CN107316095B - Regional weather drought level prediction method coupled with multi-source data - Google Patents

Abstract

The invention discloses a regional meteorological drought grade prediction method coupled with multi-source data. By performing spatial downscaling on gridded satellite remote sensing precipitation data, a high-resolution regional precipitation spatial database is constructed, and a standardized drought index is used to divide the drought grades. A large-scale meteorological factor reflecting the characteristics of atmospheric circulation was introduced as a covariate of the transition probability of drought state, and a meteorological drought grade prediction model with time-varying transition probability was constructed based on the non-stationary Markov chain model. The invention uses multi-source remote sensing information and basic underlying surface features to capture the spatial variability of regional precipitation, makes up for the lack of rainfall observed by traditional stations, and makes full use of large-scale meteorological factors that can reflect the characteristics of atmospheric circulation. The stress and precursor signals take into account the formation and development mechanism of droughts and floods to a certain extent, and are more in line with the dynamic evolution characteristics of the meteorological and hydrological systems.

Description

A regional meteorological drought grade prediction method based on coupled multi-source data

technical field

The invention belongs to the technical field of disaster forecasting and early warning, and particularly relates to a regional meteorological drought grade forecasting method coupled with multi-source data.

Background technique

Drought is a natural phenomenon of continuous shortage of water, which has the characteristics of high frequency, long duration and wide spread. my country is located in the Asian monsoon climate region, the spatial and annual distribution of precipitation is seriously uneven, and the inter-annual variation of monsoon path and intensity is large. One of the worst drought-hit countries in the world. On the basis of summarizing various definitions of drought, the American Meteorological Society divides droughts into four types, namely meteorological drought, hydrological drought, agricultural drought and socioeconomic drought. Meteorological drought refers to reduced or no precipitation, and the formation of other types of drought is directly related to meteorological drought.

Compared with the "shock" consequences caused by other extreme climate events such as floods and typhoons, the occurrence and development of droughts are significantly hidden and "creep". The beginning and end of a drought event are often difficult to define, and once the extent of its impact and degree of harm is apparent, response and remedial measures are often severely delayed. Therefore, timely and accurate drought prediction is of great significance for guiding the development of drought relief work, strengthening disaster risk emergency management, and improving disaster response levels.

Drought monitoring is the basis of drought prediction. Most of the existing methods are based on measured rainfall station data. However, due to the problem of uneven density and distribution of observation stations, it is difficult to reflect the spatial heterogeneity of rainfall distribution; , and it is even more difficult to obtain observational data. The development of remote sensing technology in recent years has provided a new way for large-scale drought monitoring. Gao Zhiqiang ^[1] invented a drought monitoring method and system based on remote sensing inversion of surface water and heat flux, which can be used to estimate the distribution of regional surface energy and evapotranspiration under different climate and terrain conditions, and provide technical support for regional agricultural drought monitoring. ^[2] invented a drought monitoring method based on HJ-1A/1B CCD data, combined with MPDI data obtained from HJ-1A/1B CCD remote sensing data and crop growth period to determine agricultural drought conditions; Feng Jie ^[3] provided a drought monitoring method based on data mining, which comprehensively considered the multi-source remote sensing spatial information in drought monitoring to spatially downscale remote sensing precipitation, and used spatial data mining technology to build a drought monitoring model. However, due to the limitation of the current technical level, remote sensing monitoring of drought still has shortcomings such as low spatial resolution.

Regional droughts and floods are usually caused by temporary anomalies in the local atmospheric moisture budget. However, due to the numerous influencing factors and complex interactions, and the lack of a comprehensive understanding of the disaster mechanism, it is still difficult to accurately assess and forecast disasters. . Instead, quantitative estimates of the relative frequency and intensity (level) of drought and flood events can be made. The existing forecasts of regional drought and flood grades mainly start from the randomness of drought and flood events, and use time series analysis tools to achieve the purpose of revealing their temporal and spatial evolution characteristics. Yang Zhiyong et al. ^[4] used the two-dimensional Copula function to construct the joint distribution of the percentage series of seasonal precipitation anomalies at representative meteorological stations in the Luanhe River Basin, and calculated the occurrence probability of two types of drought-flood combination events, such as alternation of droughts and floods and consecutive droughts and floods at each station. . Song Xinshan et al. ^[5] used the Markov model to calculate the statistical characteristics of the transition probability, duration, and recurrence time of 16 representative stations in the middle and lower reaches of the Huang-Huai-Haihai region for 540 years under different drought and flood states. Feng Ping et al. ^[6] used a three-dimensional log-linear model to establish a short-term meteorological drought grade prediction model for 21 rainfall stations in the Panjiakou Reservoir-controlled basin of the Luanhe River, and realized the meteorological drought grade forecast with a forecast period of 1 month and 2 months.

