JP2021148753A - Precipitation prediction device and precipitation prediction method - Google Patents

Abstract

To provide a precipitation prediction device and a precipitation prediction method capable of precisely predicting the maximum value of an integral rainfall amount.SOLUTION: A precipitation prediction device 10 includes: a nowcast prediction part 12 for acquiring nowcast prediction data for predicting distribution of precipitation after a short time for each lattice; a numerical value prediction part 13 for dividing an observation area in a lattice shape and acquiring numerical value prediction data for predicting a future precipitation situation for each lattice; a first filter application part 14 for setting one lattice as a reference lattice for all the lattices of an observation area divided into plural lattices of nowcast prediction data, and replacing nowcast prediction data by first filter application data; a second filter application part 15 for setting one lattice as a reference lattice for all the lattices of an observation area divided into plural lattices of numerical value prediction data and replacing numerical value prediction data by second filter application data; and a composition part 16 for compositing observation data, the first filter application data, and the second filter application data by using a composition coefficient.SELECTED DRAWING: Figure 1

Description

The present invention relates to a precipitation prediction device and a precipitation prediction method for predicting precipitation.

Conventionally, a rainfall prediction method for improving the prediction accuracy of the instantaneous precipitation distribution has been disclosed (see Patent Document 1). The invention described in Patent Document 1 uses numerical prediction data for predicting the future precipitation situation for each grid by dividing the observation area into a grid pattern and calculating the time change of the precipitation situation using the observation data, and observation data. The movement of the past precipitation area and the distribution of the current precipitation are obtained, and the nowcast prediction data that predicts the distribution of the precipitation after a short time for each grid is synthesized with the optimum synthesis coefficient.

One of the points of the invention described in Patent Document 1 is to calculate the prediction accuracy when the positional deviation is allowed for each prediction, and use the composite coefficient calculated from the ratio of the prediction data to the numerical prediction data and now. The point is that the prediction accuracy of precipitation intensity has been improved by synthesizing cast prediction data.

Japanese Unexamined Patent Publication No. 2018-31682

Kenji Kobayashi, Statistics Office, Observation Division, Japan Meteorological Agency, "Estimation of Probabilistic Rainfall and Reproduction Period", Weather Time Signal Vol. 73, No. 4-5, p51-p72, 2006

For torrential rains associated with linear precipitation zones caused by a large number of cumulonimbus clouds, it is a disaster prediction perspective to accurately predict the maximum value of rainfall accumulated in the time direction rather than the instantaneous precipitation distribution. Is important. The rainfall prediction method of Patent Document 1 considers the predicted positional deviation in the calculation of the synthesis coefficient, but since the numerical prediction data and the nowcast prediction data are combined without correcting the positional deviation, the predicted cumulative rainfall is calculated. In some cases, the maximum value of was too small.

The present invention provides a precipitation prediction device and a precipitation prediction method capable of predicting the maximum value of accumulated rainfall with high accuracy for torrential rain caused by a large number of cumulonimbus clouds as compared with the prior art. With the goal.

The precipitation prediction device according to the present invention is
At least the observation data acquisition unit that acquires the observation data of precipitation,
The observation area is divided into grids, and the movements of the past precipitation area and the distribution of current precipitation are obtained using the observation data acquired from the observation data acquisition unit, and the distribution of precipitation after a short period of time is predicted for each grid. Nowcast prediction unit that acquires Nowcast prediction data,
The observation area is divided into grids, and using the observation data acquired from the observation data acquisition unit, the time change of the precipitation situation is calculated by a computer using the physical law to predict the future precipitation situation for each grid. Numerical forecasting unit that acquires forecast data and
For all the grids in the observation region divided into a plurality of grids of the nowcast prediction data, one grid is used as a reference grid, and the maximum rainfall grid within the range of the predetermined first filter region including the reference grid. The first filter application unit that replaces the nowcast prediction data with the first filter application data by using
For all the grids in the observation region divided into a plurality of grids of the numerical prediction data, one grid is used as a reference grid, and the maximum rainfall grid within the range of the predetermined second filter region including the reference grid is used. A second filter application unit that replaces the numerical prediction data with the second filter application data by using
A synthesis unit that synthesizes the observation data, the first filter application data, and the second filter application data using the synthesis coefficient, and
It is characterized by having.

The precipitation prediction device according to the present invention is
The first filter region is smaller than the second filter region.

The precipitation prediction device according to the present invention is
In the latest prediction, the composite coefficient increases the proportion of the first filter applied data, then decreases the proportion of the first filter applied data and increases the proportion of the second filter applied data as the prediction time becomes longer. It is characterized by doing.

