Long-time series high-resolution estimation method of mountainous air temperature based on multi-source satellite remote sensing

Abstract

The present application belongs to the field of meteorological remote sensing and geographic information technology, and particularly relates to a mountainous area temperature long-time series high-resolution estimation method based on multi-source satellite remote sensing, comprising: obtaining multi-source satellite land surface temperature product data, meteorological station observation data and multi-dimensional auxiliary factor data; constructing a temperature estimation model based on different observation scenarios of the multi-source satellite land surface temperature product data, wherein the temperature estimation model is constructed based on a random forest regression model, and the multi-dimensional auxiliary factor data is input of the temperature estimation model; training the temperature estimation model by using the meteorological station observation data to generate multiple sets of temperature estimation initial value products; weighting and fusing the multiple sets of temperature estimation initial value products according to the deviation of each set of temperature estimation initial value product relative to the meteorological station observation data to generate fused temperature data; and performing spatiotemporal continuity interpolation filling on the gaps existing in the fused temperature data to obtain a spatiotemporal continuous mountainous area temperature data set.

Description

Technical Field

[0001] This invention belongs to the field of meteorological remote sensing and geographic information technology, and in particular relates to a long-term high-resolution estimation method for mountain temperature based on multi-source satellite remote sensing. Background Technology

[0002] Mountainous areas are an important part of the Earth system, serving as vital freshwater reservoirs and biodiversity hotspots, and are highly sensitive to climate change. Near-surface temperature (TA) is a fundamental variable for understanding the climate impacts of mountain ecosystems. However, due to the complex topography of mountains, existing high-altitude meteorological stations are sparse and unevenly distributed, making it difficult to capture intense microclimate changes over short distances. Although reanalysis products (such as ERA5-Land and MERRA-2) provide global coverage, their spatial resolution (typically 0.1° or coarser) is insufficient to meet the needs of refined mountain studies.

[0003] The main methods for obtaining mountain temperature data and the existing problems are as follows:

[0004] Meteorological station observations: Meteorological stations are sparse in high-altitude areas, making it difficult to capture complex spatial thermal changes.

[0005] Further analysis of the data reveals that products such as ERA5, while covering the globe, have relatively coarse spatial resolution (usually above 10km), making it impossible to accurately depict the microclimate of mountainous areas.

[0006] Satellite-based LST estimation: Satellite LST has high spatial resolution (e.g., 1 km) and is an effective proxy variable for estimating air temperature. Existing LST-TA conversion methods mainly include:

[0007] Temperature-Vegetation Index (TVX) method: This method assumes that the temperature of an infinitely thick vegetation canopy is close to the air temperature. Disadvantage: In sparsely vegetated areas (such as bare mountain terrain), its accuracy is greatly affected by soil moisture.

[0008] Physical models of surface energy balance require extensive parameter calibration and are difficult to apply in mountainous areas where data is scarce.

[0009] Establishing LST-TA relationships based on statistical regression or machine learning is the mainstream method for generating large-scale temperature products. However, existing technologies have the following main drawbacks:

[0010] Reliance on a single data source: Existing products typically rely on a single satellite sensor (such as MODIS), which is severely affected by cloud cover, resulting in a large amount of spatiotemporal gaps in cloudy mountainous areas, making it impossible to provide spatiotemporally continuous data.

[0011] Lack of multi-source fusion mechanisms: Although long-term LST observation data from different satellite platforms (such as ERS-2, ENVISAT, Sentinel-3, etc.) exist, significant differences in transit time, observation geometry, and sensitivity to land-atmosphere coupling among different sensors lead to systematic biases in the LST-temperature relationship between different source data. Current technologies lack effective strategies to integrate these complementary multi-source observation information; therefore, there is currently no global long-term mountain temperature dataset that can effectively combine multi-source satellite observations.

[0012] Insufficient utilization of diurnal differences: Existing methods often simply use the diurnal average LST without fully considering that the nighttime LST is usually more correlated with temperature than the daytime LST, and also without designing a hierarchical processing mechanism for cases where single day / night data is missing.

