Water level prediction method, water level prediction device and water level prediction program

Abstract

To reflect an impact of a statistically calculated error on a predicted water level.SOLUTION: A water level prediction device 1 predicts a water level of a river at a prediction point by using a water level prediction model and comprises: a storage unit 10 which stores the water level prediction model and a correction amount function showing relation between forecast precipitation and a forecast error; and a water level prediction unit 24 which inputs the current corrected forecast precipitation after being corrected by using the correction amount function in the water level prediction model to calculate a predicted water level of the river at the prediction point. The correction amount function is prepared on the basis of totaling results of forecast errors in the past at a plurality of points of time within a predetermined period.SELECTED DRAWING: Figure 1

Description

Technical Field

[0001] The present invention relates to a water level prediction method, a water level prediction device, and a water level prediction program for predicting the water level of a river.

Background Art

[0002] In recent years, due to the influence of global warming, heavy rain disasters have been increasing, and damage caused by flood inundation due to river runoff has been increasing. Also, in river construction work, there is a possibility of damage (such as outflow and submersion) to workers and construction equipment and materials (heavy machinery and materials) due to rising water levels. Therefore, predicting the water level in advance in river construction work and notifying the relevant construction personnel is important for protecting workers and construction equipment and materials (heavy machinery and materials).

[0003] Conventionally, predicting river runoff and notifying the river construction site has been carried out, and a flood warning system for automatically calculating the river water level has also been constructed. A configuration example of a conventional flood warning system is shown in FIG. 14. In the flood warning system shown in FIG. 14, information regarding rain is acquired from the Japan Meteorological Agency, and information regarding rivers is acquired from the Ministry of Land, Infrastructure, Transport and Tourism. The water level of the river is predicted by analyzing these pieces of information. The predicted river water level and warnings based on the prediction are notified to the construction personnel at the construction site via the Internet, mobile email, etc. In the conventional flood warning system, the river water level is predicted by a physical model (water level prediction model) based on a hydraulic formula, and for example, physical models such as (1) "distributed model", (2) "regression model", and (3) "tank model" are used (see, for example, Patent Documents 1 and 2).

[0004] The "distributed model" is a physical model that determines the water level at the construction site through distributed runoff analysis using the water level, rainfall distribution, land use, and elevation upstream of the prediction point as input values. The "regression model" is a physical model that obtains a regression equation between the observed water level at the upstream observation point and the water level at the prediction point, and predicts the water level at the prediction point from the regression equation. The "tank model" is a physical model that takes the rainfall in the basin upstream of the prediction point as an input value, simulates the infiltration into the ground and the outflow into the river by vertically connected tanks, and predicts the water level at the prediction point.

Prior Art Documents

Patent Documents

[0005]

Patent Document 1

Patent Document 2

Summary of the Invention

Problems to be Solved by the Invention

[0006] In conventional water discharge warning systems, for example, using the predicted precipitation of weather forecasts as input values and utilizing physical models to predict the water level at the prediction point, when an error occurs in the precipitation prediction, an error may also occur in the predicted water level value. That is, an error can be seen in the predicted precipitation compared to the actual measured precipitation. If the predicted precipitation is underestimated compared to the actual value, the predicted water level will also be underestimated, and there is a concern of missing water discharges. Also, if the predicted precipitation is overestimated compared to the actual value, there is a problem that the prediction of water discharges will be a false alarm.

[0007] From such a perspective, the present invention provides a water level prediction method, a water level prediction device, and a water level prediction program that can reflect the influence of errors statistically obtained with respect to the predicted water level.

Means for Solving the Problems

[0008] The water level prediction method according to the present invention is a method for predicting the water level of a river at a prediction point using a water level prediction model. This water level prediction method includes a forecast error aggregation step, a correction amount function creation step, and a water level prediction step. In the forecast error aggregation step, based on the past measured precipitation and the past forecast precipitation regarding the prediction point, the forecast error at a plurality of time points within a predetermined period is obtained, and the results are aggregated. In the correction amount function creation step, based on the aggregation result of the forecast error, a correction amount function indicating the relationship between the forecast precipitation and the forecast error is created. In the water level prediction step, the current corrected forecast precipitation after correction using the correction amount function is input into the water level prediction model, and the predicted water level of the river at the prediction point is obtained.

[0009] In the water level prediction method according to the present invention, by correcting the forecast precipitation using the correction amount function, it is possible to reflect the influence of the error statistically obtained with respect to the predicted water level. Therefore, when an error occurs in the forecast precipitation, it is possible to make the water level prediction closer to the measured water level (that is, it is possible to improve the accuracy of the water level prediction).