The precipitation prediction device according to the present invention is
It is further provided with a reproduction period calculation unit for calculating a reproduction period indicating how much the predicted cumulative rainfall is stochastically occurring at the observation point.

The precipitation prediction method according to the present invention is
At least the steps to get the precipitation observation data,
The observation area is divided into grids, and the movements of the past precipitation area and the current precipitation distribution are obtained using the observation data acquired from the observation data acquisition unit, and the precipitation distribution after a short period of time is predicted for each grid. Steps to get the Nowcast prediction data and
The observation area is divided into grids, and using the observation data acquired from the observation data acquisition unit, the time change of the precipitation situation is calculated by a computer using the physical law, and the future precipitation situation is predicted for each grid. Steps to get forecast data and
For all the grids in the observation region divided into a plurality of grids of the nowcast prediction data, one grid is used as a reference grid, and the maximum rainfall grid within the range of the predetermined first filter region including the reference grid. To replace the nowcast prediction data with the first filter application data using
For all the grids in the observation region divided into a plurality of grids of the numerical prediction data, one grid is used as a reference grid, and the maximum rainfall grid within the range of the predetermined second filter region including the reference grid is used. Using the step of replacing the numerical prediction data with the second filter application data,
A step of synthesizing the observation data, the first filter application data, and the second filter application data using the synthesis coefficient, and
It is characterized by having.

The precipitation prediction method according to the present invention is
The first filter region is smaller than the second filter region.

The precipitation prediction method according to the present invention is
The synthesis coefficient is characterized in that the ratio of the first filter applied data is increased in the first short-time prediction, and then the ratio of the second filter applied data is increased as the prediction time becomes longer.

The precipitation prediction method according to the present invention is
It is characterized by further having a step of calculating a reproduction period indicating how much the predicted cumulative rainfall is stochastically occurring at the observation point.

According to such a precipitation prediction device and a precipitation prediction method, it is possible to predict the cumulative rainfall with high accuracy for torrential rain caused by a large number of cumulonimbus clouds.

The system block diagram of the precipitation prediction apparatus of this embodiment is shown. An example of the data before applying the filter and the data after applying the filter is shown. The ratio of the predicted correct answer rate by Nowcast prediction to the predicted correct answer rate by numerical prediction is shown. The figure which divided the result of comparison of the predicted rainfall and the observed rainfall is shown. The accuracy of the 1-hour integration amount of the numerical prediction with the threshold value of the present embodiment being 50 mm is shown. The flowchart of the precipitation prediction method of this embodiment is shown. The observation data of the accumulated rainfall from 1 hour before the start of the precipitation prediction used in the precipitation prediction method of the present embodiment is shown. Each data from the time when the precipitation prediction used in the precipitation prediction method of this embodiment is started to 1 hour later is shown. The data from 1 hour to 2 hours after the start of the precipitation prediction used in the precipitation prediction method of the present embodiment are shown. Each data from 1 hour before to 2 hours after the start of the precipitation prediction used in the precipitation prediction method of the present embodiment is shown. The reproduction period of the accumulated rainfall obtained by the precipitation prediction method of this embodiment is shown.

An embodiment according to the present invention will be described with reference to the drawings. The present invention can be applied not only to torrential rain but also to torrential snow caused by snow clouds in winter. Therefore, in the present invention, precipitation means not only rain but also snow, hail, hail and the like.

FIG. 1 shows a system block diagram of the precipitation prediction device 10 of the present embodiment.

The precipitation prediction device 10 of the present embodiment includes an observation data acquisition unit 11, a nowcast prediction unit 12, a numerical prediction unit 13, a first filter application unit 14, a second filter application unit 15, and a synthesis unit 16. , Equipped with. Further, the reproduction period calculation unit 17 may be provided.

The observation data acquisition unit 11 acquires observation data such as ground observations, meteorological satellites, and radars from meteorological institutions or space agencies around the world. Since observations are made at various places and times, observation data is acquired from each acquisition destination at predetermined time intervals. Some observation data are not available due to low accuracy due to human error or equipment failure, so these data are excluded. The acquired observation data is output to the numerical prediction unit 12, the nowcast prediction unit 13, and the synthesis unit 16.

The nowcast prediction unit 12 divides the observation area into a grid pattern, and uses the observation data acquired from the observation data acquisition unit 11 to use the movement of the past precipitation area and the distribution of the current precipitation, after a short period of time. Predict the distribution of precipitation for each grid.

