Identification method of vegetation sensitivity to water changes based on remote sensing index

Abstract

The present invention discloses a method for identifying the sensitivity of vegetation to water changes based on remote sensing indexes, which relates to the technical field of identifying the sensitivity of vegetation to water changes, and includes the following steps: collecting and constructing a long-term remote sensing database with grids as units in the study area, including a vegetation remote sensing database of NDVI and a water remote sensing database of PDSI. The water sensitivity index constructed by the present invention based on the standard deviation data of NDVI and PDSI can accurately identify plots of land where vegetation is sensitive to water changes, and is operational. Secondly, the method of removing seasonal factors and retaining a complete long-term series makes full use of the advantages of remote sensing big data to discover macro-ecological laws, overcomes the problem of time period discretization, and has high efficiency and continuity. In addition, an independent and concise sensitivity index is proposed to make up for the shortcomings of existing research and provide guidance for ecosystem restoration and management.

Description

Technical Field

[0001] The invention relates to the technical field of vegetation sensitivity identification to water changes, and in particular to a method for vegetation sensitivity identification to water changes based on remote sensing indexes. Background Art

[0002] In recent years, with the intensification of climate change and global warming, water and water resources issues have gradually evolved into core issues of concern to mankind today. Studies have shown that if the current emission trend remains unchanged, the expected target of warming by the end of the 21st century will reach 3°C. As global temperatures rise, the global water resources distribution pattern will change. With global warming and changes in the water resources distribution pattern, the frequency of extreme weather events in the world has increased year by year, and the increase in extreme precipitation increases and decreases has further aggravated the phenomenon of global water anomalies. Floods and droughts, as extreme manifestations of global water anomalies, have caused huge losses to both human society and the ecosystem. The unevenness of water and water resources has evolved into an urgent global environmental issue.

[0003] Vegetation is one of the important driving factors for changes in global water resources, water environment, and water ecological patterns, and it has multiple interactive mechanisms with water. Under favorable water conditions, rapid vegetation growth will quickly consume soil moisture. When it exceeds the carrying capacity of water resources under long-term average temperature conditions, it will lead to structural overshoot and increase the risk of drought. This negative hysteresis effect includes direct and lagged effects of drought stress, temperature, and past vegetation anomalies. Temperature and precipitation in climate factors also play an important role in this process. When drought occurs, the ecosystem shows high sensitivity to synchronous precipitation anomalies, while the hysteresis effect on temperature is relatively significant. Compared with water stress, temperature stress may lead to more frequent and more severe drought events, and the depletion of soil moisture due to future vegetation increase caused by climate warming will further aggravate the frequency and intensity of hyper-drought events.

[0004] Therefore, against the backdrop of increasing global drought risks, exploring the quantitative response of vegetation to water changes and its distribution pattern, and exploring the driving mechanisms of vegetation responses to water changes from the dimensions of temperature, precipitation, soil properties, and vegetation biomass, are of great guiding significance for managers to formulate decision-making related drought prevention and disaster reduction policies and implement efficient water resources management and ecological environmental governance.

[0005] Most existing studies use drought index and vegetation condition index as causal terms for correlation or regression analysis, and analyze the trend changes, correlation coefficients and regression coefficients between drought index and vegetation index to obtain the results of the difference in sensitivity between vegetation and drought. Some studies focus on the occurrence and duration of water scarcity, and study the accumulation and lagged response mechanism of vegetation to water deficit, such as identifying drought in the growing season and different seasons through drought index and finding that it shows different changes and mechanisms to vegetation index, or calculating long-term water anomalies with drought index to identify the time and intensity of drought effects, and studying its abnormal relationship with vegetation productivity; some studies focus on the state of ecosystems during water stress, and study the differences in anti-disturbance and recovery capabilities of different vegetation and habitats.

[0006] The above studies explain the relationship between water changes and vegetation and provide guidance for ecological restoration, but they do not propose an independent and concise indicator to characterize sensitivity, which makes their explanation of vegetation water change sensitivity weak and insufficient. Some studies have proposed an independent indicator to show the sensitivity of vegetation to water changes. For example, the slope index is calculated by long-term series of NDVI and Standardized Precipitation Evapotranspiration Index (SPEI) to express sensitivity; based on the Copula probability framework, NDVI and SPIE are combined to analyze the loss of vegetation under the same drought level through conditional probability values ​​to show sensitivity; based on the VegDRI indicator, the response of vegetation to water changes is evaluated. These studies proposed an independent and concise indicator to show sensitivity, but they are all based on the original drought and vegetation time series data, or classified studies are carried out by quarter, or a certain growing season is studied, resulting in incoherent time series or data loss, and unable to reflect the common laws of long-term series, which will weaken or cover up some objective laws that can be reflected in long-term series by seasonal changes. Although discontinuous time series can obtain some objective laws, continuous time series data can better meet the actual needs of human policy making. Some studies have decomposed the original time series into trend terms, seasonal terms and residual terms by using the break for additive seasonal and trend (BFAST) method and the STL method to obtain continuous long-term data after removing seasonal changes. However, no specific indicators have been proposed in the studies to determine the sensitivity of vegetation to drought.

