periods is established.

[0031] Compared with the prior art, the present invention has the following beneficial effects:

[0032] The present invention unifies historical forest survey data and remote sensing monitoring products into the same standard system, and reconstructs the forest cover spatial distribution data set, realizing the historical archive tracing and its cross-domain integration with remote sensing big data. Through this operation, the forest definition in the historical forest survey process is unified into one standard, eliminating the statistical area error caused by different forest definitions.

[0033] The present invention integrates historical archive tracing with remote sensing big data in a cross-domain and multi-dimensional manner (combining historical archives with remote sensing technology, using big data analysis methods, across different fields and dimensions, involving comparison and analysis of surface information obtained by remote sensing technology with historical archive data, and realizing the integration of multi-faceted information), and combines spatial reverse modeling technology to achieve reconstruction of the spatiotemporal distribution of forests in historical periods. Based on this, a spatiotemporal dynamic monitoring system for forest cover changes in historical periods is established, which can provide a reference for tracing the forest disturbance process over the past century and estimating historical forest carbon reserves, and can also provide a basic data source for predicting future forest changes, assessing land-atmosphere interactions, global climate change, and land cover simulation. The present invention takes into account the spatial relationship of land cover, which helps to understand the spatial differences of historical forest cover at different time scales, understand the human activity-society-nature coupling mechanism under the dynamic changes of forests, and has the high spatiotemporal resolution required to provide basic data support for the future earth "digital twin" model.

[0034] After reconstructing the spatiotemporal distribution dataset of forest cover in the historical period, the present invention collects ancient tree information, which can better reflect the actual situation of the spatial distribution of historical forests through the ancient tree information, and verify the accuracy of the spatial distribution dataset of forest cover in the historical period based on this, thereby ensuring the accuracy of the spatiotemporal dynamic monitoring system for forest cover changes in the historical period established based on the spatiotemporal distribution dataset of forest cover in the historical period.

[0035] The present invention will be explained in detail below with reference to the accompanying drawings and specific embodiments. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 is a flow chart of the present invention;

[0037] Figure 2 It is a step diagram for unifying the forest survey data of China in the historical period into the same standard system in terms of time scale in the present invention;

[0038] Figure 3 A diagram showing the steps of unifying the land cover products monitored by remote sensing from spatial scales to the same standard system in the present invention;

[0039] Figure 4 A diagram showing the steps of combining historical archive tracing and reverse spatial modeling to reconstruct forest cover maps of historical periods in the present invention;

[0040] Figure 5 A diagram showing the steps of establishing a spatiotemporal dynamic monitoring system for forest cover changes in historical periods in the present invention;

[0041] Figure 6 These are several example diagrams of forest spatiotemporal distribution maps based on the fusion of historical archives and remote sensing big data in the present invention. DETAILED DESCRIPTION

[0042] To facilitate understanding of the present invention, the present invention will be described more comprehensively below with reference to the relevant drawings. Several embodiments of the present invention are given in the drawings, but the present invention can be implemented in different forms and is not limited to the embodiments described in the text. On the contrary, these embodiments are provided to make the content disclosed in the present invention more thorough and comprehensive.

[0043] Please refer to the attached Figure 1 , a forest spatiotemporal distribution reconstruction method based on the fusion of historical archives and remote sensing big data, including the following steps:

[0044] S1. Unify historical forest survey data and remote sensing monitoring products into the same standard system: Based on the definition of forest canopy density, unify historical forest survey data in China into the same standard system in terms of time scale; adopt a voting strategy of multiple remote sensing data assisted by the Normalized Difference Vegetation Index (NDVI) to unify remote sensing monitoring land cover products into the same standard system in terms of spatial scale;

[0045] Please refer to the attached Figure 2 In step S1, the historical Chinese forest survey data are unified into the same standard system in terms of time scale:

[0046] S111, comprehensively search national forest resource inventory data, local chronicles, ancient maps, site data, statistical reports and works, etc., to extract direct information on changes in forest area during historical periods;

[0047] S112, in view of the fact that the definition of forest has changed many times since 1900: since the 1950s, the canopy density>0.4 has been used as the standard for forested land; in the late 1970s, in order to solve the problem of inconsistent decimal selection and keep the forest resource data of the previous 30 years usable, it was clearly stipulated that the canopy density>0.3 was the standard; since 1994, the new regulation was changed to canopy density>0.2. In order to unify the standards, the forest area change rate of each province during the period when the canopy density was adjusted from 0.3 to 0.2, as well as the forest area change rate of the province before and after the two periods were calculated;

[0048] S113, check the forest area change rate of each province one by one, and calculate the forest cover area during the canopy density adjustment period based on the maximum forest area change rate of the two-period inventory data before and after the canopy density adjustment;

[0049] S114: Unify the forest definition in the inventory data of the historical period to a standard system consisting of tree species with a canopy density of 0.2 or above (including 0.2).