[0010] In the forecast error aggregation step, the forecast error may be classified using the forecast precipitation rank based on the forecast precipitation, and a frequency distribution showing the relationship between the forecast error amount and the frequency may be created for each forecast precipitation rank based on the aggregation result. By doing so, the aggregation of the forecast error is easy.

[0011] In the correction amount function creation step, a statistical criterion based on the frequency may be set in the frequency distribution, and by paying attention to the forecast error amount at that criterion, a representative forecast error amount for each forecast precipitation rank may be obtained, and a correction amount function may be created based on the representative forecast error amount. By doing so, the creation of the correction amount function is easy.

[0012] In the correction amount function creation step, a confidence interval based on the frequency may be set in the frequency distribution, and a lower representative forecast error amount, which is the lower limit value in the confidence interval, may be obtained for each forecast precipitation rank. Alternatively, an upper representative forecast error amount, which is the upper limit value in the confidence interval, may be obtained for each forecast precipitation rank. In that case, a correction amount function is created based on at least one of the lower representative forecast error amount and the upper representative forecast error amount. Doing so makes it easy to create a correction amount function. Also, when creating correction amount functions for the lower limit side and the upper limit side, it is possible to show the width of the predicted water level. That is, it is possible to show the possibility of overlooking water discharge or false alarms when the precipitation forecast is off.

[0013] It may further have a water level prediction model construction step of constructing the water level prediction model based on past measured precipitation and measured water level regarding the prediction point. Also, in the forecast error aggregation step, by preparing a plurality of forecasts with different reference times, a plurality of the forecast errors may be obtained at each time point within the predetermined period, and the results may be aggregated.

[0014] The water level prediction device according to the present invention is a device that predicts the water level of a river at a prediction point using a water level prediction model. This water level prediction device has a storage unit and a water level prediction unit. The storage unit stores a correction amount function showing the relationship between forecast precipitation and forecast error, and the water level prediction model. The water level prediction unit inputs the current corrected forecast precipitation after correction using the correction amount function into the water level prediction model to obtain the predicted water level of the river at the prediction point. The correction amount function is created based on the aggregation result of past forecast errors at a plurality of time points within a predetermined period.

[0015] In the water level prediction device according to the present invention, by correcting the forecast precipitation using the correction amount function, it is possible to reflect the influence of the error statistically obtained with respect to the predicted water level. Therefore, when an error occurs in the forecast precipitation, it is possible to bring the water level prediction closer to the measured water level (that is, it is possible to improve the accuracy of the water level prediction).

[0016] The water level prediction program according to the present invention is a program for predicting the water level of a river at a prediction point using a water level prediction model. This water level prediction program causes a computer having a correction amount function indicating the relationship between the forecast precipitation and the forecast error and a storage unit storing the water level prediction model to function as a water level prediction unit. The water level prediction unit inputs the current corrected forecast precipitation after correction using the correction amount function into the water level prediction model to obtain the predicted water level of the river at the prediction point. The correction amount function is created based on the aggregation result of past forecast errors at a plurality of time points within a predetermined period.

[0017] The water level prediction program according to the present invention is a program for predicting the water level of a river at a prediction point using a water level prediction model. This water level prediction program causes a computer that can communicate with a correction amount function indicating the relationship between the forecast precipitation and the forecast error and a storage unit storing the water level prediction model to function as a water level prediction unit. The water level prediction unit inputs the current corrected forecast precipitation after correction using the correction amount function into the water level prediction model to obtain the predicted water level of the river at the prediction point. The correction amount function is created based on the aggregation result of past forecast errors at a plurality of time points within a predetermined period.

[0018] In the water level prediction program according to the present invention, by correcting the forecast precipitation using the correction amount function, it is possible to reflect the influence of the error statistically obtained with respect to the predicted water level. Therefore, when an error occurs in the forecast precipitation, it is possible to bring the water level prediction closer to the measured water level (that is, it is possible to improve the accuracy of the water level prediction).

Effect of the Invention

[0019] According to the present invention, it is possible to reflect the influence of the error statistically obtained with respect to the predicted water level.

Brief Description of the Drawings

[0020]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Figure 12

Figure 13

Figure 14

Embodiments for Carrying Out the Invention

[0021] Hereinafter, embodiments for carrying out the present invention will be described in detail with reference to the drawings as appropriate. Each drawing only schematically shows to such an extent that the present invention can be sufficiently understood. Therefore, the present invention is not limited only to the illustrated examples. In each drawing, common components and similar components are denoted by the same reference numerals, and redundant descriptions thereof are omitted.

[0022] <Configuration of the water prediction device according to the embodiment> With reference to FIG. 1, the configuration of the water level prediction device 1 according to the embodiment will be described. FIG. 1 is a schematic configuration diagram of the water level prediction device 1 according to the embodiment. The water level prediction device 1 is a device that predicts the water level of a river at a prediction point. When predicting the water level, a forecast precipitation amount related to the prediction point is input to the water level prediction device 1, and the predicted water level of the river at the prediction point is output.