[0007] Although existing technologies have studied the relationship between vegetation and sensitivity to water changes, none of them have simultaneously studied the sensitivity relationship between vegetation and water changes based on a long-term series that completely removes seasonal changes and a concise and efficient sensitivity index. Therefore, further improvement is needed.

[0008] The above information disclosed in this Background section is only for enhancement of understanding of the background of the present disclosure and therefore it may contain information that does not constitute the prior art that is already known to one of ordinary skill in the art. Summary of the invention

[0009] The purpose of the present invention is to provide a method for identifying the sensitivity of vegetation to water changes based on remote sensing indexes, and to construct a vegetation water sensitivity index based on standard deviation data of NDVI and PDSI. The water sensitivity index can accurately identify plots of vegetation that are sensitive to water changes, and is operational. Secondly, the technical method of removing seasonal factors and retaining complete long-term series helps to give full play to the advantages of remote sensing big data and explore ecological laws at the macro level, breaking through the time period discretization problem of existing research methods, and has the advantages of high efficiency and strong continuity. Finally, an independent and concise indicator is proposed to characterize sensitivity, which makes up for the lack of existing indicators for vegetation sensitivity to water changes, and provides guidance for ecosystem restoration and management to solve the problems in the above-mentioned background technology.

[0010] In order to achieve the above object, the present invention provides the following technical solution: a method for identifying the sensitivity of vegetation to water changes based on remote sensing index, comprising the following steps:

[0011] Collect and construct a long-term remote sensing database based on grids in the study area, including the vegetation remote sensing database of NDVI and the moisture remote sensing database of PDSI;

[0012] Remove seasonal factors from long time series by using STL time series decomposition method;

[0013] Based on the standard deviation data of NDVI and PDSI, the vegetation water sensitivity index VWSI is constructed to show the sensitivity of vegetation to changes in water conditions;

[0014] The Moran index test method was used to identify the areas where the VWSI high-value concentrated plots were located, that is, the areas where vegetation was sensitive to water changes;

[0015] The response of NDVI and the changes of PDSI in the study area were analyzed based on the Granger causality test method.

[0016] Preferably, the NDVI calculated the monthly mean data during the study period and resampled it to 0.5°×0.5° using the nearest neighbor method to reveal the changes in vegetation growth conditions in each grid, and the PDSI was converted into monthly data using Matlab to reveal the changes in moisture conditions in each 0.5°×0.5° grid.

[0017] Preferably, NDVI is a vegetation index calculated based on the visible and near-infrared bands of a multispectral sensor, which is used to quantify vegetation greenness and assess changes in plant health:

[0018] When plants are healthy, they absorb red spectral radiation and reflect near-infrared radiation, and the NDVI value becomes larger, but the NDVI is less than 1;

[0019] When plants are unhealthy or affected by some stress factors, they will reflect red light and absorb near-infrared radiation. At this time, the NDVI value will become smaller, but the NDVI is greater than 0;

[0020] If there is no vegetation cover, due to the lack of green biomass, red light will be reflected in the area and no near-infrared radiation will be absorbed. At this time, NDVI continues to decrease, but NDVI is greater than -1.

[0021] Preferably, PDSI incorporates the previous and current precipitation and demand for water supply into a hydrological accounting system, which includes a two-layer bucket model for soil moisture calculation, reflects the soil moisture content within one meter of the soil surface through the monthly PDSI time series, and can reflect the spatiotemporal relationship between soil moisture and meteorology.

[0022] Preferably, the steps of removing seasonal factors from long time series by STL time series decomposition method are as follows:

[0023] The NDVI and PDSI time series of each grid of the multi-year time series data were converted from raster data to tabular data, and the forecast package in R was used to decompose the STL data, and each grid time series data was decomposed into three parts: trend, season and residual;

[0024] The trend items NDVI and PDSI data were converted from table data to raster data through ArcGIS, thus obtaining two new time series, Remove seasonal NDVI and Remove seasonal PDSI, which removed the seasonal fluctuations of the time series and only retained their own temporal changes.

[0025] The raw data items are named RawNDVI and RawPDSI to distinguish them.

[0026] Preferably, after STL decomposition removes the seasonal fluctuation factors of NDVI and PDSI, its long time series data can be expressed as the changes of regional NDVI and PDSI over many years;

[0027] For NDVI, the seasonal factors eliminated by STL decomposition include the planting and harvesting of crops, the withering and flourishing of herbs, and the changes in the leaf quality of shrubs and trees in different seasons. Therefore, NDVI reflects the changes in vegetation quality in the region over the years when seasonal factors are eliminated. For PDSI, the seasonal factors eliminated by STL decomposition include the changes in moisture differences in different periods of the year. Therefore, PDSI reflects the changes in moisture in the region over the years when seasonal factors are eliminated.

[0028] The standard deviation is used to reflect the degree of dispersion of multi-year time series data sets. Indicates the fluctuation of the NDVI sequence changes in the region over the years. Indicates the fluctuation of the PDSI series changes in the region over many years.

