Search

CN-121522643-B - Ground precipitation intensity estimation method based on phased array radar echo and geographic factors

CN121522643BCN 121522643 BCN121522643 BCN 121522643BCN-121522643-B

Abstract

The invention relates to a ground precipitation intensity estimation method based on phased array radar echoes and geographic factors, which belongs to the field of radio, and comprises the steps of firstly mapping and interpolating polar coordinate multivariable radar echoes and three-dimensional wind fields to unified three-dimensional grids to form a consistent space-time data base, then automatically dividing multiple rainfall type areas based on local density clustering, constructing a time hysteresis field according to precipitation types to realize self-adaptive time alignment of radar echoes and ground station observation, introducing three-dimensional wind fields and raindrop terminals to finish final speed calculation of precipitation drift, correcting grid point positions matched with ground station space to complete accurate space matching of radar echoes and ground station observation, combining stratum wind direction, DEM, gradient, slope direction and slope position type construction terrain elevation indexes, combining radar nonlinear precipitation potential vectors and terrain features by a physical constraint coupling model to realize quantitative estimation of ground precipitation intensity, and improving reliability and applicability of ground precipitation estimation in complex environments.

Inventors

  • LUO JICHENG
  • WANG WENMING
  • DING HONGXIN
  • YANG GANG

Assignees

  • 成都远望科技有限责任公司

Dates

Publication Date
20260505
Application Date
20260119

Claims (9)

