CN115617935A - Underground water reserve deviation downscaling method based on fusion model - Google Patents
Underground water reserve deviation downscaling method based on fusion model Download PDFInfo
- Publication number
- CN115617935A CN115617935A CN202211271859.7A CN202211271859A CN115617935A CN 115617935 A CN115617935 A CN 115617935A CN 202211271859 A CN202211271859 A CN 202211271859A CN 115617935 A CN115617935 A CN 115617935A
- Authority
- CN
- China
- Prior art keywords
- deviation
- model
- water reserve
- reserve
- degrees
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- XLYOFNOQVPJJNP-UHFFFAOYSA-N water Substances O XLYOFNOQVPJJNP-UHFFFAOYSA-N 0.000 title claims abstract description 71
- 230000004927 fusion Effects 0.000 title claims abstract description 46
- 238000000034 method Methods 0.000 title claims abstract description 46
- 238000013136 deep learning model Methods 0.000 claims abstract description 22
- 238000010801 machine learning Methods 0.000 claims abstract description 18
- 238000012952 Resampling Methods 0.000 claims abstract description 10
- 239000002352 surface water Substances 0.000 claims abstract description 6
- 239000003673 groundwater Substances 0.000 claims description 21
- 238000004088 simulation Methods 0.000 claims description 9
- 239000002689 soil Substances 0.000 claims description 8
- 238000007637 random forest analysis Methods 0.000 claims description 7
- 238000011160 research Methods 0.000 claims description 7
- 238000012549 training Methods 0.000 claims description 7
- 230000014509 gene expression Effects 0.000 claims description 6
- 238000010276 construction Methods 0.000 claims description 4
- 238000010219 correlation analysis Methods 0.000 claims description 4
- 238000012795 verification Methods 0.000 claims description 4
- 230000006870 function Effects 0.000 claims description 3
- 238000005457 optimization Methods 0.000 claims description 3
- 238000012545 processing Methods 0.000 claims description 3
- 238000001556 precipitation Methods 0.000 claims description 2
- 238000005549 size reduction Methods 0.000 description 7
- 238000002790 cross-validation Methods 0.000 description 5
- 241001494479 Pecora Species 0.000 description 3
- 230000000694 effects Effects 0.000 description 3
- 238000011156 evaluation Methods 0.000 description 3
- 238000012544 monitoring process Methods 0.000 description 3
- 238000013528 artificial neural network Methods 0.000 description 2
- 230000005484 gravity Effects 0.000 description 2
- 239000007787 solid Substances 0.000 description 2
- 238000011144 upstream manufacturing Methods 0.000 description 2
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000001364 causal effect Effects 0.000 description 1
- 238000013135 deep learning Methods 0.000 description 1
- 238000010586 diagram Methods 0.000 description 1
- 238000005516 engineering process Methods 0.000 description 1
- 239000003621 irrigation water Substances 0.000 description 1
- 238000007726 management method Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000005259 measurement Methods 0.000 description 1
- 238000012805 post-processing Methods 0.000 description 1
- 230000000306 recurrent effect Effects 0.000 description 1
- 238000012360 testing method Methods 0.000 description 1
- 238000010200 validation analysis Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F16/00—Information retrieval; Database structures therefor; File system structures therefor
- G06F16/20—Information retrieval; Database structures therefor; File system structures therefor of structured data, e.g. relational data
- G06F16/29—Geographical information databases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/18—Complex mathematical operations for evaluating statistical data, e.g. average values, frequency distributions, probability functions, regression analysis
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q50/00—Information and communication technology [ICT] specially adapted for implementation of business processes of specific business sectors, e.g. utilities or tourism
- G06Q50/06—Energy or water supply
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06V—IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING
- G06V20/00—Scenes; Scene-specific elements
- G06V20/10—Terrestrial scenes
- G06V20/182—Network patterns, e.g. roads or rivers
Landscapes
- Engineering & Computer Science (AREA)
- Physics & Mathematics (AREA)
- Theoretical Computer Science (AREA)
- General Physics & Mathematics (AREA)
- Data Mining & Analysis (AREA)
- Databases & Information Systems (AREA)
- Health & Medical Sciences (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Business, Economics & Management (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Mathematics (AREA)
- General Health & Medical Sciences (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Economics (AREA)
- Software Systems (AREA)
- Pure & Applied Mathematics (AREA)
- Bioinformatics & Computational Biology (AREA)
- Molecular Biology (AREA)
- Probability & Statistics with Applications (AREA)
- Multimedia (AREA)
- Operations Research (AREA)
- Artificial Intelligence (AREA)
