Quantitative rainfall estimation algorithm of FY4B satellite multichannel data based on machine learning

Abstract

The invention relates to a quantitative rainfall estimation algorithm of FY4B satellite multichannel data based on machine learning, and belongs to the technical field of rainfall estimation, and the calculation method provided by the invention takes the rainfall actually measured by a ground rain gauge as a standard, and evaluates and analyzes the effect of the optimization algorithm and the temporal and spatial change of an estimation result. As a whole, the method has relatively strong correlation with the ground rainfall gauge and relatively good stability, along with the increase of the average hourly rainfall of the rainfall gauge, the time is prolonged, the calculation rate of the correlation coefficient f and the average square error S is accelerated, the correlation coefficient f and the average square error S are continuously calculated according to the actual situation, and the rainfall model is continuously corrected, so that the calculation accuracy of the rainfall model is improved. And the overall trend of reduction is realized. The measurement deviation is obviously reduced, and meanwhile, the estimation precision of light rain is improved. The estimation deviation of light rain is reduced by 25%, and the estimation deviations of heavy rain and heavy rain are respectively reduced by 78% and 69%.

Description

Technical Field

[0001] The present invention relates to the technical field of precipitation estimation, and in particular to a quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning. Background Art

[0002] With the frequent occurrence of disastrous weather in recent years, radar quantitative precipitation estimation (QPE: Quantitative Precipation Estimation) information with high temporal and spatial resolution has played an increasingly important role in the short-term approach and refined early warning and forecasting of disastrous weather such as rainstorms, typhoons, and floods.

[0003] Precipitation is one of the important weather elements. Accurate and quantitative estimation of precipitation is of great significance. Using weather radar to estimate precipitation is a common problem in meteorological and hydrological services. At present, radar measurement of precipitation is mainly based on the relationship between radar reflectivity factor (echo intensity) Z and ground precipitation intensity I, that is, Z = AXI b , where A and b are empirical coefficients. The accuracy of quantitative estimation of precipitation by radar depends largely on the determination of A and b coefficients in the ZI relationship.

[0004] By combining the ground station rain gauge data and radar observation data, the ZI relationship can be determined through certain mathematical statistical methods, thereby realizing radar precipitation estimation. The empirical coefficients determined by traditional methods are generally fixed. In fact, the A and b empirical coefficients are not fixed and change with factors such as region, season, precipitation type, and distance from the radar center.

[0005] Precipitation is a key driving force for the global water-carbon-energy cycle and is of great significance to the study of global ecology and climate change. Infrared observation, passive microwave observation and active radar scanning are the three main precipitation observation methods based on satellite platforms. Among them, since infrared sensors can be carried on geostationary satellites and microwave sensors are generally only carried on low-Earth orbit satellites, infrared observations can have the advantages of high spatiotemporal resolution and high spatiotemporal coverage that other remote sensing observations are difficult to have, and have become an important source of observation data in precipitation estimation. Prior art, such as a satellite infrared multi-channel precipitation quantitative estimation method based on tomography with application number CN202311344773.7, relates to the field of precipitation prediction technology, the method includes: obtaining geographic auxiliary data of the target area and infrared observation brightness temperature of a specific band of a geostationary satellite; based on the geographic auxiliary data of the target area and the infrared observation brightness temperature of a specific band of a geostationary satellite and a precipitation prediction model, obtaining the precipitation prediction value of the target area; the precipitation prediction model is obtained by training the tomography model using a training data set, the training data set includes: geographic auxiliary data of multiple historical satellite full disk areas, infrared observation brightness temperature of a specific band of a geostationary satellite and precipitation values ​​observed by a near-orbit satellite radar, the tomography model includes multiple tomography kernels, one tomography kernel corresponds to one specific band. The empirical coefficient of the above invention is fixed, but in fact the empirical coefficient is not fixed, and changes with factors such as region, season, precipitation type, and distance from the radar center, resulting in low precipitation prediction accuracy. Summary of the invention

[0006] The present invention aims to solve the above technical problems and provides a quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning.

