Soil temperature inversion method and system considering hysteresis effect and characteristic importance

Abstract

The present invention belongs to the technical field of soil temperature inversion, and relates to a soil temperature inversion method and system that considers hysteresis effects and feature importance. Construct a basic input data set for the area to be inverted; construct a hysteresis structure input set of multi-source driving variables, and generate several time-lagged versions of input variables; evaluate the feature importance of each input variable; screen a set of dominant variables; reconstruct the final input set to train an optimized soil temperature inversion model; obtain the soil temperature estimation result at the target time in the target area; use verification data to evaluate the accuracy of the optimized soil temperature inversion model and adjust the hysteresis structure of the input variables. The present invention significantly improves the inversion accuracy by introducing an input mechanism of hysteresis structure; introduces a feature importance evaluation mechanism, dynamically selects key variables as input features, and realizes efficient modeling in variable-constrained scenarios, thereby improving the flexibility and generalization ability of the model under different surface types and meteorological conditions.

Description

Technical Field

[0001] The present invention belongs to the technical field of soil temperature inversion, and in particular relates to a soil temperature inversion method and system considering hysteresis effect and characteristic importance. Background Art

[0002] Soil temperature is crucial for crop growth, ecosystem stability, and climate research. Currently, there are two main ways to obtain soil temperature: ground-based observations and remote sensing inversion. Ground-based observations provide relatively accurate data, but their spatial coverage is limited, making it difficult to fully reflect the distribution of soil temperature in large-scale regional monitoring. For example, in vast farmland or areas with complex terrain, the sparse distribution of stations results in missing soil temperature data in many areas. Furthermore, the high cost of station construction and maintenance limits the expansion of the number of stations, making it impossible to meet the needs of large-scale and sophisticated research.

[0003] Although remote sensing inversion can achieve soil temperature measurement over a large area, due to the influence of factors such as vegetation cover, cloud obstruction and observation angle, the data obtained is often a mixed temperature of multiple underlying surface components, which is inaccurate. At the same time, cloud obstruction will cause data loss, reducing the integrity and continuity of the data.

[0004] To address these issues, estimating soil temperature using machine learning methods combined with multiple sources of data, such as meteorological and remote sensing, has become a new direction. Existing machine learning methods for soil temperature modeling suffer from two major issues: First, they ignore variable lag effects. Soil temperature changes lag behind driving factors such as air temperature and radiation, resulting in inaccurate predictions based solely on current data. Second, they ignore variable importance and directly input all variables into the model, increasing the computational burden and reducing model stability due to redundant information.

[0005] To this end, the present invention proposes a new soil temperature inversion method that fully considers the variable hysteresis effect and feature importance to improve the inversion accuracy and enhance the model performance. Summary of the Invention

[0006] In order to solve the above technical problems, the present invention provides a soil temperature inversion method and system considering hysteresis effect and feature importance.

[0007] In a first aspect, the present invention provides a soil temperature inversion method considering hysteresis effect and feature importance, comprising:

[0008] Construct the basic input dataset for the area to be inverted; the basic input dataset includes the historical time series of soil temperature labels and multi-source driving variables;

[0009] Extract the time series data of the current period and the lagged period of the multi-source driving variables, construct the lag structure input set of the multi-source driving variables, and generate several lagged versions of the input variables;

[0010] Construct a supervised learning model, use the time-lagged version of the input variables and soil temperature labels to train the supervised learning model to obtain the initial inversion model, and evaluate the feature importance of each input variable;

[0011] The input variables are screened according to the feature importance to obtain the set of dominant variables;

[0012] Obtain the time-lagged version of the input variables in the dominant variable set, reconstruct the final input set, and train the initial inversion model to obtain the optimized soil temperature inversion model;

[0013] The optimized soil temperature inversion model is used to perform soil temperature inversion and obtain the soil temperature estimation result at the target time in the target area;

[0014] The original soil temperature measured data were used as validation data. The accuracy of the optimized soil temperature inversion model was evaluated using the validation data, and the selection strategy of the input variables was adjusted according to the accuracy evaluation results.