Most of the above methods regard the evolution of drought and flood as a stable process, that is, it is considered that its statistical characteristics, such as the transition probability of drought and flood state, do not change with time, and can be obtained from past measured meteorological or hydrological sequence samples. However, due to the disturbance of climate change and human activities, the meteorological and hydrological systems have significant dynamic evolution characteristics. In order to effectively deal with the catastrophic risk of regional drought and flood events under dynamic evolution conditions, it is urgent to develop a regional drought and flood prediction method that can comprehensively consider internal causes and external stresses.

The references involved in the text are as follows:

[1] Gao Zhiqiang. Drought monitoring method and system based on remote sensing inversion of surface water heat flux. Patent No. ZL201010623662.6.

[2] Li Haohao, Chen Haibo, Yu Changhong, et al. A Drought Monitoring Method Based on HJ-1A/1B CCD Data. Patent No. ZL201310379034.1.

[3] Feng Jie, He Qisheng, Yang Zhiyong, et al. A Drought Monitoring Method Based on Data Mining. Publication No. CN105760814A.

[4] Yang Zhiyong, Yuan Zhe, Fang Hongyang, et al. Analysis of the probability characteristics of combined drought and flood events in the Luanhe River Basin based on Copula function [J]. Journal of Hydraulic Engineering, 2013, 44(5): 556-569.

[5] Song Xinshan, Yan Denghua, Wang Yuhui, et al. Analysis of the evolution of drought and flood in the central and eastern Huanghuaihai region in the past 540 years based on Markov model [J]. Journal of Hydraulic Engineering, 2013, 44(12): 1425-1432.

[6] Feng Ping, Hu Rong, Li Jianzhu. Prediction of meteorological drought grade based on three-dimensional log-linear model [J]. Journal of Hydraulic Engineering, 2014, 45(5): 505-512.

SUMMARY OF THE INVENTION

In view of the deficiencies in the prior art, the present invention provides a regional meteorological drought grade prediction method coupled with multi-source data, which constructs a high-resolution regional precipitation spatial database by spatially downscaling the gridded satellite remote sensing precipitation data. , using a standardized drought index to classify drought grades, introducing large-scale meteorological factors reflecting the characteristics of atmospheric circulation as covariates of drought state transition probability, and constructing a meteorological drought grade prediction model with time-varying transition probability based on the non-stationary Markov chain model to analyze and Predict the dynamic evolution law of regional drought state.

In order to solve the above-mentioned technical problems, the present invention adopts the following technical scheme:

A method for forecasting regional meteorological drought grades coupled with multi-source data, comprising the steps of:

Step 1: Collect basic underlying surface information in the area, and synchronous length series data of precipitation and large-scale meteorological factors;

Step 2: Construct a spatial downscaling model of remote sensing precipitation data corrected by considering the basic underlying surface factors, process the low-resolution original remote sensing monitoring precipitation data into higher-resolution precipitation data, and use the ground observation station measured precipitation data to output data. Make corrections to obtain a high-resolution regional precipitation spatial database;

Step 3, perform frequency analysis on the long series of precipitation data of each high-resolution grid in the regional gridded precipitation spatial database obtained in step 2 in turn, calculate its standardized drought index, and obtain each Grid's drought grading sequence;

Step 4: According to the drought grade sequence obtained in Step 3, the large-scale meteorological factors with a certain delay collected in Step 1 are used as the covariates of the transition probability of drought state, and the meteorological drought with time-varying transition probability is constructed based on the non-stationary Markov chain model. grade prediction model;

In step 5, the non-stationary Markov chain model obtained by the optimization in step 4 is used to predict the regional meteorological drought level, and the spatial distribution of the drought level in the whole region is obtained. Further, in step 2, a multiple linear regression method is used to couple the underlying surface information to construct a spatial downscaling model, and the original remote sensing monitoring precipitation data is scaled down.