[0007] In order to solve the above technical problems, the technical solution provided by the present invention is:

[0008] A quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning includes the following steps:

[0009] S1. Use the multi-spectral channel of the Fengyun-4 meteorological satellite to collect data and CLDAS multi-source fusion data to extract precipitation related factors;

[0010] S2, extract relevant factors of precipitation based on data and CLDAS multi-source fusion data to classify different precipitation types;

[0011] S3, establish a precipitation model and calculate the precipitation prediction formulas R(Z, D) and R(K, D) for the target area;

[0012] R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 Formula 1;

[0013] R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K) Formula 2;

[0014] S4, establishing a radar quantitative precipitation estimation model, revising and evaluating the precipitation assessment model, and calculating the correlation coefficient f;

[0015] S5, establish a precipitation recognition model, perform a final model test on the precipitation assessment model, and calculate the mean square error S;

[0016] S6. Substitute the correlation coefficient and mean square error into R(Z, D) and R(K, D) to recalculate the precipitation;

[0017] R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 ×f×S formula 3;

[0018] R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K)×f×S Formula 4.

[0019] Preferably, the collected data include national stations, regional light, near infrared, infrared and water vapor channels.

[0020] Preferably, the related factors include longitude and latitude, regional precipitation type, particle phase, raindrop spectrum parameters, radar echo data and ground rain gauge data.

[0021] Preferably, the R() is rainfall, in units of mm·h －1 , Z is the horizontal reflectivity factor, unit is dBz, D is the differential reflectivity, K is the differential propagation phase shift rate, unit is ° km -1 .

[0022] Preferably, the calculation formulas of the correlation coefficient f and the mean square error S are as follows:

[0023]

[0024] Preferably, the p is the amount of precipitation delivered to the inspection station by each inspected precipitation product, the g is the precipitation observed by the inspection station, and the n is the number of samples.

[0025] Preferably, each inspected precipitation product is a quantitative precipitation with high temporal and spatial resolution when it arrives at the inspection station, and its spatial resolution is 0.01°×0.01°.

[0026] Preferably, the false alarm rate, detection probability, critical success index and fair threat score are introduced into the precipitation model to evaluate the accuracy of the radar quantitative precipitation estimation model and predict the area ratio of the actual non-precipitation area in the precipitation area.

[0027] Preferably, the collected data is subjected to noise filtering and isolated echo filtering.

[0028] After adopting the above method, the present invention has the following advantages:

[0029] The calculation method of the present invention uses the measured rainfall of the ground rain gauge as the standard, and evaluates and analyzes the effect of the optimization algorithm and the temporal and spatial changes of the estimation results. On the whole, it shows a strong correlation with the ground rain gauge and has good stability. As the average hourly rainfall of the rain gauge increases and the time is prolonged, the calculation rate of the correlation coefficient f and the mean square error S is accelerated. The correlation coefficient f and the mean square error S are continuously calculated according to the actual situation, and the precipitation model is continuously corrected, and there is an overall trend of decreasing. The estimation deviation is significantly reduced, and the estimation accuracy of light rain is improved. The estimation deviation of light rain is reduced by 25%, and the estimation deviation of heavy rain and rainstorm is reduced by 78% and 69% respectively.

[0030] 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

[0031] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the drawings required for use in the embodiments or technical descriptions will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0032] Figure 1 It is a graph of the experimental results of the present invention. DETAILED DESCRIPTION

[0033] Specific embodiments of the present invention will now be mentioned in detail. Although the present invention is described in conjunction with these specific embodiments, it should be appreciated that it is not intended to limit the present invention to these specific embodiments. On the contrary, these embodiments are intended to cover substitutions, changes or equivalent embodiments that may be included in the spirit and scope of the invention defined by the claims. In the following description, a large number of specific details are set forth in order to provide a comprehensive understanding of the present invention. The present invention can be implemented without some or all of these specific details. In other cases, in order not to make the present invention unnecessarily obscure, well-known process operations are not described in detail.

