Method for estimating soil moisture through combined assimilation strategy considering phenology

Abstract

The invention provides a method for estimating soil moisture through a combined assimilation strategy considering phenology, and relates to the technical field of soil moisture estimation. Using the calibrated WOFOST model to obtain downscaled root zone soil moisture data of the to-be-measured area; determining assimilation frequency according to different growth stages of the crops; obtaining observation data of the to-be-measured region according to the assimilation frequency, wherein the observation data comprises soil moisture and a leaf area index; fusing the root zone soil moisture data and the observation data by using ensemble Kalman filtering and generating assimilation data; and determining the soil moisture of the to-be-detected area according to the assimilation data. The problems that in the prior art, an unnecessary assimilation process exists, efficiency is greatly reduced, and higher calculation cost is brought by model simulation on the aspect of finer resolution are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of soil moisture estimation, and in particular to a method for estimating soil moisture by a joint assimilation strategy taking phenology into consideration. Background Art

[0002] Farmland ecosystem is a major place for agricultural activities. The development of modern agriculture requires precise control of key processes in farmland. The first thing to pay attention to in farmland is the crop growth process that affects food production. Soil water is the basis for the survival of crops, and soil water itself is also an important part of water resources. Therefore, it is also necessary to study the movement of soil water. Studying the movement of soil water to improve the efficiency of agricultural water use and strengthening the control of crop growth status to ensure agricultural output are two major aspects that need to be considered in the development of precision agriculture.

[0003] Models and observations are two necessary means of scientific research. They have their own advantages and disadvantages and complement each other. Comprehensively utilizing the valuable information provided by observations and the predictive ability of models is the direction of many researchers' efforts. The development of data assimilation (DA) technology provides an opportunity to integrate models and observations. Data assimilation weighs the observed data and simulation results, continuously corrects and optimizes the model parameters, and provides the best estimate for the current state of the system, which greatly improves the model's prediction accuracy for the future. Data assimilation can also take advantage of the model's ability to simulate continuously to interpolate sparse observations, thereby obtaining variable information that is continuous in time and space. In short, data assimilation can simultaneously utilize the data value of dynamic observations and the consistency constraints of the model in terms of physical properties and time evolution laws. The dual constraints of the latest data and historical knowledge (the model is a summary of knowledge) make the results of data assimilation both continuous in time and space and avoid the blindness of a simple model.

[0004] The study of soil water movement and crop growth processes is relatively difficult. Relying solely on models or observations cannot meet the needs of modern agriculture for formulating real-time irrigation strategies, nor can it accurately assess the fluctuations in grain production under the background of global change. It is necessary to study the data assimilation technology that integrates multi-source and multi-scale agricultural-related observations such as soil moisture content, leaf area index (LAI), biomass, and evapotranspiration into soil water movement or crop growth models. It is also necessary to explore the indicative role of multi-source and multi-scale observations on key farmland processes and the constraining role of key model variables.

[0005] The phase of assimilation of remote sensing spectral information (i.e., assimilation time scale) has a great influence on the assimilation results. The existing assimilation system does not consider the phase information when using observation data and does not adjust the assimilation frequency according to the growth phenological period of plants. Therefore, there are unnecessary assimilation processes and the efficiency is greatly slowed down, which will bring higher computational costs for model simulation at a finer resolution. Summary of the invention

[0006] In order to overcome the shortcomings of the prior art, the purpose of the present invention is to provide a method for estimating soil moisture by a joint assimilation strategy taking phenology into consideration. The present invention solves the problems in the prior art of unnecessary assimilation process, greatly slowing down the efficiency and bringing higher computational cost for model simulation at a finer resolution.

[0007] To achieve the above object, the present invention provides the following solutions:

[0008] A method for estimating soil moisture using a joint assimilation strategy that considers phenology includes:

[0009] The downscaled root zone soil moisture data of the measured area are obtained using the calibrated WOFOST model;

[0010] Determine the assimilation frequency according to the different growth stages of crops;

[0011] Obtaining observation data of the area to be measured according to the assimilation frequency, wherein the observation data includes: soil moisture and leaf area index;

[0012] Ensemble Kalman filtering is used to fuse root zone soil moisture data with observation data and generate assimilated data;

[0013] The soil moisture of the area to be measured is determined according to the assimilated data.

[0014] Preferably, the calibration process of the WOFOST model is:

[0015] Determine weather data, soil data, crop management data and solar radiation for the area to be tested;

[0016] The WOFOST model is calibrated based on the weather data, soil data, crop management data and solar radiation.

[0017] Preferably, the weather data includes:

[0018] Daily maximum and minimum temperatures, rainfall, wind speed and sunshine hours.

[0019] Preferably, the WOFOST model is constructed based on a CNN convolutional neural network.

[0020] Preferably, the assimilation frequency includes:

[0021] High frequency assimilation, normal frequency assimilation and phenological period assimilation.

[0022] The present invention discloses the following technical effects:

[0023] The present invention provides a method for estimating soil moisture by a joint assimilation strategy considering phenology, including: using a calibrated WOFOST model to obtain downscaled root zone soil moisture data of a test area; determining the assimilation frequency according to different growth stages of crops; obtaining observation data of the test area according to the assimilation frequency, the observation data including: soil moisture and leaf area index, using ensemble Kalman filtering to fuse the root zone soil moisture data and the observation data and generate assimilated data; determining the soil moisture of the test area according to the assimilated data. The present invention constructs a WOFOST crop model for joint assimilation of multi-source observation data of soil moisture and leaf area index, and optimizes the assimilation model under the framework of considering the influence of different phenological assimilation amounts, thereby improving the effect of crop model simulating soil water movement process. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0025] Figure 1 A flow chart of a method for estimating soil moisture using a joint assimilation strategy taking phenology into consideration provided by an embodiment of the present invention;

[0026] Figure 2 A flowchart for producing remote sensing root zone soil moisture data provided by an embodiment of the present invention;

[0027] Figure 3 A flow chart of a method for downscaling SMAP root zone soil moisture using 2D-CNN provided in an embodiment of the present invention;

[0029] Figure 4 A flow chart of using ensemble Kalman filtering to fuse model data and observation data and generate assimilated data is provided in an embodiment of the present invention. DETAILED DESCRIPTION

[0030] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in 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. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0031] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.

[0032] like Figure 1 As shown, the present invention provides a method for estimating soil moisture by a joint assimilation strategy taking phenology into consideration, comprising:

[0033] Step 100: using the calibrated WOFOST model to obtain downscaled root zone soil moisture data of the area to be measured;

[0034] Step 200: Determine the assimilation frequency according to different growth stages of the crop;

[0035] Step 300: Obtaining observation data of the area to be measured according to the assimilation frequency, the observation data including: soil moisture and leaf area index;

[0036] Step 400: using an ensemble Kalman filter to fuse the root zone soil moisture data with the observation data and generate assimilated data;

[0037] Step 500: Determine the soil moisture of the area to be tested based on the assimilated data.

[0038] Furthermore, the calibration process of the WOFOST model is as follows:

[0039] Determine weather data, soil data, crop management data and solar radiation for the area to be tested;

[0040] The WOFOST model is calibrated based on the weather data, soil data, crop management data and solar radiation.

[0041] Specifically, the basic data input of the crop growth model includes weather data, soil data and crop management data. The meteorological data of the weather station comes from the National Meteorological Information Center of the China Meteorological Administration, including the daily maximum and minimum temperatures, rainfall, wind speed and sunshine hours. Since WOFOST requires solar radiation, the Estrom formula in the FAO-56 guide is used to convert sunshine hours into solar radiation. The soil data used in the present invention comes from the 1:1 million soil database of China. The physical and chemical properties of the soil can be obtained from the soil profile. Detailed crop management information was investigated during the field trial, including data such as planting date, planting depth and spacing, planting density, irrigation date and irrigation amount, fertilization date and fertilization amount. The parameterization and calibration of the winter wheat model in the study area uses the model provided by Huang et al. (2015b) and the detailed information on its parameterization and calibration. The present invention implements the WOFOST model in a Python environment based on the PCSE package (http: / / github.com / ajwdewit / pcse.git).

[0042] Specifically, the present invention uses meteorological data (daily maximum and minimum temperatures, rainfall, wind speed and sunshine hours) from the National Meteorological Information Center of the China Meteorological Administration and other detailed information on the model and its parameterization and calibration provided by Huang et al. (2015b) for parameterization and calibration of the winter wheat model in the study area.

[0043] Specifically, the weather data includes:

[0044] Daily maximum and minimum temperatures, rainfall, wind speed and sunshine hours.

[0045] Further, such as Figure 2-4 As shown, the SMAP product downscaling method is a CNN convolutional neural network.

[0046] Specifically, the present invention uses CNN to downscale SMAP products, making remote sensing data more effective in smaller geographic areas. The CNN convolutional neural network layer uses the Conv2D convolutional neural network layer, the number of samplers (filters) is defined as 32, the convolution stride (strides) is 3, the convolution kernel length (kernel_size) is defined as 3, the convolution kernel width is the width of the input data sample, the convolution kernel initialization (kernel_initializer) adopts the glorot processing method, the edge data padding (padding) adopts the same method, the bias initialization (bias_initializer) adopts all 0 initialization, and the activation parameters in the convolutional neural network layer adopt the Relu activation method. The batch normalization layer (batch_normalization) and the pooling layer (pooling) use default parameters, the number of neural network units (units) in the Dense fully connected layer is defined as 512, the initialization adopts the glorot processing method, and the neural network parameter amount is 6.71x107. The number of samples (batch_size) for each input training is 2048, and the number of sample iterations (epochs) in each training process is 150.

[0047] Using SMAP's root zone soil moisture data and Landsat's visible light-near red-shortwave infrared as training data, the CNN model was trained for 50 rounds to obtain a trained model, and the soil moisture of each pixel was predicted. The 9km resolution SMAP root zone soil moisture product was downscaled to obtain a 30m resolution SMAP root zone soil moisture product.

[0048] Specifically, during the preprocessing and preparation of remote sensing data, the present invention downscales the soil moisture data of the SMAP product through a convolutional neural network (CNN) to achieve finer-grained soil moisture monitoring in geographical areas. Specifically, by setting specific CNN parameters, such as 32 filters, a Conv2D layer with a step size of 3, and using a ReLU activation function, batch normalization, and a pooling layer, the method is trained using SMAP and Landsat satellite data. The model training uses 50 rounds of iterations. After the training is completed, the 9km resolution data of SMAP is downscaled and regression analysis is performed with field measurement data to obtain higher resolution soil moisture information.

[0049] Furthermore, the assimilation frequency includes:

[0050] High frequency assimilation, normal frequency assimilation and phenological period assimilation.

[0051] Specifically, the EnKF method performs the assimilation of remote sensing data and the WOFOST model, which mainly includes two processes: prediction and update. The detailed assimilation steps are as follows:

[0052] First, in the forecast part, each sample i is advanced in parallel using the forecast model and the analysis value of the previous moment:

[0053]

[0054] Where F is the prediction model, such as the crop model WOFOST. In practical applications, if it is believed that the model itself has no structural error and the uncertainty of model drivers (such as meteorological factors) has been considered when setting the input conditions of each sample, there is no need to add the noise term e m,t,i In the update part, the EnKF algorithm uses the LAI and RZSM data inverted by remote sensing to update the LAI and RZSM simulation results of the model at time t. Based on the crop model prediction value of each sample above, the prior mean is statistically calculated. and the prior covariance matrix

[0055]

[0056]

[0057] Use the Kalman filter method to update each sample and obtain the analysis value of each sample:

[0058]

[0059] Among them, d t +e obs,t,i (i=1,...,N e ) is the remote sensing observation sample i, which is in the original remote sensing observation value d t The disturbance term e is added to obs,t,i , to avoid the ensemble variance of the posterior being too small. t is the observation operator, which is assumed to be linear here, that is, there is a linear relationship between the observation vector and the state vector. t is the Kalman gain:

[0060]

[0061] Posterior mean and the posterior covariance The matrix can be approximately obtained by statistically analyzing the values ​​of each remote sensing observation sample:

[0062]

[0063]

[0064] Finally, all remote sensing information is introduced into the model through the EnKF algorithm.

[0065] More specifically, at different growth stages, the assimilation state index (LAI) has different effects on soil moisture estimation. The following three assimilation frequencies are selected: high frequency assimilation (High), normal frequency assimilation (Normal), and phenology assimilation (Phenology). High frequency assimilation means using sentinel observation data every 5 days or so, normal frequency assimilation means using sentinel observation data every 15 days or so, and phenology assimilation means using high frequency assimilation during the main growth stage of winter wheat. The motivation for using phenology assimilation is mainly based on the following considerations: First, there are differences in the sowing time of winter wheat in different regions, which means that the higher leaf area index (LAI) observed by remote sensing technology during the sowing period may not come from the winter wheat seedlings of the current season, but from the corn stalk residues that were not harvested or burned in the previous season. Second, during the wintering period of winter wheat, since the crop is basically in a state of stopping growth, the LAI during this period is relatively stable, so frequent data assimilation is not required. Third, starting from the rising and jointing stages, the leaf growth rate of winter wheat accelerates, and data assimilation during this period becomes particularly critical. By combining remote sensing data with actual field conditions and knowledge of phenology, we were able to reduce the impact of short-term remote sensing observation errors to a certain extent, while also reducing interference from other surface covers (such as residual corn stalks) during the early growth stages.

[0066] Joint assimilation technology is a comprehensive data processing method that enhances and improves the accuracy and reliability of model predictions by simultaneously integrating information from multiple different observation sources. The method of jointly assimilating root zone soil moisture and leaf area index products is particularly worthy of attention. As an indicator that directly reflects the state of soil moisture, root zone soil moisture has a direct impact on crop water use efficiency and growth conditions. The leaf area index, as an indicator that reflects the degree of vegetation coverage and growth conditions, is closely related to the photosynthesis capacity and transpiration of crops. The joint assimilation of these two products can provide information on crop growth and soil moisture conditions from different dimensions. This method of comprehensively considering vegetation and soil properties is conducive to improving the model's ability to simulate the dynamic changes of complex soil-plant systems, thereby more accurately reflecting the actual situation.

[0067] Specifically, in terms of crop model data assimilation strategy, the present invention combines the crop leaf area index (LAI) at different growth stages with soil moisture estimation, and selects different assimilation frequencies according to the specific growth stage of the crop: high frequency (every 5 days), normal frequency (every 15 days) and phenological period assimilation. Especially in the phenological period, high-frequency assimilation is adopted, taking into account the regional differences during the sowing period of winter wheat, the stability of LAI during the winter growth stagnation period, and the acceleration of leaf growth rate from the emergence stage to the jointing stage. This strategy reduces the impact of remote sensing observation errors and other surface cover interference. The present invention also adopts joint assimilation technology to integrate multi-source data such as root zone soil moisture and LAI to improve the prediction accuracy and reliability of the model for the dynamics of the soil-plant system.

[0068] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.

[0069] The principles and implementation methods of the present invention are described in this article using specific examples. The description of the above embodiments is only used to help understand the method and core idea of ​​the present invention. At the same time, for those skilled in the art, according to the idea of ​​the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.

Claims

1. A method for estimating soil moisture using a joint assimilation strategy taking phenology into account, characterized in that: include: The downscaled root zone soil moisture data of the measured area are obtained using the calibrated WOFOST model; Determine the assimilation frequency according to the different growth stages of crops; Obtaining observation data of the area to be measured according to the assimilation frequency, wherein the observation data includes: soil moisture and leaf area index; Ensemble Kalman filtering is used to fuse root zone soil moisture data with observation data and generate assimilated data; The soil moisture of the area to be measured is determined according to the assimilated data.

2. The method for estimating soil moisture by a joint assimilation strategy considering phenology according to claim 1, characterized in that: The calibration process of the WOFOST model is: Determine weather data, soil data, crop management data and solar radiation for the area to be tested; The WOFOST model is calibrated based on the weather data, soil data, crop management data and solar radiation.

3. The method for estimating soil moisture by a joint assimilation strategy considering phenology according to claim 2, characterized in that: The weather data includes: Daily maximum and minimum temperatures, rainfall, wind speed and sunshine hours.

4. The method for estimating soil moisture by a joint assimilation strategy considering phenology according to claim 2, characterized in that: The WOFOST model is built based on the CNN convolutional neural network.

5. The method for estimating soil moisture by a joint assimilation strategy considering phenology according to claim 2, characterized in that: The assimilated frequencies include: High frequency assimilation, normal frequency assimilation and phenological period assimilation.

Images