[0015] In a second aspect, the present invention provides a soil temperature inversion system that considers hysteresis effects and feature importance, comprising a first data set construction unit, an input variable generation unit, a feature importance evaluation unit, a screening unit, an input reconstruction and model training unit, an inversion unit, and an accuracy evaluation and adjustment unit;

[0016] The first dataset construction unit is used to construct a basic input dataset of the area to be inverted; the basic input dataset includes a historical time series of soil temperature labels and multi-source driving variables;

[0017] The input variable generation unit is used to extract the time series data of the current period and the lagged period of the multi-source driving variables, construct the lag structure input set of the multi-source driving variables, and generate several lagged versions of the input variables;

[0018] The feature importance evaluation unit is used to build a supervised learning model. It uses the time-lagged version of the input variables and the soil temperature label to train the supervised learning model to obtain the initial inversion model and evaluate the feature importance of each input variable.

[0019] The screening unit is used to screen the input variables according to the feature importance to obtain the dominant variable set;

[0020] The input reconstruction and model training unit is used to obtain the time-lagged version of the input variables in the dominant variable set, reconstruct the final input set, and train the initial inversion model to obtain the optimized soil temperature inversion model;

[0021] An inversion unit is used to perform soil temperature inversion using the optimized soil temperature inversion model to obtain an estimated soil temperature result at a target time in a target area;

[0022] The accuracy assessment and adjustment unit is used to use the original soil temperature measured data as verification data, use the verification data to evaluate the accuracy of the optimized soil temperature inversion model, and adjust the selection strategy of the input variables according to the accuracy assessment results.

[0023] On the basis of the above technical solution, the present invention can also be improved as follows.

[0024] Furthermore, a basic input data set of the area to be inverted is constructed, including: collecting soil temperature observation data of the area to be inverted during the target period and a period before the target period, and time series data of multi-source driving variables corresponding to the soil temperature observation data.

[0025] Furthermore, multi-source driving variables include temperature, precipitation, solar radiation, wind speed, soil moisture and vegetation index; data sources include ground meteorological stations, remote sensing products and reanalysis data.

[0026] Furthermore, when generating several time-lagged versions of input variables, the mean of the lag time of each multi-source driving variable in different lag periods is calculated based on statistical data, and the lag time with the smallest difference between the soil temperature observation data and the soil temperature prediction value corresponding to the mean is selected as the lag time of the input variable.

[0027] Furthermore, the supervised learning model is one of a random forest model, a gradient boosting tree model, and a neural network model.

[0028] Furthermore, evaluating the feature importance of each input variable includes: evaluating the importance of each input variable to the soil temperature prediction result through the variable contribution index generated during the initial inversion model training process.

[0029] Furthermore, the variable contribution index is split gain or attention weight or SHAP value or Pearson correlation coefficient.

[0030] Furthermore, the error index was calculated based on the validation data and the soil temperature inversion results, and the accuracy of the optimized soil temperature inversion model was evaluated.

[0031] Furthermore, the selection strategy for adjusting input variables includes: adjusting the number of input variables, the lag time length of input variables, and the screening threshold of feature importance; when screening input variables based on feature importance, setting the screening threshold of feature importance and retaining input variables with feature importance greater than the screening threshold; feature importance is used to characterize the contribution of input variables to soil temperature prediction.

[0032] The beneficial effects of the present invention are:

[0033] (1) By introducing a hysteresis structure input mechanism, the present invention makes up for the problem of missing information that the current input variables cannot reflect the current soil thermal state, effectively enhancing the model's ability to depict the soil thermal evolution process, thereby significantly improving the inversion accuracy;

[0034] (2) The present invention introduces a feature importance evaluation mechanism to improve the pertinence of input selection and the interpretability of the model, quantify the contribution of different variables, and dynamically select key variables as input features during the modeling process. On the one hand, this can achieve efficient modeling in variable-constrained scenarios, and on the other hand, it improves the flexibility and generalization ability of the method under different surface types and meteorological conditions. BRIEF DESCRIPTION OF THE DRAWINGS

[0035] Figure 1 Schematic diagram of the soil temperature inversion method considering hysteresis effect and feature importance provided in Example 1 of the present invention;

[0036] Figure 2 This is a system block diagram of a soil temperature inversion system that takes into account hysteresis effects and feature importance, provided in Example 2 of the present invention. DETAILED DESCRIPTION

[0037] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions of the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Generally, the components of the embodiments of the present invention described and shown in the drawings herein can be arranged and designed in various different configurations.