[0013] Therefore, there is an urgent need for a method that can integrate multi-source satellite observation data, solve the problem of missing cloud cover through a priority weighted fusion strategy, and generate a long-term, high-resolution (1km) and spatiotemporally continuous global mountain temperature estimation method. Summary of the Invention

[0014] To address the aforementioned technical challenges, this invention proposes a long-term, high-resolution estimation method for mountain air temperature based on multi-source satellite remote sensing. This method aims to construct nonlinear relationship models under different observation scenarios (dual day and night, nighttime only, and daytime only), and introduce multiple vegetation indices such as NDVI, EVI, SAVI, and DVI, as well as fine topographic factors such as altitude, slope, and aspect. Combined with a priority-based weighted fusion strategy and 3D spatiotemporal filling technology, it achieves continuous estimation of global mountain monthly average temperature with long-term, high precision, and high spatial resolution.

[0015] To achieve the above objectives, this invention provides a long-term, high-resolution estimation method for mountain air temperature based on multi-source satellite remote sensing, comprising:

[0016] Acquire multi-source satellite surface temperature product data, meteorological station observation data, and multi-dimensional auxiliary factor data;

[0017] Based on different observation scenarios of the multi-source satellite surface temperature product data, an air temperature estimation model is constructed. The air temperature estimation model is constructed based on a random forest regression model, and the multidimensional auxiliary factor data is the input of the air temperature estimation model.

[0018] The temperature estimation model is trained using the observation data from the meteorological stations to generate multiple sets of initial temperature estimation values.

[0019] Based on the deviation of each set of initial temperature estimates from meteorological station observation data, the multiple sets of initial temperature estimates are weighted and fused to generate fused temperature data.

[0020] The gaps in the fused temperature data are filled by spatiotemporal continuity interpolation to obtain a spatiotemporally continuous mountain temperature dataset.

[0021] Optionally, acquiring multi-source satellite surface temperature product data includes:

[0022] Obtain at least two land surface temperature products from ERS-2, ENVISAT, TERRA MODIS, AQUA MODIS, Sentinel-3A, and multi-sensor fusion products;

[0023] The acquired surface temperature products are uniformly resampled to the target spatial resolution to obtain the multi-source satellite surface temperature product data.

[0024] Optionally, the multidimensional auxiliary factor data includes vegetation index data and topographic factor data;

[0025] The vegetation index data is obtained by calculating the normalized vegetation index, enhanced vegetation index, soil-regulated vegetation index, and difference vegetation index.

[0026] The terrain factor data is obtained by extracting information on altitude, slope, aspect, and topographic relief from digital elevation model data.

[0027] Optionally, the vegetation index data is obtained by calculating the normalized vegetation index, enhanced vegetation index, soil-regulated vegetation index, and difference vegetation index, including:

[0028] The calculation of the Normalized Difference Vegetation Index (NDVI) includes:

[0029] ;

[0030] Calculating the Enhanced Vegetation Index (EVI) includes:

[0031] ;

[0032] Where G is the gain coefficient, and C1 and C2 are the coefficients of the reflectivity values ​​of the red band and blue band, respectively;

[0033] The calculation of the Soil Moderated Vegetation Index (SAVI) includes:

[0034] ;

[0035] in, Soil regulators;

[0036] The calculation of the Difference Vegetation Index (DVI) includes:

[0037] ;

[0038] in, , , These are the reflectance values ​​for the near-infrared band, red band, and blue band, respectively.

[0039] Optionally, the terrain factor data is obtained by extracting elevation, slope, aspect, and topographic relief information from the digital elevation model data, including:

[0040] Extracting the slope includes:

[0041] ;

[0042] Extracting the slope aspect includes:

[0043] ;

[0044] in, This represents the rate of change of elevation z along the east-west direction x. This represents the rate of change of elevation z along the north-south direction y;

[0045] Extracting the terrain relief TRI includes:

[0046] ;

[0047] in, Elevation of eight neighboring pixels. Elevation of the center pixel.

[0048] Optionally, the random forest regression model is constructed independently for each season, with model parameters established for spring, summer, autumn and winter respectively.

[0049] Optionally, based on different observation scenarios of the multi-source satellite surface temperature product data, the temperature estimation model can be constructed as follows:

[0050] The different observation scenarios include a day and night observation scenario, a night-only observation scenario, and a day-only observation scenario;

[0051] For the aforementioned day and night observation scenario, an air temperature estimation model is constructed that simultaneously utilizes nighttime and daytime surface temperatures;

[0052] For the scenario of observations only at night, an air temperature estimation model is constructed that uses only nighttime surface temperature;

[0053] For the daytime-only observation scenario, an air temperature estimation model is constructed that uses only daytime surface temperature.