[0023] As an example of precipitation forecast, there are numerical models (local model, meso model, global model, etc.) operated by the Japan Meteorological Agency. For example, in the meso numerical forecast model GPV (hereinafter sometimes referred to as "MSM") of the Japan Meteorological Agency, precipitation forecasts up to 39 hours ahead are provided. In MSM, the grid interval is subdivided into "5 km" meshes. As the input value of the water level prediction device 1, it is possible to use the forecast value by the Japan Meteorological Agency. Note that the input value of the water level prediction device 1 is not limited to the forecast value provided by the Japan Meteorological Agency.

[0024] The forecast precipitation amount often does not match the measured precipitation amount. FIG. 2 is a hyetograph comparing the measured precipitation amount at an AMeDAS observation station at a certain location in Japan and the forecast precipitation amount (39-hour ahead forecast) at the nearest MSM grid point. From FIG. 2, it can be seen that on February 20 and February 23, 2017, more than twice as much rain actually fell as the forecast precipitation amount. If the value obtained by subtracting the measured precipitation amount from the forecast precipitation amount is defined as the forecast error, this forecast error is, for example, "5 - 2 = 3 mm / h" assuming that the forecast precipitation amount is "5 mm / h" and the measured precipitation amount is "2 mm / h".

[0025] When predicting the water level of a river using a water level prediction model (for example, a numerical model, a regression model, an accumulated rainfall model, a conservation law model, a tank model, etc.), it is assumed that the forecast precipitation is used as an input value. If the forecast precipitation contains a forecast error, the prediction result of the water level will also contain an error with respect to the actual water level due to the influence of the forecast error. When the measured precipitation is greater than the forecast precipitation (that is, when the forecast precipitation deviates to the lower side), the predicted water level of the river will be lower than the actual level, causing the escape of runoff. Also, when the measured precipitation is less than the forecast precipitation (when the forecast precipitation deviates to the higher side), the predicted water level of the river will be higher than the actual level, causing a false alarm.

[0026] In the water level prediction device 1 shown in FIG. 1, the error (forecast error) between the forecast precipitation and the measured precipitation is statistically aggregated, a function considering the width of the error is obtained, and the forecast precipitation with the error difference added is used as the input value for water level prediction using the function. Therefore, the influence of the error can be reflected in the prediction of the water level. As a result, the accuracy of the predicted water level is improved. Also, for example, by outputting the water level prediction considering the error difference together with the result of the water level prediction without considering the error difference, it is possible to show the width of the predicted water level.

[0027] Referring to FIG. 3, the output result of the water level prediction device 1 will be described. FIG. 3 is an image diagram of the water level prediction result considering the error of the forecast precipitation. The upper part of FIG. 3 shows the forecast precipitation and the width of its error, and the lower part of FIG. 3 shows the predicted water level calculated using the forecast precipitation as the input value and the width of the predicted water level when considering the forecast error. The water level prediction device 1 can show the possibility of runoff or false alarm when the precipitation forecast is off by showing the width of the predicted water level.

[0028] As shown in FIG. 1, the water level prediction device 1 according to the embodiment includes a storage unit 10 and a control unit 20. The water level prediction device 1 is, for example, a personal computer (PC) operated by a person who performs water level prediction (for example, a manager of river construction work) or an application server communicably connected to the personal computer. The application server may be a device constituting a cloud system. Note that the personal computer and the application server are examples of a computer.

[0029] The storage unit 10 is composed of storage media such as a RAM (Random Access Memory), a ROM (Read Only Memory), an HDD (Hard Disk Drive), an SSD (Solid State Drive), and a flash memory. The control unit 20 is realized by program execution processing by a CPU (Central Processing Unit) or a dedicated circuit or the like. When the control unit 20 is realized by program execution processing, a program (water level prediction program) for realizing the functions of the control unit 20 is stored in the storage unit 10. Note that the water level prediction device 1 may acquire information stored in the storage unit 10 from external storage means (not shown) as necessary.

[0030] The storage unit 10 stores information necessary for predicting the water level of a river. For predicting the water level of a river, for example, a correction amount function, a water level prediction model, and input data of the water level prediction model (an example is forecast precipitation) are required. The correction amount function is a function showing the relationship between the forecast precipitation and the forecast error. The correction amount function is used for correcting the forecast precipitation which is the input data of the water level prediction model. Although details will be described later, the correction amount function is created based on the aggregation result of the forecast error.

[0031] The water level prediction model is a physical model based on a hydraulic formula, and is, for example, a numerical model, a regression model, a cumulative rainfall model, a conservation law model, a tank model, or the like. The water level prediction model is constructed using, for example, past measured precipitation and past measured water levels.

