CN116681169A - Method for evaluating influence of extreme climate on crop yield - Google Patents

Abstract

The invention relates to an evaluation method of influence of extreme climate on crop yield, which comprises the steps of selecting representative research sites in a detected area, collecting historical meteorological data, soil data and crop management measure data of each research site, and correcting a crop mechanism model APSIM; simulating crop growth period, crop yield and crop biomass of each research site under a long-time sequence by using a crop mechanism model APSIM; selecting an extreme climate index capable of expressing that the crop is influenced by an extreme climate event, and calculating the extreme climate indexes of different growth stages of the crop based on meteorological observation data; optimizing an extreme climate index combination with the best effect on simulating the crop yield through a genetic algorithm; inputting the combination of the crop biomass simulation value and the extreme climate index into a random forest model, and establishing a hybrid model based on feature selection, machine learning and a crop model; the mixed model is used for simulating crop yield of a region to be tested under extreme climate conditions, and the yield simulation level of the mixed model is evaluated.

Description

Method for assessing the impact of extreme climate on crop yield

technical field

The invention relates to a method for predicting crop yields, in particular to a method for evaluating the impact of extreme climate on crop yields.

Background technique

Among the natural resources that restrict agricultural production, climatic conditions are the most important component. The impact of climate change on agricultural production has attracted widespread attention and great attention from scientists from all over the world. With global warming, the agrometeorological disasters (including high temperature heat damage, low temperature freeze damage, drought and flood) caused by frequent extreme climate events have seriously affected crop production. Therefore, accurately simulating and assessing the impact of extreme climate events on crop yield is the premise and basis for understanding and predicting the impact of future climate change on agricultural production.

Over the past few decades, the development and development of crop models based on process mechanisms has become more and more perfect, and has been widely used to assess and quantify the impact of climate change on crop production (including phenology, yield and water consumption, etc.). However, the existing crop mechanism models mainly simulate the growth process and yield formation of crops under normal climate conditions and climate change, and lack of consideration of the response process of crop production to extreme climate events, which often makes the model simulation results underestimate the impact of climate change on crop yield. negative impact.

Contents of the invention

The purpose of the present invention is to provide a method for evaluating the impact of extreme climate on crop yield, so as to effectively evaluate the comprehensive influence of different extreme climate factors (including: high temperature, low temperature, heavy precipitation and drought) on crop yield.

The purpose of the present invention is achieved like this:

A method for assessing the impact of extreme climate on crop yield, comprising the following steps:

S1. Determine a number of research sites in the measured area, collect historical meteorological data, soil data and crop data of each research site, input the above data into the crop mechanism model APSIM, and use the trial and error method to determine the parameter values of the crop mechanism model APSIM , and based on the crop phenology and yield data observed at each research site, the determined model parameters were corrected and verified;

S2. Use the crop mechanism model APSIM with corrected parameters to simulate the crop growth period, crop yield and crop biomass of each research site under the long-term sequence, so as to obtain the simulated values of crop growth period, crop yield and crop biomass;

S3. According to the extreme climate events that crops are susceptible to in different growth stages, select the extreme climate index, and on the basis of the simulated value of crop key growth period output by the crop mechanism model APSIM, use meteorological observation data to calculate and determine the crops in different growth stages extreme climate index;

S4. Use the genetic algorithm to select the extreme climate index combination with the best crop yield simulation effect from the three extreme climate indices including extreme high temperature index, extreme low temperature index and extreme rainfall index;

S5, input the simulated value of crop biomass obtained in step S2 and the preferred extreme climate index combination of step S4 into the random forest (RF) machine learning model, and set up a hybrid model based on feature selection, machine learning and crop model; use this The hybrid model simulates the crop yield of the measured area under extreme climate conditions, and evaluates the yield simulation level of the hybrid model based on the crop yield data observed at each research site; finally, evaluates the key extreme climate index of the measured area through the output of the hybrid model Relative contribution to crop yield simulations.

Further, the extreme high temperature index includes high temperature days (HD), high temperature intensity (HSI) and high temperature duration index (HCD); the extreme low temperature index includes frost days (FD), frost intensity (CSI) and cold duration index ( FCD); the extreme rainfall index includes heavy precipitation days (R25), continuous wet index (CWD) and continuous dry index (CDD).