[0054] Optionally, weighted fusion of the multiple sets of initial temperature estimates includes:

[0055] The weighted fusion prioritizes the initial temperature estimation product corresponding to the day and night dual-observation scenario;

[0056] When products for both day and night observation scenarios are unavailable, the initial temperature estimation product corresponding to only nighttime observation scenarios is used.

[0057] When both day and night observation scenarios and night-only observation scenario products are unavailable, the initial temperature estimation product corresponding to the day-only observation scenario is used.

[0058] Compared with the prior art, the present invention has the following advantages and technical effects:

[0059] 1. Significantly improved spatiotemporal coverage: By fusing multi-source satellite data, the problem of data loss caused by cloud and rain weather affecting a single sensor is effectively solved, making it especially suitable for mountainous environments with frequent cloud and fog.

[0060] 2. High estimation accuracy: This invention extends the introduction of multiple vegetation indices such as EVI and SAVI, as well as multi-dimensional topographic factors such as slope and aspect, which can more accurately reflect the nonlinear thermal processes of complex mountain surfaces than using NDVI and altitude alone.

[0061] 3. An uncertainty-based weighted fusion strategy is adopted, and results containing nighttime information are given higher priority, making full use of the physical characteristics that the nighttime LST is more strongly correlated with temperature.

[0062] 4. Strong product applicability: The generated dataset has the characteristics of long time series (spanning more than 20 years) and high resolution (1km), which can meet the needs of refined research such as mountain climate change analysis and eco-hydrological simulation. Attached Figure Description

[0063] The accompanying drawings, which form part of this application, are used to provide a further understanding of this application. The illustrative embodiments and descriptions of this application are used to explain this application and do not constitute an undue limitation of this application. In the drawings:

[0064] Figure 1 This is a flowchart of a long-term high-resolution estimation method for mountain air temperature based on multi-source satellite remote sensing, according to an embodiment of the present invention. Detailed Implementation

[0065] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0066] It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions, and although a logical order is shown in the flowchart, in some cases the steps shown or described may be executed in a different order than that shown here.

[0067] This embodiment proposes a long-term, high-resolution estimation method for mountain air temperature based on multi-source satellite remote sensing, such as... Figure 1 As shown, the specific steps include:

[0068] Acquire multi-source satellite surface temperature product data, meteorological station observation data, and multi-dimensional auxiliary factor data;

[0069] Based on different observation scenarios of the multi-source satellite surface temperature product data, an air temperature estimation model is constructed. The air temperature estimation model is constructed based on a random forest regression model, and the multidimensional auxiliary factor data is the input of the air temperature estimation model.

[0070] The temperature estimation model is trained using the observation data from the meteorological stations to generate multiple sets of initial temperature estimation values.

[0071] Based on the deviation of each set of initial temperature estimates from meteorological station observation data, the multiple sets of initial temperature estimates are weighted and fused to generate fused temperature data.

[0072] The gaps in the fused temperature data are filled by spatiotemporal continuity interpolation to obtain a spatiotemporally continuous mountain temperature dataset.

[0073] Specifically, multi-source LST data acquisition involves acquiring multiple time-series land surface temperature (LST) products from different sensors (such as ERS-2, ENVISAT, TERRA MODIS, AQUA MODIS, Sentinel-3A, and multi-sensor fusion products released by ESA CCI) and uniformly resampling them to the target resolution (e.g., 0.01°, approximately 1km).

[0074] Site data acquisition: Acquire near-surface air (TA) observation data from high-altitude meteorological stations within the study area.

[0075] Auxiliary variable expansion extraction:

[0076] Multidimensional vegetation parameter extraction: In addition to using the Normalized Difference Vegetation Index (NDVI), we further extracted multiple vegetation parameters such as the Enhanced Vegetation Index (EVI), Soil-Adjusted Vegetation Index (SAVI), and Difference Vegetation Index (DVI) to more comprehensively characterize the differences in thermal response under different vegetation covers in mountainous areas.

[0077] Multidimensional terrain factor extraction: Based on the digital elevation model (DEM), in addition to extracting elevation, factors such as slope, aspect, and topographic relief are also calculated to correct the nonlinear influence of terrain on temperature through a random forest model.