[0034] When used in conjunction with "including," "methods comprising," or similar language in this specification and the appended claims, the singular forms "a," "an," and "the" include plural references unless the context clearly dictates otherwise. Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0035] The present invention is further described in detail below in conjunction with the full text.

[0036] Combined with Figure 1 , a quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning, including the following steps:

[0037] S1. Use the multi-spectral channel of the Fengyun-4 meteorological satellite to collect data and CLDAS multi-source fusion data to extract precipitation related factors;

[0038] S2, extract relevant factors of precipitation based on data and CLDAS multi-source fusion data to classify different precipitation types;

[0039] S3, establish a precipitation model and calculate the precipitation prediction formulas R(Z, D) and R(K, D) for the target area;

[0040] R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 Formula 1;

[0041] R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K) Formula 2;

[0042] S4, establishing a radar quantitative precipitation estimation model, revising and evaluating the precipitation assessment model, and calculating the correlation coefficient f;

[0043] S5, establish a precipitation recognition model, perform a final model test on the precipitation assessment model, and calculate the mean square error S;

[0044] S6. Substitute the correlation coefficient and mean square error into R(Z, D) and R(K, D) to recalculate the precipitation;

[0045] R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 ×f×S formula 3;

[0046] R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K)×f×S Formula 4.

[0047] The collected data include national stations, regional light, near infrared, infrared and water vapor channels.

[0048] The related factors include longitude and latitude, regional precipitation type, particle phase, raindrop spectrum parameters, radar echo data and ground rain gauge data.

[0049] The R() is the rainfall in mm·h －1 , Z is the horizontal reflectivity factor, unit is dBz, D is the differential reflectivity, K is the differential propagation phase shift rate, unit is ° km -1 .

[0050] The calculation formulas of the correlation coefficient f and the mean square error S are as follows:

[0051]

[0052] The p is the amount of precipitation delivered to the inspection station by each inspected precipitation product, the g is the precipitation observed by the inspection station, and the n is the number of samples.

[0053] In the specific implementation of the present invention, by calculating the correlation coefficient f and the mean square error S, it is solved that the statistical relationship between radar parameters and ground precipitation also has regional characteristics due to differences in precipitation types, particle phases, and raindrop spectrum parameters in different regions.

[0054] By continuously calculating the correlation coefficient f and the mean square error S, the precipitation model can be corrected to improve the accuracy of precipitation prediction.

[0055] The precipitation products tested are quantitative precipitation with high temporal and spatial resolution when they arrive at the testing station, and their spatial resolution is 0.01°×0.01°. The minute-level rain gauge data is used to perform real-time calibration and correction on the hour-level radar-estimated precipitation to improve the accuracy. The radar estimation results are verified and calibrated in combination with the ground rain gauge data to ensure accuracy. The quantitative precipitation with high temporal and spatial resolution is a quantitative precipitation product with high temporal and spatial resolution, such as a 1-hour precipitation distribution map, with a spatial resolution of up to 0.01°×0.01°.

[0056] Machine learning algorithms such as convolutional neural network (CNN), radial basis function neural network (RBF neural network), and multi-layer perceptron (MLP) are used to establish and optimize precipitation models.

[0057] The products were evaluated and corrected using ground rain gauge data and monthly climate background data to reduce system errors. The multi-source data fusion method based on radar, satellite and FY4B satellite multi-channel data can effectively improve the quality control of precipitation data. The 1-hour radar quantitative estimated precipitation of the ground automatic observation station was generated, and an automatic weather station precipitation observation system based on radar and satellite estimated precipitation products was established.

[0058] The precipitation model introduces false alarm rate, detection probability, critical success index and fair threat score to evaluate the accuracy of the radar quantitative precipitation estimation model and predict the area ratio of the actual non-precipitation area in the precipitation area.

[0059] When the present invention is implemented, the collected data is subjected to noise filtering and isolated echo filtering. The collected data after noise filtering and isolated echo filtering solves the problem of random signal interference in the process of image acquisition or transmission. Some random, discrete, and isolated pixels appearing on the image will interfere with the analysis of image information, thereby improving the accuracy of the collected data.