  1. 1. The ground precipitation intensity estimation method based on phased array radar echo and geographic factors is characterized by comprising the following steps of: s1, collecting phased array radar echo data, radar running state data, ground automatic station observation data and geographic factor data, inputting the radar running state data into a radar beam center approximate height estimation model and a map projection model, combining the obtained result with the radar echo data to construct a target three-dimensional grid, processing points in the three-dimensional grid to obtain three-dimensional wind field data sets for generating target three-dimensional grids at different moments, normalizing the phased array radar reflectivity factor data sets, differential reflectivity factor data sets and differential phase shift rate data sets to generate normalized reflectivity factor data sets, differential reflectivity factor data sets and differential phase shift rate data sets, and dividing the normalized radar running state data sets into areas of several rainfall types including light rain, medium rain, heavy rain and heavy rain by combining a local clustering model and a local density calculation model; S2, obtaining an area average reflectivity factor sequence and an area average rainfall sequence of a rainfall type area according to collected radar running state data, ground automatic station observation data, a reflectivity factor data set and a determined rainfall type area, obtaining an optimal lag time difference of the ground automatic station observation data lagging behind phased array radar observation according to a set time lag sequence, an area average reflectivity factor sequence and an area average rainfall sequence, delaying the reflectivity factor data set by the optimal lag time difference time to obtain a plurality of rainfall types, combining and unifying radar reflectivity factor data sets, and generating a differential reflectivity factor data set and a differential phase shift data set which are aligned with the ground automatic station observation data set in time and integrated with the ground automatic station observation data set by combining the differential reflectivity factor data set and the differential phase shift data set; S3, calculating the average horizontal drift distance after raindrops fall to the ground according to the collected ground automatic station observation data, the three-dimensional wind field data set and the combined and unified radar reflectivity factor data set and combining a radar reflectivity factor and a raindrop terminal end speed relation model, so as to generate a phased array radar echo data set which is subjected to space-time alignment and space matching with the ground automatic station observation data set; s4, generating a basic terrain lifting index according to the three-dimensional wind field data set and the collected geographic factor data, and constructing a terrain feature vector according to the data output in the S3, the collected geographic factor data and the basic terrain lifting index; s5, constructing a radar terrain coupling physical constraint model according to the ground automatic station observation data set and the data output in the S4, training, and finally estimating ground rainfall intensity through the model.
  2. 2. The method for estimating ground precipitation intensity based on phased array radar echo and geographic factors of claim 1, wherein the phased array radar echo data comprises radar reflectivity factor data Z0_par, differential reflectivity factor data Zdr0_par, differential phase shift data KDP0_par and three-dimensional wind field data (U0, V0, W0) subjected to four-dimensional variation assimilation inversion at different time points at spatial positions (r 0, phi0, sita 0) under polar coordinates, wherein r0 is a radial distance, phi0 is an azimuth, sita0 is an antenna elevation angle, the three-dimensional wind field data means (U0, V0, W0), U0 is a component of a horizontal wind speed eastern direction, V0 is a component of a horizontal wind speed northbound, and W0 is a vertical wind speed; The radar running state data comprises that the spatial position information of a radar station is P_par= (lon_par, lat_par and H_par), wherein lon_par, lat_par and H_par are longitude and latitude of the radar station and the height of an antenna feed source respectively; radial distance resolution delt_r_par and beam width delt_ sita _par of the radar; a radar volume scanning period delt_t and a starting time T0 of radar scanning; The ground automatic station observation data comprises rainfall Ra0_sta observed at different moments, wherein the space geographic coordinates of the rainfall Ra0_sta are (x_R0, y_R0), and x_R0 and y_R0 are position information in the directions of an X axis pointing to the east and a Y axis pointing to the north respectively; the geographic factor data comprise digital elevation data DEM, gradient data SLOPE, gradient data ASPECT and gradient type data TPI.
  3. 3. The method for estimating the ground precipitation intensity based on the phased array radar echo and the geographic factor according to claim 2, wherein the steps of inputting the radar operation state data into a radar beam center approximate height estimation model and a map projection model, constructing a target three-dimensional grid by combining the obtained result with radar echo data, processing points in the three-dimensional grid to obtain a three-dimensional wind field data set for generating the target three-dimensional grid at different moments, and the phased array radar reflectivity factor data set, the differential reflectivity factor data set and the differential phase shift rate data set specifically comprise the following steps: A1, the collected radar site space position information P_par, (R0, phi0, sita 0), delt_R_par and delt_ sita _par are input into a radar beam center approximate height estimation model together, the model is operated, and the radar beam center approximate height z0 is output; A2, inputting the collected spatial position information P_par of the radar site into a map projection model, running the model, outputting the model as plane projection coordinates (X_par, Y_par) of the radar site, constructing rays with azimuth angles phi0 in the plane coordinates according to the collected (R0, phi0, sita 0), delt_R_par and delt_ sita _par, taking a horizontal distance as s0=r0×cos (sita 