[0029] Preferably, the fluctuations of two sets of long time series are compared to As the denominator, the regional moisture fluctuation is expressed as As the numerator, it represents the regional vegetation cover fluctuation, through To reflect the vegetation coverage fluctuations under different water fluctuations in the region, Divide by To obtain the vegetation sensitivity index to water changes, namely VWSI, the specific calculation formula of standard deviation and VWSI is:

[0030]

[0031]

[0032] In the formula, n is the length of the time series, For the i The value of the time series, is the average of all time series, VWSI is the vegetation water sensitivity index, is the standard deviation of the NDVI time series after removing seasonal fluctuations, is the standard deviation of the PDSI time series after removing seasonal fluctuations.

[0033] Preferably, the Moran's index is divided into a global Moran's index and a local Moran's index. A global Moran's index is first performed for a region. The global index reflects whether clustering or outliers occur in the space. If global autocorrelation occurs, a local autocorrelation is then performed. The local Moran's index will identify outliers or clusters.

[0034] LISA cluster analysis reflects local indicators of spatial connections. The LISA cluster map divides all geographical relationships into four types: HH, LL, HL, and LH. HH refers to high-value geographical units surrounded by high-value geographical units, indicating high-value clusters; LL refers to low-value geographical units surrounded by low-value geographical units, indicating low-value clusters; HL refers to high-value geographical units surrounded by low-value geographical units, indicating high-value isolated points; LH refers to low-value geographical units surrounded by high-value geographical units, indicating low-value isolated points.

[0035] The VWSI raster data were converted into points using ArcGIS, and the Anselin Local Moran's I operation was performed. In the Conceptualization of Spatial Relationships, Inverse distance was selected, Euclidean was selected for Distance Method, Row was selected for Standardization, and the default settings were used for Distance Band or Threshold Distance. The obtained points were then converted into a 0.5° × 0.5° grid, and the HH representing the VWSI high value clusters were extracted and defined as highly sensitive areas.

[0036] Preferably, the Granger causality test is used to test whether one set of time series is another set of time series. If A is the Granger cause of B, it means that the change of A is one of the causes of the change of B; if under the condition that the past information of variables X and Y is included, the prediction effect of variable Y is better than the prediction effect of Y based on the past information of Y alone, that is, variable X helps to explain the future changes of variable Y, then variable X is considered to be the Granger cause of variable Y.

[0037] Preferably, the old and new time series of NDVI and PDSI are stationary, and then four groups of Granger one-way causality tests are carried out using the forecast package in the residual R. The four groups of Granger causality tests are RsNDVI~RsPDSI and RsPDSI~RsNDVI, RawNDVI~RawPDSI, and RawPDSI~RawNDVI. The relationship between the time series is judged by whether the p-value of the Granger one-way causality test is significant.

[0038] In the above technical solution, the technical effects and advantages provided by the present invention are:

[0039] The present invention constructs a vegetation moisture sensitivity index based on the standard deviation data of NDVI and PDSI. The moisture sensitivity index can accurately identify plots of vegetation that are sensitive to moisture changes, and is operational. Secondly, the technical method of removing seasonal factors and retaining complete long-term series helps to give full play to the advantages of remote sensing big data and explore ecological laws at the macro level, breaking through the time period discretization problem of existing research methods, and has the advantages of high efficiency and strong continuity. Finally, an independent and concise indicator is proposed to characterize sensitivity, which makes up for the lack of existing indicators for vegetation moisture change sensitivity and provides guidance for ecosystem restoration and management. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For ordinary technicians in this field, other drawings can also be obtained based on these drawings.

[0041] Figure 1 The present invention is a flow chart of a method for identifying the sensitivity of vegetation to water changes based on remote sensing indices.

[0042] Figure 2 This is a flow chart of the calculation of VWSI of the present invention.

[0043] Figure 3 It is a remote sensing spectrum.

[0044] Figure 4 This is a schematic diagram of the STL decomposition of a typical grid NDVI.

[0045] Figure 5 This is a schematic diagram of the STL decomposition of a typical grid PDSI.

[0046] Figure 6 It is a scatter plot of RsNDVI~RawNDVI.

[0047] Figure 7 This is the scatter plot of RsPDSI~RawPDSI. DETAILED DESCRIPTION

[0048] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these example embodiments are provided so that the description of the present disclosure will be more comprehensive and complete, and the concept of the example embodiments will be fully conveyed to those skilled in the art.

[0049] The present invention provides Figures 1 to 7The method for identifying the sensitivity of vegetation to water changes based on remote sensing indices includes the following steps:

[0050] Collect and construct a long-term remote sensing database based on grids in the study area, including the vegetation remote sensing database of NDVI and the moisture remote sensing database of PDSI;

[0051] NDVI (Normalized Difference vegetation index, NDVI);

[0052] The monthly mean data of NDVI during the study period were calculated and resampled to 0.5°×0.5° using the nearest neighbor method to reveal the changes in vegetation growth conditions in each grid.

