Farmland precise irrigation decision-making method based on fusion of physical process and machine learning

Abstract

The invention discloses a farmland precise irrigation decision-making method based on fusion of a physical process and machine learning, and the method comprises the following steps: S1, data collection: collecting meteorological data, soil data, fertilization and irrigation data and cotton growth data in a cotton growth period, S2, data preprocessing: carrying out the cleaning, standardization and normalization processing of an original data set, and carrying out the calculation of the data set; the method comprises the following steps: S1, preprocessing a crop mechanism model, S3, data enhancement: constructing and calibrating the crop mechanism model by using preprocessed data, S4, selecting key feature variables, and dividing the obtained data set into a training set and a test set by using an original data set and an enhanced data set, S5, model training and optimization, and S6, model verification and evaluation. According to the method, by coupling the crop growth mechanism model and the machine learning algorithm, not only can the growth conditions of crops under different moisture conditions be simulated and predicted, but also the influence of soil salinity and alkalinity on crop growth can be evaluated, so that comprehensive decision support is provided for irrigation management.

Description

Technical Field

[0001] The present invention relates to irrigation decision-making technology, and in particular to a method for precise farmland irrigation decision-making based on the fusion of physical process and machine learning. Background Art

[0002] With global climate change and population growth, the rational use of water resources and the sustainability of agricultural production have become global challenges. Agricultural irrigation is the main link in water resource consumption in agricultural production, and its efficiency is directly related to crop yield and sustainable use of water resources. Salinized soils are widely distributed in Xinjiang, seriously affecting crop growth and agricultural output. Traditional irrigation methods cannot accurately meet the water needs of crops, and fail to effectively reduce soil salinity, resulting in restricted crop growth and waste of water resources. The development of precision agricultural irrigation technology needs to consider not only the water needs of crops, but also how to reduce soil salinity through irrigation measures, thereby improving the utilization efficiency of irrigation water, reducing ineffective evaporation and deep infiltration, and maintaining the optimal water conditions required for crop growth.

[0003] Crop growth models have always played an irreplaceable role in precision agriculture because they can quantitatively describe and predict the growth and development process of crops and their dynamic relationship with the environment. They can describe the growth process and development of crops as a complex function of weather, soil and management, which can be easily explained from the perspective of plant physiology. However, the calibration of many parameters of crop models usually requires time-consuming and resource-intensive field experiments, but these parameters are difficult to generalize under different varieties and environmental conditions. Generally, the huge variability of environmental conditions, coupled with the choice of model structure and parameters, limits the predictive performance of these models, and their predictive ability may decline under variable environmental conditions. At the same time, machine learning algorithms have shown great potential in the agricultural field with their powerful data processing capabilities and predictive performance. However, due to its "black box" structure, lack of mechanism, and weak interpretability of prediction results, it has great limitations in analyzing the crop growth process. In addition, to improve the prediction accuracy of machine learning models, a sufficiently large data sample is required, but in the process of crop planting and production, it is often difficult to obtain sufficient sampling data. Summary of the invention

[0004] In view of the above-mentioned defects or deficiencies in the prior art, it is desired to provide a method for precise farmland irrigation decision-making based on the fusion of physical processes and machine learning.

[0005] According to the technical solution provided in the embodiment of the present application, the farmland precision irrigation decision-making method based on the fusion of physical process and machine learning includes the following steps:

[0006] S1. Data collection: collect meteorological data, soil data, fertilization and irrigation data, and cotton growth data during the cotton growth period to obtain an original data set, and use the original data set as the basis for model training and verification. The data used for model training is the training data set.

[0007] S2, data preprocessing, cleaning, standardizing and normalizing the original data set to obtain preprocessed data to meet the model input requirements,

[0008] In this step, the cleaning, standardization and normalization of the original data set are mainly achieved through the following methods:

[0009] Here is a simple R code example showing some basic data cleaning steps:

[0010] # Load necessary libraries

[0011] library(dplyr)

[0012] #Assume there is a database named data

[0013] # Use this line of code to read the CSV file

[0014] #data<-read.csv("path_to_your_data.csv")

[0015] #View data structure

[0016] str(data)

[0017] #1. Dealing with missing values

[0018] # You can choose to delete rows with missing values

[0019] data_clean<-data[complete.cases(data),]