The numerical weather prediction unit 13 is a portion that divides the observation region into a grid pattern and performs numerical weather prediction for each grid using the observation data acquired from the observation data acquisition unit 11. Numerical weather prediction predicts future precipitation by calculating the time change of precipitation with a computer using the laws of physics.

FIG. 2 shows an example of data before applying the filter and data after applying the filter. FIG. 2A shows the data before applying the filter, and FIG. 2B shows the data after applying the filter.

The filter used in the precipitation prediction device 10 of the present embodiment applies a filter based on a _{certain reference grid G 0} to the filter region L divided into a plurality of grids as shown in FIG. 2 (a). The previous rainfall data is replaced with the data after applying the filter as shown in FIG. 2 (b) by using _{the maximum rainfall P MAX} within the range of the filter region L with reference to the _{reference grid G 0.} This replacement is performed for all grids, as shown in FIG. 2 (c). In this way, by applying the filter even in the rainfall analysis in which heavy rain is locally present, it can be replaced by heavy rain over a wide area.

The filter region L shown in FIG. 2 is formed by a square La (km) × La (km) having a side of La (km), but it is not limited to a square and can be a figure having an area such as a circle. good. For example, in the case of the circle, the diameter and L _{R (miles),} it may be a filter region L = πL _R ^2/4 and (miles ^2). Further, in the present embodiment, the maximum rainfall _PMAX is used as an index to which the filter is applied, but other indexes may be used as long as the region of heavy rainfall can be expanded in consideration of the positional deviation. For example, the 99th percentile value of rainfall may be used.

The first filter application unit 14 applies the first filter to the nowcast prediction data obtained from the nowcast prediction unit 12. _{The first filter uses the rain nowcast prediction data of one reference grid G 0} for the data divided into a plurality of grids, and the first filter region L1 (km) × L1 (km) centered on the grid. The operation of replacing with the maximum value within the range of) is performed for all the grids, and the data is used as the first filter application data. By applying this first filter, even if Nowcast predicts a distribution in which heavy rain is locally present, it can be replaced with a distribution in which heavy rain is present over a wide range.

The second filter application unit 15 applies the second filter to the numerical prediction data obtained from the numerical prediction unit 13. The second filter applies the numerical prediction data of rain in one grid to the data divided into a plurality of grids within the range of the second filter region L2 (km) × L2 (km) centered on that grid. The operation of replacing with the maximum value of is performed for all the grids, and is used as the second filter application data. By applying this second filter, even if the numerical prediction predicts the distribution in which heavy rain is locally present, it can be replaced with the distribution in which heavy rain is present in a wide range.

The synthesizing unit 16 includes observation data acquired by the observation data acquisition unit 11 and first filter application data obtained by applying the first filter to the nowcast prediction data obtained from the nowcast prediction unit 12 by the first filter application unit 14. The numerical prediction data obtained from the numerical prediction unit 13 is combined with the second filter application data to which the second filter is applied by the second filter application unit 15.

FIG. 3 shows the ratio of the predicted correct answer rate by Nowcast prediction to the predicted correct answer rate by numerical prediction with respect to the predicted time.

As shown in FIG. 3, in the nowcast prediction, the correct answer rate is high when the prediction time is short, and the correct answer rate is low when the prediction time is long. In numerical prediction, the correct answer rate is low when the prediction time is short, and the correct answer rate is high when the prediction time is long. Therefore, it is preferable to obtain a composite coefficient that emphasizes nowcast prediction when the prediction time is short and emphasizes numerical prediction when the prediction time is long.

For example, consider the case of predicting the predicted cumulative rainfall PNh _{BLEND for N hours.} The observed data PPn ₀ h _{OBS of n 0} hours observed from n ₀ hours ago to the present is used. For the prediction from the present to n _{1 hour later, the n 1} hour first filter application data Pn _{1 h'NOWC} , in which the first filter is applied to the nowcast prediction data having a high accuracy rate of the short-time prediction, is used. For _{the prediction from n 1} hour to n ₁ + n ₂ hours later, the second filter is applied to the numerical prediction data with a high accuracy rate of the long-term prediction. n ₂ hours of the second filter application data Pn _{2 h'NWP} Use.

ただし、
PNh_BLENDは、Ｎ時間の予測積算雨量、
Pn₀h_OBSは、ｎ_０時間の観測データ、
Pn₁h’_NOWCは、ｎ_１時間積算雨量に第１フィルターを適用したデータ、
Pn₂h’_NWPは、ｎ_２時間積算雨量に第２フィルターを適用したデータ、
である。 By synthesizing these three data as shown in the following equation (1), the predicted cumulative rainfall is calculated.