- Biomedical Technology (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Evolutionary Computation (AREA)
- Evolutionary Biology (AREA)
- Algebra (AREA)
- Computing Systems (AREA)
- Bioinformatics & Cheminformatics (AREA)
- Remote Sensing (AREA)
- Public Health (AREA)
- Water Supply & Treatment (AREA)
- Human Resources & Organizations (AREA)
- Marketing (AREA)
- Primary Health Care (AREA)
- Strategic Management (AREA)
- Tourism & Hospitality (AREA)
Abstract
The invention belongs to the cross technical field of hydrology, image enhancement and the like, and provides a fusion model-based underground water reserve deviation downscaling method, which comprises the following steps: resampling each driving variable related to the underground water reserve from high resolution to low resolution by utilizing a bilinear interpolation method; constructing a statistical relationship model of the driving variables and the target variables by adopting a fusion model; obtaining the land water reserve deviation with high resolution by adopting the land water reserve deviation and the residual error obtained by the statistical relation model; and subtracting the surface water reserve deviation from the high-resolution land water reserve deviation to obtain the underground water reserve deviation. According to the method, the fusion model of multiple single-type machine learning models and the deep learning model is constructed, so that the scale reduction precision of the traditional single-type model on the underground water reserve deviation is obviously improved, and a certain reference is provided for planning and utilizing the underground water resources in the region or watershed scale.
Description
Technical Field
The invention belongs to the cross technical field of hydrology, image enhancement and the like, and particularly relates to a groundwater reserve deviation downscaling method based on a fusion model.
Background
Groundwater is the most important source of irrigation water and human life water in arid and semiarid regions. Under the background of extreme weather and increasing population of the world, the underground water exploitation amount is increased rapidly, and more attention is paid to the problem caused by underground water exhaustion, so that the method has important significance for accurately monitoring underground water change and regional underground water resource management. The traditional underground water network monitoring consumes a large amount of manpower and material resources and is difficult to apply to a larger space-time scale. A gravity measurement and climate monitoring (GRACE) satellite which is launched to lift off in 3 months in 2002 can provide earth gravity field change information with higher space-time precision, so that a great deal of research is carried out on the basis of a water balance theory and the inversion and reconstruction of underground water reserve deviation by combining GRACE satellite data and global terrestrial assimilation system (GLDAS) data. Since the actual spatial resolutions of the GRACE data after being processed by the spherical harmonics method and the point quality method are 1 ° × 1 ° and 0.5 ° × 0.5 °, to maintain the consistency of the resolutions of the two types of data, most studies resample the GLDAS data with the resolution of 0.25 ° × 0.25 ° to 0.5 ° × 0.5 ° or 1 ° × 1 °, resulting in many important spatial information loss in the GLDAS.
In order to solve the problem of inconsistent resolution of the Grace satellite data and the GLDAS data, researchers propose a plurality of downscaling methods, which are generally divided into a dynamic downscaling method and a statistical downscaling method. The dynamic downscaling method has a more solid physical foundation, but needs to consume a large amount of computing resources, and has an extremely high computing power requirement on a computer when the space-time scale is large or the resolution is improved, so the statistical downscaling method has a wider application. In many statistical downscaling methods, a machine learning model and a deep learning model can better handle nonlinear relations between driving variables and target variables, however, from the existing research, when underlying conditions are complex, causal relations captured by a single type of machine learning model or a deep learning model are different and have a limited range, so that the global effect of model downscaling is still poor. Therefore, the effective multi-model fusion means is provided, and the method has important significance for improving the data accuracy of the GRACE satellite, inverting and reconstructing the underground water reserve deviation.
Disclosure of Invention
In order to overcome the problems in the prior art, the invention provides a groundwater reserve deviation downscaling method based on a fusion model, a set of groundwater reserve deviation data set with higher spatial precision and more reliability is obtained by establishing a downscaling model of groundwater quantity deviation, and reference can be provided for planning and utilizing regional or watershed scale groundwater resources.