[0020] #Or fill missing values ​​with specific values, such as the mean value

[0021] data_clean<-data data_clean$column_name[is.na(data_clean$column_name)]<-mean(data_clean$column_name,na.rm=TRUE)

[0022] #2. Dealing with outliers

[0023] #Suppose we define outliers as data with more than 3 standard deviations

[0024] data_clean <- data_clean %>%

[0025] mutate(column_name=ifelse(abs(column_name-mean(column_name,na.rm=TRUE))>3*sd(column_name,na.rm=TRUE),NA,column_name))#Fill missing values ​​again

[0026] data_clean$column_name[is.na(data_clean$column_name)]<-mean(data_clean$column_name,na.rm=TRUE)

[0027] #3. Delete duplicate data

[0028] data_clean <- data_clean %>%

[0029] distinct()

[0030] #4. Format data

[0031] #For example, convert the date column to date format

[0032] data_clean$date_column<-as.Date(data_clean$date_column,format="%Y-%m-%d")

[0033] #5. Select the required columns

[0034] data_clean <- data_clean %>%

[0035] select(column1,column2,column3)

[0036] #6. Save the cleaned data

[0037] write.csv(data_clean,"cleaned_data.csv",row.names=FALSE)Data standardization can help improve the performance of the model, especially when using distance-based algorithms. In R, the most common are Z-score standardization and minimum-maximum standardization. The following is an example code for data standardization using R:

[0038] # Load necessary libraries

[0039] library(dplyr)

[0040] #Assume there is a cleaned database data_clean

[0041] #You can use this line of code to read the .csv file

[0042] #data_clean<-read.csv("cleaned_data.csv")

[0043] #1. Z-score standardization

[0044] #This method transforms the data into a form with zero mean and unit variance

[0045] data_standardized<-data_clean%>%

[0046] mutate(across(where(is.numeric),scale))

[0047] #View the standardized data

[0048] head(data_standardized)

[0049] #2. Min-max normalization

[0050] #This method scales the data to the interval [0,1]

[0051] data_minmax_scaled<-data_clean%>%

[0052] mutate(across(where(is.numeric),~(.-min(.,na.rm=TRUE)) / (max(.,na.rm=TRUE)-min(.,na.rm=TRUE))))

[0053] #View the minimum-maximum standardized data

[0054] head(data_minmax_scaled)

[0055] #3. Save the standardized data

[0056] write.csv(data_standardized,"standardized_data.csv",row.names=FALSE)

[0057] write.csv(data_minmax_scaled,"minmax_scaled_data.csv",row.names=FALSE)

[0058] Data normalization is scaling the data to a specific range, usually [0,1] or [-1,1], which helps improve the convergence speed and performance of some algorithms. In R, you can use the `dplyr` package to help with data normalization. The following is an example code for data normalization using R:

[0059] # Load necessary libraries

[0060] library(dplyr)

[0061] #Assume that there is already a cleaned database data_clean

[0062] #Use this line of code to read the data file .csv

[0063] #data_clean<-read.csv("cleaned_data.csv")

[0064] #1. Min-max normalization

[0065] #This method scales the data to the interval [0,1]

[0066] data_normalized<-data_clean%>%

[0067] mutate(across(where(is.numeric),~(.-min(.,na.rm=TRUE)) / (max(.,na.rm=TRUE)-min(.,na.rm=TRUE))))

[0068] #View the normalized data

[0069] head(data_normalized)

[0070] #2. L1 Normalization

[0071] #This method scales the absolute value of the data to 1

[0072] data_l1_normalized<-data_clean%>%

[0073] mutate(across(where(is.numeric),~. / sum(abs(.))))

[0074] #View the data after L1 normalization

[0075] head(data_l1_normalized)

[0076] #3. L2 Normalization

[0077] #This method scales the data to unit vectors

[0078] data_l2_normalized<-data_clean%>%

[0079] mutate(across(where(is.numeric),~. / sqrt(sum(.^2))))

[0080] #View the data after L2 normalization

[0081] head(data_l2_normalized)

[0082] #4. Save the normalized data

[0083] write.csv(data_normalized,"normalized_data.csv",row.names=FALSE)

[0084] write.csv(data_l1_normalized,"l1_normalized_data.csv",row.names=FALSE)

[0085] write.csv(data_l2_normalized,"l2_normalized_data.csv",row.names=FALSE)

[0086] In this code:

[0087] -`mutate(across(where(is.numeric),~(.-min(.,na.rm=TRUE)) / (max(.,na.rm=TRUE)-min(.,na.rm=TRUE))))`: This line of code performs min-max normalization on all numeric columns and scales the data between 0 and 1.