[0032] Since the characteristics of the forecast precipitation error are related to the rainfall characteristics, it is advisable to create a correction amount function for each region. For example, it is desirable that the correction amount function be created based on information regarding the prediction point for predicting the river water level (for example, information on the prediction point and its surroundings). The same applies to the water level prediction model. Note that the correction amount function and the water level prediction model may be prepared in advance for each region, and the correction amount function and the water level prediction model corresponding to the prediction point may be selected and used.

[0033] As shown in FIG. 1, the control unit 20 mainly includes a forecast error aggregation unit 21, a correction amount function creation unit 22, a water level prediction model construction unit 23, and a water level prediction unit 24. Note that each function of the control unit 20 shown in FIG. 1 is classified for convenience of explanation, and the classification method of the functions is merely an example. Here, an overview of each function will be described, and details of the processing of each function will be described later.

[0034] The forecast error aggregation unit 21 is a function for aggregating the difference between the forecast precipitation and the measured precipitation (forecast error). The aggregation result of the forecast error is used in the processing of the correction amount function creation unit 22. The correction amount function creation unit 22 is a function for obtaining a correction amount function based on the aggregation result of the forecast error. The correction amount function may be constant regardless of the magnitude of the forecast precipitation, or the correction amount may vary depending on the magnitude of the forecast precipitation. The obtained correction amount function is used in the processing of the water level prediction unit 24.

[0035] The water level prediction model construction unit 23 is a function for creating a water level prediction model. The created water level prediction model is used in the processing of the water level prediction unit 24. The water level prediction unit 24 is a function for predicting the river water level using the water level prediction model. The water level prediction unit 24 corrects the forecast precipitation, which is the input data of the water level prediction model, using the correction amount function, and uses the corrected forecast precipitation after correction in the processing of the water level prediction model.

[0036] <Regarding the water level prediction method according to the embodiment of the present invention> Referring to FIG. 4 (and referring to FIG. 1 as appropriate), a water level prediction method using the water level prediction device 1 according to the embodiment will be described. FIG. 4 is an example of a flowchart showing the water level prediction method according to the embodiment.

[0037] As shown in FIG. 4, the water level prediction method mainly includes a "selection step of water level prediction point (S1)", a "forecast error aggregation step (S2)", a "confidence interval setting step (S3)", a "correction amount function creation step (S4)", a "water level prediction model construction step (S5)", and a "water level prediction step (S6)". The water level prediction model construction step (S5) can be executed in parallel with steps S2 to S4, or can be executed before or after steps S2 to S4.

[0038] ≪Selection Step of Water Level Prediction Point (S1)≫ In this step, a prediction point for predicting the water level of the river is selected. For example, a human selects the water level prediction point according to the purpose of predicting the water level of the river. An example of the purpose of predicting the water level of the river is to ensure safety in river construction. In a river, there is a risk of water discharge when the water level rises significantly during heavy rain. In river construction, it is necessary to evacuate workers and construction machinery and equipment to ensure safety before such a large-scale river water discharge occurs.

[0039] In this embodiment, the case where the water level observation station in a certain river basin shown in FIG. 5 is taken as the target point for water level prediction will be exemplified and described. Note that the prediction point for predicting the water level of the river does not have to be a water level observation station, and any position on the river can be set as the prediction point. For example, the location of river construction can be selected as the prediction point.

[0040] ≪Forecast Error Aggregation Step (S2)≫ In this step, based on the past measured precipitation and the past forecast precipitation related to the prediction point, the forecast errors at a plurality of time points within a predetermined period are obtained and the results are aggregated. For example, when there are N time points within a predetermined period (an example is when the precipitation is predicted and measured 1 hour ahead, 2 hours ahead, ···, N hours ahead with 0 o'clock as the reference time point), the forecast error aggregation unit 21 obtains N forecast errors and aggregates the results.

[0041] The forecast error aggregating unit 21 may obtain a plurality of forecast errors at each time point within a predetermined period by preparing a plurality of forecasts with different reference time points, and aggregate the results. For example, when the reference time point of the forecast is changed M times due to the announcement of the forecast precipitation amount at each time point every few hours (as an example, when the precipitation amount 1 hour ahead, 2 hours ahead, ···, N hours ahead is predicted and measured at each reference time point with 0 o'clock, 1 o'clock, ··· M o'clock as the reference time points), the forecast error aggregating unit 21 obtains "M × N" forecast errors and aggregates the results.