The purpose of the invention is realized as follows:
a fusion model-based underground water reserve deviation downscaling method comprises the following steps:
resampling each driving variable related to the underground water reserve to 0.5 degrees multiplied by 0.5 degrees from 0.25 degrees multiplied by 0.25 degrees by utilizing a bilinear interpolation method, and obtaining a data set of the driving variables;
step 2, constructing a statistical relationship model of the driving variables and the target variables:
constructing a statistical relationship model of the driving variables processed in the step 1 and the target variables by adopting a fusion model; wherein the target variable is GRACE land water reserve deviation data obtained by adopting a point mass method;
the construction of the fusion model comprises the following steps:
s21, respectively training the data sets of the driving variables obtained in the step 1 by adopting m machine learning models and n deep learning models through a 3-fold cross verification method;
s22, estimating the precision of each model in the S21 by adopting a Nash efficiency coefficient (NS), a Pearson Correlation Coefficient (CC), a consistency Index (IG) and a Kling-Gupta efficiency coefficient (KG), wherein the expressions are respectively as follows:
in the formula o i And s i The ith observed value and the ith analog value are respectively; sigma s And σ o Respectively, standard deviations of the analog value and the observed value; mu.s s And mu o Respectively taking the mean values of the analog value and the observed value;
and
respectively are the mean values of the observation sequence and the simulation sequence;
forming a new array by using the output of the single type model ranked at the top by the sum of the four indexes and the output result with the minimum absolute value deviation from the original value in the m + n models;
s23, training the new array obtained by processing in the S22 by respectively using a linear fusion model and a nonlinear fusion model, and taking the fusion model with the largest sum of NS, CC, IG and KG as the statistical relationship model;
and 3, obtaining the land water reserve deviation with high resolution:
simulating land water reserve deviation with the resolution of 0.5 degrees multiplied by 0.5 degrees by using the statistical relation model constructed in the step 2, subtracting the target variable to obtain a residual error with the resolution of 0.5 degrees multiplied by 0.5 degrees, and resampling the resolution of the residual error to 0.25 degrees multiplied by 0.25 degrees by using a nearest neighbor interpolation method; then simulating land-water reserve deviation with 0.25-degree multiplied by 0.25-degree resolution by the statistical relationship model constructed in the step 2, and adding the land-water reserve deviation to obtain land-water reserve deviation after size reduction;
and 4, verifying the rationality of the downscaling result:
and (4) subtracting the surface water reserve deviation from the land water reserve deviation after the size reduction obtained in the step (3) to obtain the underground water reserve deviation of 0.25 degrees multiplied by 0.25 degrees, and further carrying out correlation analysis on the actually measured underground water level of the research area to verify the rationality of the size reduction result.
Further, the driving variables in step 1 include precipitation, surface temperature, surface runoff, subsurface runoff, actual evapotranspiration and soil humidity.
Further, in step 2, in S21, the machine learning model is constructed by using a pycart package in Python, and the optimal hyper-parameter of each model is obtained by using a tune _ model function.
Further, in step 2, in step 21, the deep learning model is constructed by adopting a tsai package in Python, and hyper-parameter optimization is performed by using a hyper package.
Further, in step 2, in step S23, a super set average model is selected as a linear fusion model, and the expression is as follows:
in the formula (S) MMSE ) t MMSE model simulation for t periodA value;
is the average value of the multi-model analog values in the t period; w is a j Weighting values for each model; f j,t And
respectively a predicted value and a predicted mean value of the jth model in the t period; z is the number of models.
Further, in step 2, in step S23, a random forest model is selected as the nonlinear fusion model.
Further, in step 4, the surface water reserve deviation includes a canopy water reserve deviation, a snow water equivalent deviation and a soil water reserve deviation.
Compared with the prior art, the invention has the advantages and beneficial effects that:
according to the groundwater reserve deviation downscaling method, multiple single-type machine learning models and a fusion model of a deep learning model are built, nash efficiency coefficients, pearson correlation coefficients, consistency indexes and Kling-Gupta efficiency coefficients are used as indexes, consistency of the models before and after downscaling is verified, spatial resolution of GRACE satellite data and GLDAS data is unified, downscaling precision of a traditional single-type model on groundwater reserve deviation is remarkably improved, and certain reference is provided for planning and utilization of regional or watershed scale groundwater resources.
Drawings
The invention is further illustrated by the following figures and examples.