[0053] During the study period, the average value of the Normalized Difference Vegetation Index (NDVI) was calculated every month, and then the data were converted to a spatial resolution of 0.5°×0.5° using the nearest neighbor resampling method, so that changes in vegetation growth conditions can be analyzed within each grid cell. This processing method allows the observation and assessment of spatiotemporal changes in vegetation cover and health status at a consistent scale and resolution over a large area.

[0054] Multispectral remote sensing images contain a combination of multiple bands that form a composite image. Using multispectral images, the individual bands in the band combination can be transformed to better highlight certain features and patterns. Image band transformation has become a common practice to generate new images from two or more image bands for information extraction. The new images generated in this way enhance the representation of vegetation and other landforms. Currently, there are hundreds of vegetation indices derived from multispectral images.

[0055] NDVI is a vegetation index calculated based on the visible and near-infrared bands of multispectral sensors. It is also one of the earliest remote sensing analysis products used to simplify the complexity of multispectral images. NDVI can be used to quantify vegetation greenness and assess changes in plant health, and is particularly useful for understanding vegetation density. The growth status of plants has a significant impact on their absorption and reflection of spectral radiation. When plants are in a healthy state, they absorb more red spectral radiation and reflect more near-infrared radiation. In this case, the NDVI value gradually approaches 1, indicating that the vegetation is growing well. On the contrary, when plants are unhealthy or affected by some stress factors, they reflect more red light and absorb less near-infrared radiation. At this time, NIR is slightly greater than red light, causing the NDVI value to get closer and closer to 0. If there is no vegetation cover, due to the lack of green biomass, red light will be reflected in large quantities in this area and almost no near-infrared radiation will be absorbed, making the NDVI value close to -1. Therefore, using NDVI when assessing the growth status of regional vegetation has a good ability. Obtaining long-term series NDVI data as an indicator through remote sensing image data can effectively reflect changes in vegetation conditions. High NDVI values ​​indicate rich and lush vegetation coverage and healthy vegetation conditions; low NDVI values ​​mean a lack of vegetation coverage or poor vegetation conditions.

[0056] NIR stands for Near Infrared Radiation. In remote sensing, NIR is a part of the electromagnetic spectrum that typically ranges in wavelength from 0.7 to 1.3 microns. Plant leaves reflect a lot of light in this band, so healthy vegetation has a high reflectivity in the NIR and a relatively low reflectivity in the red band (around 0.6 to 0.7 microns). This difference in reflectivity is the basis for calculating the NDVI (Near Infrared Radiation). NDVI uses the difference in reflectivity of red and NIR light to measure the density and health of vegetation. Healthy vegetation reflects more NIR light and absorbs more red light, while bare soil and dead vegetation have a smaller difference in reflectivity between red and NIR light, so their NDVI values ​​are close to zero or negative.

[0057] PDSI (Palmer Drought Severity Index, PDSI);

[0058] The PDSI was converted into monthly data using Matlab and used to reveal the changes in moisture conditions at each 0.5°×0.5° grid.

[0059] The Palmer Drought Index (PDSI) data were processed into monthly data using Matlab software, and then converted to a spatial resolution of 0.5°×0.5° to analyze the changes in moisture conditions within each grid cell. This processing method allows the moisture conditions and drought severity of different regions to be observed and assessed on a monthly basis at a detailed spatial scale.

[0060] PDSI is a drought index widely used in meteorology, climate, agriculture and other fields. PDSI aims to measure the cumulative deviation of surface water balance. PDSI incorporates the precipitation and demand (potential evapotranspiration) of previous and current water supply into a hydrological accounting system, which includes a two-layer bucket model for soil moisture calculation. The study found that the monthly PDSI time series can better reflect the soil moisture content within one meter of the soil surface, and can better reflect the spatiotemporal relationship between soil moisture and meteorology.

[0061] PDSI assesses the dryness and wetness of a region based on the relationship between water supply and demand. When the local water supply is insufficient to meet demand, it is a drought state; when the water supply can meet demand, it is a wet state. It ranges from about -10 (dry) to +10 (wet), and values ​​below -3 indicate severe to extreme drought. Since PDSI can reflect the time series changes in water from the perspectives of wetness and drought, and also comprehensively reflect the water conditions of the environment from the two dimensions of atmosphere and soil, especially the water conditions at the soil level have a more significant and sensitive impact on vegetation growth, PDSI is used as an indicator of water changes.

[0062] Remove seasonal factors from long time series through STL time series decomposition method (Seasonal-Trend decomposition using Loess);

[0063] Seasonal-Trend decomposition using Loess (STL) is a common algorithm in time series decomposition. Based on LOESS, the data at a certain time Decomposition into trend components (trend component), periodic component (seasonal component) and remainder (remainder component). ,STL is mainly used to remove seasonal variation factors in data.,This paper can more accurately grasp the inherent laws of data by extracting the "essential information" in the time series.,In practical applications, the STL method has been widely recognized and applied.

