Under-crown temperature inversion method, system and application
By constructing a baseline prediction model and a gating correction function in flux tower observations, the problem of missing subcanopy temperature in flux tower observations was solved, enabling the prediction of vertical temperature difference and the inversion of subcanopy temperature, thus improving data availability and consistency across sites.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- SHENYANG INST OF APPL ECOLOGY CHINESE ACAD OF SCI
- Filing Date
- 2026-01-16
- Publication Date
- 2026-05-01
AI Technical Summary
Existing technologies lack subcanopy temperature profile data in flux tower observations, making it impossible to calculate vertical temperature difference and canopy temperature buffering effect, and making it difficult to maintain consistency when applied across years and sites.
By acquiring flux tower observation data, a baseline prediction model is constructed. The stability index is used for quantile truncation and robust standardization, the gated correction function is trained, and the sub-canopy temperature is inverted to ensure that the output value meets the monotonicity and boundary conditions, suppress non-physical corrections, and enhance cross-site consistency.
This technology enables the inversion of subcanopy temperature through vertical temperature difference prediction and correction functions in the absence of subcanopy temperature data, thereby improving data availability and interpretability, reducing the risk of overfitting, and ensuring consistency across sites.
Smart Images
Figure CN121960160A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of forest ecology and microclimate inversion technology, and in particular to a method, system and application for canopy temperature inversion. Background Technology
[0002] Flux tower observations provide continuous energy flux and turbulence data for studying forest energy, water, and carbon cycles. However, in actual site construction and long-term operation, temperature profiles and surface temperature sensors are not standard configurations for flux towers: many sites lack above-canopy / below-canopy surface temperatures, or lack below-canopy temperature profiles, making it impossible to directly calculate vertical temperature differences, canopy temperature buffering effects, and their phase differences. At the same time, most flux towers have stable records of environmental factors such as net radiation, sensible heat, latent heat, and stability, creating a structural contradiction of "rich environmental factors but lack of below-canopy temperature profile measurements."
[0003] Existing methods often directly model under-canopy air temperature or soil temperature, or rely on empirical regression of a few meteorological factors. This makes it difficult to explicitly incorporate the constraint of "stability-coupling state" on vertical temperature difference into the model, leading to inconsistencies in direction and non-physical interpretations when applied across years and sites. This invention aims to establish a unified framework that "first predicts vertical temperature difference, then infers under-canopy temperature," using deployable flux tower data as the core input, to address the problem of unavailable under-canopy temperatures at missing measurement sites. Summary of the Invention
[0004] To address the above problems, this invention provides a crown temperature inversion method, system, and application, aiming to systematically solve the problems of anomalies and missing data in observation data and improve data integrity and reliability.
[0005] To achieve the above objectives, the technical solution adopted by the present invention is as follows:
[0006] A method for retrieving subcoronary temperature includes the following steps:
[0007] S1. Obtain flux tower observation data, the data including at least time information and crown temperature. And environmental factors used to predict vertical temperature differences; simultaneously acquiring or calculating stability indices reflecting the state of atmospheric stratification. The observed data undergoes quality control and missing / anomaly removal, and is then filtered and aggregated according to a preset time window to obtain a sample sequence.
[0008] S2. Based on the aforementioned environmental factors, construct and train a baseline prediction model to obtain the baseline predicted value of the vertical temperature difference. ;
[0009] S3, Based on the stability index For stability index Quantile truncation and robust normalization were performed based on the training statistics to obtain... Construct and train the gated correction function g(·);
[0010] S4. During the inference phase, the baseline prediction value is... The final predicted value of the vertical temperature difference is obtained by adding it to the output value of the gated correction function g(·). ;
[0011] S5, utilizing the crown temperature The final predicted value after subtracting the vertical temperature difference The subcorona temperature was obtained through inversion.
[0012] Furthermore, in step S3, the training objective of the gated correction function includes making its output value satisfy preset physical constraints, which include:
[0013] (a) Boundedness: The output value of the gated correction function is restricted to a preset boundary [L, U];
[0014] (b) Monotonicity: The output value of the gated correction function is monotonically non-decreasing as the stability index increases.
[0015] The gated correction function also satisfies the anchoring constraint condition: at a preset reference stability, the output value of the gated correction function is forced to zero, so that the gated correction function is used to perform relative correction in the physical direction of the baseline prediction value, rather than to compensate for the absolute deviation.
[0016] Furthermore, the boundary [L, U] of the output value is determined based on the residual quantile statistics of the baseline prediction model during the training phase; wherein, the residual is the observed value of the vertical temperature difference. Compared with the baseline prediction value difference.
[0017] Furthermore, the method for determining the boundary [L, U] includes: calculating the lower quantile Qβ and upper quantile Q(1-β) of the training set residuals; and performing symmetrical expansion with (Qβ + Q(1-β)) / 2 as the center and [(Q(1-β) - Qβ) / 2]*γ as the half-width to obtain the boundary; where γ is an expansion coefficient greater than 1.
[0018] Furthermore, in step S3, when training the gating correction function, a direction consistency constraint term is introduced into the training objective to reinforce the monotonically non-decreasing physical constraint; the direction consistency constraint term is configured as follows: for Apply a positive perturbation ε, and apply it to g( +ε) is less than g( Punishment will be imposed in cases where ( ).
[0019] Furthermore, in step S3, when training the gated correction function, a time smoothing constraint term is introduced into the training objective to suppress non-physical jumps in the output values of the gated correction function on adjacent time units; the time smoothing constraint term is configured to penalize the difference in the output values of the gated correction function of adjacent samples sorted by time.
[0020] Furthermore, after step S5, a post-processing step is also included: based on the statistical quantiles of the crown temperature observations in the training dataset, the feasible region of the retrieved crown temperature is clipped.
[0021] Furthermore, the stability index The total Richardson number; the environmental factors include at least net radiation, soil heat flux, evapotranspiration ratio, atmospheric saturated water vapor pressure difference, and time and seasonal characteristics.
[0022] A subcoronary temperature inversion system includes a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the subcoronary temperature inversion method as described in any one of claims 1 to 8.
[0023] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the subcoronal temperature inversion method as described in any one of claims 1 to 8.
[0024] The technological advancements achieved by this invention compared to existing technologies are as follows:
[0025] (1) This invention provides a method, system and application for inverting the sub-canopy temperature for flux tower scenarios. In the absence of surface temperature above / below the canopy or temperature profile observation, the vertical temperature difference is reconstructed and the sub-canopy temperature is inverted using conventional environmental elements of flux tower, thereby improving the availability, comparability and interpretability of station microclimate data.
[0026] (2) The “under-canopy temperature missing” is transformed into a deployable process of “vertical temperature difference predictable and invertible”. The non-physical correction caused by overall translation is suppressed by the “monotonically bounded + anchored” gating mechanism, which enhances cross-site directional consistency. The gating boundary is adaptively determined by the residual quantile, which reduces the risk of overfitting and extrapolation explosion, adapts to the conventional elements of flux tower, and has engineering feasibility. Attached Figure Description
[0027] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used together with the embodiments of the invention to explain the invention and do not constitute a limitation thereof.
[0028] In the attached diagram:
[0029] Figure 1 This is a schematic diagram of the overall system structure of the present invention;
[0030] Figure 2 This is a schematic diagram of the data preprocessing and daily-scale aggregation process of the present invention;
[0031] Figure 3 This is a schematic diagram of the vertical temperature difference baseline modeling process of the present invention;
[0032] Figure 4 This is a schematic diagram of the stability gating correction module of the present invention;
[0033] Figure 5 This is a schematic diagram of the crown temperature inversion and feasible region clipping of the present invention;
[0034] Figure 6 This is a schematic diagram comparing the effects of gating correction and cross-tower migration in Example 2 (left: T1 baseline vs. gating; right: T1→T3 migration vs. T3 local training). Detailed Implementation
[0035] The following specific embodiments can be combined with each other, and the same or similar concepts or processes may not be described again in some embodiments. The embodiments of the present invention will now be described with reference to the accompanying drawings.
[0036] Example 1: A method for inverting sub-canopy temperature in flux tower scenarios
[0037] S1. Data Acquisition and Sample Construction
[0038] Acquire flux tower observation data, including crown temperature and conventional environmental factors used to construct a vertical temperature difference baseline prediction; simultaneously acquire or calculate the stability index Ri reflecting the turbulent coupling / stable structure. Perform quality control and time window filtering on the data, and aggregate it according to a preset granularity to form a sample set. Define the vertical temperature difference observation values during the training phase:
[0039]
[0040] The data undergoes quality control, anomaly removal, and time window filtering, and is aggregated according to a preset granularity to form a sample set.
[0041] S2, Baseline Prediction Construction
[0042] by Using the aforementioned conventional environmental factors as the target variable, a baseline prediction model is trained to obtain the predicted value of the vertical temperature difference baseline. The baseline model can be a generalized additive model, a gradient boosting tree, or other regression model; this invention does not limit the specific form of the baseline model. The baseline model outputs the predicted vertical temperature difference:
[0043]
[0044] S3, Stability Index Preprocessing
[0045] During the training phase Winsorizing is performed to suppress the influence of extreme values, resulting in... :
[0046]
[0047] Robust normalization was performed using the median and median absolute difference (MAD) to enhance scale comparability across different sites and climatic backgrounds, resulting in the stability normalization quantity:
[0048]
[0049] in This is a numerically stable term; the quantile thresholds and robust statistics determined during the training phase are reused during the inference phase to ensure consistent transformation.
[0050] S4. Adaptive determination of gated correction amplitude boundary
[0051] Calculate the baseline residuals during the training phase:
[0052]
[0053] The upper and lower bounds of the gated correction term are determined based on the residual quantiles:
[0054]
[0055] The mechanism limits the magnitude of the correction term and reduces the risk of cross-domain extrapolation divergence.
[0056] S5, Gated Function Training
[0057] Constructing gated functions ,by The input-output gating correction terms are used to form an additive correction framework during inference:
[0058] The gate function satisfies: (1) boundedness: (2) Monotonic and non-decreasing: following Increase (3) Anchor to zero: Correct to zero at the reference stability, for example, let
[0059]
[0060] This avoids the gating term from bearing the baseline intercept offset through overall translation, thereby weakening the intercept coupling between gating and the baseline and improving migration stability.
[0061] In addition to the fitting error, the training objective includes at least one of the following: directional consistency constraints and time smoothing constraints, which are used to enhance physical consistency.
[0062] S6, Inference and Subcoronal Temperature Inversion Output
[0063] The inference phase reuses the preprocessing statistics calculated in the training phase. and output :
[0064]
[0065] The predicted subcanopy temperature was obtained based on the inversion relationship. :
[0066]
[0067] To ensure physical plausibility, it is possible to... or Apply feasible domain pruning based on training domain statistics. When stability metrics are missing, the gating correction term can be configured to zero to degenerate the method to baseline output, thereby ensuring system availability.
[0068] Example 2: Baseline temperature difference prediction + stability-gated monotonic bounded anchoring correction + subcoronal temperature inversion (consistent with script)
[0069] 2.1 Application Scenarios and Objectives
[0070] At flux tower sites, subcanopy temperatures (e.g., subcanopy / near-surface temperature) are often missing or partially missing, but canopy temperatures and conventional environmental factors for flux towers are available. This embodiment provides a subcanopy temperature inversion process on a diurnal scale (growing season, 10–14 hours per day): first predicting the vertical temperature difference. Then, the temperature under the crown is deduced from the temperature above the crown, and the consistency of the injection physical direction and migration stability are corrected by stability gating.
[0071] 2.2 Input Data
[0072] Record every half hour / hour The input must include at least:
[0073] Crown temperature: (T_surface_up)
[0074] Under-coronation temperature: (T_surface_down is available during the training / evaluation period)
[0075] Net radiation and soil heat flux: (Rn_2, G)
[0076] Latent heat and sensible heat flux: (LE, H)
[0077] VPD: (VPD)
[0078] Stability Indicators: (Ri_down is mapped to Ri)
[0079] Time fields: DoY, Hour, Date (DoY, Hour, Date)
[0080] Definition of vertical temperature difference (monitoring target):
[0081]
[0082] 2.3 Half-hour screening, daily-scale aggregation, and EF ratio-of-sums (including QC)
[0083] 2.3.1 Time Filtering
[0084] Only:
[0085] DoY ∈ [91, 305]
[0086] Hour ∈ {10, 11, 12, 13, 14} and require efficient.
[0087] 2.3.2 Daily-scale aggregation (core caliber)
[0088] Aggregated by "Tower Station × Date"; Daily temperature difference:
[0089]
[0090] 2.3.3 EF ratio-of-sums and quality control
[0091] Half an hour denominator:
[0092] Only half-hour records that meet QC standards are accumulated: (20 W / m²), and optional constraints (TRUE)
[0093] Daily scale:
[0094]
[0095] And Crop to .
[0096] 2.4 Baseline Model: Prediction (GAM / bam)
[0097] The baseline input variables are fixed as: DoY + RnG + VPD + EF. Where:
[0098]
[0099] The baseline model was fitted using mgcv::bam (GAM):
[0100]
[0101] And obtained:
[0102]
[0103] If a variable has few valid and unique values, it can degenerate into a linear term; otherwise, it is a smooth term. ,That It is determined by the number of unique values and the default upper limit.
[0104] 2.5 Stability Ri and Virtual Potential Temperature θᵥ
[0105] 2.5.1 Virtual potential temperature
[0106] Set two heights Temperature is expressed in Kelvin (K), and air pressure in Pa. Potential temperature:
[0107]
[0108] Virtual potential temperature:
[0109]
[0110] in Specific humidity (kg / kg); Pa, Take the dry air constant ratio.
[0111] 2.5.2 bulk Richardson number
[0112]
[0113] in , The average of the two heights, , , It is a stable term.
[0114] 2.6 Winsorization and Robust Normalization of Ri
[0115] The normalized parameters are estimated only on the source training set and then reused during test / external tower inference.
[0116] Quantile truncation:
[0117]
[0118] Robustness center and scale:
[0119]
[0120] standardization:
[0121]
[0122] like Then let .
[0123] 2.7 Gated correction output boundary: adaptively determined by the training residual quantiles
[0124] Calculate the residuals on the source domain training set:
[0125]
[0126] Take the quantile:
[0127]
[0128] The interval is then symmetrically extended according to the script γ=1.20.
[0129] Interval center:
[0130] Half width: (In this invention, γ is taken as an empirical value of 1.20)
[0131] Gated output boundaries:
[0132]
[0133] 2.8 Gated Function Structure: Monotonically Non-decreasing + Bounded Sigmoid + Anchored
[0134] Gating module output calibration item The final temperature difference is:
[0135]
[0136] 2.8.1 Bounded sigmoid mapping
[0137] First define the unanchored :
[0138]
[0139] in It is sigmoid. This is a learnable bias.
[0140] 2.8.2 Monotonic Non-decreasing Constraint
[0141] To ensure that it is monotonically non-decreasing, let:
[0142]
[0143] therefore Follow Monotonically increasing, Monotonic and non-decreasing.
[0144] 2.8.3 Anchoring
[0145] With reference point Anchoring:
[0146]
[0147] 2.9 Complete Loss Function
[0148] 2.9.1 Fit Term (MSE)
[0149] make ,but
[0150]
[0151] 2.9.2 Directional Consistency Item
[0152] Apply a stability positive perturbation to each sample :
[0153]
[0154] Penalize cases where "the predicted decline is reversed after the disturbance":
[0155]
[0156] 2.9.3 Time smoothing term (squared difference between adjacent days)
[0157] Sort the training samples by date to construct adjacent pairs. :
[0158]
[0159] 2.9.4 Scope of Penalty
[0160] Script for gating output Penalize the mean squared:
[0161]
[0162] 2.9.5 Total Loss
[0163]
[0164] And optimize the gating parameters using Adam. .
[0165] 2.10 Inference Output: Temperature Difference Prediction, Under-Crow Temperature Inversion and Clipping
[0166] 2.10.1 Temperature Difference Prediction
[0167] For any sample to be predicted:
[0168] Baseline output
[0169] Ri is obtained by winsorizing and robustly normalizing the training-time parameters.
[0170] Gating correction
[0171] Output:
[0172]
[0173] 2.10.2 Under-corona temperature inversion
[0174]
[0175] 2.10.3 Feasible Domain Reduction for Under-Crow Temperature
[0176] If cropping is enabled, use the source domain training set. Quantiles as upper and lower bounds:
[0177]
[0178] 2.11 Cross-tower migration operation mode
[0179] For each direction (e.g., Tower1→Tower3):
[0180] Training / testing are segmented in chronological order within the source tower.
[0181] ΔT cutoff: Take the value on the source tower training set. As the available range, this range is applied simultaneously to both the full sample from the source tower and the external sample from the target tower.
[0182] In the source tower training baseline + gating, the gating is trained only on the source tower training set.
[0183] Inference evaluation on the source tower test set and external samples of the target tower.
[0184] 2.12 Evaluation Indicators
[0185] For observation With prediction calculate:
[0186] RMSE:
[0187] MAE:
[0188] Bias:
[0189] Regression slope and :right Return to get With the coefficient of determination
[0190] 2.13 Specific Application Results
[0191] like Figure 6 As shown in the left figure: After adding the Ri gate constraint, the slope of Tower 1 changed from 0.771 to 0.883, which is closer to 1, indicating that considering the Ri gate constraint makes the prediction of vertical temperature difference more accurate;
[0192] The right figure shows that when the model trained on Tower 1 is transferred to Tower 3, its performance is slightly better than the model trained using Tower 3's own data, indicating that the Tower 1 model has better transferability.
[0193] Example 3: A computer-readable storage medium storing one or more programs including instructions that, when executed by a computing device, cause the computing device to perform the method of Example 1.
[0194] Example 4: A computing device, comprising:
[0195] One or more processors, a memory, and one or more programs, wherein the one or more programs are stored in the memory and configured to be executed by the one or more processors, and the one or more programs include instructions for performing the method of Embodiment 1.
[0196] Finally, it should be noted that the above descriptions are merely preferred embodiments of the present invention and are not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the claims of the present invention.
Claims
1. A method for retrieving subcoronal temperature, characterized in that, Includes the following steps: S1. Obtain flux tower observation data, the data including at least time information and crown temperature. And environmental factors used to predict vertical temperature differences; simultaneously acquiring or calculating stability indices reflecting the state of atmospheric stratification. ; The observation data is subjected to quality control and missing / anomaly removal, and is filtered and aggregated according to a preset time window to obtain a sample sequence; S2. Based on the aforementioned environmental factors, construct and train a baseline prediction model to obtain the baseline predicted value of the vertical temperature difference. ; S3, Based on the stability index For stability index Quantile truncation and robust normalization were performed based on the training statistics to obtain... Construct and train the gated correction function g(·); S4. During the inference phase, the baseline prediction value is... The final predicted value of the vertical temperature difference is obtained by adding it to the output value of the gated correction function g(·). ; S5, utilizing the crown temperature The final predicted value after subtracting the vertical temperature difference The subcorona temperature was obtained through inversion.
2. The method according to claim 1, characterized in that, In step S3, the training objective of the gated correction function includes making its output value satisfy preset physical constraints, which include: (a) Boundedness: The output value of the gated correction function is restricted to a preset boundary [L, U]; (b) Monotonicity: The output value of the gated correction function is monotonically non-decreasing as the stability index increases; The gated correction function also satisfies the anchoring constraint condition: at a preset reference stability, the output value of the gated correction function is forced to zero, so that the gated correction function is used to perform relative correction in the physical direction of the baseline prediction value, rather than to compensate for the absolute deviation.
3. The method according to claim 2, characterized in that, The boundary [L, U] of the output value is determined based on the residual quantile statistics of the baseline prediction model during the training phase; wherein the residual is the observed value of the vertical temperature difference. Compared with the baseline prediction value difference.
4. The method according to claim 3, characterized in that, The method for determining the boundary [L, U] includes: calculating the lower quantile Q of the training set residuals. β and upper quantile Q (1-β) ; with (Q) β + Q (1-β) Centered on ) / 2, with [(Q) (1-β) - Q β The boundary is obtained by symmetrically expanding γ to half-width; where γ is an expansion coefficient greater than 1.
5. The method according to claim 2, characterized in that, In step S3, when training the gating correction function, a direction consistency constraint term is introduced into the training objective to reinforce the monotonically non-decreasing physical constraint; the direction consistency constraint term is configured as follows: Apply a positive perturbation ε, and apply it to g( +ε) is less than g( Punishment will be imposed in cases where ( ).
6. The method according to claim 2, characterized in that, In step S3, when training the gated correction function, a time smoothing constraint term is introduced into the training objective to suppress non-physical jumps in the output values of the gated correction function on adjacent time units; the time smoothing constraint term is configured to penalize the difference in the output values of the gated correction function of adjacent samples sorted by time.
7. The method according to claim 1, characterized in that, After step S5, a post-processing step is also included: based on the statistical quantiles of the crown temperature observations in the training dataset, the feasible region of the retrieved crown temperature is clipped.
8. The method according to claim 1, characterized in that, The stability index The total Richardson number; the environmental factors include at least net radiation, soil heat flux, evapotranspiration ratio, atmospheric saturated water vapor pressure difference, and time and seasonal characteristics.
9. A subcoronal temperature inversion system, characterized in that, It includes a processor and a memory, the memory storing a computer program that, when executed by the processor, implements the subcoronal temperature inversion method as described in any one of claims 1 to 8.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the subcoronal temperature inversion method as described in any one of claims 1 to 8.