FIG. 1 is a flow chart of a fusion model-based groundwater reserve deviation downscaling method according to the present invention;
FIG. 2 is a flow chart of construction of the fusion model according to the present invention;
FIG. 3 shows cross validation results of simulation effects of the optimal machine learning model, the optimal deep learning model, the linear fusion model, and the nonlinear fusion model according to the embodiment of the present invention;
FIG. 4 is a graph comparing land water reserve deviations before and after downscaling according to embodiments of the present invention; wherein (a) is land water reserve deviation before 6 months downscaling in 2005; (b) land water reserve deviation after 6 months downscaling in 2005;
FIG. 5 is a diagram of the rationality validation result of the downscaling result according to an embodiment of the present invention; wherein, (a) is the geographical position of the middle and upper reaches of the sheep river, the guan Zhong and the black river; (b) The GWSA is the space distribution of the correlation coefficient of the geodetic GWSA of the river basin of the sheep and the underground water level of the observation well; (c) The GWSA of the downscaling of the customs area and the correlation coefficient space distribution of the underground water level of the observation well are obtained; (d) And the GWSA which is the upstream downscaling in the black river and the correlation coefficient spatial distribution of the underground water level of the observation well.
Detailed Description
The technical solutions in the embodiments of the present invention will be described clearly and completely with reference to the accompanying drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be derived by a person skilled in the art from the embodiments given herein without making any creative effort, shall fall within the protection scope of the present invention.
Example (b):
as shown in the flow chart of fig. 1, the present embodiment provides a groundwater reserve deviation downscaling method based on a fusion model, taking research of groundwater reserves in northwest regions of China as an example, and the method includes the following steps:
and resampling all driving variables related to the underground water reserve, namely rainfall, surface temperature, surface runoff, underground runoff, evapotranspiration and soil humidity to 0.5 degrees and 0.5 degrees from the resolution of 0.25 degrees and 0.25 degrees by utilizing a bilinear interpolation method, and obtaining a data set of the driving variables.
The data source and spatial accuracy of the driving variables in step 1 are shown in table 1.
TABLE 1 data Source and detailed information
Step 2, constructing a statistical relationship model of the driving variables and the target variables:
a fusion model is used to construct a statistical relationship model between the driving variables processed in the step 1 and target variables, wherein the target variables are GRACE land water reserve deviation (GRACE-TWSA) data obtained by a point mass method in Jet Propulsion Laboratory (JPL), and the specific data are obtained from Improved methods for observing Earth's time variable distribution with GRACE using road maps, published in 2015 of journal, j. Geophys. Res. Solid Earth, by Watkins et al (doi:10.1002/2014JB011547) The website is as follows:
https://podaac.jpl.nasa.gov/dataset/TELLUS_GRAC-GRFO_MASCON_GRID_RL06_V2)。
compared with the spherical harmonic coefficient method, the point quality method can reduce the leakage error of the GRACE data to the maximum extent, and does not need post-processing, so in order to reduce the uncertainty of the GRACE data, the data obtained by the point quality method is selected as the target variable.
As shown in fig. 2, the construction of the fusion model includes the following 3 steps:
s21, respectively training on the data set of the driving variables with the resolution of 0.5 degrees multiplied by 0.5 degrees obtained in the step 1 by adopting an m-machine learning model and n deep learning models through a 3-fold cross verification method (namely, 2/3 of pixel points are used as a training set and 1/3 of pixel points are used as a verification set every time).
In this embodiment, 22 machine learning models and 16 deep learning models are included, and the models are shown in tables 2 and 3 below.
TABLE 2 22 machine learning models
TABLE 3 16 deep learning models
The machine learning models are constructed by adopting a pycaret packet in Python, and the optimal hyper-parameters of the models are obtained by utilizing a tune _ model function.
The deep learning model is constructed by adopting a tsai package in Python, the package is based on Pythrch and Fastai, and the latest deep learning technology is adopted, so that the deep learning model can be used for time series classification, regression and prediction. Because the number of models and samples are more, the range of the hyper-parameters of the neural network is preset, hyper-parameter optimization is carried out by utilizing a hyper-op packet, and the type and the range of the hyper-parameters of each model are shown in the following tables 4 and 5.
