A long-term probabilistic ensemble prediction method for enteromorpha green tide

Abstract

The application discloses a kind of Enteromorpha prolifera green tide medium and long-term probabilistic ensemble prediction method, belong to Enteromorpha prolifera green tide prediction field.This method includes the following steps: a, based on historical monitoring data, the longitude, latitude and initial occurrence time of initial occurrence position of Enteromorpha prolifera green tide are obtained, potential throwing area and initial window period are set;Construct the time-space probability density function of throwing area, and generate the initial field set scheme of Enteromorpha prolifera green tide medium and long-term drift prediction;B, obtain a plurality of groups of medium and long-term marine environment prediction driving field data, and construct to form multiple driving field set scheme;C, the initial field set scheme and multiple driving field set scheme are substituted into Lagrange drift prediction model, and cycle simulation is carried out, the drift trajectory of each Enteromorpha prolifera green tide particle under all set schemes is obtained, and the medium and long-term ensemble prediction result of Enteromorpha prolifera green tide is generated.The application can carry out Enteromorpha prolifera green tide medium and long-term prediction, and significantly improve the accuracy and reliability of prediction, provide scientific basis for disaster prevention and mitigation decision-making.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of Enteromorpha prolifera green tide forecasting, in particular to an Enteromorpha prolifera green tide medium and long term probabilistic ensemble forecasting method. BACKGROUND

[0002] Large-scale outbreak of Enteromorpha prolifera green tide will cause serious impact on the marine ecological environment, coastal tourism and aquaculture of the coastal area. Under the background of ecological management of Enteromorpha prolifera green tide outbreak year after year, it is very important to carry out the medium and long term prediction of Enteromorpha prolifera green tide before the outbreak of Enteromorpha prolifera every year, and to predict the drift path of Enteromorpha prolifera in advance.

[0003] At present, for the drift prediction of Enteromorpha prolifera green tide, due to the prediction accuracy of wind and flow field data, most of them only carry out short-term drift prediction of 3-7 days, and the medium and long term prediction of more than 10 days is in the blank field. For the drift prediction of Enteromorpha prolifera green tide, the main method is to rely on the technical route of “satellite monitoring-numerical simulation”, that is, to determine the initial position by satellite image, and then to drive the water power model for forward simulation. However, if the medium and long term prediction is carried out before the Enteromorpha prolifera green tide is monitored in the same year, there are two big bottlenecks in this method: first, the uncertainty of initial condition; before the Enteromorpha prolifera is monitored, the initial discovery position and the initial discovery time of the Enteromorpha prolifera green tide in the same year cannot be confirmed. Second, the inaccuracy of driving field prediction; on the medium and long term time scale, the accuracy of atmospheric wind field and ocean current field prediction decreases sharply with the increase of prediction length. The existing method usually depends on a single prediction product, and cannot quantify the great influence of driving field error on the drift prediction result, resulting in low prediction reliability. SUMMARY

[0004] Based on the above technical problems, the present application provides an Enteromorpha prolifera green tide medium and long term probabilistic ensemble forecasting method.

[0005] The technical solution adopted by the present application is:

[0006] An Enteromorpha prolifera green tide medium and long term probabilistic ensemble forecasting method, comprising the following steps:

[0007] a. Establishing an initial field set;

[0008] Based on the historical monitoring data, the longitude, latitude and initial occurrence time of the initial occurrence position of the Enteromorpha prolifera green tide are obtained, the potential release area and the initial window period are set, the spatio-temporal probability density function of the release area is constructed, and the initial field set scheme for the medium and long term drift prediction of the Enteromorpha prolifera green tide is generated;

[0009] b. Constructing a multi-driving field set;

[0010] Obtain multiple sets of medium and long term marine environment driving field data covering the forecast area in space and time, which includes forecast data of sea surface wind field and flow field from different sources, and reanalysis data; use the above-mentioned medium and long term marine environment driving field data to construct a multi-driving field ensemble scheme for the medium and long term drift prediction of green tide;

[0011] c. generating an ensemble prediction result;

[0012] Substitute the initial field ensemble scheme obtained in step a and the multi-driving field ensemble scheme obtained in step b into the Lagrangian drift prediction model, perform cyclic simulation, obtain the drift trajectory of each green tide particle under all ensemble schemes, and generate a medium and long term ensemble prediction result of green tide.