[0060] In the radar quantitative precipitation estimation model, in a circular area with the radar center as the center, the rainfall and radar echo intensity of multiple rain gauges (Note: stations with rain gauge observations, referred to as rain gauges, the same below) in the radar scanning area are taken as a sample number, and the sample number is n.

[0061] In the specific implementation of the present invention, the false alarm rate, detection probability, critical success index and fair threat score are introduced into the precipitation model to evaluate the accuracy of the radar quantitative precipitation estimation model and predict the area ratio of the actual non-precipitation area in the precipitation area. The calculation formula for introducing the false alarm rate, detection probability, critical success index and fair threat score is as follows:

[0062]

[0063]

[0064] The meaning of the numbers is shown in the following table:

[0065] Classification table of digital meanings

[0066] precipitation Non-precipitation RF Forecast:Precipitation NA NB RF forecast: No precipitation NC ND

[0067] Among them, the false alarm rate is introduced to represent the area ratio of the actual non-precipitation area in the precipitation area predicted by the model, the detection probability represents the possibility that the model can correctly identify precipitation pixels when precipitation is actually observed, the critical success index represents the proportion of precipitation pixels in the correctly classified pixels when the correctly classified non-precipitation points are removed, and the fair threat score represents the proportion of correctly classified pixels compared with random predictions after considering unexpected situations. The smaller the false alarm rate is, the higher the detection probability, critical success index and fair threat score are, the lower the false alarm rate of precipitation events is, and the better its precipitation recognition ability is.

[0068] Embodiment 1:

[0069] Precipitation is estimated for its area within the range of 31-39°N and 89-101°E on July 24.

[0070] A quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning includes the following steps:

[0071] S1. Use the multi-spectral channel of the Fengyun-4 meteorological satellite to collect data and CLDAS multi-source fusion data to extract precipitation related factors;

[0072] S2, extract relevant factors of precipitation based on data and CLDAS multi-source fusion data to classify different precipitation types;

[0073] S3, establish a precipitation model and calculate the precipitation prediction formulas R(Z, D) and R(K, D) for the target area;

[0074] R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 Formula 1;

[0075] R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K) Formula 2;

[0076] S4, establishing a radar quantitative precipitation estimation model, revising and evaluating the precipitation assessment model, and calculating the correlation coefficient f;

[0077] S5, establish a precipitation recognition model, perform a final model test on the precipitation assessment model, and calculate the mean square error S;

[0078] S6. Substitute the correlation coefficient and mean square error into R(Z, D) and R(K, D) to recalculate the precipitation;

[0079] R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 ×f×S formula 3;

[0080] R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K)×f×S Formula 4;

[0081] Embodiment 2:

[0082] Precipitation is estimated for its area on July 28 in the range of 31-39°N and 89-101°E.

[0083] A quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning includes the following steps:

[0084] S1. Use the multi-spectral channel of the Fengyun-4 meteorological satellite to collect data and CLDAS multi-source fusion data to extract precipitation related factors;

[0085] S2, extract relevant factors of precipitation based on data and CLDAS multi-source fusion data to classify different precipitation types;

[0086] S3, establish a precipitation model and calculate the precipitation prediction formulas R(Z, D) and R(K, D) for the target area;

[0087] R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 Formula 1;

[0088] R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K) Formula 2;

[0089] S4, establishing a radar quantitative precipitation estimation model, revising and evaluating the precipitation assessment model, and calculating the correlation coefficient f;

[0090] S5, establish a precipitation recognition model, perform a final model test on the precipitation assessment model, and calculate the mean square error S;

[0091] S6. Substitute the correlation coefficient and mean square error into R(Z, D) and R(K, D) to recalculate the precipitation;

[0092] R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 ×f×S formula 3;

[0093] R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K)×f×S Formula 4;

[0094] Experimental results:

[0095] like Figure 1 As shown, b1 and b2 are the comparison of the result diagrams of Example 1, b1 is the estimated result of the rainfall during the precipitation process of the present invention, and b2 is the corresponding actual measurement result of the rain gauge; c1 and c2 are the comparison of the result diagrams of Example 2, c1 is the estimated result of the rainfall during the precipitation process of the present invention, and c2 is the corresponding actual measurement result of the rain gauge.