[0064] The NDVI and PDSI time series of each grid of multi-year time series data are converted from raster data to tabular data. The forecast package in R is used to decompose the STL data, and each grid time series data (data) is decomposed into three parts: trend, seasonal, and remainder ( Figure 4 , Figure 5 ). The trend items (trend) NDVI and PDSI data are then converted from table data to raster data through ArcGIS, thus obtaining two new time series, Remove seasonal NDVI (RsNDVI) and Remove seasonalPDSI (RsPDSI), which remove the seasonal fluctuations of the time series and only retain their own temporal changes. The original data items (data) are named Raw NDVI (RawNDVI) and RawPDSI (RawPDSI) for distinction.

[0065] Plant growth conditions will change significantly with the seasons. RawNDVI ( Figure 4 , data) has strong seasonal fluctuations. In spring, as the temperature rises and the sunshine time increases, plants begin to enter the active period and quickly restore their photosynthesis capacity. At this time, the NDVI value gradually rises, that is, the vegetation coverage is increasing. Summer is the time when plants are most lush. Under the conditions of sufficient sunshine and abundant precipitation, a large amount of chlorophyll accumulates and photosynthesizes. At this stage, the NDVI value reaches a peak and remains relatively stable. In autumn, vegetation begins to decline due to factors such as the gradual decrease in temperature, the decrease in sunshine time, and the yellowing of the leaves of deciduous trees, which leads to a gradual decrease in the NDVI value. Winter reaches the lowest point of the year. In the cold and dry environment, many herbs and shrubs are dormant or dead. Due to the lack of fresh leaves and active cell tissues for photosynthesis to produce reflective signals, the NDVI value is the lowest.

[0066] RawNDVI shows stable regular fluctuations in long time series mainly due to seasonal changes. The huge changes in temperature, moisture and phenological conditions caused by these seasonal fluctuations are too strong, making it difficult to reflect the interannual differences in the vegetation's own conditions in the long period series, which is the most intuitive indicator needed in the evaluation of ecological damage and restoration benefits. Therefore, it is believed that the seasonal fluctuations in the NDVI long time series have largely shielded the year-to-year changes in vegetation conditions, and the changes that are taken for granted in seasonal fluctuations are too stereotyped. In the process of studying the changes for 23 years, it has become a signal interference item and cannot extract the information that this article wants to obtain, so it should be eliminated.

[0067] When removing seasonal interference, comparisons of the same month in different years or between growing seasons are also commonly used research methods. However, comparing a time period in a year wastes useful data in long time series and the advantage of remote sensing data in showing objective laws through large samples. Furthermore, the specific time period chosen varies from person to person and is highly random. At the same time, due to the lag of vegetation itself, selecting a single time period for research ignores, to a certain extent, the differences in vegetation in the same time period in different years caused by the lag, and also ignores the difference in vegetation changes between the selected time period and outside the selected time period. Therefore, the RsNDVI ( Figure 4 , trend) not only removes seasonal fluctuations, but also retains the complete long-term series for 23 years, comprehensively solving the two problems of comparing the same month in different years or comparing between growing seasons. The physical meaning of RsNDVI is the monthly characteristics of NDVI in a certain grid for 23 years. Similarly, RawPDSI ( Figure 5 , data) has strong seasonal fluctuations. For example, the monsoon climate zone has the climate characteristics of more rain in summer and less rain in winter. In order to remove the seasonal effect, PDSI is also decomposed by STL to eliminate its periodic signal, and RsPDSI ( Figure 5 , trend).

[0068] Based on the standard deviation data of NDVI (Normalized Difference Vegetation Index, NDVI) and PDSI (Palmer Drought Severity Index, PDSI), the vegetation water sensitivity index VWSI (The Vegetation Water Sensitivity Index) was constructed to show the sensitivity of vegetation to changes in water conditions;

[0069] We can understand the ecological significance of VWSI by starting with the construction formula of VWSI. First, after STL decomposition removes the seasonal fluctuation factors of NDVI and PDSI, its long time series data can be expressed as the changes in regional NDVI and PDSI over many years. For NDVI, the seasonal factors eliminated by STL decomposition include the planting and harvesting of crops, the withering and flourishing of herbaceous plants, and the changes in leaf quality of shrubs and trees in different seasons. Therefore, after eliminating seasonal factors, NDVI reflects the changes in regional vegetation quality over many years; for PDSI, the seasonal factors eliminated by STL decomposition include the changes in moisture differences in different periods of the year (this change is most obvious in the monsoon climate zone with more rain in summer and less rain in winter). After eliminating seasonal factors, PDSI reflects the changes in regional moisture over many years. On this basis, the standard deviation (StandardDeviation, Sd) is used to reflect the degree of dispersion of the multi-year time series data set. If the Sd value is high, it means that the data changes more and the fluctuations over many years are more obvious. Therefore, Indicates the fluctuation of the NDVI sequence changes in the region over the years. Indicates the fluctuation of the PDSI series changes in the region over many years.