Further, in step 2, the geographic information difference method is used to further correct the output high-resolution precipitation data by using the measured precipitation at the ground observation station.

Further, in step 3, the standardized drought index uses the standardized precipitation index.

Further, in step 4, the generalized cross-entropy method is used to estimate the parameters of each non-stationary Markov chain model.

Further, in step 4, the historical drought level is calculated by back-substitution with the calibrated non-stationary Markov chain model, and the final meteorological drought level prediction model is selected by using the Akaike information quantity criterion.

Compared with the prior art, the present invention has the following advantages and effects:

1. Make full use of multi-source remote sensing information and basic underlying surface characteristics to capture the spatial variability of regional precipitation, which can make up for the inhomogeneous density and distribution of rainfall points observed by traditional stations, which cannot reflect the insufficiency of rainfall distribution characteristics on the surface.

2. A non-stationary Markov chain model with time-varying characteristics considering covariates is established, which overcomes the shortcomings of traditional methods that only consider the autocorrelation characteristics of meteorological and hydrological sequences, and is more suitable for the dynamic evolution characteristics of meteorological and hydrological systems.

3. Make full use of large-scale meteorological factors that can reflect the characteristics of atmospheric circulation, the external stress and precursor signal of the evolution of droughts and floods, and consider the formation and development mechanisms of droughts and floods to a certain extent. Lay the foundation for building a regional drought and flood warning and forecasting system.

Description of drawings

Fig. 1 is the concrete flow chart of the method of the present invention;

Figure 2 is a schematic diagram of the grid division of the watershed;

Figure 3 is a schematic diagram of precipitation probability distribution fitting;

Fig. 4 is a schematic diagram of the prediction result of drought level.

Detailed ways

The technical solutions of the present invention will be further described below with reference to the accompanying drawings.

Fig. 1 is the concrete flow chart of the inventive method, and concrete steps are as follows:

Step 1: Collect the basic underlying surface information in the area, and the synchronization length series data of precipitation and large-scale meteorological factors. This step is a conventional technique in the art.

The basic underlying surface information mainly refers to gridded, high-resolution Digital Elevation Model (DEM) data, and gridded, high-resolution land cover data (such as normalized differential vegetation index (NDVI). ), etc.; precipitation data includes actual measurements from ground-based rainfall observation sites, and gridded, low-resolution satellite remote sensing monitoring data; large-scale meteorological factors include indicator data that characterize large-scale circulation characteristics on a global scale.

Step 2, construct a spatial downscaling model of remote sensing precipitation data corrected by considering the basic underlying surface factors, and process the low-resolution original remote sensing monitoring precipitation data into higher-resolution precipitation data; Correction is performed to obtain a high-resolution regional precipitation spatial database.

In step 2, the multiple linear regression method is used to couple the regional underlying surface information to construct a spatial downscaling model of the remote sensing precipitation data, and the original remote sensing monitoring precipitation data is degraded. The output obtains a high-resolution regional precipitation spatial database for further correction. It further includes the following sub-steps:

(1) According to the spatial resolution of the underlying surface information of the region, based on the geographic information system (GIS) platform, the region is spatially discretized and divided into uniform high-resolution latitude and longitude grids. The schematic diagram of the watershed grid division is shown in Figure 2.

(2) Using the nearest neighbor interpolation method to resample the regional underlying surface information (Digital Elevation Model (DEM), Normalized Difference Vegetation Index (NDVI)) to make it consistent with the resolution of the original remote sensing monitoring precipitation data .

The nearest neighbor interpolation method is a routine technique in the art.

(3) The altitude (h), aspect (α) and slope (β) factors are extracted from the digital elevation model (DEM).