[0088] - `mutate(across(where(is.numeric),~. / sum(abs(.))))`: This line of code performs L1 normalization on all numerical columns, scaling the absolute values ​​of the data to 1.

[0089] - `mutate(across(where(is.numeric),~. / sqrt(sum(.^2))))`: This line of code performs L2 normalization on all numeric columns and scales the data to unit vectors.

[0090] Through the above methods, the original data set can be cleaned, standardized and normalized.

[0091] S3, data enhancement, using the preprocessed data to construct and calibrate a crop growth model based on physical processes, using the calibrated crop mechanism model to perform scenario analysis, and simulating crop growth data and soil data on non-sampling dates, combining daily meteorological data, soil data and irrigation and fertilization data, expanding the training data set in step S1 to obtain an enhanced data set;

[0092] S4. Select key characteristic variables, integrate the characteristic variables that affect crop growth and development using the original data set and the enhanced data set, and establish a characteristic variable table; calculate the relationship between the characteristic variables through correlation analysis, remove the characteristic variables that do not contribute significantly to crop growth and development, and retain the remaining characteristic variables as key characteristic variables;

[0093] S5, model training and optimization, dividing the enhanced data set obtained in step 3 into a training set and a test set, wherein the ratio of the training set to the test set is 8:2, and a random search method is used to optimize the hyperparameters of the machine learning model based on the training set and using the key feature variables to improve the prediction accuracy of the model;

[0094] S6. Model validation and evaluation. The prediction performance of the coupling model was evaluated using 10-fold cross validation on the test set, and indicators such as root mean square error, mean absolute error, and determination coefficient were used for evaluation.

[0095] In the present invention, further, the meteorological data is collected by an automatic weather station in the field, and the data is recorded every ten minutes, including the maximum temperature, the minimum temperature, the average temperature, the relative humidity, the wind speed, the solar radiation, and the rainfall.

[0096] In the present invention, further, the soil data includes soil bulk density, field water holding capacity, soil hydraulic characteristic parameters, soil particle size, soil basic nutrients, soil moisture content, and soil salinity.

[0097] In the present invention, further, the fertilization and irrigation data are recorded when the cotton is fertilized and irrigated, and the fertilization and irrigation time and fertilizer amount are recorded.

[0098] In summary, the beneficial effects of this application are as follows: by coupling the crop growth mechanism model and the machine learning algorithm, this method can not only simulate and predict the growth of crops under different water conditions, but also evaluate the impact of soil salinity on crop growth, thereby providing comprehensive decision support for irrigation management, reducing water waste during irrigation, and reducing the environmental impact of agricultural production. In addition, the model is highly flexible and scalable, and can adjust irrigation strategies according to climate conditions and soil types in different regions to achieve sustainable development of precision agriculture. BRIEF DESCRIPTION OF THE DRAWINGS

[0099] Other features, objects and advantages of the present application will become more apparent by reading the detailed description of non-limiting embodiments made with reference to the following drawings:

[0100] Figure 1 It is a schematic diagram of the overall steps of the present invention;

[0101] Figure 2 This is a schematic diagram comparing cotton plant heights at different time periods;

[0102] Figure 3 This is a schematic diagram comparing the cotton leaf area index in different time periods; Figure 4 This is a schematic diagram comparing the dry matter accumulation of cotton in different time periods. DETAILED DESCRIPTION

[0103] The present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It is to be understood that the specific embodiments described herein are only used to explain the relevant invention, rather than to limit the invention. It is also necessary to explain that, for ease of description, only the parts related to the invention are shown in the accompanying drawings.

[0104] It should be noted that, in the absence of conflict, the embodiments and features in the embodiments of the present application can be combined with each other. The present application will be described in detail below with reference to the accompanying drawings and in combination with the embodiments.