TABLE 4 common hyper-parameters of deep learning models and their value ranges
TABLE 5 unique associated hyper-parameters and value ranges for partial depth models
S22, estimating the precision of each model in the S21 by adopting a Nash efficiency coefficient (NS), a Pearson Correlation Coefficient (CC), a consistency Index (IG) and a Kling-Gupta efficiency coefficient (KG), wherein the expressions are respectively as follows:
in the formula o i And s i The ith observed value and the ith analog value are respectively; sigma s And σ o Respectively, standard deviations of the analog value and the observed value; mu.s s And mu o Respectively taking the mean values of the analog value and the observed value;
and
mean values for the observed and simulated sequences, respectively.
The closer each index is to 1, the higher the model accuracy. Therefore, the output of the single type model ranked at the top by the sum of the four indexes and the output result with the minimum deviation absolute value from the original value in the m + n models form a new array.
Specifically, in order to keep the simplicity of the code, the first five models and the output result with the smallest absolute value deviation from the measured value in the 38 m + n models are selected to form a new array.
The following tables 6 and 7 list the cross-validation results of the Correlation Coefficient (CC), the consistency Index (IG), the nash efficiency coefficient (NS), and the Kling-Gupta efficiency coefficient (KG) of the 16 deep learning models and the 22 machine learning models related to the present example, respectively. Compared with a machine learning model, the evaluation index mean value of the deep learning model is relatively larger, and the TWSA is simulated to have stronger learning capacity. Meanwhile, the standard deviation of each index of the deep learning model is smaller than that of the machine learning model, so that the deep learning models have more similar learning effects.
For example, LSTM, GRU and RNN all belong to the recurrent neural network. In addition, different evaluation indexes have different emphasis points, and evaluation results are greatly different. For example, in the deep learning model, the CC value and the IG value of LSTM-FNC, GRU, RNN and LSTM are large, but the NS value and the KG value are small, which indicates that although these models can simulate the variation trend of TWSA more accurately, the simulation error is large.
Table 6 cross-validation results for 22 machine learning models
Table 7 Cross-validation results for 16 deep learning models
And S23, training the new array obtained by processing in the S22 by respectively using a linear fusion model and a nonlinear fusion model, and performing scale reduction by using the fusion model with the largest sum of NS, CC, IG and KG as the statistical relationship model.
Wherein, a super set mean (MMSE) model is used as a linear fusion model, and the expression is as follows:
in the formula (S) MMSE ) t Is a simulation value of an MMSE model in a t period;
is the average value of the multi-model analog values in the t period; w is a j Weighting values for each model; f j,t And
respectively a predicted value and a predicted mean value of the jth model in the t period; z is the number of models, and 5 is taken in the embodiment; most preferablyAnd (5) reducing the model error to obtain the weight of each model.
A Random Forest (RF) model is utilized as the nonlinear fusion model.
And 3, obtaining the land water reserve deviation with high resolution:
simulating land water reserve deviation (Corse-TWSA) with 0.5 degrees multiplied by 0.5 degrees coarse resolution by using the statistical relationship model constructed in the step 2, subtracting the land water reserve deviation (GRACE-TWSA, namely the target variable) obtained by adopting a point mass method to obtain residual error with 0.5 degrees multiplied by 0.5 degrees resolution, and resampling the residual error resolution to 0.25 degrees multiplied by 0.25 degrees by using a nearest neighbor interpolation method; and then the land water reserves deviation (Fine-TWSA) with the Fine resolution of 0.25 degrees multiplied by 0.25 degrees is simulated and added with the statistical relation model constructed in the step 2 to obtain the land water reserves deviation (SIM-TWSA) after the dimension reduction.
As shown in table 8, MMSE does not improve the simulation accuracy of the single model, and the nonlinear RF model significantly improves KG and NS, reducing the single model simulation error. As shown in fig. 3, KG and NS of the RF fusion model were improved by 49%, 60%, 45%, and 51% respectively, compared to the most accurate deep learning model (ResNet) and the machine learning model (xgboost). Therefore, a nonlinear RF fusion model is adopted to carry out scale reduction on the groundwater reserve deviation in the northwest region. Taking 6 months in 2005 as an example, as shown in fig. 4, before and after the downscaling, the TWSA high values are concentrated in the west of Xinjiang, the Quercondian region, the north of Ningxia and the north of Shaanxi, and the low values are concentrated in the boundary between Xinjiang and Qinghai, the south of Qinghai province and the south of Shaanxi province. Meanwhile, the data set after the size reduction can more finely depict the TWSA distribution condition.