的经验关系：(4) Using the multiple linear regression method to construct the regional underlying surface information and the original remote sensing precipitation data

The empirical relationship of:

计算原始低分辨率下遥感降水模拟值

In the formula: b=(b ₀ , b ₁ , b ₂ , b ₃ , b ₄ , b ₅ , b ₆ ) is the parameter matrix of the multiple linear regression model, estimated by the least square method; Z=(1, X, Y, h, α, β, NDVI) is the independent variable matrix; X and Y are the longitude and latitude of the grid point center, respectively. According to the estimated parameters

Calculate the original low-resolution remote sensing precipitation simulation value

Z _LR = (1, X _LR , Y _LR , h _LR , α _LR , β _LR , NDVI _LR ) is an independent variable matrix at low resolution.

Least squares methods are routine techniques in the art.

与

的残差：(5) Calculate the original low resolution

and

The residuals of :

(6) Using the inverse distance weight interpolation method to interpolate the low-resolution grid residual to obtain the high-resolution residual value

The inverse distance weight interpolation method is a conventional technique in the art.

并由高分辨率残差值

修正得到高分辨率下遥感降水模拟值

(7) Apply the parameters of the multiple linear regression model estimated in step (4) to the high-resolution grid points to obtain the initial high-resolution remote sensing precipitation simulation value

and by the high-resolution residual values

Correction to obtain the simulated value of remote sensing precipitation under high resolution

Z _HR = (1, X _HR , Y _HR , h _HR , α _HR , β _HR , NDVI _HR ) is the independent variable matrix at high resolution.

与包含此站点的高分辨率网格点的遥感降水模拟值

之间的残差：(8) Calculate the measured precipitation for each ground observation station collected in step 1

Simulated values of remote sensing precipitation with high-resolution grid points containing this site

Residuals between:

(9) Using the inverse distance weight interpolation method to interpolate the station precipitation residual ΔP _point to obtain the high-resolution precipitation correction factor value

加上高分辨率遥感降水模拟值

得到最终高分辨率遥感降水修正值

形成高分辨率的区域降水空间数据库。(10) Precipitation correction factor with high resolution

Plus high-resolution remote sensing precipitation simulations

Obtain the final high-resolution remote sensing precipitation correction value

A high-resolution regional precipitation spatial database is formed.

Step 3, perform frequency analysis on the long series of precipitation data of each high-resolution grid in the high-resolution regional precipitation spatial database obtained in step 2 in turn, calculate its standardized drought index, and obtain the drought level division table according to the drought index. Drought grade sequence for each grid.

In step 3, the standardized drought index uses the standardized precipitation index (Standardized Precipitation Index, SPI). The calculation steps are as follows:

(1) For each corrected high-resolution grid point remote sensing precipitation sequence obtained in step 2, the Gamma distribution line is used to fit the cumulative precipitation on different time scales in each month. The probability density function of the Gamma distribution is as follows:

where a ₁ and a ₂ are the shape and scale parameters of the Gamma distribution, respectively, estimated by the maximum likelihood method; Γ( ) is the Gamma function; p is the cumulative precipitation over time, such as ten days, months, seasons, years, etc. . The fitting diagram of the precipitation probability distribution is shown in Figure 3.

Preferably, considering that the current remote sensing precipitation products are generally more credible on the monthly and above scales, but there are still large uncertainties on smaller scales, this specific implementation temporarily takes the month as the time scale. With the further improvement of the precision of remote sensing products, the method of the present invention can be applied to smaller time scales.

Maximum likelihood methods are routine techniques in the art.

(2) Calculate the cumulative probability of precipitation in a certain period:

(3) According to the principle of equal probability, the cumulative probability is converted into the quantile of the standard normal distribution, that is, SPI:

SPI=Φ ^-1 (G(p)) (9) where: Φ ^-1 (·) is the inverse function of the standard normal distribution probability distribution function.

In step 3, the drought grading of the drought index adopts the degree that its cumulative probability deviates from the normal level (50% quantile).

The classification of drought grades based on SPI is shown in Table 1.

Table 1 SPI index drought classification table

It can be seen from the table that SPI can be used not only for drought monitoring, but also for monitoring regional flood status.