[0050] Through the above processing, the statistical area errors caused by different forest definitions were solved.

[0051] Please refer to the attached Figure 3 In step S1, the land cover products of remote sensing monitoring are unified from spatial scale to the same standard system as follows:

[0052] S121, national forest inventory data and remote sensing monitoring land use / cover datasets. Due to differences in forest definitions, practicality of training datasets, uncertainty of models, etc., there have always been large differences in forest cover records between different datasets, which has led to the inability of remote sensing monitoring land cover datasets to reproduce the long-term trend of forest area reflected in China's forest inventory data. In order to unify the standards, multiple sets of remote sensing monitoring land cover datasets were automatically screened, reclassified, resampled, and reprojected to the Albers orthogonal equal-area double standard parallel conic projection;

[0053] S122, using a voting strategy involving multiple types of remote sensing monitoring land cover products to generate a potential forest cover probability map for China, wherein each pixel of the probability map is composed of a number of votes;

[0054] S123, based on the reconstructed forest resource inventory data records of each province (obtained from step S114), the annual maximum NDVI (normalized difference vegetation index) is used to assist the decision-making, and when the vote numbers are the same, the pixel with the largest NDVI is retained and defined as a forest;

[0055] S124, reconstructs the historical forest cover map composed of tree species with a canopy density of 0.2 or above (including 0.2). The forest cover map shows the detailed spatial distribution of China's forest cover during the remote sensing period.

[0056] S2. Combine historical archive tracing and reverse spatial modeling to reconstruct forest cover maps in historical periods: Based on random forest machine learning, establish a spatial probability distribution model for forest cover change; then based on hybrid geographic cellular automata, use historical forest survey data as constraints to reconstruct the spatiotemporal distribution dataset of forest cover in China in historical periods;

[0057] Please refer to the attached Figure 4 , step S2 is specifically as follows:

[0058] S211, extracting the areas where changes have occurred in the forest cover map according to the reconstructed forest cover dataset (the forest cover dataset from 2000 to 2020 reconstructed based on multi-source remote sensing land cover products obtained in step S124);

[0059] S212, select meteorological, geographical, economic and social, and spatial relationship factors, totaling 16 driving factors (temperature, precipitation, elevation, slope, aspect, shadow, soil type, population density, gross domestic product, night light data, distance to water system, distance to highway, distance to railway, distance to road, distance to city, and distance to residential area);

[0060] S213, using random forest machine learning methods to explore the relationship between driving factors and forest cover change, and constructing a spatial probability distribution model for forest cover change;

[0061] S214, based on the spatial probability distribution model of forest cover change, combined with the hybrid geographic cellular automaton model, and with the historical forest statistical data of China's six major geographical divisions as constraints, reconstructed the spatiotemporal change dataset of China's forest cover in the historical period.

[0062] The above steps S213 and S214 are reverse space modeling.

[0063] S3. Establish a spatiotemporal dynamic monitoring system for forest cover changes in historical periods: collect more than 300,000 ancient trees with detailed coordinate records and a tree age of more than 100 years for accuracy verification of historical forest spatial distribution datasets; realize spatiotemporal dynamic monitoring of China's forest cover changes at long time series, national, geographical division, provincial and pixel scales.

[0064] Please refer to the attached Figure 5 , step S3 is specifically as follows:

[0065] S311, the Kappa coefficient (a statistical indicator used to measure classification accuracy), overall accuracy, user accuracy, producer accuracy, relative entropy, and quality factor were used to test the accuracy of the reverse spatial modeling model to verify whether the model is applicable;

[0066] S312, collecting more than 300,000 ancient trees with detailed coordinate records and an age of more than 100 years for accuracy verification of the historical forest spatial distribution dataset (obtained in step S124). Ancient trees can be used as signals reflecting the historical forest spatial distribution. Assuming that each natural ancient tree corresponds to at least one hectare of forest, the accuracy of the historical forest spatial distribution can be effectively verified by checking more than 300,000 ancient trees one by one;

[0067] S321, establishing an evaluation standard for the spatiotemporal distribution of forests in the historical period, that is, based on the spatiotemporal change dataset of forest cover in China in the historical period reconstructed in step S214, analyzing the distribution and change patterns of forest cover in China in the historical period from the time scale and spatial scale;

[0068] S322, realizes spatiotemporal dynamic monitoring of forest cover changes in China at long time series, national, geographical division, provincial and pixel scales.

