Method for quantifying synchronous influence and early-stage remaining effect of extreme climate events on land carbon flux

Abstract

The invention discloses a method for quantifying synchronous influence and early-stage remaining effect of extreme climate events on land carbon flux, and relates to the technical field of data analysis and artificial intelligence. Compared with a traditional regression method, the method adopts a machine learning method, has the advantages of automatic decision making, adaptive learning, real-time calculation, reduction of human errors and the like, and can better process complexity and a nonlinear relation; a machine learning algorithm can automatically extract a mode from data and make a decision according to the mode, so that the need of manual intervention is reduced, and the working efficiency is remarkably improved; along with rapid increase of data, a machine learning system can continuously learn and update a decision or prediction model from new data, and rapidly process and analyze massive data sets; through big data analysis and pattern recognition, machine learning can reveal deep association and trends hidden in data.

Description

Technical Field

[0001] The present invention relates to the field of data analysis and artificial intelligence technology, and specifically to a method for quantitatively calculating the abnormal contribution of climate conditions and vegetation status to carbon flux of terrestrial ecosystems after extreme climate occurs based on machine learning. Background Art

[0002] Extreme climate events have significant impacts on carbon fluxes in terrestrial ecosystems, and these impacts are mainly reflected in the carbon absorption, storage and release processes of ecosystems. Extreme climate events such as droughts and heat waves may limit vegetation growth, thereby reducing the absorption of atmospheric CO2 by plants, and may also indirectly promote carbon release by affecting ecosystem structure and function; while fires, storms, etc. may directly destroy vegetation, causing carbon stored in plants and soil to be released into the atmosphere. In the past 50 years, about 25% to 30% of anthropogenic CO2 on Earth has been absorbed by ecosystems, but extreme climate events may reduce this proportion. Before an extreme event occurs, the ecosystem may be in a carbon absorption state; but after the event, due to vegetation destruction and soil carbon decomposition, ecosystem carbon absorption often weakens, and may even turn into a carbon release state. However, the impact of extreme climate events on carbon flux is complex and far-reaching, and the scientific community needs to strengthen research on the global carbon cycle to better understand the impact mechanism of extreme climate events on carbon flux.

[0003] The quantitative contribution of extreme climate to carbon flux anomalies is a complex and in-depth issue because it involves a variety of extreme climate types, such as heat waves, droughts, storms, etc., as well as their immediate and lagged effects on the carbon cycle of ecosystems. The current challenges and uncertainties in quantification are as follows: ① Data acquisition is difficult. Quantifying the impact of extreme climate events on carbon flux anomalies requires a large amount of field observation data and model simulation results. However, due to the difficulty and high cost of data acquisition and limited model accuracy, the current quantitative results still have large uncertainties. ② The impact of extreme climate events on carbon flux is often nonlinear, that is, minor climate changes may lead to significant carbon flux anomalies, which is related to the resilience of terrestrial vegetation ecosystems, but this nonlinear relationship obviously increases the complexity and uncertainty of quantification.

[0004] In summary, quantifying the impact of extreme climate events on carbon flux anomalies is a challenging task. Therefore, we propose a method to quantify the simultaneous impact of extreme climate events on terrestrial carbon flux and its legacy effects in order to solve the above problems. Summary of the invention

[0005] The purpose of the present invention is to provide a method for quantifying the contemporaneous impact of extreme climate events on terrestrial carbon flux and the previous legacy effects, so as to solve the current market problems raised by the above background technology.

[0006] To achieve the above object, the present invention provides the following technical solution: a method for quantifying the contemporaneous impact of extreme climate events on terrestrial carbon flux and the previous residual effects, comprising the following steps:

[0007] Step 1: Preparation work, including relevant climate data and carbon flux data, unified time scale to monthly scale, using nearest neighbor method to resample to uniform spatial resolution; unified output according to variables and months as easy-to-read code files; select climate variables according to the categories of extreme events such as high temperature and drought events, storms and rainstorms, rainstorms and floods, remove the long-term linear trend of the data, that is, the impact of global warming, remove the climate state, that is, the monthly average, and obtain abnormal values ​​to determine the specific latitude and longitude range of the study area; unify the time scale according to the time series of all data;

[0008] Step 2: Based on the principle that changes in terrestrial ecosystem carbon flux are related to the climate conditions of the same period, the previous conditions, and the vegetation status, taking the gross primary productivity (GPP) as an example, a regression equation with climate conditions and vegetation status is constructed on a monthly scale based on a machine learning algorithm:

[0009] GPP~Pre c +SM c +TAS c +VPD c +SR c

[0010] +Pre l1 +SM l1 +TAS l1 +VPD l1 +SR l1 +GPP l1

[0011] +Pre l2 +SM l2 +TAS l2 +VPD l2 +SR l2 +GPP l2

[0012] +Pre l3 +SM l3 +TAS l3 +VPD l3 +SR l3 +GPP l3 +ε

[0013] Among them, Pre c Pre refers to the precipitation in the same month as the event. l1 Pre refers to the precipitation in the previous month. l2Pre refers to the precipitation in the first two months. l3 It refers to the precipitation in the previous three months. The subscripts of other variables have the same meaning. ε represents the error term;

[0014] Step 3: Select the XGBoost machine learning algorithm, based on Python, debug the code according to the relationship described in step 2, and the latitude and longitude range and time scale of the study area determined in step 1; specifically, divide the original data into training set and validation set according to 9:1, train the regression model, and keep the invariance of the fitting model by fixing the random number; debug the model parameters, select the optimal model, and output the evaluation parameters of the fitting model, such as R 2 , RMSE;

[0015] Step 4: After determining the accuracy of the model based on the evaluation parameters of the fitted model, change the input data of the original model and conduct a control experiment. The principle of the control experiment is that the climate state represents the average state of past climate conditions or vegetation. By comparing the state of climate and vegetation when an extreme event occurs with the average state, the impact of the extreme event can be obtained.

[0016] Compared with the prior art, the present invention has the following beneficial effects:

[0017] (1) Compared with traditional regression methods, the present invention adopts a machine learning method, which has the advantages of automated decision-making, adaptive learning, real-time calculation, and reduction of human errors, and can better handle complex and nonlinear relationships; machine learning algorithms can automatically extract patterns from data and make decisions based on them, thereby reducing the need for human intervention and significantly improving work efficiency; with the rapid growth of data, machine learning systems can continuously learn from new data, update decision or prediction models, and quickly process and analyze massive data sets; through big data analysis and pattern recognition, machine learning can reveal deep connections and trends hidden in the data.

[0018] (2) This method also takes into account the lagged or residual effects of extreme climate events on the carbon cycle. Traditional studies often focus on the real-time impacts of extreme events on vegetation, while studies on residual effects focus on their long-term effects. This method considers both the real-time and lagged effects of extreme events on terrestrial ecosystems, which helps to more fully understand the ecological consequences of extreme climate events and provide a scientific basis for the formulation of long-term response strategies.

[0019] (3) This method is calculated based on the pixel scale. As the basic unit in remote sensing images, pixels can provide detailed spatial information. Analysis based on the pixel scale can capture more subtle changes in geographical features, thereby more accurately reflecting the impact of extreme climate events on carbon flux. Different regions may be significantly affected by extreme climate events in terms of the degree and manner in which they are affected. For example, in arid areas, extreme drought events may lead to vegetation dieback and soil carbon release, while in humid areas, extreme precipitation events may promote vegetation growth and carbon absorption. These differences can be clearly seen based on the pixel scale, which facilitates statistical analysis of the differences in the impact of different vegetation types, soil types, topography and other factors, improves the accuracy and reliability of the research, and provides a scientific basis for formulating targeted ecological protection and carbon management strategies.

[0020] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] Figure 1 Spatial map of temperature and soil moisture anomalies for August 2022. DETAILED DESCRIPTION

[0022] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0023] A method for quantifying the concurrent impact of extreme climate events on terrestrial carbon flux and the residual effects of previous events, comprising the following steps:

[0024] Step 1: Preparation work, including relevant climate data and carbon flux data, unified time scale to monthly scale, using nearest neighbor method to resample to uniform spatial resolution; unified output according to variables and months as easy-to-read code files; select climate variables according to the categories of extreme events such as high temperature and drought events, storms and rainstorms, rainstorms and floods, remove the long-term linear trend of the data, that is, the impact of global warming, remove the climate state, that is, the monthly average, and obtain abnormal values ​​to determine the specific latitude and longitude range of the study area; unify the time scale according to the time series of all data;

[0025] Step 2: Based on the principle that changes in terrestrial ecosystem carbon flux are related to the climate conditions of the same period, the previous conditions, and the vegetation status, taking the gross primary productivity (GPP) as an example, a regression equation with climate conditions and vegetation status is constructed on a monthly scale based on a machine learning algorithm:

[0026] GPP~Pre c +SM c +TAS c +VPD c +SR c

[0027] +Pre l1 +SM l1 +TAS l1 +VPD l1 +SR l1 +GPP l1

[0028] +Pre l2 +SM l2 +TAS l2 +VPD l2 +SR l2 +GPP l2

[0029] +Pre l3 +SM l3 +TAS l3 +VPD l3 +SR l3 +GPP l3 +ε

[0030] Among them, Pre c Pre refers to the precipitation in the same month as the event. l1 Pre refers to the precipitation in the previous month. l2 Pre refers to the precipitation in the first two months. l3 refers to the precipitation in the previous three months, and the subscripts of other variables have the same meaning. ε represents the error term;