For the three extreme climate indices, calculate their values in different crop growth stages; the calculation period of the extreme high temperature index is: jointing to flowering (JF) and flowering to maturity (FM); the calculation period of the extreme low temperature index is There are two stages: from seeding to jointing (SJ) and from jointing to flowering (JF); the calculation period of extreme rainfall index is: from seeding to jointing (SJ), from jointing to flowering (JF) and from flowering to maturity (FM). According to the combination of three types of extreme climate indices and different growth stages of crops, a total of 21 extreme climate indices were obtained.

Further, the crop biomass simulation value and the preferred extreme climate index combination are used as predictor variables, and the observed yield is used as the target value to train the random forest (RF) machine learning model to determine the random forest (RF) machine learning model. A hybrid model based on feature selection, machine learning and crop model is constructed.

Further, the crop yield simulation level of the hybrid model in step S5 is evaluated by using root mean square error (RMSE) and correlation coefficient (R ² );

The calculation formula of S5-1 root mean square error (RMSE) is:

Among them, X _obs,i is the observed yield of crops, X _model,i is the simulated yield of crops, and i is the number of research sites (1,2,...,N);

The calculation formula of S5-2 correlation coefficient (R ² ) is:

in, is the average of the observed yields

With its advantages in dealing with nonlinear problems, machine learning algorithms have been widely used in crop yield prediction modeling. The machine learning algorithm takes data as the object. It extracts the data features, discovers the knowledge in the data and abstracts the model to make predictions on the data. Before carrying out a machine learning task, it is usually necessary to perform feature selection on the acquired data to remove the allowable features and reduce the difficulty of the learning task. The present invention uses a genetic algorithm to perform feature selection on candidate extreme climate indices.

Genetic Algorithm (GA) is a method to search for the optimal solution in the process of biological evolution based on natural selection and genetic mechanism. The algorithm starts from a candidate solution set composed of a group of coded individuals, and evolves from generation to generation to produce better and better approximate solutions according to the principles of survival of the fittest and survival of the fittest. In each generation, individuals are selected according to their fitness in the problem domain, and combined crossover and mutation are performed with the help of genetic operators to generate a population representing a new solution set. This process is repeated until some convergence metric is met. Feature extraction was performed using the Caret package in the R language software, where the population size (popSize) was set to 50, the crossover probability (pcrossover) was set to 0.8, and the mutation probability (pmutation) was set to 0.1.

Plant Mechanism Model (APSIM) is a simulation model of agricultural production system that can simulate the main components of the agricultural system. This model can be used to simulate the crop growth process and the dynamics of soil water and nitrogen in the agricultural system. The potential and yield benefits of farming practices are affected by climate fluctuations and environmental changes. The APSIM model allows users to easily configure their own crop models by selecting a range of crops, soils, and other sub-modules. The logical relationship between modules can be specified very simply through the "plug-in" function of the modules. Because of its flexibility and operability, the APSIM model is considered as a flexible software environment for a model system.

The Random Forest (RF) model is a machine learning algorithm that combines multiple uncorrelated decision trees through ensemble learning. The RF model uses the Bootstrap resampling method to extract multiple samples from the original data, and constructs a base predictor (such as CART) for each Bootstrap sample, then combines the prediction results of all base predictors, and obtains the final result by voting. result. The RF modeling was performed using the randomforest package in the R language software. In the RF model, there are two main parameters that need to be set: the number of trees (ntree) and the number of variables selected by tree nodes (mtry). The model is tuned by selecting a set of candidate tuning parameter values.

The method based on feature selection in the present invention optimizes the key extreme climate factors that affect crop growth, combined with machine learning algorithms and crop mechanism model simulation, can fully consider the impact of extreme climate conditions on crop yield, and then improve the simulation of crop yield under the background of climate change ability.

The present invention drives the APSIM crop mechanism model by using meteorological data, soil data and crop management measure data of research stations, and simulates crop yields under long-term sequences of different research stations; and uses meteorological observation data to calculate extreme climate indices, and utilizes genetic algorithm features The selection method optimizes the extreme climate index combination with the best simulation effect on crop yield; input the simulated value of crop biomass and the optimized extreme climate index combination into the random forest model, and establish a hybrid model based on feature selection, machine learning and crop model. Mixed model simulations considering crop yields under extreme climate conditions and assessing the contribution of key extreme climate indices are of great significance for understanding and predicting the impact of future climate change on agricultural production and proposing countermeasures for adapting to climate change.