However,
PNh _BLEND is the estimated cumulative rainfall for N hours,
Pn ₀ h _OBS is the observation data for _{n 0 hours,}
Pn _{1 h'NOWC} is the data obtained by applying the first filter to the accumulated rainfall for _{n 1 hour.}
Pn _{2 h'NWP} is the data obtained by applying the second filter to the accumulated rainfall for _{n 2 hours.}
Is.

ここで、
NTは予測開始時刻から予測終了時刻までの時間方向の分割数、
Wは合成係数やバイアス補正などの演算を行う重み関数、
Sは空間フィルターを適用する関数、
Pは積算雨量を示し、
Fは時刻nにおけるナウキャストと数値予測のW, S, Pから成る関数である。 (1) can be described more generally as follows in the calculation formula of the predicted cumulative rainfall.

here,
NT is the number of divisions in the time direction from the prediction start time to the prediction end time,
W is a weighting function that performs operations such as synthesis coefficient and bias correction.
S is a function that applies a spatial filter,
P indicates the accumulated rainfall,
F is a function consisting of Nowcast and numerical prediction W, S, P at time n.

この(3)式において、

とすると(1)式になる。 There are variations in the order in which W and S are applied to P, and as one form, an example in which the spatial filter S is applied to P and then the weight function W is applied is shown below.

In this equation (3),

Then, it becomes equation (1).

In this way, the first filter is applied in the appropriate first filter area L1 (km) × L1 (km), and the second filter is applied in the appropriate second filter area L2 (km) × L2 (km). Therefore, even if the predicted rainfall is misaligned, it is possible to predict the maximum value of the predicted integrated rainfall synthesized by the synthesis unit 16.

Next, a method of determining the filter region L of the present embodiment will be described.

The precipitation prediction device 10 of the present embodiment uses category verification as an example when determining the filter region L. The precipitation prediction device 10 calculates the predicted rainfall with a filter applied to various grids. Then, the predicted rainfall and the observed rainfall acquired by the observation data acquisition unit 11 are compared.

Here, the amount of rainfall for comparison may be set according to the target phenomenon. For example, when targeting torrential rain associated with a linear precipitation zone, it is preferable to set a threshold value of rainfall having a cumulative rainfall of about 20 mm to 50 mm per hour.

FIG. 4 shows a divided view of the results of comparison between the predicted rainfall and the observed rainfall.

The precipitation prediction device 10 of the present embodiment compares the predicted rainfall with the observed rainfall for various filter regions L. Then, the result of comparing the predicted rainfall and the observed rainfall is divided as shown in FIG.

When both the predicted rainfall and the observed rainfall in the filter region L are within the threshold range, it is determined to be suitable (A). When the predicted rainfall in the filter region L is within the threshold range and the observed rainfall is not within the threshold range, it is determined to be a missed swing (B). When the predicted rainfall in the filter region L is not in the threshold range and the observed rainfall is in the threshold range, it is determined to be overlooked (C). If both the predicted rainfall and the observed rainfall in the filter region L are not within the threshold range, it is determined as other (D).

Based on this division result, an index such as a supplement rate (POD), a missed swing rate (FAR), a threat score (CSI), or an equitable threat score (ETS) is calculated.

The supplement rate (POD) is A / (A + C), the missed swing rate (FAR) is B / (A + B), the threat score (CSI) is B / (A + B + C), and the equitable threat score (ETS) is (A-). Arand) / (A + B + C-Arand). However, Arand is {(A + B) · (A + C)} / (A + B + C + D).

FIG. 5 shows the accuracy of the 1-hour integration amount from 1 hour ahead to 2 hours ahead of the numerical prediction with the threshold value of the present embodiment set to 50 mm.

As shown in FIG. 5, if the second scale L2 of the second region used for the numerical prediction data is lengthened, the supplement rate increases. However, if the second scale L2 is lengthened, the missed swing rate (FAR) also increases. Therefore, the filter region L may be determined depending on which index is emphasized.

For example, if you want to emphasize the supplement rate (POD) and reduce the oversight (C), set a threshold value for the supplement rate (POD) in advance, satisfy the threshold value, and set L2 that maximizes the threat score (CSI). You can select it. In the present embodiment, in FIG. 5, the threshold value of the supplement rate (POD) is set to POD ≧ 0.7, and the second scale of the second region of the numerical prediction that maximizes the threat score (CSI) is L2 = 11 km. ..