[0013] The beneficial technical effects of the present application are as follows:

[0014] The present application proposes a medium and long term probabilistic ensemble prediction method for green tide, which generates an initial field ensemble by constructing a spatiotemporal probability density function, constructs a multi-driving field ensemble, and finally outputs a probabilistic prediction product by coupling simulation of the initial field ensemble and the multi-driving field ensemble. The present application can perform medium and long term drift prediction of green tide, and significantly improve the accuracy and reliability of the prediction, providing a scientific basis for disaster prevention and mitigation decision-making.

[0015] Specifically, the prediction method of the present application has the following advantages in view of the medium and long term uncertainty of the initial field and the driving field:

[0016] 1. Scientific quantification of uncertainty: by constructing the "double ensemble" of initial field ensemble and multi-driving field ensemble, the two main sources of uncertainty in prediction (initial condition, boundary condition) are probabilized for the first time, making the prediction from "deterministic" to "probabilistic", which is more in line with scientific laws.

[0017] 2. Improve the reliability of the prediction: the initial field ensemble, by introducing the spatiotemporal probability density function, breaks through the limitations of traditional fixed position release, more truly reflects the natural randomness of the occurrence and development of green tide from the source, significantly improves the rationality and accuracy of the ensemble prediction of green tide drift path; the multi-driving field super ensemble, by introducing driving fields of different models, different physical processes and climate states, effectively smooths the systematic bias and random error of a single model, and its ensemble effect is usually better than any single model, greatly improving the stability and reliability of medium and long term prediction.

[0018] 3. Provide decision support information: the output probabilistic products (such as impact probability, scenario clustering) not only tell "what is most likely to happen", but also tell "how likely it is to happen" and "what other possibilities there are", with rich information dimensions, greatly improving the decision support value of the prediction product. BRIEF DESCRIPTION OF DRAWINGS

[0019] Figure 1 A flowchart of the long-term probabilistic ensemble prediction method of Enteromorpha green tide of the present application;

[0020] Figure 2 A spatial probability density map of the initial discovery position of Enteromorpha green tide in a specific application example of the present application;

[0021] Figure 3 A spatial probability distribution prediction map of Enteromorpha green tide particle points in a specific application example of the present application. DETAILED DESCRIPTION

[0022] The present application provides a long-term probabilistic ensemble prediction method of Enteromorpha green tide. The method comprises the following steps: obtaining the longitude, latitude and initial occurrence time of the initial occurrence position of Enteromorpha green tide based on historical Enteromorpha monitoring data, and setting a potential release area and an initial window period; wherein the monitoring data comprises ship monitoring data, satellite or unmanned aerial vehicle remote sensing interpretation data; constructing a spatiotemporal probability density function of the release area, and designing an initial field ensemble scheme for long-term drift prediction of Enteromorpha green tide; obtaining multiple sets of environmental driving field data, and designing different driving field ensemble schemes; wherein the driving field data comprises multiple source sea surface wind field, flow field prediction data and reanalysis data covering the drift range of Enteromorpha green tide in time and space; substituting the multiple sets of initial field ensemble schemes and driving field ensemble schemes into a Lagrangian drift prediction model, predicting the drift trajectory of each Enteromorpha particle under all ensemble schemes, and calculating and producing probabilistic long-term drift prediction products. The present application quantifies the uncertainty of the initial time and place of Enteromorpha green tide occurrence and the inaccuracy of driving field prediction by constructing an ensemble containing multiple possible initial conditions (position and time) and an ensemble containing multiple environmental driving fields, so as to provide more accurate long-term drift prediction information of Enteromorpha green tide.

[0023] The present application will be further described below in conjunction with the drawings and specific embodiments.

[0024] As shown in the drawings, Figure 1 A long-term probabilistic ensemble prediction method of Enteromorpha green tide, comprising the following steps:

[0025] (1) Data source preparation:

[0026] Monitoring data: collect historical Enteromorpha outbreak initial data (usually from early April to mid-to-late May), ship monitoring data, unmanned aerial vehicle monitoring data, satellite data with high spatiotemporal resolution and suitable for floating vegetation detection.