[0070] "Sensitive" means "very keen on something or something, quick to perceive, and able to judge or react quickly." By comparing the fluctuations of two sets of long time series, As the denominator, the regional moisture fluctuation is expressed as As the numerator, it represents the fluctuation of regional vegetation cover. If there is a certain causal relationship between the two (, ,Y~X), then It reflects the vegetation coverage fluctuations under different water fluctuations in the region. Small and Big, at this time will be very large, indicating that the region presents large vegetation changes while small water fluctuations occur, that is, vegetation changes are highly sensitive to water changes. In summary, under the premise that the causal relationship can be proved, It can reflect the sensitivity of vegetation changes to water changes, so Divide by To obtain the vegetation sensitivity index to water changes, namely VWSI, the specific calculation formula of standard deviation and VWSI is:

[0071]

[0072]

[0073] In the formula, n is the length of the time series, is the value of the i-th time series, is the average value of all time series, VWSI is the water sensitivity index of vegetation in this paper, is the standard deviation of the NDVI time series after removing seasonal fluctuations, It is the standard deviation of the PDSI time series after removing seasonal fluctuations; the standard deviation can reflect the dispersion and fluctuation of the time series. According to the formula ; If a grid The smaller the value, The larger the value, the larger the VWSI value of the grid. This indicates that a large change in vegetation conditions occurs at the same time as a small change in water content in the long time series, that is, the change in NDVINDVI of the grid is highly sensitive to the response of PDSIPDSI. Since PDSI has positive and negative values, the physical meaning of VWSI can be reflected in the ability of vegetation to resist disasters under drought conditions when PDSI is negative, and can reflect the ability of vegetation to recover when water conditions improve when PDSI is positive.

[0074] The Moran index test method was used to identify the areas where the VWSI high-value concentrated plots were located, that is, the areas where vegetation was sensitive to water changes;

[0075] Moran's index is divided into global Moran's index (Global Moran's I) and local Moran's index (Local Moran's I). Usually, we first make a global Moran's index for a region. The global index tells us whether there are clusters or outliers in the space. If there is global autocorrelation, we then make a local autocorrelation. The local Moran's index will tell us where outliers or clusters appear.

[0076] LISA cluster analysis (Local indicators of spatial association, LISA) reflects the local indicators of spatial connection. LISA cluster diagram divides all geographical relationships into four types: HH, LL, HL and LH. HH refers to high-value geographical units surrounded by high-value geographical units, indicating high-value clusters; LL refers to low-value geographical units surrounded by low-value geographical units, indicating low-value clusters; HL refers to high-value geographical units surrounded by low-value geographical units, indicating high-value isolated points; LH refers to low-value geographical units surrounded by high-value geographical units, indicating low-value isolated points.

[0077] The VWSI raster data was converted to points using ArcGIS, and the Anselin Local Moran's I operation was performed. In the Conceptualization of Spatial Relationships, Inversedistance was selected, Euclidean was selected for Distance Method, and Row was selected for Standardization. The default settings were used for Distance Band or Threshold Distance. The obtained points were then converted to a 0.5°×0.5° grid. The HHs representing the high-value clusters of VWSI in the results were extracted and defined as highly sensitive areas.

[0078] The response of NDVI and the changes of PDSI in the study area were analyzed based on the Granger causality test method;

[0079] Granger causality test is used to test whether one set of time series is the cause of another set of time series. If A is the Granger cause of B, it means that the change of A is one of the causes of the change of B. If the prediction effect of variable Y is better than the prediction effect of Y based on the past information of Y alone under the condition of including the past information of variables X and Y, that is, variable X helps to explain the future changes of variable Y, then variable X is considered to be the Granger cause of variable Y. In more popular terms, that is, the goal is to predict the change of Y, and the prediction result of X is better than the prediction result of Y alone, then it is said that there is a Granger causal relationship between X and Y. In other words, the change of X can explain the change of Y.

[0080] The Granger causality test results show whether the causal relationship between two variables is unidirectional or bidirectional. The specific formula is:

[0081]

[0082]

[0083] There are four results:

[0084] ① When The whole is not zero, The whole is zero. At this time, x is the cause of the change in y, that is, there is a one-way causal relationship from x to y;

[0085] ② When The whole is zero, The overall value is not zero. At this time, y is the cause of the change in x, that is, there is a one-way causal relationship from y to x;

[0086] ③When and The total is not zero. At this time, x and y are mutually causal, that is, there is a one-way causal relationship from x to y, and there is also a one-way causal relationship from y to x;

[0087] ④ When and The total is zero. At this time, x and y are independent, or there is no causal relationship between x and y.

[0088] Since the Granger time series test itself has taken into account the lag effect between the two time series, it is not considered to deal too much with the possible lag between the two time series. This paper first conducts time series stationarity on the old and new time series of NDVI and PDSI, and then uses the forecast package in R to carry out four groups of Granger causality tests. The four groups of Granger causality tests are RsNDVI~RsPDSI (dependent variable Y~independent variable X) and RsPDSI~RsNDVI, RawNDVI~RawPDSI, and RawPDSI~RawNDVI. The relationship between the time series is judged by whether the p value of the Granger one-way causality test is significant.