[0078] Digital Elevation Model (DEM) is essentially equivalent to altitude in some respects, which can be disregarded here; it can be understood as a type of altitude data.

[0079] Furthermore, acquiring multi-source satellite surface temperature product data includes:

[0080] Obtain at least two land surface temperature products from ERS-2, ENVISAT, TERRA MODIS, AQUA MODIS, Sentinel-3A, and multi-sensor fusion products;

[0081] The acquired surface temperature products are uniformly resampled to the target spatial resolution to obtain the multi-source satellite surface temperature product data.

[0082] Furthermore, the multidimensional auxiliary factor data includes vegetation index data and topographic factor data;

[0083] The vegetation index data is obtained by calculating the normalized vegetation index, enhanced vegetation index, soil-regulated vegetation index, and difference vegetation index.

[0084] The terrain factor data is obtained by extracting information on altitude, slope, aspect, and topographic relief from digital elevation model data.

[0085] Furthermore, the vegetation index data is obtained by calculating the Normalized Difference Vegetation Index (NDVI), Enhanced Vegetation Index (EDI), Soil-Regulated Vegetation Index (SDI), and Difference Vegetation Index (DVI), including:

[0086] Normalized Difference Vegetation Index (NDVI) is used to reflect vegetation growth status and cover. Calculation of NDVI includes:

[0087] ;

[0088] Enhanced Vegetation Index (EVI): Introduces the blue light band to reduce the impact of atmospheric aerosols and corrects for canopy background signals, improving sensitivity in high biomass areas and overcoming the problem of NDVI saturation. Calculation of the Enhanced Vegetation Index includes:

[0089] ;

[0090] in, , , These are the reflectance values ​​for the near-infrared, red, and blue bands, respectively. The coefficients are typically set to a value of [value missing]. .

[0091] Soil-modified vegetation index (SAVI): Based on NDVI, soil modifiers are introduced. This method reduces the impact of soil background brightness and is particularly suitable for sparsely vegetated mountainous environments. Calculating the soil-modified vegetation index includes:

[0092] ;

[0093] in, The soil conditioning factor is typically set to 0.5;

[0094] Difference Vegetation Index (DVI): Directly reflects the absolute amount of vegetation cover change and is relatively sensitive to soil background. Calculation of DVI includes:

[0095] .

[0096] Furthermore, the terrain factor data is obtained by extracting elevation, slope, aspect, and topographic relief information from the digital elevation model data, including:

[0097] Slope: Reflects the steepness of a land surface unit. Extracting the slope includes:

[0098] ;

[0099] Slope aspect: Reflects the direction of the projection of the surface normal onto the horizontal plane, determining the amount of sunlight received by the slope (with true north as 0 degrees, clockwise direction). Extracting the slope aspect includes:

[0100] ;

[0101] in, This represents the rate of change of elevation z along the east-west direction x. This represents the rate of change of elevation z along the north-south y direction. This represents the rate of change of elevation y along the east-west direction x;

[0102] Topographic relief (TRI): Reflects the degree of elevation variation and roughness within a local window (using the root mean square error model). Extraction of the topographic relief includes:

[0103] ;

[0104] in, Elevation of eight neighboring pixels. Elevation of the center pixel.

[0105] Furthermore, the temperature estimation model is constructed based on a random forest regression model.

[0106] Specifically, regarding the random forest model here, only the construction process is described, but the calibration process of the random forest model itself does not need to be considered. This is because the random forest model is a black box model; it automatically captures the nonlinear relationship between terrain and temperature. The model's consideration of this nonlinear relationship also achieves calibration in another sense.

[0107] Build process:

[0108] Sample resampling: For each decision tree, the Bootstrap method is used to extract samples with replacement from the original training set to construct a differentiated training subset.

[0109] Random feature selection: When splitting at each node of the decision tree, a subset of features (such as LST, NDVI, altitude, slope, etc.) are randomly selected as the candidate feature set.

[0110] Node splitting and growth: Traverse candidate features and their splitting points, calculate mean squared error (MSE) or purity gain, select the optimal feature for node splitting, until the stopping condition (such as maximum depth or minimum number of leaf node samples) is met.

[0111] Integrated output: The final temperature forecast is determined by the average of all decision tree forecasts.

[0112] Model expression: Let the input feature vector be... The temperature estimation model can then be expressed as:

[0113] ;

[0114] in, The total number of decision trees, Indicates the first A decision tree, For the first The independent and identically distributed random vectors of the trees (which determine the sample sampling and feature selection of the tree).

[0115] Furthermore, the random forest regression model is constructed independently for each season, with model parameters established for spring, summer, autumn, and winter respectively.

[0116] Furthermore, based on different observation scenarios of the multi-source satellite surface temperature product data, the temperature estimation model is constructed as follows:

[0117] The different observation scenarios include a day and night observation scenario, a night-only observation scenario, and a day-only observation scenario;

[0118] For the aforementioned day and night observation scenario, an air temperature estimation model is constructed that simultaneously utilizes nighttime and daytime surface temperatures;

[0119] For the scenario of observations only at night, an air temperature estimation model is constructed that uses only nighttime surface temperature;

[0120] For the daytime-only observation scenario, an air temperature estimation model is constructed that uses only daytime surface temperature.

[0121] Specifically, for each LST product, the scenario-based LST-TA modeling constructs a Random Forest (RF) regression model for three observation scenarios based on the validity of daytime and nighttime LST data at the pixel scale.

[0122] Regarding the specific technical aspects of the model:

[0123] Seasonal Dynamic Modeling Strategy: To accurately capture the differences in thermal dynamics across different seasons, this invention does not construct a single global model, but rather independently constructs LST-TA relationship models for each season. Specifically, it divides the seasons into spring (March-May), summer (June-August), autumn (September-November), and winter (December-February). A corresponding RF model is trained for each season to adapt to seasonal fluctuations in vegetation phenology and solar radiation intensity.

[0124] Multi-source data maximization strategy: To maximize the use of satellite observation information when constructing the training set, this invention employs an inclusive data utilization strategy: When constructing the "daytime-only observation model," not only are samples with data available only during the daytime used, but the daytime LST component and its corresponding station data from the "dual observation scenario" are also included in the training set. Similarly, when constructing the "nighttime-only observation model," the nighttime LST component from the "dual observation scenario" is also included. This strategy significantly increases the training sample size of the single-observation-scenario model and improves the model's generalization ability.

[0125] 5-fold Cross-Validation and Model Optimization: A 5-fold cross-validation strategy is employed for model training and evaluation. The entire site observation dataset is randomly divided into 5 subsets, with 4 subsets used alternately as the training set and the remaining subset as the validation set. Key parameters of the random forest (such as the number of decision trees) are optimized by minimizing the root mean square error (RMSE) of the validation set. (e.g., minimum number of samples for node splitting) to ensure the model's predictive stability on unseen sites.

[0126] Three scenarios:

[0127] Dual observation scenario (Day+Night): When both daytime and nighttime LSTs are valid, the model is constructed as follows: ;

[0128] Day-only observation scenario: When LST is only valid during the day, the model is constructed as follows: ;

[0129] Night-only observation scenario: When LST is only valid at night, the model is constructed as follows: ;

[0130] in, , , These are the estimated monthly average temperatures under scenarios of dual observation, nighttime observation only, and daytime observation only, respectively. For random forest regression function; This refers to the daytime surface temperature. This refers to the surface temperature at night. , , These are altitude, slope, and aspect, respectively. , These are latitude and longitude, respectively.

[0131] The above model was trained and cross-validated using meteorological station data to generate initial temperature estimates for each LST product.

[0132] Furthermore, the weighted fusion of the multiple sets of initial temperature estimates includes:

[0133] The weighted fusion prioritizes the initial temperature estimation product corresponding to the day and night dual-observation scenario;

[0134] When products for both day and night observation scenarios are unavailable, the initial temperature estimation product corresponding to only nighttime observation scenarios is used.

[0135] When both day and night observation scenarios and night-only observation scenario products are unavailable, the initial temperature estimation product corresponding to the day-only observation scenario is used.

[0136] Specifically, a priority-based weighted fusion method integrates the temperature estimates from multiple sources generated in step two. Considering that nighttime LST is typically more correlated with temperature, and that the dual-observation scenario contains the most comprehensive information, the following fusion priorities are set:

[0137] First priority: Estimation results for dual-observation scenarios;

[0138] Second priority: Estimation results for nighttime observation scenarios only;

[0139] Third priority: Estimation results for daytime observation scenarios only.

[0140] The fusion formula uses a weighted average method, with the weights determined based on the bias of each LST product. The smaller the bias, the greater the weight, thus generating the fused temperature data (FusTA).

[0141] The specific details regarding the weighted fusion formula are as follows.

[0142] For any pixel, the weighted average of the blended temperatures is calculated using the following formula:

[0143] ;

[0144] Where, N DN , N N , and N D These represent the estimated TA for the six products. DN , TA N , and TA D Total number of available cells; TA DN,i , TA N,i , and TA D,i TA obtained from the i-th TA product DN , TA N , and TA D Value correspondence; W DN,i W N,i , and W D,k The dynamic weights are represented by the following formula:

[0145] ;

[0146] ;

[0147] ;

[0148] Among them, δD,i and δ N,i These represent the absolute deviations of diurnal and nighttime surface temperatures in the i-th product, respectively. These deviation values ​​are provided in the form of a BIAS layer in the original surface temperature product. If a pixel has no valid TA estimate, it should be assigned a missing value. This strategy enhances the robustness of fused GMTA products by effectively utilizing available thermal information from satellite observations while reducing the uncertainty caused by insufficient observation coverage.

[0149] Filling the gaps in time and space:

[0150] For the few data gaps that still exist after fusion, a 3D smoothing algorithm (DCT-PLS, Discrete Cosine Transform Penalized Least Squares) is used to fill the gaps by interpolation using the spatiotemporal continuity characteristics of the data, and finally a spatiotemporally seamless mountain temperature dataset is generated.

[0151] A detailed analysis was conducted using a global mountain monthly mean temperature dataset from 1997 to 2020:

[0152] Data preparation:

[0153] LST data sources: Six LST products provided by ESA CCI were selected, including ERS-2 (1997–2003), ENVISAT (2002–2012), MODIS TERRA (2000–2018), MODIS AQUA (2002–2018), Sentinel-3A (2016–2020), and a multi-sensor fusion product (1997–2020).

[0154] Ground observations: Monthly average temperature data were collected from 6,842 high-altitude meteorological stations worldwide, including C3S, NOAA, and CMDC.

[0155] Auxiliary data expansion:

[0156] Multiple vegetation indices, such as NDVI, EVI, SAVI, and DVI, were calculated using MODIS reflectance data to construct a multidimensional feature space for vegetation conditions.

[0157] Based on ETOPO or SRTM DEM data, calculate elevation, slope, and aspect to construct a terrain feature space.

[0158] Model building and training:

[0159] For the aforementioned six LST products, random forest regression models were constructed for each. Taking the "dual-observation scenario" as an example, the input feature vector is... The target variable is the measured temperature at the station. ,in, The measured monthly average temperature at the station is used. The model parameters were optimized using the 5-fold cross-validation method, which not only utilized the thermal infrared information of LST, but also corrected for the influence of the complex underlying surface of the mountainous area by extended vegetation and topographic factors.

[0160] Priority-weighted fusion:

[0161] For a specific pixel and a specific month, temperature is integrated. The calculation formula is as follows:

[0162] ;

[0163] Among them, weight It is determined by the reciprocal of the absolute deviation (Bias) of the data for each product and site. The sum of the weights of all products participating in the integration (i.e.) In this embodiment, the data source containing the nighttime LST is set to be used preferentially. and This is because experiments have shown that the thermal coupling between the Earth's surface and the atmosphere is more stable at night.

[0164] 3D Gap Filling:

[0165] For any missing pixels remaining after fusion (e.g., all satellites are obscured by clouds in the current month), 3D smoothing is performed using DCT-PLS (Discrete Cosine Transform Penalized Least Squares). This method utilizes the continuity of temperature in adjacent spatial locations and adjacent time months to reconstruct missing values, ultimately outputting a complete dataset of global mountain monthly average temperatures at 1km resolution from 1997 to 2020.

[0166] The above are merely preferred embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A high-resolution long-time series estimation method for mountain air temperature based on multi-source satellite remote sensing, characterized in that, include: Acquire multi-source satellite surface temperature product data, meteorological station observation data, and multi-dimensional auxiliary factor data; The multidimensional auxiliary factor data includes vegetation index data and topographic factor data; The vegetation index data is obtained by calculating the normalized vegetation index, enhanced vegetation index, soil-regulated vegetation index, and difference vegetation index. The terrain factor data is obtained by extracting elevation, slope, aspect, and terrain relief information from digital elevation model data; Based on different observation scenarios of the multi-source satellite surface temperature product data, an air temperature estimation model is constructed. The air temperature estimation model is constructed based on a random forest regression model, and the multidimensional auxiliary factor data is the input of the air temperature estimation model. The temperature estimation model is trained using the observation data from the meteorological stations to generate multiple sets of initial temperature estimation values. Based on the deviation of each set of initial temperature estimates from meteorological station observation data, the multiple sets of initial temperature estimates are weighted and fused to generate fused temperature data. The gaps in the fused temperature data are filled by spatiotemporal continuity interpolation to obtain a spatiotemporally continuous mountain temperature dataset.

2. The method for long-term high-resolution estimation of mountain air temperature based on multi-source satellite remote sensing according to claim 1, characterized in that, Acquiring multi-source satellite surface temperature product data includes: Obtain at least two land surface temperature products from ERS-2, ENVISAT, TERRA MODIS, AQUA MODIS, Sentinel-3A, and multi-sensor fusion products; The acquired surface temperature products are uniformly resampled to the target spatial resolution to obtain the multi-source satellite surface temperature product data.

3. The method for long-term high-resolution estimation of mountain air temperature based on multi-source satellite remote sensing according to claim 1, characterized in that, The vegetation index data are obtained by calculating the normalized vegetation index, enhanced vegetation index, soil-regulated vegetation index, and difference vegetation index, including: The calculation of the Normalized Difference Vegetation Index (NDVI) includes: ； Calculating the Enhanced Vegetation Index (EVI) includes: ； Where G is the gain coefficient, and C1 and C2 are the coefficients of the reflectivity values ​​of the red band and blue band, respectively; The calculation of the Soil Moderated Vegetation Index (SAVI) includes: ； in, Soil regulators; The calculation of the Difference Vegetation Index (DVI) includes: ； in, , , These are the reflectance values ​​for the near-infrared band, red band, and blue band, respectively.

4. The method for long-term high-resolution estimation of mountain air temperature based on multi-source satellite remote sensing according to claim 1, characterized in that, The terrain factor data is obtained by extracting elevation, slope, aspect, and topographic relief information from digital elevation model data, including: Extracting the slope includes: ； Extracting the slope aspect includes: ； in, This represents the rate of change of elevation z along the east-west direction x. This represents the rate of change of elevation z along the north-south y direction. This represents the rate of change of elevation y along the east-west direction x; Extracting the terrain relief TRI includes: ； in, Elevation of eight neighboring pixels. Elevation of the center pixel.

5. The method for long-term high-resolution estimation of mountain air temperature based on multi-source satellite remote sensing according to claim 1, characterized in that, The random forest regression model is constructed independently for each season, with model parameters established for spring, summer, autumn, and winter respectively.

6. The method for long-term high-resolution estimation of mountain air temperature based on multi-source satellite remote sensing according to claim 5, characterized in that, Based on different observation scenarios of the multi-source satellite surface temperature product data, the temperature estimation model is constructed as follows: The different observation scenarios include a day and night observation scenario, a night-only observation scenario, and a day-only observation scenario; For the aforementioned day and night observation scenario, an air temperature estimation model is constructed that simultaneously utilizes nighttime and daytime surface temperatures; For the scenario of observations only at night, an air temperature estimation model is constructed that uses only nighttime surface temperature; For the daytime-only observation scenario, an air temperature estimation model is constructed that uses only daytime surface temperature.

7. The method for long-term high-resolution estimation of mountain air temperature based on multi-source satellite remote sensing according to claim 6, characterized in that, The weighted fusion of the multiple sets of initial temperature estimates includes: The weighted fusion uses the initial temperature estimation product corresponding to the day and night dual-observation scenario; When products for both day and night observation scenarios are unavailable, the initial temperature estimation product corresponding to only nighttime observation scenarios is used. When both day and night observation scenarios and night-only observation scenario products are unavailable, the initial temperature estimation product corresponding to the day-only observation scenario is used.

Images