[0105] like Figure 1 As shown in the figure, a method for farmland precision irrigation decision-making based on the integration of physical process and machine learning is shown in the figure. The specific steps are as follows:

[0106] 1. Data Collection

[0107] In this embodiment, the data is divided into sampling data and simulation data. The sampling data mainly includes meteorological data during the cotton growth period, basic soil nutrient data, soil texture data, soil water and salt content data, crop growth data, and fertilization and irrigation data. The simulation data mainly refers to the soil water and salt data, cotton growth and yield data predicted by the crop mechanism model. The sampling data specifically includes:

[0108] 1.1 Meteorological data

[0109] The data is collected from the automatic weather station in the field and recorded every ten minutes, including maximum temperature, minimum temperature, average temperature, relative humidity, wind speed, solar radiation, rainfall, etc.

[0110] 1.2 Soil data

[0111] (1) Soil bulk density, field water holding capacity, and soil hydraulic characteristic parameters: Before the experiment, soil bulk density, field water holding capacity, and soil moisture characteristic curve were measured using the ring knife method. The ring knife was taken at a distance of 0-20, 20-40, 40-60, and 60-80 cm from the ground surface, and repeated 5 times.

[0112] (2) Soil particle size: Before the experiment, the soil was evenly collected after air drying and sieving, and the soil particle size and proportion were measured using a laser particle size analyzer. This was repeated 5 times.

[0113] (3) Basic soil nutrients: Before sowing, samples were taken every 20 cm in the 0-80 cm soil layer. The test soil was air-dried and sieved for measuring total nitrogen, available nitrogen, phosphorus, potassium, conductivity, and organic matter. The total nitrogen content was determined by AA3 flow analyzer, the available potassium by flame photometry, the available nitrogen by alkaline diffusion method, the available phosphorus by sodium bicarbonate extraction method, and the extract conductivity EC was determined by DDSJ-308A (Shanghai Yidian) portable conductivity meter. 1:5 , organic matter was determined by colorimetry.

[0114] (4) Soil moisture: Soil moisture is measured by soil drying. Soil is collected before each irrigation and at the beginning and end of each growth stage. The soil sampling locations are the center of the film room and two locations below the drip irrigation belt. The sampling depths are 0-20, 20-40, 40-60, and 60-80 cm, respectively. Soil is collected once before each irrigation, and at the end of each growth period, the soil moisture content after irrigation is measured. The soil moisture content after irrigation is also measured once after irrigation in the middle of the flowering and boll period.

[0115]

[0116] Where: θ: soil moisture content (%)

[0117] m0: mass of the dried empty aluminum box, g;

[0118] m1: mass of aluminum box and soil sample before drying, g;

[0119] m2: Mass of aluminum box and soil sample after drying, g.

[0120] (5) Soil salt content: Soil salt content is characterized by conductivity value, and the sampling method is the same as soil moisture content. The soil sample is dried and crushed, and 20g of the soil sample that has passed a 1mm sieve is weighed into a triangular flask, and 100mL of distilled water is added. The mixture is shaken for 10min, and then filtered with filter paper after standing for 15min to prepare a soil water extract with a soil-water mass ratio of 1:5. The soil salt content (g·kg -1 ), and converted into soil salt mass fraction after calibration (y=3.4021x+1.9535, R 2 =0.9569; y is soil salt content, g·kg -1 , x is the conductivity, dS·m -1 ).

[0121] 1.3 Fertilization and irrigation data

[0122] It is mainly recorded when cotton is fertilized and irrigated, mainly recording the irrigation and fertilization time, amount, etc.

[0123] 1.4 Crop Data

[0124] (1) Leaf area index: The leaf area index (LAI) is calculated every 15 days after the cotton plant enters the seedling stage. The length and maximum width of the leaves of the marked cotton plants are measured with a ruler. The leaf area is determined by the coefficient method. The cotton leaf area index (LAI) is calculated using the formula:

[0125]

[0126] Where, LAI: leaf area index;

[0127] K: leaf area conversion coefficient, taken as 0.75;

[0128] ρ: cotton density (actual effective number of plants);

[0129] m: number of plants measured;

[0130] n: number of leaves of the ith plant;

[0131] L ij , W ij : are the maximum length and maximum width of the j-th leaf of the ith cotton plant respectively.

[0132] (2) Biomass: Destructive sampling was used to measure the dry matter mass. Six representative plants were randomly selected from each treatment. The fresh weights of the plants were measured after they were separated into stems, leaves and reproductive organs. The plants were then placed in an oven at 105°C for 30 min and dried at 75°C until constant weight was reached. The dry matter mass of each organ and the dry matter mass ratio of each organ were then calculated based on the cotton planting density.