[0031] Step 3: Select the XGBoost machine learning algorithm, based on Python, debug the code according to the relationship described in step 2, and the latitude and longitude range and time scale of the study area determined in step 1; specifically, divide the original data into training set and validation set according to 9:1, train the regression model, and keep the invariance of the fitting model by fixing the random number; debug the model parameters, select the optimal model, and output the evaluation parameters of the fitting model, such as R 2 , RMSE;

[0032] Step 4: After determining the accuracy of the model based on the evaluation parameters of the fitted model, change the input data of the original model and conduct a control experiment. The principle of the control experiment is that the climate state represents the average state of past climate conditions or vegetation. By comparing the state of climate and vegetation when an extreme event occurs with the average state, the impact of the extreme event can be obtained.

[0033] The principle of this method is based on the regression relationship between carbon flux and vegetation status and climate conditions. Machine learning can fit this nonlinear relationship well. When constructing the relationship, the concurrent and lagged effects of extreme events on vegetation are considered at the same time. The carbon fluxes commonly considered in the study include net ecosystem productivity (NEP), gross primary productivity (GPP), and terrestrial ecosystem respiration (TER). The climate factors mainly consider precipitation (Pre), soil moisture (SM), surface air temperature (TAS), atmospheric saturated water vapor pressure (VPD), and solar radiation (SR); the previous vegetation state is represented by the previous carbon flux. Since the response of vegetation to extreme climate events takes time, it is considered to be studied on a monthly scale. By setting up multiple groups of control experiments, the concurrent impact and previous legacy of the event are determined and the contribution of different factors is quantitatively calculated.

[0034] Specifically, a total of 5 groups of control experiments were conducted, taking the impact of extreme events occurring in August on carbon flux during the same period as an example, and the experimental contents are listed in the table below.

[0035] Table 1. The first set of controlled experiments used to obtain the concurrent and residual impacts of extreme events

[0036]

[0037] Table 2. The second set of control experiments used to obtain the contribution of different climate factors in the same period

[0038]

[0039]

[0040] Table 3. The third set of control experiments for obtaining the contributions of different climate factors and vegetation states in the first month of the previous period

[0041]

[0042] Table 4. The fourth set of control experiments for obtaining the contributions of different climate factors and vegetation states in the second month of the previous period

[0043]

[0044] Table 5. The fifth set of control experiments for obtaining the contributions of different climate factors and vegetation states in the third month of the previous period

[0045]

[0046]

[0047] Output the model results according to the several groups of experiments in the method principle table. Both the original data and the carbon flux data output by the model need to remove the climate state and trend to obtain the carbon flux anomaly value.

[0048] Compare the results of different experiments and quantify the contribution of different variables. Among them, the results of experiment 1 in the first group of experiments are compared with the original data results to verify the accuracy of model simulation. Through the first group of experiments, the influence of climate conditions in the same period and the residual influence of the previous climate background and vegetation status can be obtained. Specifically, the results of experiment 2 and experiment 3 are subtracted to obtain the influence of vegetation and climate in May on GPP in August, the results of experiment 3 and experiment 4 are subtracted to obtain the influence of vegetation and climate in June on GPP in August, and the results of experiment 4 and experiment 5 are subtracted to obtain the influence of vegetation and climate in July on GPP in August, while experiment 5 obtains the influence of climate conditions in August on GPP in August. Similarly, by comparing the results of the second, third, fourth, and fifth groups of experiments with the results of experiment 1 in the first group of experiments, the specific contributions of different variables can be quantified and the dominant factors can be obtained by comparison.

[0049] Example

[0050] This method is further explained using the impact of the extreme combined high temperature and drought event that occurred in the Yangtze River Basin of China in August 2022 on the gross primary productivity (GPP) of vegetation as an example.

[0051] (1) Select the climate data of ERA5-Land,

[0052] https: / / cds.climate.copernicus.eu / cdsapp#! / dataset / reanalysis-481era5-land-monthly-means?tab=overview;

[0053] Including TAS, SM, Pre, SR, and dewpoint temperature (TD). VPD is calculated based on TAS and TD:

[0054]

[0055] Select FluxSat's GPP data.

[0056] https: / / avdc.gsfc.nasa.gov / pub / tmp / FluxSat_GPP;

[0057] The spatial resolution of the two data is unified to 0.1°×0.1°, and the time is from January 2000 to December 2022, with monthly scale data. Because we want to quantitatively calculate the contribution of different factors to the GPP anomaly in August, we extract the data sets of each variable from May, June, July and August 2000 to 2022. The temperature and soil moisture data in August are detrended and declimatized to obtain the anomalies of the two. The scope of the study area is determined according to the spatial distribution of the temperature and soil moisture anomalies in August 2022, that is, 25°-33°N, 100°-123°E (such as Figure 1 shown).