Step 4: According to the drought grade sequence obtained in step 3, a large-scale meteorological factor with a certain time lag is used as a covariate of the transition probability of drought state, and a meteorological drought grade prediction model with time-varying transition probability is constructed based on the non-stationary Markov chain model.

Taking the entire study area as the overall system, and the high-resolution grid points in the area as the calculation unit, the drought and flood structure in the system is the ratio of the number of grid points in different drought levels at a certain time t to the total number of calculation units (that is, the disaster-affected area). The ratio A(t)=(A _1t , A _2t , . . . , A _7t )) is represented, and it evolves with time, and is often described by the Markov chain model.

The stationary Markov chain model assumes that the probability π _ij that the drought level state of any computing unit in the system is transformed from i to j at time t+1 in the future (i,j=1,2,...,7 in this specific implementation) is only related to the current The state i known at time t is related to the previous state, and does not change with time, that is:

In the formula: i and j are the state values, S is the state domain, and T is the time domain. Then the evolution process of the proportion of the affected area can be expressed as:

A(t+1)=A(t)π (11)

Where: π is the state transition matrix.

In fact, limited by the limited observation samples of drought-flood state transition, it often leads to unreasonable zero-value phenomenon in the state transition experience matrix. At the same time, regional precipitation is controlled by a variety of factors, which is a typical manifestation of the joint action of the sea-land-atmosphere system. Its state transition law is not only related to the current state, but also changes with the change of the external environment (such as atmospheric stress). Therefore, the present invention adopts a non-stationary time-varying Markov chain model to describe the transition process of drought and flood state, assuming that the transition probability of drought and flood state changes with time, and introduces external explanatory variables to construct a quantitative relationship between it and the transition probability:

π _ij (t)=f _ij (z _ij (t), η _ij ) (12)

In the formula: z _ij (t) is the explanatory variable matrix, η _ij is the regression parameter, and f _ij (·) is the function linking the covariate and transition probability. A linear regression equation is often used to construct an empirical relationship between explanatory variables and transition probabilities:

Φ ^-1 (π _ij (t))=η _ij z _ij (t) (13)

In the formula: Φ ^-1 (·) is the inverse function of the probability distribution function of the standard normal distribution. The quantile transformation is used to convert the dependent variable from the probability interval [0,1] to a continuous real number space.

In step 4, the types and delays of large-scale meteorological factors as covariates are preliminarily selected from the candidate large-scale meteorological factors through correlation analysis; the parameters of each non-stationary Markov chain model are estimated by the generalized cross-entropy method; the pass rate The predetermined non-stationary Markov chain model is used to calculate the historical drought level, and the Akaike Information Criterion (AIC criterion) is used to select the final meteorological drought level prediction model. It further includes the following sub-steps:

(1) Take the long series of index data that characterize the circulation characteristics of large-scale global scale collected in step 1 as the candidate large-scale meteorological factor set, and set the maximum lag time for the response of regional drought grade evolution to the fluctuation of large-scale meteorological factors Lag _max and minimum lag time Lag _min , for all the candidate large-scale meteorological factors, gradually increase from the minimum delay time to the maximum delay time, and use the correlation coefficient to carry out the correlation analysis and test of the asynchronous series of the precipitation series and the candidate large-scale meteorological factors respectively , according to the magnitude of the correlation coefficient, several groups of large-scale meteorological factor covariates that have a significant impact on the evolution of drought and flood are selected.

(2) The generalized cross-entropy method is used to estimate the parameters of each non-stationary Markov chain model.

(3) Calculate the historical drought grade by back-substitution of the calibrated non-stationary Markov chain model, and use the AIC criterion to select the final meteorological drought grade prediction model.

In step 4, the set of candidate large-scale meteorological factors can be the North Atlantic Oscillation Index (NAO), the Arctic Oscillation Index (AO), the Pacific Decadal Oscillation (PDO), the Southern Oscillation Index (SOI), and the multivariate ENSO index. (MEI), one or more of the North Atlantic Decadal Oscillation (AMO), etc.

In step 4, this specific implementation takes months as the time step, the minimum lag time Lag _min is 1 month, and the maximum lag time Lag _max is 12 months.