The region is determined separately according to the prediction accuracy of the nowcast prediction data and the numerical prediction data. The misalignment of the prediction becomes large as the prediction time is long. Therefore, the prediction accuracy in a long prediction time is lowered.

Therefore, the misalignment of the nowcast prediction used in the short-term prediction is smaller than the misalignment of the numerical prediction used in the long-term prediction. That is, the first scale L1 of the first region used for the nowcast prediction data and the second scale L2 of the second region used for the numerical prediction data may have the relationship of the following equation (2).
L1 <L2 (5)

Therefore, in the case of the present embodiment, the first scale L1 of the first region used for the nowcast prediction data may be set smaller than the second scale L2 = 11 km of the second region used for the numerical prediction data.

The filter area L may be calculated from a large number of past statistical data. Further, the filter region L may be calculated from the prediction accuracy of several hours ago in near real time as described in Patent Document 1. Further, the filter area L may be calculated by combining these methods.

Next, an example of the precipitation prediction method of the precipitation prediction device 10 of the present embodiment will be described. In the present embodiment, this example is used to compare the actual 3-hour integrated precipitation observation data at the point marked with a star with the prediction data using the precipitation prediction device 10 of the present embodiment.

FIG. 6 shows a flowchart of the precipitation prediction method of the present embodiment.

First, in step 1, the precipitation prediction device 10 acquires the observation data by the observation data acquisition unit 11 (ST1).

FIG. 7 shows the observation data of the accumulated rainfall from one hour before the start of the precipitation prediction used in the precipitation prediction method of the present embodiment. In the figures shown in FIGS. 7 to 10 of the present embodiment, the horizontal axis indicates the east longitude and the vertical axis indicates the north latitude, and the dark-colored area indicates that the accumulated rainfall is large.

As for the observation data, the observation data acquisition unit 11 acquires the observation data from several hours before the start of the precipitation prediction. The observation data acquisition unit 11 of this example acquired the observation data of the accumulated rainfall from one hour before the start of the precipitation prediction.

Next, the predicted cumulative rainfall after several hours is calculated based on the observation data. As shown in FIG. 3, the nowcast prediction data cannot consider the occurrence, development, and weakness of rain clouds, so that the predicted correct answer rate after a short time is high and the predicted correct answer rate after a long time is low. In addition, although the numerical prediction data can predict the occurrence or development of rain clouds, the prediction correct answer rate after a short time is low and the predicted correct answer rate after a long time is high because a spin-up time is required. Therefore, it is preferable to use nowcast prediction data for prediction after a short time and synthetic prediction using numerical prediction data for prediction after a long time.

In this example, the nowcast prediction unit 12 predicted the amount of rainfall from the time when the precipitation prediction was started to one hour later. In addition, the numerical prediction unit 13 predicted the amount of rainfall from 1 hour to 2 hours later. The method of calculating the nowcast prediction data and the numerical prediction data may be a generally known method, for example, the method described in Patent Document 1.

Following step 1, in step 2, the nowcast prediction unit 12 calculates the nowcast prediction data from the observation data acquired in step 1 (ST2). Subsequently, in step 3, the first filter application unit 14 applies the first filter to the nowcast prediction data acquired in step 2 and calculates the first filter application data (ST3).

FIG. 8 shows each data from the time when the precipitation prediction used in the precipitation prediction method of the present embodiment is started to one hour later. FIG. 8A shows actual observation data of the accumulated rainfall from the time when the precipitation prediction is started to 1 hour later. FIG. 8B shows nowcast prediction data from the time when the precipitation prediction is started to one hour later. FIG. 8C shows the first filter application data calculated by applying the first filter from the time when the precipitation prediction is started to one hour later.

Compared with the actual observation data shown in FIG. 8 (a), the nowcast prediction data shown in FIG. 8 (b) is predicted so that the maximum value of accumulated rainfall is shifted to the east. As a result, the integrated rainfall of Nowcast is greatly underestimated at the point marked with a star where the maximum integrated rainfall is recorded in the observation.

Therefore, the first filter application unit 14 applies the first filter to the nowcast prediction data to calculate the first filter application data. Compared with the nowcast prediction data shown in FIG. 8 (b), the first filter application data shown in FIG. 8 (c) predicts a large amount of accumulated rainfall around the points marked with stars. In this way, by applying the first filter in consideration of the positional deviation, it is possible to prevent the maximum value of the integrated rainfall in the prediction data from being underestimated.

Subsequently, in step 4, the numerical prediction unit 13 calculates the numerical prediction data from the observation data acquired in step 1 (ST4). Subsequently, in step 5, the second filter application unit 15 applies the second filter to the numerical prediction data acquired in step 4, and calculates the second filter application data (ST5).