[0038] Example 1

[0039] As an example, Figure 1 As shown, to solve the above technical problems, this embodiment provides a soil temperature inversion method that considers hysteresis effect and feature importance, including:

[0040] Construct the basic input dataset for the area to be inverted; the basic input dataset includes the historical time series of soil temperature labels and multi-source driving variables;

[0041] Extract the time series data of the current period and the lagged period of the multi-source driving variables, construct the lag structure input set of the multi-source driving variables, and generate several lagged versions of the input variables;

[0042] Construct a supervised learning model, use the time-lagged version of the input variables and soil temperature labels to train the supervised learning model to obtain the initial inversion model, and evaluate the feature importance of each input variable;

[0043] The input variables are screened according to the feature importance to obtain the set of dominant variables;

[0044] Obtain the time-lagged version of the input variables in the dominant variable set, reconstruct the final input set, and train the initial inversion model to obtain the optimized soil temperature inversion model;

[0045] The optimized soil temperature inversion model is used to perform soil temperature inversion and obtain the soil temperature estimation result at the target time in the target area;

[0046] The original soil temperature measured data were used as validation data. The accuracy of the optimized soil temperature inversion model was evaluated using the validation data, and the selection strategy of the input variables was adjusted according to the accuracy evaluation results.

[0047] The present invention proposes a soil temperature inversion method that simultaneously considers the hysteresis effect of input variables and the importance of features, which can improve the physical consistency of the supervised learning model for the soil thermal response process, the effectiveness of the input structure, and the prediction accuracy under various environmental conditions.

[0048] The present invention significantly improves soil temperature inversion accuracy by introducing a hysteresis structure input mechanism. Traditional soil temperature modeling methods mostly use only input variables at the current moment, ignoring the lagged response characteristics of soil temperature to external environmental drivers. This significantly increases model prediction errors, especially in the context of rapidly changing temperature and radiation. By constructing a variable hysteresis structure, the present invention introduces input variables from several past moments (such as temperature, solar radiation, and soil moisture) into the modeling process. This overcomes the problem of missing information that the current input variables cannot reflect the current soil thermal state, effectively enhancing the model's ability to depict the soil thermal evolution process, and thus significantly improving inversion accuracy.

[0049] The present invention introduces a variable importance assessment mechanism to improve the pertinence of input variable selection and the interpretability of the model. In traditional methods, all variables are input into the model at the same time without effective screening, which not only introduces redundant information, but also increases the computational cost and training difficulty. The present invention embeds feature importance assessment, utilizes built-in or external interpretation methods of the model to quantify the contribution of different variables, and dynamically selects key variables as input features during the modeling process. On the one hand, efficient modeling can be achieved in variable-constrained scenarios, and on the other hand, the flexibility and generalization ability of the method under different surface types and meteorological conditions are improved.

[0050] The present invention also proposes a strategy for selecting input variables, that is, screening the input variables, selecting the variable combination with the best prediction effect based on the importance assessment of the input variables, and removing input variables that are invalid or have a negative impact on the prediction results.

[0051] Optionally, constructing a basic input data set for the area to be inverted includes collecting soil temperature observation data for the area to be inverted during a target period and a period before the target period, and time series data of multi-source driving variables corresponding to the soil temperature observation data.

[0052] Optionally, multi-source driving variables include temperature, precipitation, solar radiation, wind speed, soil moisture and vegetation index; data sources include ground meteorological stations, remote sensing products and reanalysis data.

[0053] Optionally, when generating several lagged versions of input variables, the mean of the lag durations of each multi-source driving variable in different lag periods is calculated based on statistical data, and the lag duration with the smallest difference between the soil temperature observation data and the soil temperature prediction value corresponding to the mean is selected as the lag duration of the input variable.

[0054] Optionally, the supervised learning model is one of a random forest model, a gradient boosting tree model, and a neural network model.

[0055] The nonlinear mapping relationship between input variables and soil temperature labels is fitted through a supervised learning model.

[0056] Optionally, evaluating the feature importance of each input variable includes: evaluating the importance of each input variable to the soil temperature prediction result through a variable contribution index generated during the initial inversion model training process.

[0057] Feature importance refers to the variable contribution index generated during the initial inversion model training process, which evaluates the importance of each input variable to the soil temperature inversion result.

[0058] Optionally, the variable contribution metric is split gain or attention weight or SHAP value (SHapley Additive Explanations or Pearson correlation coefficient).

[0059] Specifically, feature importance is determined based on supervised learning models, such as the split gain of tree models, the attention weight of neural networks, and the SHAP value of posterior analysis methods.

[0060] During the tree model construction process, data is continuously split to reduce node impurity and improve the model's ability to classify or predict data. Split gain metrics include the Gini index and information gain. When training a tree model, each feature is selected for splitting and the gain values after each feature split are calculated. For example, for a classification problem, the Gini index is used to calculate the change in node impurity after each feature split. The greater the reduction in impurity, the greater the split gain, indicating that the feature is more effective in distinguishing different categories and contributes more to the model. After training is complete, the number of times each feature is used for splitting in all trees, as well as the total gain from these splits, are counted. The more times a feature is used for splitting, the greater the total split gain, and the higher the feature importance.

[0061] In neural network models with an attention mechanism, attention weights can be used to measure feature importance. Taking the Transformer model in natural language processing as an example, the self-attention mechanism dynamically assigns different weights to each position in the input sequence. During training, the model automatically learns these weights, giving greater weight to important features. For input text data, each word vector interacts with other word vectors to generate an attention score. This score is then converted into attention weights using a softmax function. These weights reflect the importance of each word for the current position. By summing the attention weights for a feature across all positions, we can determine its importance. A higher weight indicates that the feature receives more attention when the model processes information and has a greater impact on the model's output.

[0062] The SHAP value is based on the concept of Shapley values in cooperative game theory. It considers all possible feature combinations and assigns an importance value to each feature. To calculate the SHAP value, the model must first be evaluated multiple times. For a dataset containing n features, all possible feature subsets are considered. For each feature subset, the difference between the model's prediction on that subset and the prediction on the empty set is calculated. This difference represents the contribution of that feature subset to the model's prediction. This contribution is then distributed to each feature in the subset using methods such as sampling approximation. This process is repeated until the contribution of all features in all possible subsets is calculated, ultimately resulting in a SHAP value for each feature. A larger SHAP value indicates a greater impact of the feature on the model's prediction and a higher feature importance.

[0063] Optionally, an error index is calculated based on the validation data and the soil temperature inversion results to evaluate the accuracy of the optimized soil temperature inversion model.

[0064] Error metrics include root mean square error (RMS), mean absolute error (MAE), and coefficient of determination (CDR). To calculate the RMS error, first find the square of the difference between the predicted and measured values for each sample point. These squared values are then added up and divided by the total number of samples to obtain the mean square error (MSE). The square root is then taken to obtain the RMS error. A smaller RMS error indicates that the model's predicted values are closer to the measured values, and the model's prediction accuracy is higher. To calculate the MAE, first find the absolute value of the difference between the predicted and measured values for each sample point. These absolute values are then added up and divided by the total number of samples. A smaller MAE indicates a smaller average error between the model's predicted and measured values, and a better model prediction. A CDR closer to 1 indicates a better fit of the model to the data and a higher proportion of the variation in the dependent variable that the model can explain, indicating a stronger predictive ability. A CDR closer to 0 indicates a poorer fit of the model to the data and weaker predictive ability.

[0065] Optionally, adjusting the selection strategy of input variables includes: adjusting the number of input variables, the lag time length of the input variables, and the screening threshold of feature importance; when screening input variables based on feature importance, setting the screening threshold of feature importance and retaining input variables with feature importance greater than the screening threshold; feature importance is used to characterize the contribution of input variables to soil temperature prediction.

[0066] Adjusting the number of input variables, the lag time length of input variables, and the screening threshold of feature importance based on the verification results can further improve the adaptability and robustness of the soil temperature inversion model in different climatic zones, surface types, or observation densities.

[0067] Example 2

[0068] Based on the same principle as the method shown in Example 1 of the present invention, as shown in the attached Figure 2 As shown, an embodiment of the present invention further provides a soil temperature inversion system that considers hysteresis effect and feature importance, including a first data set construction unit, an input variable generation unit, a feature importance evaluation unit, a screening unit, an input reconstruction and model training unit, an inversion unit, and an accuracy evaluation and adjustment unit;

[0069] The first dataset construction unit is used to construct a basic input dataset of the area to be inverted; the basic input dataset includes a historical time series of soil temperature labels and multi-source driving variables;

[0070] The input variable generation unit is used to extract the time series data of the current period and the lagged period of the multi-source driving variables, construct the lag structure input set of the multi-source driving variables, and generate several lagged versions of the input variables;

[0071] The feature importance evaluation unit is used to build a supervised learning model. It uses the time-lagged version of the input variables and the soil temperature label to train the supervised learning model to obtain the initial inversion model and evaluate the feature importance of each input variable.

[0072] The screening unit is used to screen the input variables according to the feature importance to obtain the dominant variable set;

[0073] The input reconstruction and model training unit is used to obtain the time-lagged version of the input variables in the dominant variable set, reconstruct the final input set, and train the initial inversion model to obtain the optimized soil temperature inversion model;

[0074] An inversion unit is used to perform soil temperature inversion using the optimized soil temperature inversion model to obtain an estimated soil temperature result at a target time in a target area;

[0075] The accuracy assessment and adjustment unit is used to use the original soil temperature measured data as verification data, use the verification data to evaluate the accuracy of the optimized soil temperature inversion model, and adjust the selection strategy of the input variables according to the accuracy assessment results.

[0076] Optionally, constructing a basic input data set for the area to be inverted includes collecting soil temperature observation data for the area to be inverted during a target period and a period before the target period, and time series data of multi-source driving variables corresponding to the soil temperature observation data.

[0077] Optionally, multi-source driving variables include temperature, precipitation, solar radiation, wind speed, soil moisture and vegetation index; data sources include ground meteorological stations, remote sensing products and reanalysis data.

[0078] Optionally, when generating several lagged versions of input variables, the mean of the lag durations of each multi-source driving variable in different lag periods is calculated based on statistical data, and the lag duration with the smallest difference between the soil temperature observation data and the soil temperature prediction value corresponding to the mean is selected as the lag duration of the input variable.

[0079] Optionally, the supervised learning model is one of a random forest model, a gradient boosting tree model, and a neural network model.

[0080] Optionally, evaluating the feature importance of each input variable includes: evaluating the importance of each input variable to the soil temperature prediction result through a variable contribution index generated during the initial inversion model training process.

[0081] Optionally, the variable contribution indicator is split gain, attention weight, SHAP value, or Pearson correlation coefficient.

[0082] Optionally, an error index is calculated based on the validation data and the soil temperature inversion results to evaluate the accuracy of the optimized soil temperature inversion model.

[0083] Optionally, adjusting the selection strategy of input variables includes: adjusting the number of input variables, the lag time length of the input variables, and the screening threshold of feature importance; when screening input variables based on feature importance, setting the screening threshold of feature importance and retaining input variables with feature importance greater than the screening threshold; feature importance is used to characterize the contribution of input variables to soil temperature prediction.

[0084] The above are merely preferred embodiments of the present invention and are not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations of the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention shall be included within the scope of protection of the present invention.

Claims

1. A soil temperature inversion method considering hysteresis effect and characteristic importance, characterized in that: include: Construct the basic input data set of the area to be inverted; The basic input dataset includes historical time series of soil temperature labels and multi-source driving variables; Extract the time series data of the current period and the lagged period of the multi-source driving variables, construct the lag structure input set of the multi-source driving variables, and generate several lagged versions of the input variables; Construct a supervised learning model, use the time-lagged version of the input variables and soil temperature labels to train the supervised learning model to obtain the initial inversion model, and evaluate the feature importance of each input variable; The evaluation of the feature importance of each input variable includes: evaluating the importance of each input variable to the soil temperature prediction result through the variable contribution index generated during the initial inversion model training process; the variable contribution index is split gain, attention weight, SHAP value or Pearson correlation coefficient; The input variables are screened according to the feature importance to obtain the set of dominant variables; Obtain the time-lagged version of the input variables in the dominant variable set, reconstruct the final input set, and train the initial inversion model to obtain the optimized soil temperature inversion model; The optimized soil temperature inversion model is used to perform soil temperature inversion and obtain the soil temperature estimation result at the target time in the target area; The original soil temperature measured data were used as validation data. The accuracy of the optimized soil temperature inversion model was evaluated using the validation data, and the selection strategy of the input variables was adjusted according to the accuracy evaluation results.

2. The soil temperature inversion method considering hysteresis effect and characteristic importance according to claim 1 is characterized in that: Construct the basic input data set of the area to be inverted, including: collecting soil temperature observation data of the area to be inverted during the target period and a period before the target period, and time series data of multi-source driving variables corresponding to the soil temperature observation data.

3. The soil temperature inversion method considering hysteresis effect and characteristic importance according to claim 1 is characterized in that: Multi-source driving variables include temperature, precipitation, solar radiation, wind speed, soil moisture and vegetation index; data sources include ground meteorological stations, remote sensing products and reanalysis data.

4. The soil temperature inversion method considering hysteresis effect and characteristic importance according to claim 1 is characterized in that: When generating several lagged versions of input variables, the mean of the lag time of each multi-source driving variable in different lag periods is calculated based on statistical data, and the lag time with the smallest difference between the soil temperature observation data and the soil temperature prediction value corresponding to the mean is selected as the lag time of the input variable.

5. The soil temperature inversion method considering hysteresis effect and characteristic importance according to claim 1 is characterized in that: The supervised learning model is one of the random forest model, gradient boosting tree model and neural network model.

6. The soil temperature inversion method considering hysteresis effect and characteristic importance according to claim 1, characterized in that: The error index was calculated based on the verification data and the soil temperature inversion results, and the accuracy of the optimized soil temperature inversion model was evaluated.

7. The soil temperature inversion method considering hysteresis effect and characteristic importance according to claim 1, characterized in that: The selection strategy for adjusting input variables includes: adjusting the number of input variables, the lag time length of input variables, and the screening threshold of feature importance; when screening input variables based on feature importance, setting the screening threshold of feature importance and retaining input variables with feature importance greater than the screening threshold; feature importance is used to characterize the contribution of input variables to soil temperature prediction.

8. A soil temperature inversion system considering hysteresis effect and characteristic importance, characterized in that: It includes a first data set construction unit, an input variable generation unit, a feature importance evaluation unit, a screening unit, an input reconstruction and model training unit, an inversion unit, and an accuracy evaluation and adjustment unit; The first dataset construction unit is used to construct a basic input dataset of the area to be inverted; the basic input dataset includes a historical time series of soil temperature labels and multi-source driving variables; The input variable generation unit is used to extract the time series data of the current period and the lagged period of the multi-source driving variables, construct the lag structure input set of the multi-source driving variables, and generate several lagged versions of the input variables; The feature importance evaluation unit is used to build a supervised learning model. It uses the time-lagged version of the input variables and the soil temperature label to train the supervised learning model to obtain the initial inversion model and evaluate the feature importance of each input variable. The evaluation of the feature importance of each input variable includes: evaluating the importance of each input variable to the soil temperature prediction result through the variable contribution index generated during the initial inversion model training process; the variable contribution index is split gain, attention weight, SHAP value or Pearson correlation coefficient; The screening unit is used to screen the input variables according to the feature importance to obtain the dominant variable set; The input reconstruction and model training unit is used to obtain the time-lagged version of the input variables in the dominant variable set, reconstruct the final input set, and train the initial inversion model to obtain the optimized soil temperature inversion model; An inversion unit is used to perform soil temperature inversion using the optimized soil temperature inversion model to obtain an estimated soil temperature result at a target time in a target area; The accuracy assessment and adjustment unit is used to use the original soil temperature measured data as verification data, use the verification data to evaluate the accuracy of the optimized soil temperature inversion model, and adjust the selection strategy of the input variables according to the accuracy assessment results.

Images