[0027] Driving field data: obtain driving field data of different prediction centers and different numerical models. The driving field data comprises sea surface wind field, flow field prediction data and historical reanalysis data covering the drift range of Enteromorpha green tide in time and space.

[0028] (2) Establish the initial field set:

[0029] Based on historical monitoring data, extract the geographical location (latitude and longitude coordinates) and date of the first discovery of Enteromorpha each year.

[0030] (21) Construct the spatial probability function ;

[0031] Superimpose the "first discovery location" of all years on a map, and the spatial distribution will naturally concentrate in a specific sea area. Define this concentrated area as a regular potential release area, and divide it into a fine and regular spatial grid (for example, each grid cell is 0.1° x 0.1°).

[0032] For each grid point in the grid , calculate the probability of the occurrence of past first discovery locations within the range of the grid point. Calculation method: spatial probability = (the number of historical locations falling within the range of the grid point ) / (the total number of Enteromorpha locations in all years). In this way, the grid points corresponding to the historical high-frequency occurrence area will obtain high probability values, and the low-frequency area will obtain low probability values, thereby obtaining the spatial probability density distribution of the initial location of Enteromorpha.

[0033] Spatial probability function Calculation formula:

[0034] ;

[0035] Where, is the total number of times the Enteromorpha location falls within the th grid; is the total number of Enteromorpha locations in all years.

[0036] (22) Construct the time probability function ;

[0037] Time probability function , the calculation formula is as follows:

[0038] ;

[0039] Where is the total number of times the first discovery of Enteromorpha occurs in this time period, is the number of years of statistics.

[0040] (23) Construct the spatiotemporal probability density function ;

[0041] The uncertainty of initial field comes from both space and time. Therefore, a joint spatio-temporal probability density function is defined to represent the probability of initial occurrence of Enteromorpha prolifera under the joint definition of "spatial grid " and "time period ":

[0042]

[0043] where is the spatial probability function, is the temporal probability function, is the grid position, defined within the delineated "potential release area". is the time period, defined within the initial window of Enteromorpha prolifera occurrence.

[0044] The meaning of this function is: the probability of initial occurrence of Enteromorpha prolifera at position , time period is equal to the spatial probability of its occurrence at position , multiplied by the temporal probability of its occurrence at time period , i.e. the initial distribution of space and time is independent of each other.

[0045] (24) Generate a set of initial fields based on the probability distribution;

[0046] Convert the spatio-temporal probability density function obtained in step (23) into a set of specific initial conditions that can be used by the particle tracking model. The core logic is to allocate simulation particles in proportion according to the probability size, thereby generating a large number of different initial scenarios to cover the uncertainty of the initial occurrence of Enteromorpha prolifera. The entire process follows the following steps:

[0047] (241) Disperse the large Enteromorpha prolifera bloom into a certain number of Enteromorpha particles, and set the total number of particles as Z;

[0048] First, it is necessary to determine how many "Enteromorpha particles" are used to represent the total biomass in the entire ensemble prediction. This total number is denoted as Z (for example, Z = 1,000,000). This number is a total quantity, which is the total number of particles in all spatio-temporal units , which will be allocated in all release of all ensemble members.

[0049] Secondly, calculate the theoretical number of particles for each spatio-temporal unit:

[0050] A large number of "spatio-temporal units" are defined by the spatial grid and the time period . Each unit corresponds to an initial scenario: at a certain time period , at a certain position​ Seaweed began to appear.

[0051] For each spacetime unit According to their joint probability Calculate the theoretically required number of particles.

[0052] The calculation formula is: theoretical number of particles ;

[0053] For example: Suppose a grid located at the center of a radiation sandbar. ,That =5% (meaning there is a 5% historical probability that this was discovered here for the first time); a certain period of time (e.g., the third week of April) =20% (meaning that historically, 20% of the years saw the first discovery in that week). What is the joint probability of this spatiotemporal unit? =5% × 20% = 1%. If the total number of particles Z = 1,000,000, then the theoretical number of particles in this unit is... =1,000,000 × 1% = 10,000.

[0054] Then, generate members:

[0055] Each spatiotemporal unit that passes the above screening Each of these will generate an independent member.

[0056] Each member contains three key attributes:

[0057] Drop location: The spatial grid corresponding to this element. Uniformly distributed within.

[0058] Release time: The time period corresponding to this unit The time series is divided into 1-hour intervals.