FIG. 9 shows each data from 1 hour to 2 hours after the start of the precipitation prediction used in the precipitation prediction method of the present embodiment. FIG. 9A shows actual observation data of the accumulated rainfall from 1 hour to 2 hours after the start of precipitation prediction. FIG. 9B shows numerical prediction data from 1 hour to 2 hours after the start of precipitation prediction. FIG. 9C shows the second filter application data calculated by applying the second filter from 1 hour to 2 hours after the start of precipitation prediction.

Compared with the actual observation data shown in FIG. 9A, the numerical prediction data shown in FIG. 9B is predicted so that the maximum value of accumulated rainfall is shifted to the north. As a result, the integrated rainfall of Nowcast is greatly underestimated at the point marked with a star where the maximum integrated rainfall is recorded in the observation.

Therefore, the second filter application unit 15 applies the second filter to the numerical prediction data and calculates the second filter application data. Compared with the numerical prediction data shown in FIG. 9B, the second filter application data shown in FIG. 9C predicts a large amount of accumulated rainfall around the point marked with a star. In this way, by applying the second filter in consideration of the positional deviation, it is possible to prevent the maximum value of the integrated rainfall in the prediction data from being underestimated.

Next, in step 6, the synthesis unit 15 calculates the composite data from the observation data acquired in step 1, the first filter application data acquired in step 3, and the second filter application data acquired in step 5. (ST6).

FIG. 10 shows each data from 1 hour before to 2 hours after the start of the precipitation prediction used in the precipitation prediction method of the present embodiment. FIG. 10A shows actual observation data of 3-hour cumulative rainfall from 1 hour before to 2 hours after the start of precipitation prediction. FIG. 10 (b) shows a conventional composite data of 3-hour integrated rainfall, which is a combination of the nowcast prediction data of FIG. 8 (b) and the numerical prediction data of FIG. 9 (b). FIG. 10 (c) shows the filter-applied composite data of the 3-hour integrated rainfall, which is a composite of the first filter-applied data of FIG. 8 (c) and the second filter-applied data of FIG. 9 (c).

As shown in FIG. 10A, in the actual observation data, the 3-hour cumulative rainfall at the observation point marked with a star indicates 150 mm or more. On the other hand, the conventional composite data of the 3-hour cumulative rainfall shown in FIG. 10 (b) without using the first filter and the second filter was determined to be 50 mm-100 mm, and the nowcast of FIG. 8 (b) was determined. Similar to the forecast data and the numerical forecast data of FIG. 9B, the accumulated rainfall is underestimated.

In the filter-applied composite data shown in FIG. 10 (c), the 3-hour cumulative rainfall at the observation point marked with a star is predicted to be 150 mm or more, and data close to the observation data shown in FIG. 10 (a) can be obtained. .. By synthesizing the first filter application data and the second filter application data in consideration of the positional deviation in this way, it is possible to prevent the prediction data of the accumulated rainfall from being underestimated.

Next, in step 7, the reproduction period of the predicted accumulated rainfall data is calculated (ST7).

FIG. 11 shows a reproduction period of the accumulated rainfall obtained by the precipitation prediction method of the present embodiment. FIG. 11A shows a reproduction period based on actual observation data. FIG. 11B shows a reproduction period based on conventional synthetic data. FIG. 11C shows a reproduction period based on the filtered composite data. In the figure shown in FIG. 11 of the present embodiment, the horizontal axis represents the east longitude and the vertical axis represents the north latitude, indicating that the darker the color, the longer the reproduction period of the accumulated rainfall.

The precipitation prediction device of the present embodiment can calculate a reproduction period indicating how many years the predicted cumulative rainfall is stochastically occurring at the observation point. In general, stochastic rainfall varies from region to region. For example, even if the amount of rainfall is the same, an area with a high frequency of heavy rainfall may be determined to be once every 50 years, and an area with a low frequency of heavy rainfall may be determined to be once every 100 years. In this way, it may be difficult to determine the degree of disaster risk from the cumulative rainfall alone. The precipitation prediction device of the present embodiment can also predict information close to the risk of disaster by calculating the reproduction period. The reproduction period can be calculated by the method described in Non-Patent Document 1.

As shown in FIG. 11A, the actual observation data determines that the reproduction period of the 3-hour cumulative rainfall at the observation point marked with a star is once every 100 years or more. On the other hand, during the reproduction period calculated using the conventional composite data shown in FIG. 11B without using the first filter and the second filter, the accumulated rainfall is underestimated, so that the observation point marked with a star is used. The reproduction period of the 3-hour cumulative rainfall is also short, and it is judged that it occurs once in less than 30 years.