Description of drawings

Fig. 1 is a flowchart of the evaluation method of the present invention.

Fig. 2 is the simulation performance figure of crop model and mixed model; Wherein, (a) is the verification figure of crop model simulation winter wheat yield of the present invention, (b) is the verification figure of mixed model simulation winter wheat yield.

Fig. 3 is a time series diagram of APSIM model simulation values of growth period, yield and biomass of winter wheat.

Figure 4 is a time series graph of 21 extreme climate indices.

Figure 5 is a plot of the relative contribution of key extreme climate indices to crop yield simulations.

Detailed ways

The present invention will be described in further detail below in conjunction with the accompanying drawings and embodiments.

Example:

As shown in Figure 1, taking North China as the measured area, the impact of extreme climate in this area on winter wheat yield is evaluated, and the evaluation method is as follows:

S1. Determine the following 32 research sites (see Table 1) in North China, and collect historical meteorological data, soil data and crop data of each research site. Historical meteorological data are detailed climate data from 1981 to 2009, including daily maximum temperature, minimum temperature, radiation and rainfall, etc. Soil data include soil organic matter content, field water capacity, etc. Crop data include winter wheat phenology, yield, variety, fertilization and irrigation management measures, etc.

Table 1: Study Site Information

According to the needs of the crop mechanism model APSIM, the daily meteorological data, soil data and field management measure data (sowing methods, irrigation, fertilization, harvesting, etc.) of the corresponding years of the above research sites are input into the crop mechanism model APSIM, and the trial and error method is used to determine The parameter values of crop varieties in the crop mechanism model (the main parameters of winter wheat varieties include tt_end_of_juvenile, Startgf_to_mat, Vern_sens, Photop_sens, Potential_grain_filling_rate, Grains_per_gram_stem, Max_grain_size), and according to the winter wheat phenology and yield data observed at each research site, the determined model parameters were calculated. Calibration and verification.

According to the comparative analysis of the simulated yield and the measured yield at various research stations in North China, the correlation coefficient R ² is 0.70, as shown in Figure 2(a), indicating that the determined model parameter values are well applicable to the research stations in North China sex and representation.

S2. Using the crop mechanism model APSIM with corrected parameters, and inputting meteorological data, soil data and field management measure data strictly according to the observation data, the winter wheat growth period, winter wheat yield and winter wheat biomass of each research site under long-term sequence were analyzed. Simulation, output the simulated values of winter wheat growth period, winter wheat yield and winter wheat biomass in North China from 1981 to 2009 (see Figure 3).

S3. According to the extreme climate events that winter wheat is susceptible to in different growth stages, three extreme high temperature indices are designed, including the number of high temperature days (HD), high temperature intensity (HSI) and high temperature duration index (HCD); three extreme low temperature indices, including frost Day Number (FD), Frost Intensity (CSI), and Cold Persistence Index (FCD); and three extreme rainfall indices, including Heavy Rain Days (R25), Continuously Wet Index (CWD), and Continuously Dry Index (CDD). Among them, the calculation period of extreme high temperature index is: jointing to flowering (JF) and flowering to maturity (FM); the calculation period of extreme low temperature index is: sowing to jointing (SJ) and jointing to flowering (JF) Two stages; the calculation period of the extreme rainfall index is three stages: from seeding to jointing (SJ), from jointing to flowering (JF) and from flowering to maturity (FM). According to the combination of three types of extreme climate indices and different growth stages of winter wheat, a total of 21 extreme climate indices were obtained. The definitions of extreme climate indices in different growth stages of winter wheat are shown in Table 2. Based on the crop mechanism model APSIM, the simulated value of the key growth period of winter wheat is output, and the extreme climate index of winter wheat at different growth stages is calculated using meteorological observation data. The calculation results of the extreme climate index in North China are shown in Figure 4.