[0059] Number of particles released: The final number of particles in this unit during this time period after probability calculation. (such as 10,000 in the example above).

[0060] Each of these members represents one possibility for the initial occurrence of seaweed.

[0061] This leads to the formation of the initial field set:

[0062] By combining all the generated members, the final initial field set is formed.

[0063] The total number of members in this set, M = i × j, is equal to the total number of all spatiotemporal units (the number of combinations of spatial points i and time periods j).

[0064] Finally, this set of M members, as well as the specific attributes of each member (position, time, particle number), is passed to the Lagrangian particle tracking model for subsequent parallel simulation.

[0065] The essence of this process is to discretize the continuous, theoretical probability density distribution into a set of finite, well-defined initial conditions with clear physical meaning. Instead of giving a single "most likely" answer, this set scientifically and quantitatively "spreads a net" to include all initial scenarios with a certain probability of occurrence, laying the most solid foundation for subsequent probabilistic prediction. This method greatly enhances the ability of the prediction system to describe the uncertainty of initial conditions.

[0066] (3) Multi-driving field set construction:

[0067] Obtain N sets of medium and long-term marine environmental driving field data from different domestic and foreign prediction centers, which at least include sea surface wind field, flow field prediction data and reanalysis data covering the range of green tide drift; the N sets of driving field data and reanalysis data together constitute a driving field set, which is used to represent the uncertainty of medium and long-term environmental prediction. The driving field set can be selected from:

[0068] (31) authoritative weather prediction model products and ocean dynamic prediction model products released by different domestic and foreign operational prediction centers.

[0069] Weather prediction models include global high-resolution weather prediction models released by the European Center for Medium-Range Weather Forecasts (ECMWF), the Global Forecast System (GFS) of the National Oceanic and Atmospheric Administration (NOAA) of the United States, and regional meteorological models such as WRF, which provide key meteorological driving factors such as sea surface wind field for particle tracking; ocean dynamic prediction models include global ocean prediction models provided by the Copernicus Marine Environment Monitoring Service (CMEMS) in Europe, the HYCOM global hybrid coordinate ocean model in the United States, and high-resolution regional ocean models (such as ROMS, FVCOM) for China's coastal waters, etc. The sea current field data output by these models constitutes the ocean dynamic basis for green tide drift and diffusion simulation.

[0070] In order to reduce the influence of systematic bias of the driving field data (sea surface wind field, flow field) on the stability of medium and long-term prediction, a general systematic bias correction method can be further used to process the original prediction field to improve its reliability as a model driving field.

[0071] Specifically, the systematic bias is calculated by the following method:

[0072] The systematic bias is calculated by comparing the driving field data from the numerical prediction model with the climatological mean field of the same period. The driving field data is corrected by the method of corrected prediction field = original prediction field - systematic bias, and the corrected data is taken as an independent member into the driving field set.

[0073] (32) Multiple prediction products obtained by using different physical process parameterization schemes for a single numerical model. In addition to integrating model products from multiple different prediction centers, a set of different prediction members can be generated based on a certain numerical model by purposefully changing its internal key physical process parameterization schemes to intuitively reflect the prediction risk introduced by the lack of understanding of physical processes. Taking the mesoscale weather prediction model WRF as an example, multiple WRF model instances can be run in parallel by replacing and combining different atmospheric boundary layer turbulence exchange schemes, cumulus parameterization schemes, and near-surface layer parameterization schemes, thereby obtaining a set of sea surface wind field prediction sequences that are reasonably dispersed due to different physical process descriptions based on the same model framework. Taking the ocean model FVCOM as an example, multiple FVCOM model instances can be run in parallel by replacing and combining different vertical mixing process parameterization schemes, horizontal diffusion and mixing parameterization schemes, bottom boundary layer friction and lateral open boundary condition parameterization schemes, thereby obtaining a set of reasonably dispersed sea current field prediction sequences that are consistent in physical framework but different in physical process description. Similarly, in order to reduce the impact of the systematic bias of the driving field data (sea surface wind field, current field) on the stability of the medium and long-term prediction, the prediction results are corrected for systematic bias, as described above.

[0074] (33) Climatological mean field data calculated based on historical reanalysis data; specifically, all data of the same month (or the same day) of the same year in the reference period are arithmetically averaged at each spatial grid point, thereby obtaining a set of climatological fields that can represent the long-term average state of the element.