Table 8 cross-validation results of two fusion models
And 4, verifying the rationality of the downscaling result:
and (3) subtracting surface water reserve deviation (crown layer water reserve deviation, snow water equivalent deviation and soil water reserve deviation) from the land water reserve deviation (SIM-TWSA) obtained after the size reduction in the step (3) to obtain the underground water reserve deviation of 0.25 degrees multiplied by 0.25 degrees, wherein the underground water reserve deviation (GWSA) is equal to the underground water level deviation multiplied by the soil water supply degree, but no space distribution data of the soil water supply degree exists at present, so that the rationality of the size reduction result can be verified by carrying out correlation analysis on the actually measured underground water level of the research area.
In the embodiment, the downscaled groundwater reserve deviation is obtained based on the step 4. And verifying the rationality of the scale reduction result by performing correlation analysis on typical actually measured underground water levels of the river basin, the Guanzhong area and the upstream in the black river. And when a plurality of observation wells are distributed on one grid point, calculating the deviation mean value of the underground water levels of the plurality of wells.
As shown in fig. 5, the Pearson correlation coefficient of each typical region is relatively high, and most of the grid points pass the significance test of 0.01, wherein the correlation coefficient of the central area ranges from 0.08 to 0.88, and the mean value is 0.66; the related coefficient range of the river basin of the sheep is-0.04-0.72, and the mean value is 0.48; the correlation coefficient range of the black river basin is 0.05-0.93, and the mean value is 0.45, which shows that the GWSA after size reduction can effectively capture the underground water level change of the research area.
Finally, it should be noted that the above is only for illustrating the technical solution of the present invention and not for limiting, and although the present invention is described in detail with reference to the preferred arrangement, it should be understood by those skilled in the art that the technical solution of the present invention (such as the application of various formulas, the sequence of steps, etc.) can be modified or equivalently replaced without departing from the spirit and scope of the technical solution of the present invention.
Claims (7)
1. A fusion model-based underground water reserve deviation downscaling method is characterized by comprising the following steps:
step 1, drive variable resampling:
resampling each driving variable related to the underground water reserve to 0.5 degrees multiplied by 0.5 degrees from 0.25 degrees multiplied by 0.25 degrees by utilizing a bilinear interpolation method, and obtaining a data set of the driving variables;
step 2, constructing a statistical relationship model of the driving variables and the target variables:
constructing a statistical relationship model of the driving variables processed in the step 1 and target variables by adopting a fusion model, wherein the target variables are GRACE land water reserve deviation data obtained by adopting a point mass method;
the construction of the fusion model comprises the following steps:
s21, respectively training the data sets of the driving variables obtained in the step 1 by adopting m machine learning models and n deep learning models through a 3-fold cross verification method;
s22, estimating the precision of each model in the S21 by adopting a Nash efficiency coefficient (NS), a Pearson Correlation Coefficient (CC), a consistency Index (IG) and a Kling-Gupta efficiency coefficient (KG), wherein the expressions are respectively as follows:
in the formula o i And s i Respectively an ith observed value and an ith analog value; sigma s And σ o Respectively, standard deviations of the analog value and the observed value; mu.s s And mu o Respectively taking the mean values of the analog value and the observed value;
and
respectively mean values of the observation sequence and the simulation sequence;
the output of the single type model ranked at the top by the sum of the four indexes and the output result with the minimum absolute value deviation from the original value in the m + n models form a new array;
s23, training the new array obtained by processing the data in the S22 by respectively using a linear fusion model and a nonlinear fusion model, and taking the fusion model with the largest sum of NS, CC, IG and KG as the statistical relationship model;
and 3, obtaining the land water reserve deviation with high resolution:
simulating land water reserve deviation with the resolution of 0.5 degrees multiplied by 0.5 degrees by using the statistical relation model constructed in the step 2, subtracting the target variable to obtain a residual error with the resolution of 0.5 degrees multiplied by 0.5 degrees, and resampling the resolution of the residual error to 0.25 degrees multiplied by 0.25 degrees by using a nearest neighbor interpolation method; then the land water reserve deviation with the resolution of 0.25 degrees multiplied by 0.25 degrees is simulated and added with the statistical relation model constructed in the step 2 to obtain the land water reserve deviation after the dimension reduction;
and 4, verifying the rationality of the downscaling result:
and (3) subtracting the surface water reserve deviation from the land water reserve deviation after the downscaling obtained in the step (3) to obtain the groundwater reserve deviation of 0.25 degrees multiplied by 0.25 degrees, and further verifying the rationality of the downscaling result by carrying out correlation analysis on the actually measured groundwater level of the research area.