[0089] The following are specific examples of implementation methods:

[0090] A remote sensing database from 2000 to 2022 was constructed and applied to 3538 grids across the country. The seasonal effect in the long-term series was eliminated by using the STL decomposition time series method (Seasonal-Trend decomposition using LOESS, STL). On this basis, the long-term changes of NDVI (Normalized DifferenceVegetation Index, NDVI) and PDSI (Palmer Drought Severity Index, PDSI) after eliminating the seasonal effect were analyzed using the standard deviation indicator. The sensitive response of NDVI changes to PDSI changes was defined as the vegetation water sensitivity index (TheVegetation Water Sensitivity Index, VWSI), and an efficient water sensitivity index was proposed. The spatial distribution and aggregation of the VWSI index were evaluated nationwide, and the scientific rationality was verified based on the Granger time series test method.

[0091] STL decomposition results. By comparing the scatter plot linear regression coefficients of the Raw time series and the Rs time series, it was found that there was no significant difference in the spatial distribution of NDVI and PDSI before and after STL decomposition ( Figure 6 , Figure 7). The slope of the RsNDVI~RawNDVI scatter plot is 0.999, R2>0.999, while the slope of the RsPDSI~RawPDSI scatter plot is 0.998, R2>0.99. This shows that the Raw time series and the Rs time series basically satisfy the linear relationship of y=x. Both sets of data show a high degree of fitting characteristics before and after STL decomposition, which fully demonstrates the scientificity of STL in processing long time series data.

[0092] Through the Granger test of 3538 grids across the country, it was found that when there was seasonal cycle interference, the Granger test did not show universal significant characteristics, which indicates that when the long-term series is still disturbed by seasonal changes, there is no universal causal relationship between the vegetation index and the moisture index in the long-term series. After removing the interference of seasonal cycle changes, the comparison test items RsPDSI~NDVI also did not show universal significance; only RsNDVI~RsPDSI showed obvious results nationwide, with a total of 2582 grids meeting the Granger significance requirements, accounting for 72.98% of all grids, of which 2242 grids met the extremely significant requirements, accounting for 63.37% of all grids. This shows that the PDSI changes in 72.98% of China's regions during 2000-2022 are one of the reasons for the changes in NDVI. When the long-term series removes seasonal changes, the changes in NDVI caused by PDSI in the study area are meaningful, and verify that the VWSI presentation proposed in this paper is not an accidental result, but an objective law in statistics, which is ecologically meaningful.

[0093] The present invention constructs a vegetation moisture sensitivity index based on the standard deviation data of NDVI and PDSI. The moisture sensitivity index can accurately identify plots of vegetation that are sensitive to moisture changes, and is operational. Secondly, the technical method of removing seasonal factors and retaining complete long-term series helps to give full play to the advantages of remote sensing big data and explore ecological laws at the macro level, breaking through the time period discretization problem of existing research methods, and has the advantages of high efficiency and strong continuity. Finally, an independent and concise indicator is proposed to characterize sensitivity, which makes up for the lack of existing indicators for vegetation moisture change sensitivity and provides guidance for ecosystem restoration and management.

[0094] The above description is only by way of illustration of certain exemplary embodiments of the present invention. It is undoubted that those skilled in the art can modify the described embodiments in various ways without departing from the spirit and scope of the present invention. Therefore, the above drawings and descriptions are illustrative in nature and should not be construed as limiting the scope of protection of the claims of the present invention.

Claims

1. A method for identifying vegetation sensitivity to water changes based on remote sensing indices, characterized in that: The following steps are involved: Collect and construct a long-term remote sensing database based on grids in the study area, including the vegetation remote sensing database of NDVI and the moisture remote sensing database of PDSI; Remove seasonal factors from long time series by using STL time series decomposition method; Based on the standard deviation data of NDVI and PDSI, the vegetation water sensitivity index VWSI is constructed to show the sensitivity of vegetation to changes in water conditions; After STL decomposition removes the seasonal fluctuation factors of NDVI and PDSI, its long-term series data can be expressed as the changes of regional NDVI and PDSI over many years. For NDVI, the seasonal factors eliminated by STL decomposition include the planting and harvesting of crops, the withering and flourishing of herbs, and the changes in the leaf quality of shrubs and trees in different seasons. Therefore, NDVI reflects the changes in vegetation quality in the region over the years when seasonal factors are eliminated. For PDSI, the seasonal factors eliminated by STL decomposition include the changes in moisture differences in different periods of the year. Therefore, PDSI reflects the changes in moisture in the region over the years when seasonal factors are eliminated. The standard deviation is used to reflect the degree of dispersion of multi-year time series data sets. Indicates the fluctuation of the NDVI sequence changes in the region over the years. It indicates the fluctuation of the PDSI series changes in the region over the years; Compare the fluctuations of two sets of long time series to As the denominator, the regional moisture fluctuation is expressed as As the numerator, it represents the regional vegetation cover fluctuation, through To reflect the vegetation coverage fluctuations under different water fluctuations in the region, Divide by To obtain the vegetation sensitivity index to water changes, namely VWSI, the specific calculation formula of standard deviation and VWSI is: ; ; In the formula, n is the length of the time series, For the i The value of the time series, is the average of all time series, VWSI is the vegetation water sensitivity index, is the standard deviation of the NDVI time series after removing seasonal fluctuations, is the standard deviation of the PDSI time series after removing seasonal fluctuations; The Moran index test method was used to identify the areas where the VWSI high-value concentrated plots were located, that is, the areas where vegetation was sensitive to water changes; The response of NDVI and the changes of PDSI in the study area were analyzed based on the Granger causality test method.