[0075] Select N driving field products from the above product categories as the final driving field set members, generally N > 3.

[0076] (4) Coupling simulation and probabilistic prediction product generation:

[0077] The M members in the initial field set and the N driving fields in the driving field set are substituted into the Lagrangian drift prediction model for cyclic simulation, i.e., M x N times of forward numerical simulation, to generate a set of prediction results. The drift paths of all particle points are integrated and analyzed to calculate the occurrence probability of Enteromorpha in different sea areas and at different time points, and to generate probabilistic medium and long-term prediction products and spatial probability distribution prediction maps.

[0078] The drift sweeping sea range of Enteromorpha particles is set as a regular grid (for example, each grid size is 0.1°*0.1°). For each simulation (i.e. each of M*N times), the longitude and latitude coordinates of all Enteromorpha particles at the target time T are extracted. All particle coordinates of M*N times of simulation at time T are superimposed on the regular grid of the above step. Each grid is traversed , the total number of particles falling into the grid is counted , the spatial probability is calculated :

[0079] ;

[0080] In the formula, is the number of particles falling into the grid, N is the number of driving field sets, Z is the total number of particles in the initial field set, is the number of hours in the time period (every 1 hour is thrown); the number of samples falling into a certain grid divided by the total number of samples, that is, the probability of the occurrence of Enteromorpha in the grid.

[0081] In summary, the present application adopts a dynamic throwing mechanism, obtains the historical initial occurrence position and time of Enteromorpha green tide based on historical monitoring data of Enteromorpha green tide, constructs a spatiotemporal probability density function describing the initial occurrence regularity of Enteromorpha by statistical method, combines a multi-driving field coupling strategy to form a driving field set from model products from different forecast centers, model products of different physical processes, and climatological average field data, reduce the inaccuracy of medium and long-term prediction, and finally generate a probabilistic prediction product, integrate and analyze the ensemble prediction results, calculate the occurrence probability of Enteromorpha in different sea areas and different time points, and generate a probabilistic medium and long-term prediction product.

[0082] The present application will be further described below in combination with specific application examples.

[0083] 1) Data acquisition:

[0084] The HY-1C / D / E, GF-1 / 6, MODIS satellite data and ship monitoring data of Enteromorpha outbreak in the early stage from 2008 to 2025 are counted to obtain the longitude and latitude position of the first discovered Enteromorpha and the corresponding time.

[0085] The wind field, flow field prediction products of global meteorological models such as ECMWF and CFS and global current models such as CMEMS and FVCOM, and HYCOM reanalysis flow field data and ERA5 reanalysis wind field data are obtained.

[0086] 2) Initial field set:

[0087] The initial position of the aggregation area is designated as a regular potential release area by counting the first discovery of Enteromorpha prolifera in the past 18 years. The initial period window (April 10 to May 30) is designated by counting the first discovery date range in all years.

[0088] In the potential release area, a fine grid (0.125° x 0.125°) is created, with a total of 221 grids. The vector position of the first observation of Enteromorpha prolifera in the past 18 years is superimposed on the grid, and the probability of the occurrence of Enteromorpha prolifera in each grid is calculated to obtain the spatial probability function . Figure 2 The spatial probability density map of the initial discovery position of Enteromorpha prolifera is shown.

[0089] The initial period window (April 10 to May 30) is divided into 3 time periods (each time period is 1 month long), which are April 10 to May 10, April 20 to May 20, and April 30 to May 30. The number of times the first discovery date appears in the above 3 time periods is counted to obtain the time probability density function .

[0090] Through statistical analysis, we obtain

[0091] = 20% in the first time period, = 20% in the second time period, and

[0092] = 20% in the third time period. = 60% in the first time period, = 60% in the second time period, and

[0093] = 20% in the third time period. = 20% in the first time period, = 20% in the second time period, and

[0094] = 20% in the third time period. According to the spatial probability function and the time probability function, a joint probability density function is constructed.

[0095] ;

[0096] For example, in the first time period, the 35th grid has = 20%, = 50%, and the joint probability density function = 50% x 20% = 10%.