[0042] In this embodiment, it is assumed that the measured precipitation amount uses the precipitation amount of the AMeDAS observation station, and the forecast precipitation amount uses the precipitation amount of the MSM. Then, a case will be described in which the forecast error (ΔR = Rp - Rm) is aggregated for the measured precipitation amount (Rm) of the Shingu observation station (see FIG. 5), which is the AMeDAS observation station closest to the Aikawa water level observation station, and the nearest MSM forecast precipitation amount (Rp). MSM is distributed every 3 hours (0 o'clock, 3 o'clock, 6 o'clock ···), and each distribution includes forecasts 1 hour ahead to 39 hours ahead. The forecast error is aggregated for forecasts for several years, for example, the forecasts 1 hour ahead to 39 hours ahead at 0 o'clock and the forecasts 1 hour ahead to 39 hours ahead at 3 o'clock.

[0043] Subsequently, the forecast error aggregating unit 21 classifies the forecast error (ΔR) using the forecast precipitation rank based on the forecast precipitation amount (Rp), and creates a frequency distribution showing the relationship between the forecast error amount and the frequency for each forecast precipitation rank based on the aggregation result. The forecast precipitation rank is, for example, a classification of the forecast precipitation amount (Rp) every "several mm / h". It is desirable to use the cumulative frequency distribution as the frequency distribution because it facilitates the processing in the confidence interval setting step (S3).

[0044] For example, a case will be described where the forecast precipitation amount (Rp) is ranked every "2 mm / h", and a cumulative frequency distribution of the forecast error (ΔR) for each forecast precipitation amount (Rp) is created. FIG. 6 is an example of the cumulative frequency distribution, and it is the cumulative frequency distribution for the forecast precipitation amount rank of "6 to 8 mm / h (6 < Rp ≤ 8 mm / h)". As shown in FIG. 6, the horizontal axis of the cumulative frequency distribution is the forecast error (ΔR) [mm / h], and the vertical axis is the cumulative frequency [%]. Although illustration is omitted, in addition to the cumulative frequency distribution shown in FIG. 6, cumulative frequency distributions for forecast precipitation amount ranks of "0 to 2 mm / h (0 < Rp ≤ 2 mm / h)", "2 to 4 mm / h (2 < Rp ≤ 4 mm / h)", "4 to 6 mm / h (4 < Rp ≤ 6 mm / h)", "8 to 10 mm / h (8 < Rp ≤ 10 mm / h)",... are similarly created.

[0045] <<Step of setting confidence interval (S3)>> In this step, at least one statistical criterion based on frequency (an example is the lower limit or upper limit of the confidence interval) is set for the frequency distribution, and by paying attention to the amount of forecast error at that criterion, the representative amount of forecast error for each forecast precipitation amount rank is obtained. It is preferable that the criteria set for the frequency distribution are common to all forecast precipitation amount ranks.

[0046] A case will be described where the representative amount of forecast error is obtained using the upper and lower limits of the interval as criteria by setting an "80% confidence interval" for the cumulative frequency distribution (an example is the one shown in FIG. 6). The confidence interval means the range of errors to be considered, and the "80% confidence interval" shown in FIG. 6 is obtained by excluding 10% from both sides of the "100% confidence interval".

[0047] In this case, the forecast error ΔR that is the lower limit value of the 80% confidence interval (refer to reference sign F1 in FIG. 6) and the forecast error ΔR that is the upper limit value of the 80% confidence interval (refer to reference sign F2 in FIG. 6) become the representative amounts of forecast error for the forecast precipitation amount rank of "6 to 8 mm / h (6 < Rp ≤ 8 mm / h)". The forecast error ΔR indicated by reference sign F1 is the lower limit value of the 80% confidence interval, so it is particularly referred to as the "lower-side representative amount of forecast error", and the forecast error ΔR indicated by reference sign F2 is the upper limit value of the 80% confidence interval, so it is particularly referred to as the "upper-side representative amount of forecast error".

[0048] When determining the possibility that the water level prediction deviates to the lower side (that is, the possibility that the actual water level is higher than the predicted water level, which may be the upper part K1 of the predicted water level width in Fig. 3), the criterion set for the frequency distribution is set in the range where the forecast error ΔR is a negative value (ΔR < 0). On the other hand, when determining the possibility that the water level prediction deviates to the upper side (that is, the possibility that the actual water level is lower than the predicted water level, which may be the lower part K2 of the predicted water level width in Fig. 3), the criterion set for the frequency distribution is set in the range where the forecast error ΔR is a positive value (ΔR > 0). Note that different criteria may be set for the negative side and the positive side. For example, it is also possible to set a criterion that excludes 5% on the negative side and 10% on the positive side.