2. The method for identifying vegetation sensitivity to water changes based on remote sensing index according to claim 1, characterized in that: The monthly mean data of NDVI during the study period were calculated and resampled to 0.5°×0.5° using the nearest neighbor method to reveal the changes in vegetation growth conditions in each grid. The PDSI was converted into monthly data using Matlab to reveal the changes in moisture conditions in each 0.5°×0.5° grid.

3. The method for identifying vegetation sensitivity to water changes based on remote sensing index according to claim 1, characterized in that: NDVI is a vegetation index calculated based on the visible and near-infrared bands of a multispectral sensor, used to quantify vegetation greenness and assess changes in plant health: When plants are healthy, they absorb red spectral radiation and reflect near-infrared radiation, and the NDVI value becomes larger, but the NDVI is less than 1; When plants are unhealthy or affected by some stress factors, they will reflect red light and absorb near-infrared radiation. At this time, the NDVI value will become smaller, but the NDVI is greater than 0; If there is no vegetation cover, due to the lack of green biomass, red light will be reflected in the area and no near-infrared radiation will be absorbed. At this time, NDVI continues to decrease, but NDVI is greater than -1.

4. The method for identifying vegetation sensitivity to water changes based on remote sensing index according to claim 1, characterized in that: PDSI incorporates the previous and current precipitation and demand for water supply into a hydrological accounting system, which includes a two-layer bucket model for soil moisture calculation. The monthly PDSI time series reflects the soil moisture content within one meter of the soil surface and can reflect the spatiotemporal relationship between soil moisture and meteorology.

5. The method for identifying vegetation sensitivity to water changes based on remote sensing index according to claim 1, characterized in that: The steps to remove seasonal factors from long time series by STL time series decomposition method are as follows: The NDVI and PDSI time series of each grid of the multi-year time series data were converted from raster data to tabular data, and the forecast package in R was used to decompose the STL data, and each grid time series data was decomposed into three parts: trend, season and residual; The trend items NDVI and PDSI data were converted from table data to raster data through ArcGIS, thus obtaining two new time series, Remove seasonal NDVI and Remove seasonal PDSI, which removed the seasonal fluctuations of the time series and only retained their own temporal changes. The raw data items are named RawNDVI and RawPDSI to distinguish them.

6. The method for identifying vegetation sensitivity to water changes based on remote sensing index according to claim 1, characterized in that: The Moran index is divided into global Moran index and local Moran index. First, make a global Moran index for a region. The global index reflects whether there are clusters or outliers in the space. If there is global autocorrelation, then make a local autocorrelation. The local Moran index will identify outliers or clusters. LISA cluster analysis reflects local indicators of spatial connections. The LISA cluster map divides all geographical relationships into four types: HH, LL, HL, and LH. HH refers to high-value geographical units surrounded by high-value geographical units, indicating high-value clusters; LL refers to low-value geographical units surrounded by low-value geographical units, indicating low-value clusters; HL refers to high-value geographical units surrounded by low-value geographical units, indicating high-value isolated points; LH refers to low-value geographical units surrounded by high-value geographical units, indicating low-value isolated points. The VWSI raster data were converted into points using ArcGIS, and the Anselin Local Moran's I operation was performed. In the Conceptualization of Spatial Relationships, Inverse distance was selected, Euclidean was selected for Distance Method, Row was selected for Standardization, and the default settings were used for Distance Band or Threshold Distance. The obtained points were then converted into a 0.5° × 0.5° grid, and the HH representing the VWSI high value clusters were extracted and defined as highly sensitive areas.

7. The method for identifying vegetation sensitivity to water changes based on remote sensing index according to claim 1, characterized in that: Granger causality test is used to test whether one set of time series is another set of time series. If A is the Granger cause of B, it means that the change of A is one of the causes of the change of B. If the prediction effect of variable Y is better than the prediction effect of Y based on the past information of Y alone, that is, variable X helps to explain the future changes of variable Y, then variable X is considered to be the Granger cause of variable Y.

8. The method for identifying vegetation sensitivity to water changes based on remote sensing index according to claim 1, characterized in that: The new and old time series of NDVI and PDSI were stationary, and then four groups of Granger one-way causality tests were carried out using the forecast package in R. The four groups of Granger causality tests were RsNDVI~RsPDSI, RsPDSI~RsNDVI, RawNDVI~RawPDSI, and RawPDSI~RawNDVI. The relationship between the time series was judged by whether the p-value of the Granger one-way causality test was significant.

Images