[0097] The large Enteromorpha prolifera green tide was discretized into particle points, and the total number of particles released per hour was given as 3000. According to the spatiotemporal probability distribution density, the number of particles released per time period in each grid was determined. For example, during the first time period (April 10 to May 10), 3000 particles were released every 1 hour in the 35th grid. =3000 0.1=300 particle points, and the number of particles released in the jth time period in the ith grid was determined in turn. The model was simulated by cyclic substitution.

[0098] According to this, 221 grids (i = 221) were set in the potential release area, and 3 time periods (j = 3) were set in the initial period window, generating an initial field set of M = i x j = 663 spatiotemporal members.

[0099] 3) Driving field set:

[0100] N1 = 3 sets of wind field driving field data were selected, including: GFS, ECMWF wind field forecast products after climate bias correction, and climate wind field products calculated from ERA5 reanalysis data.

[0101] N2 = 3 sets of flow field driving field data were selected, including: CMEMS, HYCOM flow field forecast products after climate bias correction, and climate flow field products calculated from HYCOM reanalysis data.

[0102] Thus, a driving field set containing N = N1 x N2 = 9 members was constructed.

[0103] 4) Coupling simulation and forecast product production;

[0104] The Lagrangian particle tracking model was used as the Enteromorpha prolifera drift prediction model. A large number of virtual particles were released in each grid according to the spatiotemporal probability function, and the Enteromorpha prolifera particles would drift on the sea surface under the action of sea surface wind and sea surface flow. For a certain Enteromorpha prolifera particle, the distance it drifted in a unit of time The calculation method is as follows:

[0105] ;

[0106] wherein is the drag coefficient of 10 m height sea surface wind on Enteromorpha prolifera, which is taken as 0.03 according to the field experiment results; is the drag coefficient of sea surface flow on Enteromorpha prolifera, which is taken as 1 according to the field experiment results; is the 10 m height wind speed vector, is the sea surface flow velocity vector, is the model calculation time step.

[0107] Set 9 driving field combination schemes with initial values of M initial field set members as models, run 663x9=5967 times of forward drift simulation, and carry out drift prediction of Enteromorpha prolifera green tide for 120 days (medium and long term) in the future.

[0108] According to the drift prediction results, a spatial probability distribution prediction map at a specific time (July 1, 2025) is made:

[0109] The drift sweeping range of Enteromorpha prolifera is divided into a regular Euler grid (0.1°x0.1°), and integrated analysis is carried out on all 5967 simulation results: all MxN simulations are superimposed on the target time (July 1, 2025 12:00:00) longitude and latitude coordinates of the Euler grid, the number of particles falling into each grid is counted, and the spatial probability at the time is obtained by dividing the total number of samples.

[0110] Figure 3 The spatial probability distribution prediction map of Enteromorpha prolifera particle points on July 1, 2025 (reported on April 10, 2025). By comparing with the satellite remote sensing interpretation results (black line range in the figure) of the same day, it can be seen that the spatial probability distribution prediction results can accurately give the drift position information of Enteromorpha prolifera particles.

[0111] Although embodiments of the present application have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the present application, and the scope of the present application is defined by the appended claims and their equivalents.