[0058] (2) XGBoost is a machine learning algorithm based on gradient boosting tree. It has been widely used and recognized in the field of machine learning due to its advantages such as high performance, efficiency, accuracy, interpretability, flexibility and scalability. Therefore, in this experiment, XGBoost is selected as the machine learning algorithm. Based on the python debugging code, the original data is divided into training set and validation set according to 9:1 to train the regression model. The regression relationship here is as follows:

[0059] GPP8~Pre8+SM8+TAS8+VPD8+SR8

[0060] +Pre7+SM7+TAS7+VPD7+SR7+GPP7

[0061] +Pre6+SM6+TAS6+VPD6+SR6+GPP6

[0062] +Pre5+SM5+TAS5+VPD5+SR5+GPP5+ε

[0063] The subscripts in the formula represent months; the RMSE of the validation data is 0.6953 gC m -2 day -1 , R 2 The value is 0.97, so the model is good for further experiments.

[0064] (3) Based on the regression model in step (2), five groups of comparative experiments were completed to obtain GPP results under different scenarios.

[0065] (4) The GPP values ​​obtained from different experiments were detrended and declimatized, and the outliers were obtained by experiment. The GPP outliers simulated by the model obtained from experiment 1 in the first group of experiments were subtracted from the results of different experiments to obtain the specific contribution of each variable.

[0066] It is calculated that the Yangtze River Basin ( Figure 1 The GPP in the rectangle shows a significant decrease, and the GPP abnormal value is -0.3588PgC yr -1 The GPP anomaly obtained by model simulation is -0.3577PgC yr -1 , of which -0.1272PgCyr -1 The contribution of July, June and May to the GPP anomaly in August is -0.0951 PgC yr respectively. -1 , -0.02PgC yr -1 and -0.1154PgC yr -1 From this, we can find that the previous legacy effect is -0.2305PgC yr -1 , which is greater than the climate impact during the same period. In addition, the contributions of different factors are shown in Table 6. It can be seen that among the climate factors during the same period, VPD and TAS played a major role. In addition, GPP in July, that is, the vegetation status in the previous month, played an even greater role.

[0067] Table 6. Contribution of different factors to the negative anomaly of GPP in August

[0068]

[0069] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, unless they are contradictory.

[0070] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A method for quantifying the contemporaneous impact of extreme climate events on terrestrial carbon flux and its residual effects, characterized in that: The steps include: Step 1: Preparation work, including relevant climate data and carbon flux data, unifying the time scale to the monthly scale, and using the nearest neighbor method to resample to a uniform spatial resolution; Output uniformly according to variables and months into easy-to-read code files; select climate variables according to the categories of extreme events such as high temperature and drought events, storm events, and rainstorm and flood events, remove the long-term linear trend of the data, that is, the impact of global warming, remove the climate state, that is, the monthly average, and obtain abnormal values ​​to determine the specific latitude and longitude range of the study area; unify the time scale according to the time series of all data; Step 2: Based on the principle that the change of carbon flux in terrestrial ecosystems is related to the climate conditions of the same period, the previous conditions and the vegetation status, taking GPP as an example, a regression relationship between it and the climate conditions and vegetation status is constructed on a monthly scale based on a machine learning algorithm: GPP~Pre c +SM c +TAS c +VPD c +SR c +Pre l1 +SM l1 +TAS l1 +VPD l1 +SR l1 +GPP l1 +Pre l2 +SM l2 +TAS l2 +VPD l2 +SR l2 +GPP l2 +Pre l3 +SM l3 +TAS l3 +VPD l3 +SR l3 +GPP l3 +ε Among them, Pre c Pre refers to the precipitation in the same month as the event. l1 Pre refers to the precipitation in the previous month. l2 Pre refers to the precipitation in the first two months. l3 refers to the precipitation in the previous three months, and the subscripts of other variables have the same meaning. ε represents the error term; Step 3: Select the XGBoost machine learning algorithm, based on Python, debug the code according to the relationship described in step 2, and the latitude and longitude range and time scale of the study area determined in step 1; specifically, divide the original data into training set and validation set according to 9:1, train the regression model, and keep the invariance of the fitting model by fixing the random number; debug the model parameters, select the optimal model, and output the evaluation parameters of the fitting model, such as R 2 , RMSE; Step 4: After determining the accuracy of the model based on the evaluation parameters of the fitted model, change the input data of the original model and conduct a control experiment. The principle of the control experiment is that the climate state represents the average state of past climate conditions or vegetation. By comparing the state of climate and vegetation when an extreme event occurs with the average state, the impact of the extreme event can be obtained.

Images