Table 2: Extreme climate indices of winter wheat at different growth stages

serial number name extreme climate index definition growth stage 1 HD_JF High temperature days Number of days with daily maximum temperature ≥ 30°C jointing to flowering 2 HD_FM High temperature days Number of days with daily maximum temperature ≥ 30°C bloom to maturity 3 HSI_JF high temperature strength The average maximum temperature of the number of days with daily maximum temperature ≥ 30°C jointing to flowering 4 HSI_FM high temperature strength The average maximum temperature of the number of days with daily maximum temperature ≥ 30°C bloom to maturity 5 HCD_JF High Temperature Persistence Index Number of days with daily maximum temperature ≥ 30°C for 3 consecutive days or more jointing to flowering 6 HCD_FM High Temperature Persistence Index Number of days with daily maximum temperature ≥ 30°C for 3 consecutive days or more bloom to maturity 7 FD_SJ frost days Number of days with daily minimum temperature < 0°C sowing to jointing 8 FD_JF frost days Number of days with daily minimum temperature < 0°C jointing to flowering 9 CSI_SJ frost intensity The opposite number of the average minimum temperature of the number of days with the daily minimum temperature < 0°C sowing to jointing 10 CSI_JF frost intensity The opposite number of the average minimum temperature of the number of days with the daily minimum temperature < 0°C jointing to flowering 11 FCD_SJ cold duration index The number of days with daily minimum temperature <0°C for 3 consecutive days or more sowing to jointing 12 FCD_JF cold duration index The number of days with daily minimum temperature <0°C for 3 consecutive days or more jointing to flowering 13 R25_SJ Heavy precipitation days Number of days with daily precipitation ≥ 25mm sowing to jointing 14 R25_JF Heavy precipitation days Number of days with daily precipitation ≥ 25mm jointing to flowering 15 R25_FM Heavy precipitation days Number of days with daily precipitation ≥ 25mm bloom to maturity 16 CWD_SJ persistent humidity index Number of days with daily precipitation ≥ 0.1mm for 3 consecutive days sowing to jointing 17 CWD_JF persistent humidity index Number of days with daily precipitation ≥ 0.1mm for 3 consecutive days jointing to flowering 18 CWD_FM persistent humidity index Number of days with daily precipitation ≥ 0.1mm for 3 consecutive days bloom to maturity 19 CDD_SJ Sustained drying index The longest consecutive days with daily precipitation < 5mm sowing to jointing 20 CDD_JF Sustained drying index The longest consecutive days with daily precipitation < 5 mm jointing to flowering twenty one CDD_FM Sustained drying index The longest consecutive days with daily precipitation < 5mm bloom to maturity

S4. Using Genetic Algorithm (GA) to select the extreme climate index combination with the best simulation effect on winter wheat yield from the above-mentioned 21 extreme climate indices. Set the population size (popSize) of GA to 50, the crossover probability (pcrossover) to 0.8, the mutation probability (pmutation) to 0.1, and take the minimum root mean square error (RMSE) as the fitness function to calculate and find the optimal extreme climate index combination . The optimal combination is 10 indexes including HD_FM, HSI_FM, HCD_FM, FD_SJ, CSI_SJ, CSI_JF, FCD_SJ, R25_SJ, CWD_FM and CDD_JF.

S5, the winter wheat biomass simulation value obtained in step S2 and the 10 extreme climate indices selected in step S4 are used as predictor variables, and the observed yield is used as the target value to train the random forest model (RF), and determine the key parameter values in the RF model. In the RF model, there are two main parameters that need to be set: 1. The number of trees (ntree); 2. The number of variables selected from the tree nodes (mtry). The value of the number of trees (ntree) is set to 100–1500, and the step size is 100; the value of the number of variables selected by tree nodes (mtry) is set to 1–15, and the step size is 1.

The RF model is tuned by selecting a set of candidate tuning parameter values composed of different ntrees and mtry. When ntree=400 and mtry=7, the constructed hybrid model based on feature selection, machine learning and crop model has the smallest root mean square error (RMSE) and the best effect in simulating winter wheat yield.

The hybrid model was used to simulate winter wheat yield under extreme climate conditions in North China, and the yield simulation level of the hybrid model was evaluated based on the crop yield data observed at each research site. Specifically, root mean square error (RMSE) and correlation coefficient (R ² ) were used to evaluate the simulation level of the mixed model for winter wheat yield.

