CN-121766621-B - Ecological system productivity estimation method for predicting vegetation optimal temperature based on LSTM

CN121766621BCN 121766621 BCN121766621 BCN 121766621BCN-121766621-B

Abstract

The invention discloses an ecological system productivity estimation method based on LSTM prediction vegetation optimal temperature, which relates to the technical field of ecological information and comprises the steps of obtaining historical meteorological, topographic and future climate data, constructing a spatial neighborhood of a target pixel, quantifying the communication cost of the climate based on the topography and a wind field, generating anisotropic spatial features, resolving dynamic space-time confidence factors representing microclimate complexity, inputting the anisotropic spatial features and the historical air temperature into a long-short-term memory network, regulating fusion weights through a dynamic gating mechanism by utilizing the confidence factors, predicting the vegetation optimal temperature, constructing a decoupling model containing thermal kinetic energy and nonlinear heat stress items, and calculating productivity based on the predicted values and the future air temperature.

Inventors

WU MINGCHUN
ZHANG ZHAOYING
ZHANG YONGGUANG

Assignees

南京大学

Dates

Publication Date: 20260508
Application Date: 20260304

Claims (7)

1. The ecological system productivity estimation method for predicting the optimal vegetation temperature based on LSTM is characterized by comprising the following steps: acquiring historical weather driving data, geographical terrain data and future weather situation data at least comprising future air temperature data of a target area, wherein the historical weather driving data comprises a time sequence of historical air temperatures and a historical wind field vector, and the future weather situation data also comprises future carbon dioxide concentration data and future precipitation data; Constructing a corresponding space neighborhood set aiming at each target pixel in the target region, and quantifying weather communication cost of each neighborhood pixel in the space neighborhood set relative to the target pixel based on the geographic terrain data and the historical wind field vector to generate anisotropic space characteristics; calculating the distribution discrete degree of the climate communication cost in the space neighborhood set, and calculating a dynamic space-time confidence factor for representing the space microclimate complexity of the area where the target pixel is located based on the distribution discrete degree; inputting the time sequence of the anisotropic spatial features and the historical air temperature into a long-short-term memory network, and adaptively regulating and controlling the fusion weight of the anisotropic spatial features and the historical air temperature in network state updating based on the dynamic space-time confidence factor through a dynamic gating mechanism to generate a vegetation optimal temperature predicted value of the target pixel; Constructing a decoupling estimation model comprising a thermal energy term and a nonlinear heat stress term, and calculating the ecological system productivity of the target area through the decoupling estimation model based on the vegetation optimal temperature predicted value and the future air temperature data; The decoupling estimation model comprising a thermal energy term and a nonlinear heat stress term is constructed, and based on the vegetation optimal temperature predicted value and the future air temperature data, the ecological system productivity of the target area is calculated through the decoupling estimation model, and specifically the following formula is adopted: ; Wherein, the In order to be productive in an ecological system, For future air temperature data as a thermal kinetic energy term, Is an index corresponding to a nonlinear heat stress term, For the data of the concentration of carbon dioxide in the future, For the data of the precipitation in the future, As the coefficient of regression of the coefficient of the data, Is a residual term; Calculating the ecosystem productivity of the target area through the decoupling estimation model, further comprising: Monitoring an index corresponding to the nonlinear heat stress item in real time, and comparing the index with a preset air pore conductivity closing threshold value, wherein the air pore conductivity closing threshold value represents a physiological critical point of vegetation triggering an air pore closing mechanism due to extreme heat stress; When the index is lower than the air hole conductivity closing threshold, generating an air hole conductivity limiting factor, and executing numerical attenuation processing on a coefficient corresponding to future carbon dioxide concentration data and a coefficient corresponding to future precipitation data in the regression coefficient by utilizing the air hole conductivity limiting factor to generate a dynamic regression coefficient; And using the dynamic regression coefficient to replace the original regression coefficient, and executing the calculation of the ecological system productivity to represent the physiological reduction of vegetation water coupling efficiency under the extreme heat stress state.
2. The method for estimating the productivity of an ecosystem based on LSTM predicted vegetation optimal temperature of claim 1, wherein quantifying the weather connectivity cost of each neighborhood pixel in the spatial neighborhood set relative to the target pixel, generating anisotropic spatial features comprises the steps of: Acquiring the altitude and slope vectors of each neighborhood pixel and the target pixel in the space neighborhood set; Calculating the topographic impedance factor of each neighborhood pixel relative to the target pixel, wherein the topographic impedance factor is obtained by weighting and summing the altitude difference between the neighborhood pixel and the target pixel and the gradient difference degree calculated based on the gradient vector; Constructing a relative position vector pointing to the target pixel from each neighborhood pixel, projecting a historical wind field vector of the neighborhood pixel onto the relative position vector, and calculating airflow conveying gain; And generating anisotropic weights of all neighborhood pixels relative to the target pixels based on the difference value of the terrain impedance factor and the airflow conveying gain, and carrying out weighted aggregation on time sequences of the historical air temperatures in the space neighborhood set by utilizing the anisotropic weights to obtain the anisotropic spatial characteristics.
3. The method for estimating the productivity of an ecosystem based on LSTM predicted vegetation optimal temperature according to claim 2, wherein the calculating the dynamic space-time confidence factor comprises: Calculating the space neighborhood entropy of the space neighborhood set by utilizing an information entropy formula based on the anisotropic weight, wherein the space neighborhood entropy is used for representing the uniformity of the climate communication cost on space distribution; inputting the spatial neighborhood entropy into a nonlinear activation function containing preset learnable parameters, and mapping to obtain normalized dynamic space-time confidence factors; And when the spatial neighborhood entropy is higher than a preset reference, the dynamic space-time confidence factor approaches to 1, and the long-term and short-term memory network is indicated to increase the dependency weight on the time sequence of the historical air temperature.
4. The method for estimating the productivity of an ecosystem based on LSTM predicted vegetation optimal temperature according to claim 1, wherein the time series of the anisotropic spatial feature and the historical air temperature is input into a long-short-term memory network, and the fusion weight of the anisotropic spatial feature and the historical air temperature in network state updating is adaptively regulated and controlled based on the dynamic space-time confidence factor through a dynamic gating mechanism, and the method comprises the following steps: Configuring a feature projection layer, and mapping the anisotropic spatial feature in a scalar form into a spatial feature vector by utilizing the feature projection layer, wherein the dimension of the spatial feature vector is consistent with the hidden layer dimension of the long-term and short-term memory network; inputting the time sequence of the historical air temperature and the hidden state of the long-short-period memory network at the last moment into the long-short-period memory network, and calculating the candidate memory state of the time channel at the current moment; Using the dynamic space-time confidence factor as a mutual exclusion weighting coefficient, and carrying out linear weighted fusion on the time channel candidate memory state and the space feature vector to obtain a fused updated information stream; And combining the forgetting gate output of the long-period memory network with the fused updating information flow to update the cell state of the long-period memory network at the current moment.
5. The method for estimating an ecosystem productivity based on an LSTM predicted vegetation optimum temperature according to claim 4, wherein the calculation formula for updating the cell state at the present time is: ; Wherein, the As the state of the cell at the present moment, In order to forget the door, In order to be the state of the cell at the previous time, In order to be able to enter the door, As a dynamic space-time confidence factor, For the candidate memory state of the time channel, Is a spatial feature vector.
6. The method for estimating the productivity of an ecosystem based on the predicted optimal temperature of vegetation by using LSTM according to claim 1, wherein the nonlinear heat stress term characterizes the nonlinear inhibition effect on the productivity by utilizing the deviation of the predicted optimal temperature value of vegetation and the future air temperature data, and specifically comprises the following steps: calculating a difference value between the future air temperature data and the vegetation optimal temperature predicted value; when the difference is greater than zero, calculating a nonlinear heat stress index by using an exponential decay function to represent nonlinear inhibition of the productivity by the heat stress; calculating a nonlinear heat stress index using a linear function to characterize the low temperature limit on productivity when the difference is less than or equal to zero, the nonlinear heat stress index The calculation formula of (2) is as follows: ; Wherein, the For the future air temperature data, Is the predicted value of the optimal temperature of the vegetation, And Is a preset constant.
7. The method for estimating the productivity of an ecosystem based on the optimal temperature of vegetation predicted by LSTM according to claim 1, wherein generating the air pore conductance limiting factor and executing the numerical attenuation processing by using the air pore conductance limiting factor comprises the following steps: Constructing a nonlinear mapping function based on the ratio of the index to the air hole air conductivity closing threshold value, and generating the air hole air conductivity limiting factor with a value interval of zero to one, wherein the nonlinear mapping function is configured to enable the air hole air conductivity limiting factor to be in a nonlinear acceleration descending trend when the index approaches zero; and multiplying the coefficient corresponding to the future carbon dioxide concentration data and the coefficient corresponding to the future precipitation data with the air hole conductivity limiting factor to obtain the corrected dynamic regression coefficient.

Description

Ecological system productivity estimation method for predicting vegetation optimal temperature based on LSTM Technical Field The invention relates to the technical field of ecological information, in particular to an ecological system productivity estimation method for predicting optimal vegetation temperature based on LSTM. Background The total primary productivity (GPP) of the land ecosystem is the maximum carbon flux of the atmosphere to the biosphere, drives various key ecosystem functions, accurately estimates the change trend of the land ecosystem in future climate situations, and has important significance for understanding global carbon dynamics and formulating climate coping strategies. In the existing mainstream estimation system, the optimal temperature of vegetation photosynthesis is generally regarded as one of the core parameters for determining the accuracy of GPP estimation, and the conventional remote sensing estimation method or the earth system model is mostly used for determining the parameter based on a fixed photosynthesis-temperature response curve, namely, assuming that the optimal growth temperature of vegetation is kept relatively static in a specific biological community or long time scale, or simply performing linear extrapolation according to the air temperature, so as to be used as a basic basis for simulating productivity variation. However, plant physiology researches show that vegetation does not passively accept the environmental temperature, but has a remarkable thermal adaptation mechanism, namely, the optimal temperature is adjusted to adapt to climate warming, so that photosynthesis efficiency is maintained at a higher temperature, the adaptation process does not occur in isolation or even but is deeply restricted by complex space-time environmental elements, in an actual natural geographic environment, micro-climate environments often show high spatial heterogeneity due to factors such as terrain obstruction and atmospheric transportation, physiological adjustment of vegetation has remarkable time lag and accumulation effects, an existing model is lack of capturing capacity of the complex space-time coupling mechanism, uniform processing logic is often adopted, real thermal adaptation track of vegetation cannot be accurately reflected when facing complex terrain or rapid climate warming, and the lack of the modeling of the physiological adjustment mechanism directly causes serious underestimation of productivity growth potential of vegetation in a future warming background, so that a forecasting result of GPP has systematic deviation in an extreme or complex environment. Therefore, how to break through the limitation of static parameters in the model, and establish an estimation method capable of truly reflecting the vegetation thermal adaptation dynamic mechanism in the complex space-time environment becomes a key problem to be solved in order to improve future GPP prediction reliability. Disclosure of Invention Aiming at the defects existing in the prior art, the invention provides an ecological system productivity estimation method for predicting the optimal vegetation temperature based on LSTM. In order to achieve the above object, the technical scheme of the present invention is as follows: the invention discloses an ecological system productivity estimation method for predicting vegetation optimal temperature based on LSTM, which comprises the following steps: Acquiring historical meteorological driving data, geographical terrain data and future climate scene data at least comprising future air temperature data of a target area, wherein the historical meteorological driving data comprises a time sequence of historical air temperatures and a historical wind field vector; Constructing a corresponding space neighborhood set aiming at each target pixel in the target region, and quantizing weather connection cost of each neighborhood pixel in the space neighborhood set relative to the target pixel based on geographic topography data and historical wind field vectors to generate anisotropic space characteristics; calculating the distribution discrete degree of the climate communication cost in the space neighborhood set, and calculating a dynamic space-time confidence factor for representing the space microclimate complexity of the region where the target pixel is located based on the distribution discrete degree; Inputting the time sequence of the anisotropic spatial characteristics and the historical air temperature into a long-short-term memory network, and adaptively regulating and controlling the fusion weight of the anisotropic spatial characteristics and the historical air temperature in network state updating based on a dynamic space-time confidence factor through a dynamic gating mechanism to generate a vegetation optimal temperature predicted value of a target pixel; And constructing a decoupling estimation model comprising a thermal energy term and a nonline