0), taking projection points of ray center lines on the horizontal plane as x0=X_par+s0×sin (phi 0), and combining the points (X0, Y0) with the acquired z0, and converting polar coordinates of radar echo data into three-dimensional space coordinates (X0, Y0, z 0); A3, setting the spatial resolutions of the horizontal grids in the X and Y directions as delt_x and delt_y respectively, setting the resolution of the vertical height as delt_z, thus constructing a target three-dimensional grid, for a certain point M in the target three-dimensional grid, calculating the radial distance r_m and the azimuth angle phi_m from the point M to the plane projection coordinates (X_par, Y_par) of the radar site, wherein the spatial sitting marks are (x_m, y_m and z_m); for any point I in (X0, Y0, Z0), the sitting sign is (x_i, y_i, z_i), the reflectivity factor data corresponding to the point I and the three-dimensional wind field data are extracted according to the collected radar echo data and respectively marked as Z0 par_i and (U0_i, V0_i, W0_i), wherein U0_i is a component of the horizontal wind speed at the point I, V0_i is a component of the horizontal wind speed at the point I, W0_i is a component of the horizontal wind speed at the point I, the vertical wind speed at the point I is calculated, the coordinate difference of the point I relative to the point M is marked as (delta_x_i, delta_y_i, delta_z_i), delta_x_i and delta_y_i are decomposed into transverse directions which are consistent with the beam broadening directions along the radial directions, and are marked as (delta_r_i, delta_b_i); A4, setting radial direction weight, transverse direction weight and vertical direction weight coefficients as apha _r, apha_b and apha _z respectively, calculating the anisotropic distance of the I point as distance_i= apha _r×delta_r_i 2 +apha_b×delta_b_i 2 +apha_z×delta_z_i 2 , further calculating the inverse distance weight index as W_i=1/distance_i, and calculating the reflectivity factor Z_par_m and three-dimensional wind field data (U0_i_m, V0_i_m, W0_i_m) at the midpoint M of the target three-dimensional grid by using a weighted average method, wherein U0_i_m is a component of horizontal wind speed eastward at the M point, V0_i_m is a component of horizontal wind speed northward at the M point, and W0_i_m is a vertical wind speed at the M point; A5, applying the steps of A3 and A4 to all points in the target three-dimensional grid at different moments, and generating a phased array radar reflectivity factor data set Z_par and a three-dimensional wind field data set (U, V, W) of the target three-dimensional grid at different moments, wherein U, V and W are an eastward component of the horizontal wind speed, an northward component of the horizontal wind speed and a vertical wind speed of the target three-dimensional grid at different moments respectively; A6, the reflectivity factor data set in the step A2-A5 is replaced by a differential reflectivity factor data set Zdr _par and a differential phase shift rate data set KDP0_par in sequence, and a phased array radar differential reflectivity factor data set Zdr _par and a differential phase shift rate KDP_par data set of a target three-dimensional grid at different moments are generated.
  4. 4. The method for estimating the ground rainfall intensity based on the phased array radar echo and the geographic factor of claim 3, wherein the generating the normalized reflectivity factor data set, the differential reflectivity factor data set and the differential phase shift rate data set after normalization and dividing the area into a plurality of rainfall types by combining a local clustering model and a local density calculation model specifically comprises the following steps: Searching the maximum value of the generated reflectivity factor data set Z_par, dividing each reflectivity factor data by the maximum value of the reflectivity factor to generate normalized reflectivity factor data, and applying the same processing method to the differential reflectivity factor data set and the differential phase shift rate data set to generate normalized differential reflectivity factor data set and differential phase shift rate data set; Inputting the normalized reflectivity factors, the differential reflectivity factors and the differential phase shift rate data set into a local clustering model based on density, setting an initial searching radius r_orig around data points and a minimum point value MinPts of clusters, operating the local clustering model, generating preliminarily divided areas with different rainfall intensity types, counting average reflectivity factors in each area, and preliminarily marking each rainfall area as a small rain area, a medium rain area, a heavy rain area and a heavy rain area according to the relation between the average reflectivity factors and the rainfall intensity classification; Inputting the generated reflectivity factor data set Z_par into a local density calculation model, running the local density calculation model, generating a local density value of each reflectivity factor data point, and respectively counting the local density data in the areas preliminarily marked with light rain, medium rain, heavy rain and heavy rain according to the local density value of each reflectivity factor data point to obtain the average local density rho of each precipitation area; According to an optimization calculation formula of the search radius based on the local density, the search radius r_opt after the optimization of the regions of light rain, medium rain, heavy rain and heavy rain is obtained respectively, and then the initial search radius r_orig around the data points set in the local clustering model is updated to r_opt, the local clustering model is restarted, and the regions of the types of light rain, medium rain, heavy rain and heavy rain are accurately divided.