2. A fusion model based groundwater reserve deviation downscaling method according to claim 1, wherein the driving variables in step 1 include precipitation, surface temperature, surface runoff, subsurface runoff, actual evapotranspiration and soil humidity.
3. A fusion model-based underground water reserve deviation downscaling method according to claim 1, wherein in step 2, S21, the machine learning model is constructed by using PyCaret package in Python, and the optimal hyper-parameter of each model is obtained by using tune _ model function.
4. The fusion model-based groundwater reserve deviation downscaling method according to claim 1, wherein in step 2, in step S21, the deep learning model is constructed by using a tsai package in Python, and hyper-parameter optimization is performed by using a hyper-op package.
5. The fusion model-based groundwater reserve deviation downscaling method according to claim 1, wherein in step 2, in S23, a super set average model is selected as a linear fusion model, and an expression is as follows:
in the formula (S) MMSE ) t Is a simulation value of an MMSE model in a t period;
is the average value of the multiple model analog values in the t time period; w is a j Weighting values for each model; f j,t And
respectively a predicted value and a predicted mean value of the jth model in the t period; z is the number of models.
6. The fusion model-based groundwater reserve deviation downscaling method according to claim 1, wherein in step 2, in step S23, a random forest model is selected as the nonlinear fusion model.
7. A fusion model-based groundwater reserve deviation downscaling method according to claim 1, wherein in step 4, the surface water reserve deviation comprises a canopy water reserve deviation, a snow water equivalent deviation and a soil water reserve deviation.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211271859.7A CN115617935A (en) | 2022-10-18 | 2022-10-18 | Underground water reserve deviation downscaling method based on fusion model |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202211271859.7A CN115617935A (en) | 2022-10-18 | 2022-10-18 | Underground water reserve deviation downscaling method based on fusion model |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN115617935A true CN115617935A (en) | 2023-01-17 |
Family
ID=84862318
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202211271859.7A Pending CN115617935A (en) | 2022-10-18 | 2022-10-18 | Underground water reserve deviation downscaling method based on fusion model |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN115617935A (en) |
Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108106979A (en) * | 2017-12-21 | 2018-06-01 | 深圳先进技术研究院 | A kind of PM2.5 inversion methods merged based on MODIS and machine learning model |
| CN108268735A (en) * | 2018-01-29 | 2018-07-10 | 浙江大学 | Soil moisture NO emissions reduction method based on multi-source remote sensing satellite fused data |
| CN109635309A (en) * | 2018-10-17 | 2019-04-16 | 广州地理研究所 | A kind of surface temperature space NO emissions reduction method |
| CN110442937A (en) * | 2019-07-24 | 2019-11-12 | 武汉大学 | A kind of valley water security management method merging satellite remote sensing and machine learning techniques |
| CN111078678A (en) * | 2019-12-18 | 2020-04-28 | 中国气象局乌鲁木齐沙漠气象研究所 | Satellite precipitation data correction method based on multi-source information fusion and scale reduction |
| CN111898660A (en) * | 2020-07-17 | 2020-11-06 | 武汉大学 | Hydrological simulation method for averagely fusing multi-source data based on Bayesian mode |
| CN112949158A (en) * | 2020-12-23 | 2021-06-11 | 中国空间技术研究院 | Method for improving spatial resolution and precision of underground water level variable quantity |
| CN113486000A (en) * | 2021-09-08 | 2021-10-08 | 中国测绘科学研究院 | Surface evapotranspiration data downscaling method based on multi-source data and deep learning |
| CN114120101A (en) * | 2021-11-02 | 2022-03-01 | 武汉大学 | Soil moisture multi-scale comprehensive sensing method |
| CN114218853A (en) * | 2021-11-18 | 2022-03-22 | 中国空间技术研究院 | Method for improving water reserve resolution and precision based on machine learning space downscaling |
-
2022
- 2022-10-18 CN CN202211271859.7A patent/CN115617935A/en active Pending
Patent Citations (10)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN108106979A (en) * | 2017-12-21 | 2018-06-01 | 深圳先进技术研究院 | A kind of PM2.5 inversion methods merged based on MODIS and machine learning model |