In step 4, the generalized cross entropy method is used to estimate the parameters of the non-stationary Markov model, which can be attributed to the information carried by the posterior estimation and prior estimation of the transition probability of the model under the premise of making full use of all the information without adding redundancy. problem with the smallest amount of difference.

(1) Objective function (cross entropy):

是先验转移概率矩阵，为求解后验转移概率提供信息。where:

is the prior transition probability matrix, which provides information for solving the posterior transition probability.

(2) Constraints:

可由平稳Markov链模型得到的经验转移矩阵代替。prior transition probability matrix

It can be replaced by the empirical transition matrix obtained from the stationary Markov chain model.

In step 5, the non-stationary Markov chain model obtained by the optimization in step 4 is used to predict the regional meteorological drought level, and the spatial distribution of the drought level in the whole region is obtained.

In step 5, the optimal drought level prediction model is applied to each high-resolution grid point in turn to obtain the probability that the drought level of this grid point will transfer to different levels at the next moment, and the level state with the largest transition probability is used as the At the next moment, the drought level of the grid point will be forecast, and the affected area of each level will be estimated. Figure 4 shows a schematic diagram of the prediction result of drought level at a certain time.

The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited to this, any person skilled in the art can easily think of changes or replacements within the technical scope of the present invention, All should be included within the protection scope of the present invention. Therefore, the protection scope of the present invention should be subject to the protection scope of the claims.

Claims (6)

1. a regional meteorological drought grade prediction method of coupling multi-source data, is characterized in that, comprises the steps: Step 1: Collect basic underlying surface information in the area, and synchronous length series data of precipitation and large-scale meteorological factors; Step 2: Construct a spatial statistical downscaling model of remote sensing precipitation data corrected by considering the high-resolution basic underlying surface factors, process the low-resolution original remote sensing monitoring precipitation data into high-resolution remote sensing precipitation data, and use the ground observation site to measure the precipitation data. Calculate the residual correction factor, correct the output data, obtain a high-resolution regional gridded precipitation spatial database, and quantitatively simulate the spatial heterogeneity of regional precipitation distribution; Step 3, perform frequency analysis on the long series of precipitation data of each high-resolution grid in the regional gridded precipitation spatial database obtained in step 2 in turn, calculate its standardized drought index, and obtain each Grid's drought grading sequence; Step 4: According to the drought level sequence obtained in Step 3, it is assumed that the transition probability between drought levels changes with time, and the large-scale meteorological factor sequence with a certain lag collected in Step 1 is used as the external explanatory variable for the transition probability of drought state. The stationary Markov chain model describes the quantitative functional relationship between the external explanatory variables and the transition probability of drought state, and builds a meteorological drought grade prediction model with time-varying transition probability; Step 5: Use the non-stationary Markov chain model optimized in step 4 to predict the meteorological drought level of each high-resolution grid point in the region, and obtain the spatial distribution of the drought level in the whole region and the evolution process of the proportion of the affected area. 2. a kind of regional meteorological drought grade prediction method of coupling multi-source data as claimed in claim 1 is characterized in that: In step 2, the multiple linear regression method is used to couple the underlying surface information to construct a spatial downscaling model, and the original remote sensing monitoring precipitation data is scaled down. 3. a kind of regional meteorological drought grade prediction method of coupling multi-source data as claimed in claim 1 or 2, is characterized in that: In step 2, the geographic information difference method is used to further correct the output high-resolution precipitation data by using the precipitation measured by the ground observation station. 4. a kind of regional meteorological drought grade prediction method of coupling multi-source data as claimed in claim 1 is characterized in that: In step 3, the standardized drought index uses the standardized precipitation index. 5. a kind of regional meteorological drought grade prediction method of coupling multi-source data as claimed in claim 1 is characterized in that: In step 4, the generalized cross-entropy method is used to estimate the parameters of each non-stationary Markov chain model. 6. A kind of regional meteorological drought grade prediction method of coupling multi-source data as claimed in claim 1 or 5, is characterized in that: In step 4, the historical drought level is calculated by back-substitution with the calibrated non-stationary Markov chain model, and the final meteorological drought level prediction model is selected using the Akaike information criterion.

Images