1. The calculation formula of root mean square error (RMSE) is:

Among them, X _obs,i is the observed yield of winter wheat, X _model,i is the simulated yield of winter wheat, and i is the number of research stations (1,2,...,N);

2. The formula for calculating the correlation coefficient (R ² ) is:

in, is the average of the observed yields.

As shown in Figure 2(b), through ^the calculation of evaluation indicators, the simulation effect of the mixed model on winter wheat yield is much higher than that of the APSIM crop model. The root square error RMSE was reduced from 934.6 kg/ha in the crop mechanism model APSIM to 425.5 kg/ha in the mixed model.

As shown in Figure 5, the relative contribution of the key extreme climate indices to the simulation of winter wheat yield was evaluated through the output of the mixed model. The extreme climate indices are CSI_SJ, FD_SJ, HSI_FM, FCD_SJ and HD_FM in sequence.

Claims (4)

1. A method for assessing the effect of extreme weather on crop yield comprising the steps of:

s1, determining a plurality of research sites in a detected area, collecting historical meteorological data, soil data and crop data of each research site, inputting the data into a crop mechanism model APSIM, determining parameter values of the crop mechanism model APSIM by adopting a trial-and-error method, and correcting and verifying the determined model parameters according to crop weathers and yield data observed by each research site;

s2, simulating the crop growth period, the crop yield and the crop biomass of each research site under a long time sequence by using a crop mechanism model APSIM with corrected parameters so as to obtain simulation values of the crop growth period, the crop yield and the crop biomass;

s3, selecting an extreme climate index according to extreme climate events which are easy to be received by crops in different growth stages, and calculating the extreme climate index of the crops in different growth stages by using meteorological observation data on the basis of outputting a crop key growth period simulation value by using an APSIM (advanced programmable logic array) of a crop mechanism model;

s4, selecting an extreme climate index combination with the best effect on simulating the crop yield from three extreme climate indexes including an extreme high temperature index, an extreme low temperature index and an extreme rainfall index by using a genetic algorithm;

s5, inputting the crop biomass simulation value obtained in the step S2 and the extreme climate index combination optimized in the step S4 into a Random Forest (RF) machine learning model, and establishing a hybrid model based on feature selection, machine learning and crop model; simulating crop yield of the detected area under extreme climate conditions by using the mixed model, and evaluating the yield simulation level of the mixed model according to crop yield data observed by each research site; finally, the results are output through the mixed model, and the relative contribution of the key extreme climate indexes of the tested area to the crop yield simulation is estimated.

2. The method of assessing the effect of extreme weather on crop yield of claim 1 wherein said extreme high temperature index comprises high Wen Rishu (HD), high temperature strength (HSI) and high temperature persistence index (HCD); the extremely low temperature index includes Frost Date (FD), frost intensity (CSI), and cold persistence index (FCD); the extreme rainfall indices include strong precipitation days (R25), continuous wetting index (CWD), and continuous drying index (CDD);

calculating the values of the three extreme climate indexes in different crop growth stages according to the three extreme climate indexes; the calculation period of the extreme high temperature index is: two stages, jointing to flowering (JF) and flowering to maturity (FM); the calculation period of the extreme low temperature index is: sowing to the jointing (SJ) and jointing to the flowering (JF); the calculation period of the extreme rainfall index is: sowing to the Stage of Jointing (SJ), jointing to flowering (JF) and flowering to maturity (FM); according to the combination of three types of extreme climate indexes and different growth stages of crops, 21 extreme climate indexes are obtained.

3. The method of claim 1, wherein the combination of crop biomass simulation values and preferred extreme climate indices are used as predictive variables, and the Random Forest (RF) machine learning model is trained with observed yield as target values to determine key parameter values in the Random Forest (RF) machine learning model, and a hybrid model based on feature selection, machine learning and crop model is constructed.

4. The method for evaluating the effect of extreme weather on crop yield according to claim 1, wherein the crop yield simulation level of the mixed model in step S5 is a method using Root Mean Square Error (RMSE) and correlation coefficient (R ² ) Evaluating;

the calculation formula of the S5-1 Root Mean Square Error (RMSE) is as follows:

wherein X is _obs,i Is the observed yield of crops, X _model,i Is crop simulated yield, i is the number of study sites (1, 2.,. The.,. N.;

s5-2 correlation coefficient (R ² ) The calculation formula of (2) is as follows:

where X is the average of the observed yields.