River channel water temperature change trend prediction method based on remote sensing and water temperature inversion

Abstract

The invention discloses a riverway water temperature change trend prediction method based on remote sensing and water temperature inversion, and belongs to the field of riverway water temperature change trend prediction.The method comprises the steps that hydrological data of a riverway are obtained, thermal infrared remote sensing data, hydrological data and actually-measured water temperature data of all riverway sub-segments are aligned based on a time sequence and measuring point positions, and therefore the water temperature change trend of the riverway is obtained; meanwhile, adding air temperature data of each time to obtain a plurality of groups of time sequence alignment data sequences; performing filling processing on each group of time sequence alignment data sequences to obtain a plurality of groups of water temperature inversion data sequences; according to each group of water temperature inversion data sequences, discrete values are eliminated, and the average water temperature of each river subsection at each time is solved; iteratively solving the weight of each water temperature influence factor by using a BP neural network to obtain a water temperature prediction model; and obtaining a to-be-measured real condition, and obtaining a water temperature prediction result by using the water temperature prediction model. According to the method, the problem that water temperature inversion is greatly influenced by atmosphere, earth surface and solar radiation, so that the accuracy of a water temperature prediction result is not enough is solved.

Description

Technical Field

[0001] The invention belongs to the field of river water temperature change trend prediction, and in particular relates to a river water temperature change trend prediction method based on remote sensing and water temperature inversion. Background Art

[0002] Thermal infrared remote sensing satellites use thermal infrared sensors to detect and measure thermal radiation from the Earth's surface (including land and water). These satellites can provide valuable information on surface temperature, vegetation status, water characteristics, urban heat island effect, etc. Up to now, the main thermal infrared remote sensing satellites are:

[0003] Landsat series: The Landsat project is one of the most famous satellites in the history of remote sensing. Landsat 7 and Landsat 8 are equipped with thermal infrared sensors. For example, the ETM+ sensor carried by the Landsat 7 satellite launched in 1999 has a thermal infrared band resolution of 60 meters and is used to capture spectral data and thermal data. MODIS (Moderate Resolution Imaging Spectroradiometer): This is an instrument on the satellites Terra and Aqua that can provide data in the thermal infrared band for monitoring the earth's atmosphere, oceans and land. Advanced Very High Resolution Radiometer (AVHRR): This sensor is used to monitor changes in land, oceans and atmosphere. GOES (Geostationary Operational Environmental Satellites): A geostationary orbit environmental satellite that provides data in the thermal infrared band. Sentinel series: The Sentinel-3 satellite is equipped with a thermal infrared sensor to monitor the earth's surface temperature and the sea surface temperature. ASTER (Advanced Spaceborne Thermal Emission and Reflection Radiometer): This is a sensor installed on the Terra satellite that can provide high-resolution thermal infrared band images. ECOSTRESS (Ecosystem Spaceborne Thermal Radiometer Experiment on Space Station): Used to measure the temperature of the Earth's surface. Suomi NPP (National Polar-orbiting Partnership): Carried with sensors for monitoring the Earth's atmosphere, oceans and land, including thermal infrared bands.

[0004] The thermal infrared single-channel algorithm uses the single-channel data received by the satellite in the atmospheric window, and uses the atmospheric transmittance / radiation program to correct the attenuation and emission of the atmosphere. The atmospheric profile data needs to be input. Then, under the condition of known surface emissivity, reverse calculation is performed to obtain the surface temperature. Accurately inverting the surface temperature using this method requires a high-quality atmospheric transmittance / radiation program to estimate the atmospheric parameters, and also requires known channel emissivity and accurate atmospheric profiles, and the influence of terrain needs to be considered. Generally, the accuracy of the atmospheric transmittance / radiation program is mainly affected by the atmospheric radiation transfer model used in the program and the uncertainty of the atmospheric molecular absorption coefficient and aerosol absorption coefficient (Wan, 1999). The most common atmospheric radiation transfer models, such as the MODTRAN series model and the 4A / OP model, have been widely used for atmospheric correction or satellite thermal infrared data simulation. Studies have shown that within the atmospheric window, such as 3.4-4.1μm and 8-13μm, the accuracy of different atmospheric radiation transfer models is between 0.5% and 2%, which will cause the inverted surface temperature error to be between 0.4 and 1.5K. It is worth noting that even if the atmospheric radiation transfer model itself has no errors, the incompleteness of the atmospheric profile itself will cause a serious problem when correcting the atmospheric absorption and path radiation. Related studies have also shown that a 1% error in the surface emissivity will cause an error of 0.3K in the surface temperature in a humid and hot atmosphere, and the error is as high as 0.7K in a dry and cold atmosphere. Since the wavelength range of a single channel is usually around 10μm, most of the surface emissivities within this range are within a few percent, and the uncertainty of the emissivity will cause an error of 1-2K in the surface temperature. However, if the surface emissivity of a certain wavelength is known, the inversion error comes entirely from the atmospheric correction. The atmospheric profile is usually measured by ground-based radiosonde equipment, satellite vertical sounders and meteorological prediction models. Since atmospheric water vapor changes dramatically over time and space, using ground-based radiosonde equipment to detect the atmosphere far away from the target area or to detect the atmosphere far away from the satellite transit time may cause large errors in the surface temperature inversion results.

[0005] It should be noted that the single-channel algorithm is a simple deformation of the radiation transfer equation, which is based on the premise that the surface emissivity and atmospheric profile are known. Although these methods can accurately invert the surface temperature in theory, it is difficult to obtain high-precision surface emissivity in practical applications.

[0006] The use of a single-channel algorithm requires the knowledge of the surface emissivity of each pixel, the atmospheric radiation transfer model, and the accurate atmospheric profile. These conditions are difficult or impossible to meet in most practical situations. In order to use satellite thermal infrared data to obtain high-precision surface temperature at a global or regional scale, other methods must be developed. A method for ocean temperature inversion (i.e., the split window algorithm) utilizes the different characteristics of water vapor absorption in two channels with a central wavelength between 11 and 12 μm. It was first proposed by McMillin (1975). This method does not require any atmospheric profile information. After that, a variety of split window algorithms were proposed and modified and successfully used for sea surface temperature inversion. Inspired by the successful use of the split window algorithm for sea surface temperature remote sensing inversion, scholars at home and abroad have been trying to expand it for surface temperature inversion since the 1980s. There are mainly linear split window algorithms, nonlinear split window algorithms, linear or nonlinear multi-channel algorithms, temperature emissivity separation (TES), etc.

[0007] It should be noted that the mid-infrared data received by the satellite at the top of the atmosphere during the day include direct solar radiation reflected by the surface and the emitted radiation from the surface and the atmosphere itself, and the error in the solar correction term will also affect the final surface temperature inversion accuracy, especially in arid and semi-arid areas with high reflectivity in the mid-infrared band. In addition, adding an extra channel will also come at the cost of increased measurement errors. In addition, errors related to measurement instrument noise and other uncertainties will also affect the final surface temperature inversion accuracy. In the mid-infrared channel and the 8.7μm channel, the range of emissivity and various uncertainties of natural and artificial surfaces are higher than those used in the traditional split-window algorithm, which further limits the widespread use of these channels in operational applications. Summary of the invention

[0008] In view of the above-mentioned deficiencies in the prior art, the present invention provides a method for predicting river water temperature change trends based on remote sensing and water temperature inversion, which solves the problem that water temperature inversion is greatly affected by the atmosphere, surface and solar radiation, resulting in insufficient accuracy of water temperature prediction results.

[0009] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a method for predicting the change trend of river water temperature based on remote sensing and water temperature inversion comprises the following steps:

[0010] S1. Obtain hydrological data of a river channel and set a drop threshold, dividing the river channel into a number of river channel sub-sections based on the drop threshold, wherein the riverbed elevation difference between the starting section and the ending section of each of the river channel sub-sections is greater than or equal to the drop threshold; the hydrological data includes flow and water level;

[0011] S2, collecting thermal infrared remote sensing data and measured water temperature data of each river channel subsection, and aligning the thermal infrared remote sensing data, hydrological data and measured water temperature data of each river channel subsection based on time series and measurement point positions, and adding the air temperature data at each time to obtain several groups of time series aligned data sequences;

[0012] S3, performing filling processing on each group of time-series aligned data sequences to obtain several groups of water temperature inversion data sequences;

[0013] S4, according to each group of water temperature inversion data sequences, for the water temperature inversion data within the same time period, the discrete values ​​are eliminated to obtain the average water temperature of each river channel subsection at each time period;

[0014] S5, according to each group of water temperature inversion data series and the average water temperature of each river section at each time, using BP neural network, iteratively solve the weight of each water temperature influencing factor to obtain a water temperature prediction model;

[0015] S6. Obtain the thermal infrared remote sensing data, hydrological data and air temperature data of the actual situation to be measured, and use the water temperature prediction model to obtain the water temperature prediction result.

[0016] The beneficial effects of the present invention are as follows: based on the pre-collected data, the data is first filled to make up for the problem of insufficient data volume, and based on the neural network model, the season (temperature data), infrared remote sensing data, hydrology (flow and water level) and river hydrological changes after the construction of water conservancy projects are considered to solve the synergistic relationship of various influencing factors from multiple factors and aspects, thereby improving the accuracy of water temperature prediction.

[0017] Furthermore, the expression of the timing alignment data sequence in step S2 is:

[0018] W i ={t:[T t,i,j ,R t,i,j ,Q t,i ]:Temp t}

[0019] Among them, W i is the time-series alignment data sequence of the i-th river channel sub-segment; i is the river channel sub-segment number; t is the time; T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; j is the sampling position number, which is numbered from the starting section to the ending section of the river channel subsection; R t,i,j is the thermal infrared remote sensing data of the jth sampling position at time t in the i-th river channel subsection; Q t,i Temp is the hydrological data of the ith river channel subsection at time t; t is the temperature data at time t.

[0020] The beneficial effects of the above further scheme are: collecting early data, using time as the number of training samples, and preparing for the later stage.

[0021] Furthermore, the step S3 is specifically as follows:

[0022] S301, obtaining the vacant positions in each group of timing alignment data sequences;

[0023] S302, obtaining a time-aligned data sequence at a vacant position as a current data group;

[0024] S303, determine whether the measured water temperature data of the current data group is empty, if so, fill in the measured water temperature data, otherwise, directly go to step S304;

[0025] S304, determine whether the thermal infrared remote sensing data of the current data group is vacant, if so, fill in the thermal infrared remote sensing data, otherwise, directly go to step S305;

[0026] S305, determining whether the hydrological data of the current data group is missing, if so, performing data fitting according to the hydrological data of adjacent moments to fill in the hydrological data, otherwise, directly proceeding to step S306;

[0027] S306. Determine whether the temperature data of the current data group is missing. If so, perform data fitting based on the temperature data at adjacent moments to fill in the temperature data. Otherwise, return to step S301 to obtain the next missing position until all missing positions are filled to obtain several groups of water temperature inversion data sequences.

[0028] The beneficial effects of the above further scheme are: based on the known water temperature data, the measured water temperature data is first filled in, the results are determined first, and the subsequent influencing factors are adjusted based on the results, which can reduce the difficulty of adjustment; at the same time, since the thermal infrared remote sensing data has little to do with the time series, the use of the water temperature inversion model to obtain the thermal infrared remote sensing data can improve the data quality.

[0029] Furthermore, the formula for filling the measured water temperature data in step S303 is:

[0030]

[0031] Among them, T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; T t,i,j-1 is the measured water temperature data of the j-1th sampling position at time t in the i-th river channel subsection; X t,i,j is the horizontal projection coordinate of the jth sampling position in the ith river channel sub-segment at time t; X t,i,j-1 is the horizontal projection coordinate of the j-1th sampling position at time t in the ith river channel subsegment; Xt,i,j+1 is the horizontal projection coordinate of the j+1th sampling position at time t in the ith river channel subsegment; T t,i,j+1 It is the measured water temperature data of the j+1th sampling position at time t in the ith river section.

[0032] The beneficial effect of the above further solution is that the temperature transition at the sampling position is taken into account to ensure the quality of the temperature filling data.

[0033] Furthermore, the step S304 is specifically as follows:

[0034] S3041, determine whether the thermal infrared remote sensing data of the current data group is vacant, if so, directly go to step S3042, otherwise, directly go to step S305;

[0035] S3042, performing data fitting according to the atmospheric parameters in the thermal infrared remote sensing data at adjacent moments of the current sampling position, and solving the atmospheric parameters of the current data set;

[0036] S3044. Calculate the thermal infrared spectrum radiation brightness received by the satellite sensor at the top of the atmosphere according to the measured water temperature data of the current data group and the atmospheric parameters of the current data group, as the thermal infrared remote sensing data of the current data group:

[0037]

[0038] δ(R t,i,j )=-γ(R t,i,j )·R t,i,j +T(R t,i,j )

[0039]

[0040] Among them, R t,i,j is the thermal infrared spectrum radiation brightness of the jth sampling position in the ith river channel subsection at time t; T t,i,j is the measured water temperature data of the jth sampling location at time t in the i-th river channel subsection; δ(R t,i,j ) and γ(R t,i,j ) are intermediate variables determined by Planck’s formula; Temp t,1 Temp t,2 and Temp t,3 are atmospheric functions obtained based on the atmospheric parameters of the current data set; ε is the water body emissivity; T(R t,i,j ) is the brightness temperature detected by the sensor; c2 and c1 are constants; λ is the equivalent wavelength in the thermal infrared band.

[0041] The beneficial effect of the above further scheme is: using the water temperature inversion model to reverse the thermal infrared remote sensing data can improve the data quality.

[0042] Furthermore, the step S4 is specifically as follows:

[0043] S401, fitting the measured water temperature data of each river channel subsection at each time according to each group of water temperature inversion data sequences, to obtain a water temperature fitting line segment of each river channel subsection at each time;

[0044] S402, according to each group of water temperature inversion data sequences, based on the water temperature fitting line segment of each river channel sub-segment at each time, the discrete points whose distance from the fitting line segment is greater than the discrete threshold are eliminated to obtain the valid data group of each river channel sub-segment at each time;

[0045] S403, according to the valid data group of each river channel sub-section at each time, the measured water temperature data are summed and averaged to obtain the average water temperature of each river channel sub-section at each time.

[0046] The beneficial effects of the above further scheme are: eliminating discrete points can improve the accuracy of the average water temperature. At the same time, calculating the average water temperature prepares for the calculation of the subsequent prediction error benchmark value.

[0047] Furthermore, the step S5 is specifically as follows:

[0048] S501. According to each group of water temperature inversion data sequences and the average water temperature of each river channel subsection at each time, the prediction error benchmark value of each river channel subsection at each time is solved:

[0049]

[0050] Among them, ε t,i is the prediction error benchmark value of the ith river channel subsection at time t; T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; is the average water temperature of the ith river section at time t; N j is the number of sampling locations;

[0051] S502, according to each group of water temperature inversion data sequences, split the water temperature influencing factors to obtain several groups of training data:

[0052] A1={t:[x=R t,i,j ; y = T t,i,j ]}

[0053]

[0054]

[0055] Among them, A1 is the first set of training data; A2 is the second set of training data; A3 is the third set of training data; x is the independent variable; Qt,i (l t,i ) is subject to l i The hydrological data of the ith river channel subsection affected by the change at time t; l i is the length of the river from the center of the ith river section to the water conservancy project; y is the dependent variable; t is time; R t,i,j is the thermal infrared remote sensing data of the jth sampling position at time t in the i-th river channel subsection; T t,i,j Temp is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; t is the temperature data at time t; is the parameter weight of thermal infrared remote sensing data, which is an unknown quantity; is the parameter weight of hydrological data, which is an unknown quantity; θ is the correction coefficient of thermal infrared remote sensing data, which is an unknown quantity;

[0056] S503, according to the first set of training data, using BP neural network to iteratively solve the parameter weights of thermal infrared remote sensing data;

[0057] S504, according to the parameter weights of the thermal infrared remote sensing data, based on the second set of training data, using the BP neural network to solve Q t,i (l i ) parameter weight and The correction coefficient θ;

[0058] S505, according to Q t,i (l i ) parameter weight and The correction coefficient θ is solved iteratively using the BP neural network based on the third set of training data and the prediction error benchmark value of each river section at each time. The correction factor, Correction factor and Temp t The parameter weights are used to obtain the water temperature prediction model.

[0059] The beneficial effect of the above further scheme is: the first group of training data is single-factor influence data. When obtaining the single-factor parameter weights, new influencing factors are introduced and the single-factor weights of the previous round are retained. This can strengthen the association between factors while retaining the uniqueness of the factors and reflect the mutual influence relationship between the factors.

[0060] Furthermore, the loss function of the water temperature prediction model in step S505 is:

[0061]

[0062] Among them, LOSS is the loss function of the water temperature prediction model; is the predicted water temperature at the jth sampling location in the ith river subsegment at time t.

[0063] The beneficial effects of the above further scheme are: the introduction of losses ensures the prediction accuracy of the water temperature prediction model and guides the iteration direction.

[0064] Furthermore, the objective function of the water temperature prediction model in step S505 is:

[0065] O=min|LOSS-∑ i ε t,i |

[0066] Among them, O is the objective function of the water temperature prediction model; min is the minimum value; || is the absolute value; ε t,i is the prediction error benchmark value of the ith river channel subsegment at time t.

[0067] The beneficial effects of the above further scheme are: ensuring the prediction accuracy of the water temperature prediction model and guiding the termination of iteration. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] Figure 1 The figure is a flow chart of the method of the present invention. DETAILED DESCRIPTION

[0069] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.

[0070] like Figure 1 As shown, in one embodiment of the present invention, a method for predicting the change trend of river water temperature based on remote sensing and water temperature inversion includes the following steps:

[0071] S1. Obtain hydrological data of a river channel and set a drop threshold, dividing the river channel into a number of river channel sub-sections based on the drop threshold, wherein the riverbed elevation difference between the starting section and the ending section of each of the river channel sub-sections is greater than or equal to the drop threshold; the hydrological data includes flow and water level;

[0072] S2, collecting thermal infrared remote sensing data and measured water temperature data of each river channel subsection, and aligning the thermal infrared remote sensing data, hydrological data and measured water temperature data of each river channel subsection based on time series and measurement point positions, and adding the air temperature data at each time to obtain several groups of time series aligned data sequences;

[0073] S3, performing filling processing on each group of time-series aligned data sequences to obtain several groups of water temperature inversion data sequences;

[0074] S4, according to each group of water temperature inversion data sequences, for the water temperature inversion data within the same time period, the discrete values ​​are eliminated to obtain the average water temperature of each river channel subsection at each time period;

[0075] S5, according to each group of water temperature inversion data series and the average water temperature of each river section at each time, using BP neural network, iteratively solve the weight of each water temperature influencing factor to obtain a water temperature prediction model;

[0076] S6. Obtain the thermal infrared remote sensing data, hydrological data and air temperature data of the actual situation to be measured, and use the water temperature prediction model to obtain the water temperature prediction result.

[0077] In this embodiment, the time parameter is added to index the data sequence using time as a number.

[0078] The expression of the timing alignment data sequence in step S2 is:

[0079] W i ={t:[T t,i,j ,R t,i,j ,Q t,i ]:Temp t}

[0080] Among them, W i is the time-series alignment data sequence of the i-th river channel sub-segment; i is the river channel sub-segment number; t is the time; T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; j is the sampling position number, which is numbered from the starting section to the ending section of the river channel subsection; R t,i,j is the thermal infrared remote sensing data of the jth sampling position at time t in the i-th river channel subsection; Q t,i Temp is the hydrological data of the ith river channel subsection at time t; t is the temperature data at time t.

[0081] In this embodiment, hydrological factors are taken into account in order to take into account the temperature diffusion under river scouring, and air temperature data is taken into account in order to take into account the influence of seasons; thermal infrared remote sensing data is indispensable data for water temperature inversion, based on which the water temperature can be calculated; but at the same time, due to the influence of the atmosphere, solar radiation, etc., the results of water temperature inversion will be biased, and therefore, the water temperature inversion model is only used to reversely derive thermal infrared remote sensing data when filling in data.

[0082] The step S3 is specifically as follows:

[0083] S301, obtaining the vacant positions in each group of timing alignment data sequences;

[0084] S302, obtaining a time-aligned data sequence at a vacant position as a current data group;

[0085] S303, determine whether the measured water temperature data of the current data group is empty, if so, fill in the measured water temperature data, otherwise, directly go to step S304;

[0086] S304, determine whether the thermal infrared remote sensing data of the current data group is vacant, if so, fill in the thermal infrared remote sensing data, otherwise, directly go to step S305;

[0087] S305, determining whether the hydrological data of the current data group is missing, if so, performing data fitting according to the hydrological data of adjacent moments to fill in the hydrological data, otherwise, directly proceeding to step S306;

[0088] S306. Determine whether the temperature data of the current data group is missing. If so, perform data fitting based on the temperature data at adjacent moments to fill in the temperature data. Otherwise, return to step S301 to obtain the next missing position until all missing positions are filled to obtain several groups of water temperature inversion data sequences.

[0089] The formula for filling the measured water temperature data in step S303 is:

[0090]

[0091] Among them, T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; T t,i,j-1 is the measured water temperature data of the j-1th sampling position at time t in the i-th river channel subsection; X t,i,j is the horizontal projection coordinate of the jth sampling position in the ith river channel sub-segment at time t; X t,i,j-1 is the horizontal projection coordinate of the j-1th sampling position at time t in the ith river channel subsegment; X t,i,j+1 is the horizontal projection coordinate of the j+1th sampling position at time t in the ith river channel subsegment; T t,i,j+1 It is the measured water temperature data of the j+1th sampling position at time t in the ith river section.

[0092] The step S304 is specifically as follows:

[0093] S3041, determine whether the thermal infrared remote sensing data of the current data group is vacant, if so, directly go to step S3042, otherwise, directly go to step S305;

[0094] S3042, performing data fitting according to the atmospheric parameters in the thermal infrared remote sensing data at adjacent moments of the current sampling position, and solving the atmospheric parameters of the current data set;

[0095] S3044. Calculate the thermal infrared spectrum radiation brightness received by the satellite sensor at the top of the atmosphere according to the measured water temperature data of the current data group and the atmospheric parameters of the current data group, as the thermal infrared remote sensing data of the current data group:

[0096]

[0097] δ(R t,i,j )=-γ(R t,i,j )·R t,i,j +T(R t,i,j )

[0098]

[0099] Among them, R t,i,j is the thermal infrared spectrum radiation brightness of the jth sampling position in the ith river channel subsection at time t; T t,i,j is the measured water temperature data of the jth sampling location at time t in the i-th river channel subsection; δ(R t,i,j ) and γ(R t,i,j ) are intermediate variables determined by Planck’s formula; Temp t,1 Temp t,2 and Temp t,3 are atmospheric functions obtained based on the atmospheric parameters of the current data set; ε is the water body emissivity; T(R t,i,j ) is the brightness temperature detected by the sensor; c2 and c1 are constants; λ is the equivalent wavelength in the thermal infrared band.

[0100] The step S4 is specifically as follows:

[0101] S401, fitting the measured water temperature data of each river channel subsection at each time according to each group of water temperature inversion data sequences, to obtain a water temperature fitting line segment of each river channel subsection at each time;

[0102] S402, according to each group of water temperature inversion data sequences, based on the water temperature fitting line segment of each river channel sub-segment at each time, the discrete points whose distance from the fitting line segment is greater than the discrete threshold are eliminated to obtain the valid data group of each river channel sub-segment at each time;

[0103] S403, according to the valid data group of each river channel sub-section at each time, the measured water temperature data are summed and averaged to obtain the average water temperature of each river channel sub-section at each time.

[0104] The step S5 is specifically as follows:

[0105] S501. According to each group of water temperature inversion data sequences and the average water temperature of each river channel subsection at each time, the prediction error benchmark value of each river channel subsection at each time is solved:

[0106]

[0107] Among them, ε t,i is the prediction error benchmark value of the ith river channel subsection at time t; T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; is the average water temperature of the ith river section at time t; N j is the number of sampling locations;

[0108] S502, according to each group of water temperature inversion data sequences, split the water temperature influencing factors to obtain several groups of training data:

[0109] A1={t:[x=R t,i,j ; y = T t,i,j ]}

[0110]

[0111]

[0112] Among them, A1 is the first set of training data; A2 is the second set of training data; A3 is the third set of training data; x is the independent variable; Q t,i (l t,i ) is subject to l i The hydrological data of the ith river channel subsection affected by the change at time t; l i is the length of the river from the center of the ith river section to the water conservancy project; y is the dependent variable; t is time; R t,i,j is the thermal infrared remote sensing data of the jth sampling position at time t in the i-th river channel subsection; T t,i,j Temp is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; t is the temperature data at time t; is the parameter weight of thermal infrared remote sensing data, which is an unknown quantity; is the parameter weight of hydrological data, which is an unknown quantity; θ is the correction coefficient of thermal infrared remote sensing data, which is an unknown quantity;

[0113] S503, according to the first set of training data, using BP neural network to iteratively solve the parameter weights of thermal infrared remote sensing data;

[0114] S504, according to the parameter weights of the thermal infrared remote sensing data, based on the second set of training data, using the BP neural network to solve Q t,i (l i ) parameter weight and The correction coefficient θ;

[0115] S505, according to Q t,i (l i ) parameter weight and The correction coefficient θ is solved iteratively using the BP neural network based on the third set of training data and the prediction error benchmark value of each river section at each time. The correction factor, Correction factor and Temp t The parameter weights are used to obtain the water temperature prediction model.

[0116] The loss function of the water temperature prediction model in step S505 is:

[0117]

[0118] Among them, LOSS is the loss function of the water temperature prediction model; is the predicted water temperature at the jth sampling location in the ith river subsegment at time t.

[0119] The objective function of the water temperature prediction model in step S505 is:

[0120] O=min|LOSS-∑ i ε t,i |

[0121] Among them, O is the objective function of the water temperature prediction model; min is the minimum value; || is the absolute value; ε t,i is the prediction error benchmark value of the ith river channel subsegment at time t.

Claims

1. A method for predicting river water temperature change trends based on remote sensing and water temperature inversion, characterized in that: The following steps are involved: S1. Obtain hydrological data of a river channel and set a drop threshold, dividing the river channel into a number of river channel sub-sections based on the drop threshold, wherein the riverbed elevation difference between the starting section and the ending section of each of the river channel sub-sections is greater than or equal to the drop threshold; the hydrological data includes flow and water level; S2, collecting thermal infrared remote sensing data and measured water temperature data of each river channel subsection, and aligning the thermal infrared remote sensing data, hydrological data and measured water temperature data of each river channel subsection based on time series and measurement point positions, and adding the air temperature data at each time to obtain several groups of time series aligned data sequences; S3, performing filling processing on each group of time-series aligned data sequences to obtain several groups of water temperature inversion data sequences; S4, according to each group of water temperature inversion data sequences, for the water temperature inversion data within the same time period, the discrete values ​​are eliminated to obtain the average water temperature of each river channel subsection at each time period; S5, according to each group of water temperature inversion data series and the average water temperature of each river section at each time, using BP neural network, iteratively solve the weight of each water temperature influencing factor to obtain a water temperature prediction model; S6. Obtain the thermal infrared remote sensing data, hydrological data and air temperature data of the actual situation to be measured, and use the water temperature prediction model to obtain the water temperature prediction result.

2. The method for predicting river water temperature change trend based on remote sensing and water temperature inversion according to claim 1 is characterized in that: The expression of the timing alignment data sequence in step S2 is: W i ={t:[T t,i,j ,R t,i,j ,Q t,i ]:Temp t } Among them, W i is the time-series alignment data sequence of the i-th river channel sub-segment; i is the river channel sub-segment number; t is the time; T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; j is the sampling position number, which is numbered from the starting section to the ending section of the river channel subsection; R t,i,j is the thermal infrared remote sensing data of the jth sampling position at time t in the i-th river channel subsection; Q t,i Temp is the hydrological data of the ith river channel subsection at time t; t is the temperature data at time t.

3. The method for predicting river water temperature change trend based on remote sensing and water temperature inversion according to claim 1 is characterized in that: The step S3 is specifically as follows: S301, obtaining the vacant positions in each group of timing alignment data sequences; S302, obtaining a time-aligned data sequence at a vacant position as a current data group; S303, determine whether the measured water temperature data of the current data group is empty, if so, fill in the measured water temperature data, otherwise, directly go to step S304; S304, determine whether the thermal infrared remote sensing data of the current data group is vacant, if so, fill in the thermal infrared remote sensing data, otherwise directly go to step S305; S305, determining whether the hydrological data of the current data group is missing, if so, performing data fitting according to the hydrological data of adjacent moments to fill in the hydrological data, otherwise, directly proceeding to step S306; S306. Determine whether the temperature data of the current data group is missing. If so, perform data fitting based on the temperature data at adjacent moments to fill in the temperature data. Otherwise, return to step S301 to obtain the next missing position until all missing positions are filled to obtain several groups of water temperature inversion data sequences.

4. The method for predicting river water temperature change trend based on remote sensing and water temperature inversion according to claim 3 is characterized in that: The formula for filling the measured water temperature data in step S303 is: Among them, T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; T t,i,j-1 is the measured water temperature data of the j-1th sampling position at time t in the i-th river channel subsection; X t,i,j is the horizontal projection coordinate of the jth sampling position in the ith river channel sub-segment at time t; X t,i,j-1 is the horizontal projection coordinate of the j-1th sampling position at time t in the ith river channel sub-segment; X t,i,j+1 is the horizontal projection coordinate of the j+1th sampling position at time t in the ith river channel subsegment; T t,i,j+1 It is the measured water temperature data of the j+1th sampling position at time t in the ith river section.

5. The method for predicting river water temperature change trend based on remote sensing and water temperature inversion according to claim 2 is characterized in that: The step S304 is specifically as follows: S3041, determine whether the thermal infrared remote sensing data of the current data group is vacant, if so, directly go to step S3042, otherwise, directly go to step S305; S3042, performing data fitting according to the atmospheric parameters in the thermal infrared remote sensing data at adjacent moments of the current sampling position, and solving the atmospheric parameters of the current data set; S3044. Calculate the thermal infrared spectrum radiation brightness received by the satellite sensor at the top of the atmosphere according to the measured water temperature data of the current data group and the atmospheric parameters of the current data group, as the thermal infrared remote sensing data of the current data group: δ(R t,i,j )=-γ(R t,i,j )·R t,i,j +T(R t,i,j ) Among them, R t,i,j is the thermal infrared spectrum radiation brightness of the jth sampling position in the ith river channel subsection at time t; T t,i,j is the measured water temperature data of the jth sampling location at time t in the i-th river channel subsection; δ(R t,i,j ) and γ(R t,i,j ) are intermediate variables determined by Planck’s formula; Temp t,1 Temp t,2 and Temp t,3 are atmospheric functions obtained based on the atmospheric parameters of the current data set; ε is the water body emissivity; T(R t,i,j ) is the brightness temperature detected by the sensor; c2 and c1 are constants; λ is the equivalent wavelength in the thermal infrared band.

6. The method for predicting river water temperature change trend based on remote sensing and water temperature inversion according to claim 1 is characterized in that: The step S4 is specifically as follows: S401, fitting the measured water temperature data of each river channel subsection at each time according to each group of water temperature inversion data sequences, to obtain a water temperature fitting line segment of each river channel subsection at each time; S402, according to each group of water temperature inversion data sequences, based on the water temperature fitting line segment of each river channel sub-segment at each time, the discrete points whose distance from the fitting line segment is greater than the discrete threshold are eliminated to obtain the valid data group of each river channel sub-segment at each time; S403, according to the valid data group of each river channel sub-section at each time, the measured water temperature data are summed and averaged to obtain the average water temperature of each river channel sub-section at each time.

7. The method for predicting river water temperature change trend based on remote sensing and water temperature inversion according to claim 1 is characterized in that: The step S5 is specifically as follows: S501. According to each group of water temperature inversion data sequences and the average water temperature of each river channel subsection at each time, the prediction error benchmark value of each river channel subsection at each time is solved: Among them, ε t,i is the prediction error benchmark value of the ith river channel subsection at time t; T t,i,j is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; is the average water temperature of the ith river section at time t; N j is the number of sampling locations; S502, according to each group of water temperature inversion data sequences, split the water temperature influencing factors to obtain several groups of training data: A1={t:[x=R t,i,j ;y=T t,i,j ]} Among them, A1 is the first set of training data; A2 is the second set of training data; A3 is the third set of training data; x is the independent variable; Q t,i (l t,i ) is subject to l i The hydrological data of the ith river channel subsection affected by the change at time t; l i is the length of the river from the center of the ith river section to the water conservancy project; y is the dependent variable; t is time; R t,i,j is the thermal infrared remote sensing data of the jth sampling position at time t in the i-th river channel subsection; T t,i,j Temp is the measured water temperature data of the jth sampling position at time t in the i-th river channel subsection; t is the temperature data at time t; is the parameter weight of thermal infrared remote sensing data, which is an unknown quantity; is the parameter weight of hydrological data, which is an unknown quantity; θ is the correction coefficient of thermal infrared remote sensing data, which is an unknown quantity; S503, according to the first set of training data, using BP neural network to iteratively solve the parameter weights of thermal infrared remote sensing data; S504, according to the parameter weights of the thermal infrared remote sensing data, based on the second set of training data, using the BP neural network to solve Q t,i (l i ) parameter weight and The correction coefficient θ; S505, according to Q t,i (l i ) parameter weight and The correction coefficient θ is solved iteratively using the BP neural network based on the third set of training data and the prediction error benchmark value of each river section at each time. The correction factor, Correction factor and Temp t The parameter weights are used to obtain the water temperature prediction model.

8. The method for predicting river water temperature change trend based on remote sensing and water temperature inversion according to claim 7 is characterized in that: The loss function of the water temperature prediction model in step S505 is: Among them, LOSS is the loss function of the water temperature prediction model; is the predicted water temperature at the jth sampling location in the ith river subsegment at time t.

9. The method for predicting river water temperature change trend based on remote sensing and water temperature inversion according to claim 7 is characterized in that: The objective function of the water temperature prediction model in step S505 is: O=min|LOSS-∑ i e t,i | Among them, O is the objective function of the water temperature prediction model; min is the minimum value; || is the absolute value; ε t,i is the prediction error benchmark value of the ith river channel subsegment at time t.

Images