[0069] Please refer to the attached Figure 6 , Figure 6 The spatiotemporal distribution map of forests based on the fusion of historical archives and remote sensing big data in this embodiment is shown (1900, 1950, 1980, 2000, 2010, and 2020).

[0070] from Figure 6 It is not difficult to see that the present invention realizes the spatiotemporal dynamic monitoring of forest cover changes in China at the scale of long time series, country, geographical division, province and pixel. It also has the spatial resolution of 1km required to provide basic data support for the future earth "digital twin" model. Clarifying the spatiotemporal change trend of China's forests since the 20th century not only provides a reference for tracing the forest disturbance process over the past century and estimating historical forest carbon storage, but also provides a basic data source for predicting future forest changes, evaluating land-atmosphere interactions, global climate change and land cover / use simulation.

[0071] The above is an exemplary description of the present invention in combination with the accompanying drawings. It is obvious that the specific implementation of the present invention is not limited to the above-mentioned method. As long as such non-substantial improvements are made by adopting the method concept and technical solution of the present invention, or the concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A method for reconstructing forest spatiotemporal distribution based on the fusion of historical archives and remote sensing big data, characterized by: The steps include: S1. Unify historical forest survey data and remote sensing monitoring products into the same standard system, and reconstruct the historical forest cover spatial distribution dataset; S2, combining the data set obtained in step S1 with the inverse spatial modeling to reconstruct the spatiotemporal distribution data set of forest cover in the historical period; S3. Collect information on ancient trees to verify the accuracy of the forest cover spatial distribution dataset in the historical period; establish a spatiotemporal dynamic monitoring system for forest cover changes in the historical period; Step S1 includes: S11. Based on the definition of forest canopy density, the historical forest survey data should be unified into the same standard system in terms of time scale; S12. Use a variety of remote sensing data voting strategies assisted by the Normalized Difference Vegetation Index to unify the land cover products monitored by remote sensing from spatial scales to the same standard system; Step S11 specifically includes: S111. Comprehensively search national forest resource inventory data, local chronicles, ancient maps, site data, statistical reports and works to extract direct information on changes in forest area during historical periods; S112. Calculate the forest area change rate during the canopy density adjustment period for each province, as well as the forest area change rate before and after the two periods for each province; S113. Check the forest area change rate of each province one by one, and calculate the forest cover area during the canopy density adjustment period based on the maximum forest area change rate of the two-period inventory data before and after the canopy density adjustment; S114. Unify the definition of forests in the historical inventory data to a standard system consisting of tree species with a canopy density of 0.2 or above; Step S12 specifically includes: S121, automatically screen multiple sets of remote sensing monitoring land cover datasets, uniformly reclassify, resample, and reproject to Albers orthogonal equal-area double standard parallel conic projection; S122, using a voting strategy involving multiple types of remote sensing monitoring land cover products to generate a potential forest cover probability map, wherein each pixel of the probability map is composed of a number of votes; S123, based on the reconstructed forest resource inventory data records of each province, the annual maximum normalized difference vegetation index is used to assist the decision-making, and when the votes are the same, the pixel with the largest NDVI is defined as the forest; S124. Reconstruct the forest cover map composed of tree species with a canopy density of 0.2 or above, and generate a dataset of spatial distribution of forest cover in the historical period; Step S2 is specifically as follows: S21, extracting the areas where changes have occurred in the forest cover map based on the reconstructed forest cover spatial distribution dataset; S22, select 16 driving factors from four categories: meteorology, geography, economic society, and spatial relations; S23. Use random forest machine learning methods to explore the relationship between driving factors and forest cover change, and construct a spatial probability distribution model for forest cover change; S24. Based on the spatial probability distribution model of forest cover change, combined with the hybrid geographic cellular automaton model, the historical forest statistical data of geographical divisions are used as constraints to reconstruct the spatiotemporal distribution dataset of forest cover in the historical period; The ancient tree information collected in step S3 is used to verify the accuracy of the forest cover spatial distribution dataset in the historical period, including: S311, Kappa coefficient, overall accuracy, user accuracy, producer accuracy, relative entropy, and quality factor were used to test the accuracy of the reverse spatial modeling model; S312. Collect a large number of ancient trees with detailed coordinate records and an age of more than 100 years for accuracy verification of forest spatial distribution datasets in historical periods; The establishment of a spatiotemporal dynamic monitoring system for forest cover changes in historical periods in step S3 is specifically as follows: based on the spatiotemporal distribution dataset of forest cover in historical periods reconstructed in step S2, the distribution and change patterns of forest cover in historical periods are analyzed from the time scale and spatial scale, and an evaluation standard for the spatiotemporal distribution of forests in historical periods is established.

Images