The 3-hour cumulative rainfall reproduction period at the star-marked observation points calculated using the filter-applied composite data shown in FIG. 11 (c) is determined to be once every 100 years or more, and the observation data shown in FIG. 11 (a). We have obtained data close to. By synthesizing the first filter application data and the second filter application data in consideration of the positional deviation in this way, the reproduction period of the accumulated rainfall can be predicted with high accuracy. Therefore, the distribution of the risk of disaster can be predicted with high accuracy for the torrential rain associated with the linear precipitation zone, etc., and it can be used as a judgment material when evacuating.

As described above, in the precipitation prediction device 10 of the present embodiment, the precipitation prediction device 10 is acquired from the observation data acquisition unit 11 by dividing the observation area into a grid pattern and the observation data acquisition unit 11 that acquires at least the observation data of precipitation. Using the observation data, the movement of the past precipitation area and the distribution of the current precipitation are obtained, and the nowcast prediction unit 12 that acquires the nowcast prediction data that predicts the distribution of the precipitation after a short time for each grid, and the observation area. Is divided into grids, and using the observation data acquired from the observation data acquisition unit 11, the time change of the precipitation situation is calculated by a computer using the physical law, and the future precipitation situation is predicted for each grid. For all the grids of the observation region divided into a plurality of grids of the numerical prediction unit 13 and the nowcast prediction data, one grid is used as a reference grid, and a predetermined first filter region L1 including the reference grid is used. For the first filter application unit 14 that replaces the nowcast prediction data with the first filter application data using the maximum rainfall grid within the range of, and for all the grids of the observation area divided into a plurality of grids of the numerical prediction data. , A second filter application that replaces the numerical prediction data with the second filter application data by using one lattice as the reference lattice and using the maximum rainfall grid for the grid within the predetermined second filter area L2 including the reference grid. A unit 15 and a composition unit 16 that synthesizes observation data, first filter application data, and second filter application data using a composition coefficient are provided. Therefore, it is possible to predict the maximum value of accumulated rainfall with high accuracy for torrential rainfall caused by a large number of cumulonimbus clouds.

Further, in the precipitation prediction device 10 of the present embodiment, the first filter region L1 is smaller than the second filter region L2. Therefore, it is possible to predict the accumulated rainfall with higher accuracy.

Further, in the precipitation prediction device 10 of the present embodiment, the composite coefficient increases the ratio of the first filter application data in the latest prediction, and then decreases the ratio of the first filter application data as the prediction time becomes longer. 2 Increase the ratio of filter application data. Therefore, it is possible to predict the accumulated rainfall with higher accuracy.

Further, the precipitation prediction device 10 of the present embodiment further includes a reproduction period calculation unit 17 for calculating a reproduction period indicating how much the predicted cumulative rainfall stochastically occurs at the observation point. Therefore, the distribution of the degree of risk can be predicted with high accuracy for torrential rains associated with linear precipitation zones and the like, and can be used as a judgment material when evacuating.

Further, in the precipitation prediction method of the present embodiment, at least the step of acquiring the observation data of precipitation and the movement of the past precipitation area by dividing the observation area into a grid and using the observation data acquired from the observation data acquisition unit. And the step of acquiring the nowcast prediction data that predicts the distribution of precipitation after a short time for each grid, and the observation data obtained by dividing the observation area into grids and acquiring from the observation data acquisition unit. It is divided into a step of acquiring numerical prediction data for predicting the future precipitation situation for each grid by calculating the time change of the precipitation situation with a computer using the physical law, and a plurality of grids of nowcast prediction data. For all the grids in the observation area, one grid is used as the reference grid, and the nowcast prediction data is applied to the first filter using the maximum rainfall grid within the range of the predetermined first filter region L1 including the reference grid. For the step of replacing with data and all the grids of the observation region divided into multiple grids of the numerical prediction data, one grid is used as the reference grid, and it is within the range of the second filter region L2 predetermined including the reference grid. It has a step of replacing the numerical prediction data with the second filter application data using the maximum rainfall grid, and a step of synthesizing the observation data, the first filter application data, and the second filter application data using the synthesis coefficient. Therefore, it is possible to predict the cumulative rainfall with high accuracy for torrential rainfall caused by a large number of cumulonimbus clouds.

Further, in the precipitation prediction method of the present embodiment, the first filter region L1 is smaller than the second filter region L2. Therefore, it is possible to predict the accumulated rainfall with higher accuracy.