  5. 5. The method for estimating the ground precipitation intensity based on phased array radar returns and geographical factors as set forth in claim 4, wherein S2 comprises the following steps: B1, constructing a phased array radar observation time sequence ts=t0+k×delt_t according to the collected delt_t and T0, k=0, 1,2,..K, wherein K is the total number of radar scanning periods, extracting a phased array radar reflectivity factor data set corresponding to the time sequence ts according to a phased array radar reflectivity factor data set Z_par at a target three-dimensional grid at different output moments, namely Z_par_ts, extracting the phased array radar reflectivity factor data set corresponding to a region determined to be a small rain type from the Z_par_ts, namely Z_par_ts_light, extracting a region average reflectivity factor sequence of a small rain type region corresponding to the time sequence ts according to a space average method, and namely Z_par_ts_light_m; B2, generating ground automatic station rainfall observation data on a time sequence ts by adopting a linear interpolation method in time according to collected ground automatic station observation data R0_sta at different moments, wherein the ground automatic station rainfall observation data is marked as R0_sta_ts, a ground automatic station observation data set corresponding to a region determined to be a small rain type is extracted from the R0_sta_ts and is marked as R0_sta_ts_light, and then the region average rainfall rate sequence of the region of the small rain type corresponding to the time sequence ts is marked as R0_sta_ts_ ligh _m according to a space average method; B3, setting a time delay sequence with a step length of delt_T, wherein m=0, 1,2, & M-1, wherein M is the total number of delays, selecting a sequence corresponding to the time delay value tau from the R0_sta_ts_ ligh _m sequence, marking the sequence as R0_sta_ts_ ligh _m1, and calculating a correlation coefficient between Z_par_ts_light_m and R0_sta_ts_ ligh _m1 for a certain delay value tau in the time delay sequence; B4, sequentially updating the hysteresis value tau to all M values in the time delay, repeating the step B3, respectively calculating correlation coefficients between Z_par_ts_light_m and the rainfall data of the ground automatic station after the M hysteresis values to form a correlation coefficient sequence, marking as CC, identifying the maximum value in the CC sequence, extracting a time hysteresis value corresponding to the maximum value, namely, marking as the optimal hysteresis time difference of the ground automatic station observation data lagging the phased array radar observation, marking as the time delay light, delaying Z_par_ts_light by the time delay light, and forming a radar reflectivity factor data set of the phased array aligned with the ground automatic station observation data R0_sta_ts_light in a small rain area, namely marking as Z1_par_ts_light; B5, respectively applying the steps of B1-B4 to the medium rain, heavy rain and storm areas, and then calculating the optimal lag time difference (time delay_middle, time delay_middle and time delay_store) corresponding to the medium rain, heavy rain and storm areas, so as to generate a ground automatic station observation data set (R0_sta_ts_middle, R0_sta_ts_store) and a phased array radar reflectivity factor data set (Z1_par_ts_middle, Z1_par_ts_store) aligned with the medium rain, heavy rain and storm areas, wherein R0_sta_ts_store, R0_sta_store are respectively the ground automatic station data in the medium rain, heavy rain and storm areas, and Z1_par_ts_store are respectively aligned with the phased array radar reflectivity factor data in the medium rain, heavy rain and storm areas; B6, the radar reflectivity factor data sets [ Z1_par_ts_light ] under different rainfall intensities, z1_par_ts_middle, Z1_par/u ts_middle, and the corresponding ground automatic station observation data R0 sta ts light, r0_sta_ts_middle, r0_sta_ts_leave, r0_sta_ts_storm ] is integrated into a ground automatic station rainfall dataset r1_sta_ts arranged by precipitation intensity; B7, respectively extracting differential reflectivity factors and differential phase shift rates of corresponding rainfall areas according to the determined regions of light rain, medium rain, heavy rain and heavy rain by combining the phased array radar differential reflectivity factor dataset Zdr _par and the differential phase shift rate dataset KDP_par of the target three-dimensional grid at different output moments; B8, utilizing the obtained optimal delay time differences (time_delay_middle, time_delay_mean, time_delay_store) corresponding to the medium, heavy and heavy rain areas to respectively delay the differential reflectivity factors and differential phase shift of each rainfall area, then merging and unifying the processed data, and finally generating a differential reflectivity factor dataset Zdr1_par_ts and a differential phase shift data set KDP1_par_ts which are aligned with the ground automatic station observation data set in time.
  6. 6. The method for estimating the ground precipitation intensity based on phased array radar returns and geographical factors of claim 5, wherein S3 comprises the following steps: C1, selecting a certain station G from the collected ground automatic station observation data, wherein the geographic coordinates of the station G are (x_R0_G, y_R0_G), and then combining the output Z1_par_ts, and finding out a nearest phased array radar echo grid point P right above the station G by using a nearest space matching method, wherein the position information on the horizontal plane is recorded as (x_par_P, y_par_P); C2, setting a horizontal search window of inclined precipitation as N_h×N_h, wherein N_h is the width of the window size, the number of matching layers in the vertical direction of inclined precipitation as N_v, determining that an adjacent airspace above a station G is Vol_space=N_h×N_h×N_v, respectively calculating the average value of U and V components on an ith altitude layer according to a three-dimensional wind field data set (U, V, W) and Z1_par_ts and combining the range of the adjacent airspace Vol_space, and recording the average value of reflectivity factor data on the ith altitude layer as Z_i, i=1, 2,., N_v; C3, inputting Z_i into a relation model of a reflectivity factor and a terminal velocity of the raindrops, running the model, generating terminal velocity of the raindrops on the ith Height layer, recording as Vt_i, obtaining height_i=i×delt_z of the grid of the ith Height layer according to the set delt_z, combining the obtained Vt_i, obtaining a time difference of the raindrops on the ith Height layer falling onto the ground as delta_t_i=height_i/Vt_i, respectively calculating distances delta_X_i=U_i×delta_t_i of the raindrops in X and Y directions in the time difference delta_t_i according to the obtained U_i and V_i, delta_Y_i=V_i×t_i, and calculating a reflectivity factor weighting coefficient apha _i=Z/sum of the obtained all N_v Height layers