[0049] The lower limit of the 80% confidence interval shown in Fig. 6 exists in the range where the forecast error ΔR is a negative value (ΔR < 0). Therefore, the representative forecast error amount on the lower limit side can be used as information for creating a correction amount function for determining the possibility that the water level prediction deviates to the lower side (that is, the possibility that the actual water level is higher than the predicted water level). Also, the upper limit of the 80% confidence interval exists in the range where the forecast error ΔR is a positive value (ΔR > 0). Therefore, the representative forecast error amount on the upper limit side can be used as information for creating a correction amount function for determining the possibility that the water level prediction deviates to the upper side (that is, the possibility that the actual water level is lower than the predicted water level).

[0050] ≪Correction Amount Function Creation Step (S4)≫ In this step, based on the aggregated result of the forecast error, a correction amount function showing the relationship between the forecast precipitation amount Rp and the forecast error ΔR is created. In this embodiment, using the lower limit value (representative forecast error amount on the lower limit side) of the 80% confidence interval obtained for each forecast precipitation amount rank, a correction amount function used when the forecast is too small compared to the actual measurement is created. Also, using the upper limit value (representative forecast error amount on the upper limit side) of the 80% confidence interval obtained for each forecast precipitation amount rank, a correction amount function used when the forecast is too large compared to the actual measurement is created.

[0051] In the case where ensuring safety in river works is assumed for the purpose of predicting the water level of a river, it is more important to know the possibility of missing the emergence of water than to know the possibility of a false alarm. Therefore, in the following, the processing of the lower limit value (lower-side representative forecast error amount) of the 80% confidence interval particularly related to the possibility of missing the emergence of water will be described. Note that the processing of the upper limit value (upper-side representative forecast error amount) of the 80% confidence interval related to the possibility of a false alarm is the same as the processing of the lower limit value.

[0052] An example of creating a correction amount function using the lower limit value (lower-side representative forecast error amount) of the 80% confidence interval will be described. As shown in FIG. 7, the lower limit value (lower-side representative forecast error amount) of the 80% confidence interval is plotted in association with the forecast precipitation rank. FIG. 7 is a graph showing the relationship between the forecast precipitation Rp and the lower limit value of the 80% confidence interval. In FIG. 7, the lower limit value (lower-side representative forecast error amount) of the 80% confidence interval when the forecast precipitation rank is "0 to 2 mm / h (0 < Rp ≤ 2 mm / h)" is plotted at the position of "Rp = 1 mm / h", which is the median within the rank. For other forecast precipitation ranks, they are also plotted at the positions of the medians within the ranks in the same way.

[0053] From FIG. 7, the lower limit value (lower-side representative forecast error amount) in the 80% confidence interval of the forecast error ΔR with respect to the forecast precipitation Rp is minimized when the forecast precipitation Rp is "4 to 6 mm / h", and approaches the hit (ΔR = 0 mm / h)·overestimated side (ΔR > 0 mm / h) for forecast precipitation Rp larger than that.

[0054] The formula (correction amount function f(Rp)) representing the error relationship obtained in FIG. 7 is defined as in formula (1). The correction amount function f(Rp) in formula 1 is an approximate curve of the lower-side representative forecast error amount plotted in FIG. 7, and is a function that asymptotically approaches "ΔR = 0" as the forecast precipitation Rp increases. This correction amount function f(Rp) is a formula that determines the error difference that becomes too small with respect to the magnitude of the forecast precipitation Rp. Correction amount function f(Rp) = {-2.75 × exp(-0.15 × Rp)} × Rp ··· Formula (1) Note that since the characteristics of the forecast precipitation error are related to the rainfall characteristics, it is advisable to create a relational expression (correction amount function) for each region. The correction amount function is not limited to an exponential function, and appropriate functions such as a logarithmic function, an n-th order function (n is an integer of 1 or more), and a step function can be used.

[0055] ≪Water level prediction model construction process (S5)≫ In this process, a water level prediction model (for example, a numerical model, a regression model, a cumulative rainfall model, a conservation law model, a tank model, etc.) is constructed. For example, a water level prediction model for predicting the water level of the Soga observation station is constructed using past measured precipitation and measured water levels.

[0056] ≪Water level prediction process (S6)≫ In this process, the water level of the river is predicted using the water level prediction model. Specifically, the water level prediction unit 24 corrects the forecast precipitation, which is the input data of the water level prediction model, using the correction amount function f(Rp), and inputs the corrected forecast precipitation after correction into the water level prediction model to predict the water level of the river. Based on the provided water level prediction, the administrator evacuates and cures heavy machinery and equipment in the construction yard.

[0057] As described above, in the water level prediction device 1 and the water level prediction method according to the embodiment, by correcting the forecast precipitation Rp using the correction amount function f(Rp), even when an error occurs in the forecast precipitation Rp, the influence of the error on the water level prediction can be reflected. Therefore, when an error occurs in the forecast precipitation, it is possible to bring the water level prediction closer to the measured water level (that is, it is possible to improve the accuracy of the water level prediction). Further, when a confidence interval is set in the frequency distribution to obtain a lower limit value and an upper limit value, and correction amount functions f(Rp) on the lower limit side and the upper limit side are created, it is possible to show the width of the predicted water level as shown in FIG. 3. That is, it is possible to show the possibility of water discharge and false alarm when the precipitation forecast is off.