Further, in the precipitation prediction method of the present embodiment, the composite coefficient increases the ratio of the first filter application data in the first short-time prediction, and then increases the ratio of the second filter application data as the prediction time becomes longer. .. Therefore, it is possible to predict the accumulated rainfall with higher accuracy.

Further, the precipitation prediction method of the present embodiment further includes a step of calculating a reproduction period indicating how much the predicted cumulative rainfall is stochastically occurring at the observation point. Therefore, the distribution of the risk of disaster can be predicted with high accuracy for the torrential rain associated with the linear precipitation zone, etc., and it can be used as a judgment material when evacuating.

The present invention is not limited to this embodiment. That is, in the description of the embodiment, many specific detailed contents are included for illustration purposes, but those skilled in the art may make various variations and changes to these detailed contents.

10 ... Precipitation prediction device 11 ... Observation data acquisition unit 12 ... Nowcast prediction unit 13 ... Numerical prediction unit 14 ... First filter application unit 15 ... Second filter application unit 16 ... Synthesis unit 17 ... Reproduction period calculation unit

Claims (8)

At least the observation data acquisition unit that acquires the observation data of precipitation,
The observation area is divided into grids, and the movements of the past precipitation area and the distribution of current precipitation are obtained using the observation data acquired from the observation data acquisition unit, and the distribution of precipitation after a short period of time is predicted for each grid. Nowcast prediction unit that acquires Nowcast prediction data,
The observation area is divided into grids, and using the observation data acquired from the observation data acquisition unit, the time change of the precipitation situation is calculated by a computer using the physical law to predict the future precipitation situation for each grid. Numerical forecasting unit that acquires forecast data and
For all the grids in the observation region divided into a plurality of grids of the nowcast prediction data, one grid is used as a reference grid, and the maximum rainfall grid within the range of the predetermined first filter region including the reference grid. The first filter application unit that replaces the nowcast prediction data with the first filter application data by using
For all the grids in the observation region divided into a plurality of grids of the numerical prediction data, one grid is used as a reference grid, and the maximum rainfall grid within the range of the predetermined second filter region including the reference grid is used. A second filter application unit that replaces the numerical prediction data with the second filter application data by using
A synthesis unit that synthesizes the observation data, the first filter application data, and the second filter application data using the synthesis coefficient, and
Precipitation prediction device characterized by being equipped with. The precipitation prediction device according to claim 1, wherein the first filter region is smaller than the second filter region. In the latest prediction, the composition coefficient increases the proportion of the first filter applied data, then decreases the proportion of the first filter applied data and increases the proportion of the second filter applied data as the prediction time becomes longer. The precipitation prediction device according to claim 1 or 2, wherein the data is set. The invention according to any one of claims 1 to 3, further comprising a reproduction period calculation unit that calculates a reproduction period indicating how much the predicted cumulative rainfall stochastically occurs at the observation point. Precipitation forecaster. At least the steps to get the precipitation observation data,
The observation area is divided into grids, and the movements of the past precipitation area and the current precipitation distribution are obtained using the observation data acquired from the observation data acquisition unit, and the precipitation distribution after a short period of time is predicted for each grid. Steps to get the Nowcast prediction data and
The observation area is divided into grids, and using the observation data acquired from the observation data acquisition unit, the time change of the precipitation situation is calculated by a computer using the physical law, and the future precipitation situation is predicted for each grid. Steps to get forecast data and
For all the grids in the observation region divided into a plurality of grids of the nowcast prediction data, one grid is used as a reference grid, and the maximum rainfall grid within the range of the predetermined first filter region including the reference grid. To replace the nowcast prediction data with the first filter application data using
For all the grids in the observation region divided into a plurality of grids of the numerical prediction data, one grid is used as a reference grid, and the maximum rainfall grid within the range of the predetermined second filter region including the reference grid is used. Using the step of replacing the numerical prediction data with the second filter application data,
A step of synthesizing the observation data, the first filter application data, and the second filter application data using the synthesis coefficient, and
Precipitation prediction method characterized by having. The precipitation prediction method according to claim 5, wherein the first filter region is smaller than the second filter region. 5. The composition coefficient is characterized in that the ratio of the first filter applied data is increased in the first short-time prediction, and then the ratio of the second filter applied data is increased as the prediction time becomes longer. Or the precipitation prediction method according to 6. The precipitation prediction method according to any one of claims 5 to 7, further comprising a step of calculating a reproduction period indicating how much the predicted cumulative rainfall is stochastically occurring at the observation point. ..

Images