according to a reflectivity factor data average value Z_i, wherein the distances delta_X_i=U_i and delta_i are a delta_m (sum) function; C4, calculating weighted average distances delta_x=sum (delta_x_i× apha _i) and delta_y=sum (delta_y_i× apha _i) of adjacent airspace raindrops drifting in the X and Y axis directions above the ground automation station G according to the obtained apha _i, delta_x_i and delta_y_i, and according to delta_x and delta_y, combining position information (x_par_p, y_par_p) on the horizontal plane of the nearest phased array radar echo grid point P above the ground automation station G, and adjusting the P-position to generate phased array radar echo grid point Q which is exactly matched with the sky above the ground automation station G, wherein the horizontal position information is (x_par_p+delta_x, y_par_p+y); C5, the steps of C1-C4 are applied to all stations of the ground automatic station, and phased array radar reflectivity factors, differential reflectivity factors and differential phase shift rate data sets which are subjected to time alignment and space matching with the ground automatic station observation data set R1_sta_ts are respectively marked as Z1_par_ts_match, ZDr1_par_ts_match and KDp1_par_ts_match.
  7. 7. The method for estimating a ground level water level based on phased array radar returns and geographic factors of claim 5 wherein said generating a base terrain elevation index from the three-dimensional wind field dataset and the collected geographic factor data comprises: Extracting a horizontal wind field (U_low, V_low) of the lowest high layer according to the target three-dimensional wind field data set (U, V, W), wherein U_low is an eastern component of the horizontal wind speed of the lowest high layer, V_low is a north component of the horizontal wind speed of the lowest high layer, and a near-stratum wind direction W_dir=atan2 (U_low, V_low) is calculated, wherein atan2 () returns a clockwise angle of the wind direction relative to the north; according to collected slope data ASPECT, calculating an included angle delta_angle=ASPECT-W_dir between the slope direction and the wind direction, and further calculating a windward slope factor F_wind=max (0, cos (delta_angle×pi/180)), wherein max () is a maximum function; Searching the maximum value of the collected gradient data SLOPE, recording as max_SLOPE, calculating a gradient factor F_slope=SLOPE/max_SLOPE, setting a neighborhood search radius R_DEM according to the collected digital elevation data DEM, respectively calculating the digital elevation average value in each search radius with each DEM lattice point as the center and the radius R_DEM, further calculating the elevation difference of each lattice point relative to the digital elevation average value in each search radius, recording as delta_DEM, carrying out normalization processing on the delta_DEM by utilizing a hyperbolic tangent function to generate F_DEM=0.5× (delta_DEM/DEM 0)), wherein tan () is a hyperbolic tangent function, and DEM0 is a scale parameter; According to the collected slope type data TPI, a slope weight coefficient w_tpi is generated based on a slope type mapping weight rule, and according to the generated f_wind, f_slope, f_dem and w_tpi, a base terrain lift index terrain _lift=w_tpi×f_wind×f_slope×f_dem is generated.
  8. 8. The method for estimating the ground precipitation intensity based on the phased array radar echo and the geographic factor of claim 7, wherein the constructing the terrain feature vector based on the data output in the step S3, the collected geographic factor data and the basic terrain elevation index specifically comprises the following steps: Structured particle size distribution index feature f2= Zdr1 construction particle size distribution index feature f2= Zdr1_par_ts_match/z1_par_ts_match, the Z1_par_ts_match, the Zdr1_par_ts_match and the KDP1_par_ts_match are phased array radar reflectivity factors, differential reflectivity factors and differential phase shift data sets after time alignment and space matching respectively; constructing a nonlinear radar precipitation potential vector R_radar= [ Z1_par_ts_match, zdr _par_ts_match, KDP1_par_ts_match, F1, F2], and according to collected DEM, SLOPE, ASPECT, TPI and terrain _lift, firstly extracting digital elevation, gradient, SLOPE direction, SLOPE position and basic topography lifting index data sets which are spatially matched with the R1_sta_ts data sets according to a nearest neighbor principle and respectively marking as DEM1, SLOPE1, ASPECT1, TPI1 and terrain _lift1; The SLOPE and SLOPE dynamics feature t_d= [ SLOPE1, sin (ASPECT 1), cos (ASPECT 1) ], the topography and elevation feature t_m= [ TPI1, terrain _lift1], and then the topography feature vector t_ terrain = [ DEM1, t_d, t_m ] are constructed.
  9. 9. The method for estimating the ground precipitation intensity based on phased array radar returns and geographical factors of claim 8, wherein S5 comprises the following steps: According to the output R1_sta_ts, R_radar and T_ terrain, firstly constructing an end-to-end radar terrain coupling physical constraint MODEL MODEL formed by three sub-MODELs, wherein the sub-MODEL 1 is formed by a multi-layer perceptron MODEL, an input data set and an output data set of the sub-MODEL 1 during training are respectively R_radar and R1_sta_ts, training the sub-MODEL 1, and outputting the training result to ignore radar precipitation potential P_radar under the influence of terrain after training is completed; the submodel 2 is composed of XGBoost models, an input data set and an output data set are respectively T_ terrain and R1_sta_ts when the submodel 2 is trained, and a precipitation terrain modulation factor M_ terrain is output after training is finished; The submodel 3 is formed by mapping a physical constraint function, the output of the submodel 3 is R_out=P_radar×exp (M_ terrain), wherein exp () is a logarithmic function, a loss function of the radar terrain coupling physical constraint MODEL MODEL is set as a mean square error loss function, and the radar terrain coupling physical constraint MODEL MODEL is trained; After the radar terrain coupling physical constraint MODEL MODEL is trained, R1 st ts and R radar, T terrain are input into the trained radar terrain coupling physical constraint MODEL MODEL again, and the MODEL is operated to realize ground precipitation intensity estimation based on phased array radar echo and geographic factors.