[0096] exist Figure 1 In the figure, the estimated results can correspond well to the rain gauge in areas with more rainfall and areas with less rainfall, which further illustrates that the estimated results of the optimization algorithm of the present invention have good consistency with the ground rain gauge. As the average hourly rainfall of the rain gauge increases and the time is prolonged, the calculation rate of the correlation coefficient f and the mean square error S is accelerated, and the deviation tends to decrease overall. In summary, the optimization algorithm of the present invention shows a strong correlation with the ground rain gauge and has good stability. As the average hourly rainfall of the rain gauge increases, the correlation coefficient f and the mean square error S are continuously calculated according to the actual situation, and tend to decrease overall.

[0097] The present invention and its embodiments are described above, and such description is not restrictive. What is shown in the full text is only one of the embodiments of the present invention, and the actual structure is not limited thereto. In short, if ordinary technicians in the field are inspired by it, without departing from the purpose of the invention, they can design a structure and embodiment similar to the technical solution without creativity, which should belong to the protection scope of the present invention.

Claims

1. A quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning, It is characterized in that The following steps are involved: S1. Use the multi-spectral channel of the Fengyun-4 meteorological satellite to collect data and CLDAS multi-source fusion data to extract precipitation related factors; S2, extract relevant factors of precipitation based on data and CLDAS multi-source fusion data to classify different precipitation types; S3, establish a precipitation model and calculate the precipitation prediction formulas R(Z, D) and R(K, D) for the target area; R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 Formula 1; R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K) Formula 2; S4, establishing a radar quantitative precipitation estimation model, revising and evaluating the precipitation assessment model, and calculating the correlation coefficient f; S5, establish a precipitation recognition model, perform a final model test on the precipitation assessment model, and calculate the mean square error S; S6. Substitute the correlation coefficient and mean square error into R(Z, D) and R(K, D) to recalculate the precipitation; R(Z, D) = 0.00786(Z) 0.978 ×D -4.96 ×f×S formula 3; R(K, D)=136×|K| 0.806 ×(Z) -2.86 ×sign(K)×f×S Formula 4.

2. The quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning according to claim 1, Features: The collected data include national stations, regional light, near infrared, infrared and water vapor channels.

3. The quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning according to claim 1, Features: The related factors include longitude and latitude, regional precipitation type, particle phase, raindrop spectrum parameters, radar echo data and ground rain gauge data.

4. The quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning according to claim 1, Features: The R() is the rainfall in mm·h －1 , Z is the horizontal reflectivity factor, unit is dBz, D is the differential reflectivity, K is the differential propagation phase shift rate, unit is °·km -1 .

5. The quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning according to claim 1, Features: The calculation formulas of the correlation coefficient f and the mean square error S are as follows:

6. The quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning according to claim 5, Features: The p is the amount of precipitation delivered to the inspection station by each inspected precipitation product, the g is the precipitation observed by the inspection station, and the n is the number of samples.

7. The quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning according to claim 6, Features: Each of the inspected precipitation products will be converted into quantitative precipitation with high temporal and spatial resolution when it arrives at the inspection station, and its spatial resolution is 0.01°×0.01°.

8. The quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning according to claim 1, Features: The precipitation model introduces false alarm rate, detection probability, critical success index and fair threat score to evaluate the accuracy of the radar quantitative precipitation estimation model and predict the area ratio of the actual non-precipitation area in the precipitation area.

9. The quantitative precipitation estimation algorithm based on FY4B satellite multi-channel data based on machine learning according to claim 1, Features: The collected data is subjected to noise filtering and isolated echo filtering processing.

Images