[0058] In order to confirm the effects of the water level prediction device 1 and the water level prediction method according to the embodiment, a verification test using actual data will be described. The verification test assumes a case of monitoring water discharge.

[0059] In the verification test, the Aikawa water level observation station in the Kumano River basin shown in Fig. 5 was set as the target point for water level prediction. Also, for the actual measured precipitation, the precipitation at the AMeDAS observation station was used, and for the forecast precipitation, the precipitation of MSM was used. The forecast error (ΔR = Rp - Rm) was tabulated for the actual measured precipitation (Rm) at the Shingu observation station, which is the AMeDAS observation station closest to the Aikawa water level observation station, and the nearest MSM forecast precipitation (Rp). The tabulation was for approximately 14 years from 2006 to 2019. As a result of tabulating the forecast error, the correction amount function “f(Rp) = {-2.75 × exp(-0.15 × Rp)} × Rp” shown in Equation (1) was obtained.

[0060] A water level prediction model for predicting the water level at the Aikawa observation station was constructed using past actual measured precipitation and actual measured water levels. In the verification test, two methods, the tank model and the regression equation, were adopted. Since the water level prediction model in the verification test is assumed to be applied at the time of water discharge, the model was constructed for 12 water discharge events that exceeded the provisional warning water level of 8.0 m (T.P.) between 2020 and 2022. Fig. 8 shows the relationship between the actual measured water level and the predicted water level (the input value of the predicted water level is the actual measured precipitation) at the Aikawa water level observation station by the tank model. It can be seen that the tank model used in the verification test has an error of approximately “+1 m” between the predicted water level and the actual measured water level, and reproduces the water level with relatively high accuracy.

[0061] For water level prediction, the tank model was used to predict the water level with the forecast precipitation of all 12 events as the input value. An example of the prediction result is shown in Fig. 9. Fig. 9 is a comparison of the water level prediction results of the tank model with and without considering the forecast error. At the top of Fig. 9, the upper limit value of the forecast precipitation considering the actual measured precipitation, forecast precipitation, and error is shown by a dotted line, and at the bottom, the actual measured water level, predicted water level, and upper limit value of the predicted water level are shown by a solid line.

[0062] The predicted water level without considering the forecast error (thin solid line) is significantly lower than the measured water level (thick solid line). However, the upper limit of the predicted water level considering the error (medium-thickness solid line) is higher than the predicted water level without considering the forecast error ΔR and approaches the measured water level. It can be seen that in the case where the forecast precipitation is smaller than the measured precipitation, the error of the predicted water level can be reduced by applying the correction amount function.

[0063] The frequencies of the predicted water level being "hit", "missed", "false alarm", and "no flood" were tabulated for all 12 events when the correction amount function of the forecast precipitation was applied and when it was not applied. As shown in Fig. 10, the tabulation method is as follows: when the measured water level is 8.0 m or higher, if the predicted water level is 8.0 m or higher, it is considered a hit (see Fig. 10(a)), and if the predicted water level is less than 8.0 m, it is considered a miss (see Fig. 10(b)). Also, when the measured water level is less than 8.0 m, if the predicted water level is 8.0 m or higher, it is considered a false alarm (see Fig. 10(c)), and if the predicted water level is less than 8.0 m, it is considered no flood (see Fig. 10(d)). Fig. 10 is a diagram for explaining the relationship between the predicted water level and the measured water level based on the warning water level.

[0064] The tabulation results are shown in Fig. 11. Fig. 11 shows the water level prediction results with and without considering the forecast error. It targeted all the forecasts distributed during 12 flood events over the three years from 2020 to 2022. Since the MSM forecast is distributed every three hours, when targeting the entire period of events 1 to 12, the number of MSM forecasts (i.e., the number of water level predictions) is 262 times. The numbers described in the graph of Fig. 11 are the number of times of each result (hit, miss, etc.). By considering the forecast error ΔR, the number of misses decreased by approximately 1 / 4. From the above, the effect of reducing misses by the water level prediction method described in this embodiment was shown.

[0065] The aggregated results of the forecast errors obtained in the verification test are shown in FIG. 12. As shown in FIG. 12, in the range where the forecast error ΔR is underestimated (ΔR ≦ -1), the relatively high proportion is for the forecasts with relatively small forecast precipitation amounts Rp of about "2 to 8 mm / h". Also, in the range where the forecast error ΔR is accurate (-1 < ΔR ≦ 1), the relatively high proportion is for the forecasts with forecast precipitation amounts Rp of "0 to 2 mm / h". Further, in the range where the forecast error ΔR is overestimated (ΔR > 1), the proportion increases as the forecast precipitation amount Rp increases.

[0066] Also, the relationship between the lower limit value of the forecast error ΔR and the forecast precipitation amount Rp in the 60% to 80% confidence interval is shown in FIG. 13. In the embodiment, the case of setting the 80% confidence interval for the frequency distribution was exemplified, but it is also possible to set other setting intervals (for example, 60% confidence interval or 70% confidence interval) for the frequency distribution and use the correction amount function related to its lower limit value and upper limit value.

[0067] As described above, the embodiments of the present invention have been explained, but the present invention is not limited thereto, and can be implemented without changing the gist of the claims.

Explanation of Signs

[0068] 1 Water level prediction device 10 Storage unit 20 Control unit 21 Forecast error aggregation unit 22 Correction amount function creation unit 23 Water level prediction model construction unit 24 Water level prediction unit

Claims

1. A water level prediction method for predicting the water level of a river at a prediction location using a water level prediction model, comprising: a forecast error aggregation step of obtaining forecast errors at a plurality of time points within a predetermined period based on past measured precipitation and past forecast precipitation regarding the prediction location, and aggregating the results; a correction amount function creation step of creating a correction amount function showing the relationship between the forecast precipitation and the forecast error based on the aggregation result of the forecast errors; a water level prediction step of inputting the current corrected forecast precipitation after correction using the correction amount function into the water level prediction model to obtain the predicted water level of the river at the prediction location. A water level prediction method characterized by the above.

2. In the forecast error aggregation step, the forecast errors are classified using a forecast precipitation rank based on the forecast precipitation, and a frequency distribution showing the relationship between the forecast error amount and the frequency is created for each forecast precipitation rank based on the aggregation result. The water level prediction method according to claim 1, characterized by the above.

3. In the correction amount function creation step, a statistical criterion based on the frequency is set for the frequency distribution, and a representative forecast error amount for each forecast precipitation rank is obtained by paying attention to the forecast error amount at the criterion, and a correction amount function is created based on the representative forecast error amount. The water level prediction method according to claim 2, characterized by the above.

4. In the correction amount function creation step, a confidence interval based on the frequency is set for the frequency distribution, a lower-side representative forecast error amount which is the lower limit value in the confidence interval is obtained for each forecast precipitation rank, or an upper-side representative forecast error amount which is the upper limit value in the confidence interval is obtained for each forecast precipitation rank, and a correction amount function is created based on at least one of the lower-side representative forecast error amount and the upper-side representative forecast error amount. The water level prediction method according to claim 2, characterized by the above.

5. Further comprising a water level prediction model construction step of constructing the water level prediction model based on past measured precipitation and measured water level regarding the prediction location. The water level prediction method according to claim 1, characterized by the above.

6. In the forecast error aggregation step, a plurality of forecasts with different reference time points are prepared to obtain a plurality of the forecast errors at each time point within the predetermined period, and the results are aggregated. The water level prediction method according to claim 1, characterized by the above.

7. A water level prediction apparatus for predicting the water level of a river at a prediction location using a water level prediction model, comprising: A correction amount function showing the relationship between the predicted precipitation and the prediction error, and a storage unit for storing the water level prediction model, A water level prediction unit that inputs the current corrected predicted precipitation after correction using the correction amount function into the water level prediction model to obtain the predicted water level of the river at the prediction point, wherein the correction amount function is created based on the aggregation result of past prediction errors at a plurality of time points within a predetermined period. A water level prediction device characterized by the above.

8. A water level prediction program for predicting the water level of a river at a prediction point using a water level prediction model, wherein a computer having a correction amount function showing the relationship between the predicted precipitation and the prediction error and a storage unit for storing the water level prediction model is caused to function as a water level prediction unit that inputs the current corrected predicted precipitation after correction using the correction amount function into the water level prediction model to obtain the predicted water level of the river at the prediction point, wherein the correction amount function is created based on the aggregation result of past prediction errors at a plurality of time points within a predetermined period. A water level prediction program characterized by the above.

9. A water level prediction program for predicting the water level of a river at a prediction point using a water level prediction model, wherein a computer capable of communicating with a correction amount function showing the relationship between the predicted precipitation and the prediction error and a storage unit for storing the water level prediction model is caused to function as a water level prediction unit that inputs the current corrected predicted precipitation after correction using the correction amount function into the water level prediction model to obtain the predicted water level of the river at the prediction point, wherein the correction amount function is created based on the aggregation result of past prediction errors at a plurality of time points within a predetermined period. A water level prediction program characterized by the above.

Images