[0133] (3) Yield: After stopping irrigation, the yield was measured when the cotton boll opening exceeded 80%. A sample plot of 2.28 × 2.93 m in size was randomly selected in each plot, the harvest density was observed, the cotton bolls were picked, the weight of 100 bolls and the boll opening seed cotton yield were weighed, and the yield components such as the number of plants, the number of bolls, the remaining bolls, and the weight of 100 bolls were recorded. The theoretical yield of the remaining bolls was calculated based on 0.75 times the weight of a single boll. The 50 bolls that were weighed were deseeded to obtain the lint yield and calculate the cotton lint percentage.

[0134] (4) Water use efficiency: WUE ET (kg·m -3 ) is calculated as:

[0135] WUE ET =0.1y / ETa

[0136] Where: y represents seed cotton yield (kg·hm -2 ); ETa is the water consumption of cotton fields during the growth period (mm2), and 0.1 is the unit conversion coefficient.

[0137] 2. Construct the original data set and the enhanced data set from the sorted data set, and divide the sampled data set into a training set and a test set, as follows:

[0138] 2.1 Construction of original data set: The sampled data is divided into training set and test set (8:2), the collected data is cleaned, outliers and missing data are removed, the training set and test set are normalized using minimum-maximum normalization, and saved in corresponding Excel files respectively.

[0139] 2.2 Enhanced data set construction: In order to reduce the workload of manual sampling and processing and provide sufficient training data sets for machine learning, the data corresponding to the unsampled dates are expanded through the crop mechanism model to simulate the data of non-sampling dates before harvest, and combined with daily meteorological data and fertilization and irrigation data to form a partial data set. On the other hand, scenario analysis is carried out to expand the training set data. The test set is still the original data set, and the training set is normalized using minimum-maximum normalization and saved in the corresponding Excel files.

[0140] 2.3 Feature extraction of original training set and enhanced training set: The Pearson correlation coefficient is used to calculate the correlation between the input parameters of the original training set and the enhanced training set and the crop growth index, and the significantly correlated input parameters are selected as the input features of the machine learning model. The model is constructed for the original data set and the enhanced data set according to the selected significantly correlated features. The randomized search (RandomizedSearchCV) method is used to optimize the hyperparameters of the machine learning model based on the training set and the key feature variables. The machine learning model is trained using the optimized hyperparameters, and the prediction accuracy and generalization ability of the model are evaluated using 10-fold cross validation on the test set.

[0141] 2.4 Introduce automatic parameter adjustment tools: Use machine learning models to establish a relationship between the input data of the mechanism model and the calibration parameters, limit the parameter adjustment range according to the training data set, introduce the PEST automatic parameter adjustment tool, determine the optimal soil parameters and crop parameters, simplify the parameter adjustment process of the mechanism model, reduce the calculation complexity, and improve the applicability of the model.

[0142] 3. Model performance evaluation: The experimental performance of single models and coupled models on the test set is evaluated by root mean square error (RMSE), mean absolute error (MAE) and coefficient of determination (R2).

[0143]

[0144] where y i represents the measured value, shi is the predicted value, yˉ represents the average value of the reference sample, and n is the total number of samples.

[0145] 4. Irrigation management decisions

[0146] In the present invention, irrigation management decisions are made based on the prediction results of a coupled model that comprehensively considers crop growth conditions, soil moisture content, and soil salinity.

[0147] 4.1 Crop growth monitoring

[0148] Use remote sensing technology to monitor crop growth conditions in real time, including plant height, leaf area index, biomass, etc. ( Figure 2 ), based on the prediction results of the coupling model, evaluate whether the crop growth is in line with expectations (the evaluation basis is whether the crop growth condition is better than the actual crop growth condition under the W2 irrigation quota).

[0149] 4.2 Analysis of soil moisture content

[0150] Field sensors are used to monitor the soil moisture content at 0-80 cm in real time. The coupling model will adjust the irrigation strategy based on the real-time data and prediction results of the soil moisture content to ensure that the crops receive appropriate water supply. The lower limit of moisture used is a mathematical model calculated based on the soil moisture balance principle described in the invention patent (CN 107945042 A), which includes the formula:

[0151] Minimum soil moisture content W min =WP+0.75*(FC-WP),

[0152] Maximum soil moisture content W max =ST,

[0153] Among them, WP is soil moisture wilting coefficient, W min ——Minimum allowable soil water storage, W max ——Maximum allowable soil water storage capacity, FC——Soil field water holding capacity, ST——Soil saturated water content,

[0154] When the water storage in the planned wet layer is close to or less than W min , irrigation is needed at this time to replenish water.