| CN108268735A (en) * | 2018-01-29 | 2018-07-10 | 浙江大学 | Soil moisture NO emissions reduction method based on multi-source remote sensing satellite fused data |
| CN109635309A (en) * | 2018-10-17 | 2019-04-16 | 广州地理研究所 | A kind of surface temperature space NO emissions reduction method |
| CN110442937A (en) * | 2019-07-24 | 2019-11-12 | 武汉大学 | A kind of valley water security management method merging satellite remote sensing and machine learning techniques |
| CN111078678A (en) * | 2019-12-18 | 2020-04-28 | 中国气象局乌鲁木齐沙漠气象研究所 | Satellite precipitation data correction method based on multi-source information fusion and scale reduction |
| CN111898660A (en) * | 2020-07-17 | 2020-11-06 | 武汉大学 | Hydrological simulation method for averagely fusing multi-source data based on Bayesian mode |
| CN112949158A (en) * | 2020-12-23 | 2021-06-11 | 中国空间技术研究院 | Method for improving spatial resolution and precision of underground water level variable quantity |
| CN113486000A (en) * | 2021-09-08 | 2021-10-08 | 中国测绘科学研究院 | Surface evapotranspiration data downscaling method based on multi-source data and deep learning |
| CN114120101A (en) * | 2021-11-02 | 2022-03-01 | 武汉大学 | Soil moisture multi-scale comprehensive sensing method |
| CN114218853A (en) * | 2021-11-18 | 2022-03-22 | 中国空间技术研究院 | Method for improving water reserve resolution and precision based on machine learning space downscaling |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN110738252B (en) | Space autocorrelation machine learning satellite precipitation data downscaling method and system | |
| Fan et al. | Temporal and spatial change detecting (1998–2003) and predicting of land use and land cover in Core corridor of Pearl River Delta (China) by using TM and ETM+ images | |
| Nkuna et al. | Filling of missing rainfall data in Luvuvhu River Catchment using artificial neural networks | |
| CN110619432A (en) | Feature extraction hydrological forecasting method based on deep learning | |
| CN101853290A (en) | Meteorological service performance evaluation method based on geographical information system (GIS) | |
| CN113902580B (en) | Historical farmland distribution reconstruction method based on random forest model | |
| Fang et al. | Application of long short-term memory (LSTM) on the prediction of rainfall-runoff in karst area | |
| CN111199298A (en) | Flood forecasting method and system based on neural network | |
| CN115544889A (en) | Numerical mode precipitation deviation correction method based on deep learning | |
| CN112100922A (en) | Wind resource prediction method based on WRF and CNN convolutional neural network | |
| CN114676621A (en) | Method for improving accuracy of land water reserve abnormity based on deep learning weight load | |
| Yin et al. | Improving the resolution of GRACE-based water storage estimates based on machine learning downscaling schemes | |
| CN201716727U (en) | Geographical simulation system based on remote sensing and GIS | |
| Chen et al. | Comparison of deep learning models and a typical process-based model in glacio-hydrology simulation | |
| CN116796799A (en) | Method for creating small-river basin flood rainfall threshold model in area without hydrologic data | |
| CN117194926A (en) | Method and system for predicting hoisting window period of land wind power base | |
| CN115392128B (en) | Method for simulating river basin runoff by utilizing space-time convolution LSTM network | |
| CN102880753A (en) | Method for converting land utilization spatial characteristic scale based on fractal dimension | |
| Wang et al. | A data‐driven approach for flood prediction using grid‐based meteorological data | |
| CN115952743A (en) | Multi-source precipitation data collaborative downscaling method and system coupled with random forest and HASM | |
| CN115617935A (en) | Underground water reserve deviation downscaling method based on fusion model | |
| Sun et al. | A surrogate modeling method for distributed land surface hydrological models based on deep learning | |
| Chen et al. | Urban inundation rapid prediction method based on multi-machine learning algorithm and rain pattern analysis | |
| Zhang et al. | High resolution annual irrigation water use maps in China based-on input variables selection and convolutional neural networks | |
| Dong et al. | A hybrid prediction model for wind speed using support vector machine and genetic programming in conjunction with error compensation |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |