CN-121997232-A - Surface temperature anomaly detection and attribution analysis method based on thermal infrared data

CN121997232ACN 121997232 ACN121997232 ACN 121997232ACN-121997232-A

Abstract

The invention relates to the technical field of geothermal detection, in particular to a ground surface temperature anomaly detection and attribution analysis method based on thermal infrared data, which comprises the following steps of S1, collecting MODIS thermal infrared bright temperature data to form a sequence input sample of each city, S2, constructing a prediction model, training the prediction model to obtain a reconstruction error and a regression residual error of the prediction model, S3, obtaining a combined anomaly score of each month of each city, setting a judgment threshold, indicating that the corresponding month is an anomaly month when the combined anomaly score is larger than the judgment threshold, otherwise, obtaining a plurality of core variables for each anomaly month of each city, carrying out standardization processing and multiple linear regression on the plurality of core variables, quantifying the contribution degree of each core variable to temperature anomaly, improving the ground surface temperature prediction precision and the accuracy of ground surface temperature anomaly screening, and carrying out quantitative analysis on anomaly temperature causes.

Inventors

WANG FANG
YUAN GUOLIANG
Peng Peisong
ZHUO XIAO
MA QIAN
SHI XUMING
Dai fuxing
JIANG RUIQI
HU WEIDA

Assignees

中国科学院上海技术物理研究所

Dates

Publication Date: 20260508
Application Date: 20260127

Claims (10)

1. The earth surface temperature anomaly detection and attribution analysis method based on the thermal infrared data is characterized by comprising the following steps of: S1, acquiring MODIS thermal infrared bright temperature data, dividing according to four wave bands of the MODIS, and forming a sequence input sample of each city; s2, constructing a prediction model, wherein the constructed prediction model structure comprises an LSTM coder, a pooling module, an LSTM decoder and a regression head, processing a sequence input sample to obtain a time sequence data set, training the prediction model by using the time sequence data set, and thus obtaining a learned prediction model, and obtaining a reconstruction error and a regression residual error of the prediction model; S3, obtaining a combined abnormal score of each month of each city according to the reconstruction error and the regression residual error of the prediction model, setting a judgment threshold, comparing the combined abnormal score with the judgment threshold, and indicating that the corresponding month is an abnormal month when the combined abnormal score is larger than the judgment threshold, otherwise, the corresponding month is a normal month; s4, screening out the abnormal months and the cities corresponding to the abnormal months judged in the step S3, obtaining a plurality of core variables for each abnormal month of each city, forming a local time window in the abnormal month and k months before the abnormal month, and then carrying out standardized processing and multiple linear regression on the plurality of core variables to quantify the contribution degree of each core variable to temperature abnormality.
2. The method for detecting and analyzing surface temperature anomalies and attributions based on thermal infrared data according to claim 1, wherein S2 specifically comprises the steps of: s21, setting hidden layer dimension of the LSTM model, and for any time step t, the LSTM model firstly uses the hidden state of the last moment Computing forgetting gate together with current input x Input door And output door Forgetting door Input door And output door All are processed by Sigmoid activating function, and the value range is between 0 and 1 For determining the memory cell at the previous time Dimension to be preserved, input gate For determining candidate memories Write to the current memory cell, output gate The method is used for controlling the degree to which the memory state after nonlinear transformation is exposed to be a hidden state; S22, updating the memory unit and the hidden state, and outputting a final memory unit and a final hidden state; s23, final memory unit of encoder output And a final hidden state After passing through the pooling module, the data are input into the regression head and decoder at the same time, the pair in the regression head 、 Obtaining a predicted value of the average temperature of the month after treatment At the same time, the true value y of the month average temperature is obtained, and the decoder will 、 As the initial state of the decoder, the original 12 month input sequence is reconstructed by taking the zero vector as the decoding input and generating the sequence step by step to obtain the reconstructed sequence Obtaining the real sequence Predicted value according to average temperature of month True value y of month average temperature, and reconstruction sequence True sequence Obtaining an overall optimization target of the prediction model in the training process; S24, obtaining a month average temperature predicted value in the recognition stage by using the predicted model trained in the step S2 True value of average temperature per month Reconstruction sequence True sequence And obtaining a reconstruction error and a regression residual error.
3. The method for detecting and analyzing abnormal surface temperature based on thermal infrared data according to claim 2, wherein in step S22, the memory unit is updated by the following formula: ; In the above-mentioned method, the step of, Representing the memory cell corresponding to time step t.
4. The method for detecting and analyzing surface temperature anomalies based on thermal infrared data according to claim 3, wherein in step S22, the hidden state is updated by: ; In the above-mentioned method, the step of, And representing the hidden state corresponding to the time step t.
5. The method for detecting and analyzing surface temperature anomalies and attributions based on thermal infrared data according to claim 2, wherein S23 specifically comprises the steps of: s231, according to the reconstruction sequence True sequence Obtaining reconstruction loss; S232, predicting value according to average temperature of month Calculating the temperature regression loss of the corresponding month according to the true value y of the month average temperature; S233, obtaining an overall optimization target of the model in the training process according to the reconstruction loss and the temperature regression loss, and calculating by the following formula: ; In the above-mentioned method, the step of, Representing the overall optimization objective(s), Indicating a loss of reconstruction and, 。
6. The method for detecting and attributing analysis to surface temperature anomaly based on thermal infrared data according to claim 5, wherein the reconstruction loss and the temperature regression loss are specifically calculated by the following formula: ; ; In the above-mentioned method, the step of, Representing a time step.
7. The method for detecting and analyzing surface temperature anomalies and attributions based on thermal infrared data according to claim 2, wherein S24 specifically comprises the steps of: s241, according to the reconstruction sequence True sequence The reconstruction error of the LSTM decoder is obtained and is used for describing the structural deviation degree of the sequence and the history mode, and the reconstruction error is calculated by the following formula: ; In the above-mentioned method, the step of, Representing a reconstruction error; s242, calculating a regression residual error by the following formula: ; In the above-mentioned method, the step of, Representing the regression residual.
8. The method for detecting and analyzing surface temperature anomalies and attributions based on thermal infrared data as set forth in claim 6, wherein S3 comprises the steps of: s31, carrying out normalization processing on the reconstruction error and the regression residual error, and specifically carrying out the normalization processing according to the following formula: ; ; In the above-mentioned method, the step of, Representing the reconstructed error after normalization, Representation of The standard deviation on the training set is used, Representing the normalized regression residual of the model, Representation of Standard deviation on the training set; S32, calculating a combined anomaly score according to the normalized reconstruction error and the normalized regression residual, and specifically calculating by the following formula: ; In the above-mentioned method, the step of, Representing the combined anomaly score(s), A first weight is indicated and a second weight is indicated, Representing a second weight, the first weight and the second weight adding to 1; s33, setting a judgment threshold, wherein the acquisition method of the judgment threshold comprises the following steps of processing an average value of joint scores corresponding to samples which are most deviated from a normal mode by 5% in a training set to obtain the judgment threshold: ; In the above-mentioned method, the step of, Represents a judgment threshold value, The number of bits representing 95% of the number of bits, Mean value of the joint scores corresponding to the samples of the training set that deviate most from the normal mode by 5%; S34, judging whether the average month temperature of the month in the city is an abnormal month or a normal month.
9. The method for detecting and analyzing abnormal surface temperature based on thermal infrared data according to claim 1, wherein in step S4, the core variables include population density, NDVI, and evaporation capacity, wherein population density is obtained for the whole year of the corresponding city, and corresponding values of NDVI and evaporation capacity are obtained for the abnormal month.
10. The method for detecting and analyzing surface temperature anomalies and attribution based on thermal infrared data according to claim 9, wherein S4 specifically comprises the steps of: S41, for each abnormal month, constructing an analysis window expressed as: , a set of temperature values representing the ith abnormal month; S42, for each abnormal month, in the analysis window In, construction comprises multiple linear regression expressed as: ; In the above-mentioned method, the step of, Represents the normalized temperature, P represents the normalized population density, The amount of evaporation is normalized and is indicated, The expression is given for a standardized NDVI, The term of the constant is represented by a term, 、、、、、 All of which represent the regression coefficients of the model, Representing an error term; S43, calculating the contribution quantity of each core variable according to multiple linear regression, wherein the larger the contribution quantity is, the larger the contribution of the core variable to temperature abnormality is, and the contribution quantity is specifically calculated by the following formula: ; In the above-mentioned method, the step of, Representing the contribution of the jth core variable, j takes 1 to 3, Representing the regression coefficients of the jth core variable in the multiple regression model, The regression coefficients showing the first variable are represented, where l ranges from 1 to q, q representing the total number of core variables incorporated into the regression model.

Description

Surface temperature anomaly detection and attribution analysis method based on thermal infrared data Technical Field The invention relates to the technical field of geothermal detection, in particular to a method for detecting and attributing to an abnormal earth surface temperature based on thermal infrared data. Background The surface temperature is taken as an important physical parameter reflecting the surface energy exchange condition, can intuitively represent the urban heat environment characteristics, and is a key index for carrying out urban heat island effect evaluation and space diversity analysis. Related researches show that the earth surface temperature information obtained by utilizing the remote sensing means has important application value in urban scale thermal environment monitoring, abnormal region identification and environment response analysis. Especially, on the area scale and the long-time scale, the construction of a stable and continuous surface temperature sequence based on thermal infrared remote sensing data has important significance for revealing the evolution rule of the urban thermal environment. Currently, the acquisition of surface temperature relies primarily on multi-source earth observation satellite data. The medium resolution imaging spectrometer MODIS can provide basic data support for surface temperature inversion and atmospheric parameter estimation by setting a plurality of thermal infrared surface measurement wave bands. The formed surface temperature product has good continuity and consistency in time scale and space coverage, and is suitable for long-term thermal environment monitoring research in regional and even global scale. In addition to MODIS, data from the land satellite (Landsat) series and the sentencel satellite (Sentinel) are also widely used in surface environmental research. However, different satellite systems differ in terms of spatial resolution, revisit period, and sensor design, making their applicability in long-time sequence analysis different. In the aspect of a surface temperature analysis method, the prior art mainly focuses on inversion and anomaly identification of the surface temperature driven by a physical model or a statistical method. The method generally depends on a radiation transmission model, a single window algorithm or an empirical formula, has strong dependence on intermediate parameters such as earth surface emissivity, atmospheric water vapor content and the like, has a complex parameter acquisition process, and is easy to accumulate in multiple links. In addition, along with the continuous increase of the time span and the data volume of remote sensing observation data, the traditional method is difficult to combine the calculation efficiency and the time sequence structure expression capability when processing long-time sequence data. In recent years, machine learning and deep learning methods are gradually introduced into the field of remote sensing time series analysis to mine time series patterns and potential change features in data. However, the existing partial model still has a certain limitation in terms of long-sequence modeling capability, parameter stability or computational complexity, and it is difficult to ensure sequential continuity expression and simultaneously consider wide-range application requirements. The related patent documents: The publication No. CN120654148A, publication No. 2025.09.16 discloses a geothermal anomaly point identification system based on remote sensing, which comprises a data layer, a processing layer, an analysis layer and an output layer, wherein the data layer is used for collecting thermal infrared remote sensing data, multispectral/hyperspectral remote sensing data, radar data, geological and topographic data, meteorological data and geophysical prospecting data, the processing layer is used for extracting key features, capturing dynamic changes of geothermal anomalies, distinguishing continuous geothermal anomalies from transient events and combining hyperspectral data with magnetotelluric sounding data to construct the earth surface The analysis layer is used for learning the diffusion rule of geothermal anomalies along a fracture zone through the graph propagation model, utilizing a physical information neural network to constraint a geothermal Tian Re conduction equation as a loss function, jointly training a temperature field prediction model, constructing a deep three-dimensional geological structure based on gravity and geophysical prospecting data, constructing a geothermal anomaly detection model through the characteristics provided by the processing layer to identify geothermal anomalies, and the output layer is used for generating a geothermal anomaly probability graph. The prior art represented by the foregoing documents has at least the following technical problems or drawbacks that have not been solved: In the prior art represented by CN120654148A, the metho