[0155] 4.3 Control of soil salinity and alkali content

[0156] Field sensors are used to monitor the soil salt content at 0-80 cm in real time. The coupling model considers the salt thresholds of cotton in each growth period shown in Table 1 as the upper limit of cotton affected by salt stress. According to the model prediction results, an irrigation plan to reduce soil salt is formulated. The relationship between soil desalination rate and irrigation quota is:

[0157] y=0.2875x-19.781(R 2 =0.92)

[0158] Among them, y is the desalination rate (%), x is the irrigation quota (mm)

[0159] Table 1 Salt thresholds of cotton at different growth stages

[0160]

[0161] 4.4 Irrigation management decisions

[0162] The coupled model provides decision support for irrigation time and amount by integrating crop growth conditions, soil moisture content, and soil salt content. During the irrigation process, the response of crops and soil is monitored in real time, and the irrigation strategy is adjusted when necessary. After irrigation, the crop growth index, soil moisture and salt content are measured again to evaluate the irrigation effect, and the evaluation results are fed back to the coupled model for model calibration and optimization. At the same time, considering climate change and changes in crop growth cycles, the irrigation strategy is dynamically adjusted to adapt to changing environmental conditions.

[0163] The above description is only an explanation of the preferred embodiments of the present application and the technical principles and other schemes used. At the same time, the scope of the invention involved in the present application is not limited to the technical scheme formed by a specific combination of the above technical features, but also should cover other technical schemes formed by any combination of the above technical features or their equivalent features without departing from the inventive concept. For example, the above features are replaced with the technical features with similar functions disclosed in the present application (but not limited to) by each other to form a technical scheme.

Claims

1. A method for precision irrigation decision-making based on the integration of physical processes and machine learning, which is characterized by , including the following steps, S1. Data collection: collect meteorological data, soil data, fertilization and irrigation data, and cotton growth data during the cotton growth period to obtain an original data set, and use the original data set as the basis for model training and verification. The data used for model training is the training data set. S2. Data preprocessing: cleaning, standardizing and normalizing the original data set to obtain preprocessed data to meet the model input requirements. S3, data enhancement, using the preprocessed data to construct and calibrate a crop growth model based on physical processes, using the calibrated crop mechanism model to perform scenario analysis, and simulating crop growth data and soil data on non-sampling dates, combining daily meteorological data, soil data and irrigation and fertilization data, expanding the training data set in step S1 to obtain an enhanced data set; S4. Select key characteristic variables, use the original data set and the enhanced data set to integrate the characteristic variables that affect crop growth and development, and establish a characteristic variable table; Through correlation analysis, the relationship between each characteristic variable is calculated, and the characteristic variables that do not contribute significantly to crop growth and development are removed, and the remaining characteristic variables are retained as key characteristic variables; S5, model training and optimization, dividing the enhanced data set obtained in step 3 into a training set and a test set, wherein the ratio of the training set to the test set is 8:2, and a random search method is used to optimize the hyperparameters of the machine learning model based on the training set and using the key feature variables to improve the prediction accuracy of the model; S6. Model validation and evaluation. The prediction performance of the coupling model was evaluated using 10-fold cross validation on the test set, and indicators such as root mean square error, mean absolute error, and determination coefficient were used for evaluation.

2. The method for farmland precision irrigation decision-making based on the fusion of physical process and machine learning according to claim 1 is characterized by: The meteorological data are collected by an automatic weather station in the field, and the data are recorded every ten minutes, including the maximum temperature, the minimum temperature, the average temperature, the relative humidity, the wind speed, the solar radiation, and the rainfall.

3. The method for farmland precision irrigation decision-making based on the integration of physical process and machine learning according to claim 1 is characterized by: The soil data include soil bulk density, field water holding capacity, soil hydraulic characteristic parameters, soil particle size, soil basic nutrients, soil moisture content, and soil salinity.

4. The method for farmland precision irrigation decision-making based on the integration of physical process and machine learning according to claim 1 is characterized by: The fertilization and irrigation data are recorded when cotton is fertilized and irrigated, and record the irrigation and fertilization time and fertilizer usage.

Images