Description

Ground precipitation intensity estimation method based on phased array radar echo and geographic factors Technical Field The invention relates to the field of radio, in particular to a ground precipitation intensity estimation method based on phased array radar echo and geographic factors. Background The ground rainfall intensity is a key parameter for monitoring, early warning and risk assessment of storm flood, mountain flood geological disasters, urban inland inundation and the like. The phased array weather radar has the characteristics of high body scanning speed, high time resolution and capability of acquiring multi-elevation angle multivariable three-dimensional echoes, and provides an important data source for quantitative precipitation estimation with high space-time resolution. However, radar observation reflects the scattering characteristics of precipitation particles in a certain height range, ground precipitation is also influenced by the falling process, wind field transportation, terrain lifting and the like, and obvious space-time inconsistency exists between the ground precipitation and the wind field transportation and the terrain lifting, so that a high-precision ground precipitation estimation method facing the characteristics of the phased array radar is urgently needed. The prior phased array radar ground precipitation estimation method mainly comprises the steps that 1, a first type of method is a quantitative precipitation estimation method based on an empirical relationship, and the method generally utilizes an empirical relationship (such as a Z-R relationship) between a reflectivity factor and precipitation intensity or combines dual polarization parameters such as a differential reflectivity factor, a differential phase shift rate and the like to construct an empirical or semi-empirical formula. The method has simple structure and high calculation efficiency, but the parameters of the method generally depend on specific areas, specific precipitation types and statistical samples, once the micro-physical characteristics of precipitation or environmental conditions change, the estimation error is obviously increased, and the method is difficult to adapt to the complex precipitation process. 2. The second type of method is a physical inversion method based on dual polarization variables. The method enhances the characterization capability of the particle size distribution and the water content of the raindrops by introducing variables such as differential reflectivity factors, differential phase shift rates and the like, and improves the estimation accuracy under the condition of strong precipitation to a certain extent. However, such methods are still mainly aimed at radar to observe the precipitation characteristics of high layers, and do not fully consider the space-time evolution and the terrain influence of precipitation particles in the falling process, so that the method has limited applicability to mountainous areas or complex terrain areas. 3. The third class of methods are precipitation estimation methods based on machine learning or deep learning. In recent years, some studies have attempted to map radar observation variables directly to ground precipitation intensities using neural networks, random forests, and other models. The fitting capacity is improved to a certain extent by the method, but most researches only use radar variables as input, and neglect systematic influences of terrain lifting, windward slope effect and wind field conveying on precipitation distribution in the precipitation process. Meanwhile, the existing method is used for carrying out simple time or space matching on the radar and the ground station, so that the problems of time delay and horizontal drift between radar echo and ground precipitation are not fully solved, sample noise is large, and model generalization capability is limited. Furthermore, the advantages of high time resolution of phased array radars have not been fully exploited in the current research. Most methods still use the processing thought of the traditional body scanning radar, lack of self-adaptive time lag modeling and space matching mechanisms aiming at areas with different rainfall intensities, and are difficult to finely describe the dynamic relationship between radar echoes and ground rainfall responses under different rainfall types. Disclosure of Invention The invention aims to overcome the defects of the prior art, provides a ground precipitation intensity estimation method based on phased array radar echo and geographic factors, and solves the defects in the prior art. The invention aims at realizing the ground precipitation intensity estimation method based on phased array radar echo and geographic factors, which comprises the following steps: s1, collecting phased array radar echo data, radar running state data, ground automatic station observation data and geographic factor data, inputting the radar running s