Claims

1. A long-term probabilistic ensemble forecasting method for *Ulva prolifera* green tides, characterized in that... Includes the following steps: a. Establish the initial field set; Based on historical monitoring data, the longitude, latitude, and initial occurrence time of the initial appearance of the Ulva prolifera green tide are obtained, and potential release areas and initial window periods are set. The spatiotemporal probability density function of the release area is constructed, and an initial field set scheme for predicting the long-term drift of the Ulva prolifera green tide is generated. b. Construct a set of multiple driving fields; Multiple sets of medium- and long-term marine environmental driving field data covering the forecast area in time and space were acquired. These data included forecast data of sea surface wind field and current field from different sources, as well as reanalysis data. Using the above-mentioned medium- and long-term marine environmental driving field data, a multi-driving field ensemble scheme for predicting the medium- and long-term drift of Ulva prolifera green tide was constructed. c. Generate ensemble forecast results; Substitute the initial field ensemble scheme obtained in step a and the multi-driving field ensemble scheme obtained in step b into the Lagrange drift prediction model, perform cyclic simulation, obtain the drift trajectory of each Ulva green tide particle under all ensemble schemes, and generate the long-term ensemble forecast results of Ulva green tide. In step a, constructing the spatiotemporal probability density function of the release area includes the following steps: a1. Constructing the spatial probability function ; By overlaying the locations of the first discovery of green tides of seaweed over the years onto a single map, their spatial distribution will cluster in one area. This area will be designated as a potential release zone and divided into a fine, regular spatial grid. For each grid point in the grid Calculate the probability of the location being the first previously discovered location occurring within this grid point range; spatial probability function. The calculation formula is as follows: ； in, The position of the seaweed falls into the first Total number of times within each grid, This represents the total number of locations of *Ulva prolifera* across all years. a2. Constructing the time probability function ; The dates of the first discovery of *Ulva prolifera* green tides over the years were statistically analyzed, and the initial window of the *Ulva prolifera* green tide occurrence was divided into j time periods with a certain temporal resolution; for each time period... Calculate the frequency with which the first discovery date falls within this time period; time probability function. The calculation formula is as follows: ； in, This represents the total number of times the first discovery of *Ulva prolifera* occurred within this time period. The number of years for which statistics are collected; a3. Construct the spatiotemporal probability density function ; The uncertainty of the initial field arises simultaneously from space and time; the spatiotemporal probability density function The calculation formula is as follows: ； in, It is a spatial probability function. It is a time probability function. For grid positions, For a time period.

2. The long-term probabilistic ensemble forecasting method for *Ulva prolifera* green tides according to claim 1, characterized in that, In step a: the historical monitoring data includes ship monitoring data and satellite or UAV remote sensing interpretation data.

3. The method for long-term probabilistic ensemble forecasting of *Ulva prolifera* green tides according to claim 1, characterized in that, Step a, generating the initial field set scheme for predicting the long-term drift of *Ulva prolifera* during green tides includes the following steps: a41. Discretize the green algae of *Ulva prolifera* into a certain number of *Ulva prolifera* particles, and set the total number of particles Z; a42. Calculate the theoretical number of particles in each spacetime unit; By spatial grid and time period They collectively define numerous spatiotemporal units, each corresponding to an initial scenario: the appearance of seaweed at a certain time and location. For each spatiotemporal unit According to the spatiotemporal probability density function Calculate the theoretical number of particles that should be allocated; the calculation formula is as follows: Theoretical number of particles ; a43. Form the initial field set; Each spacetime unit And based on the theoretical number of particles calculated according to step a42, a member is formed; each member represents a possibility of the initial occurrence of seaweed, including three key attributes: release location, release time, and number of released particles; all members are summarized to form the initial field set.

4. The long-term probabilistic ensemble forecasting method for *Ulva prolifera* green tides according to claim 1, characterized in that, In step b, the forecast data of sea surface wind field and flow field from different sources include: weather forecast model products and ocean dynamic forecast model products released by different domestic and foreign operational forecast centers, as well as multiple forecast products obtained by applying different physical process parameterization schemes to a single numerical model.

5. The long-term probabilistic ensemble forecasting method for *Ulva prolifera* green tides according to claim 4, characterized in that... Step b also includes the following steps: bias correction of forecast data of sea surface wind field and flow field from different sources to obtain corrected forecast data; climatological averaging of reanalysis data of sea surface wind field and flow field to obtain climatological average data; and constructing a multi-driving field ensemble scheme for predicting long-term drift of seaweed green tide by combining the corrected forecast data and climatological average data.

6. The long-term probabilistic ensemble forecasting method for *Ulva prolifera* green tides according to claim 1, characterized in that, Step c also includes the following steps: The drift sweep range of the seaweed particles is set as a regular spatial grid; for each simulation, the latitude and longitude coordinates of all seaweed particles at the target time are extracted; the coordinates of all seaweed particles at the target time from all simulations are superimposed onto the spatial grid; each grid is traversed. Count the total number of particles that fall into the grid. Calculate spatial probability : ； In the formula, The number of particles falling into this grid is N, where N is the number of driving field sets and Z is the total number of particles in the initial field set. Time period The number of hours within the time frame is determined by releasing the particles every hour; the drift paths of all particles are integrated and analyzed to calculate the probability of Ulva prolifera appearing in different sea areas and at different times, generating probabilistic medium- and long-term forecast products and creating spatial probability distribution prediction maps.

Images