Soil moisture monitoring method based on NSGA-BP neural network model

Abstract

The application relates to a high-precision monitoring method for ground soil moisture, in particular to a soil moisture monitoring method based on an NSGA-BP neural network model, a soil moisture model of multiple data fusion is constructed, and the surface environment data used include longitude and latitude information, NDVI, rainfall, air temperature, land cover type, slope, slope surface, elevation and shadow. Since the relationship between the above-mentioned surface environment data and soil moisture is a complex nonlinear mode, it is difficult to fuse multiple data types and draw a soil moisture map by using a traditional linear statistical regression algorithm. Compared with a traditional algorithm, a machine learning technology performs well in processing complex nonlinear problems. In particular, a genetic algorithm back propagation neural network model optimized by a genetic algorithm has high stability and good fitting property. The method can obtain high-precision soil moisture data without field monitoring, and greatly reduces the cost of obtaining high-precision soil moisture data.

Description

TECHNICAL FIELD

[0001] The application relates to a high-precision monitoring method for ground soil moisture, in particular to a soil moisture monitoring method based on an NSGA-BP neural network model. BACKGROUND

[0002] Soil moisture is an important parameter of global land surface water cycle and a physical surface quantity that people have been interested in for a long time. Large-scale monitoring of soil moisture is of great significance to agriculture, hydrology and geographical environment, and soil moisture also plays an important role in the climate system and extreme weather (such as drought, flood and submergence). The duration of extreme weather is relatively short, compared with atmospheric moisture, soil moisture has higher memory, and soil moisture can be used to judge extreme weather. Therefore, it is of great practical significance to monitor soil moisture with high precision and low cost. The precision of soil moisture measured by instruments in the field is high, but the field measurement needs to consume manpower and material resources, and the cost is high, and it is not very convenient. SUMMARY

[0003] The application provides a soil moisture monitoring method based on an NSGA-BP neural network model to realize high-precision and low-cost monitoring of soil moisture.

[0004] The application is implemented by using the following technical scheme: the soil moisture monitoring method based on the NSGA-BP neural network model has the following steps.

[0005] Establishing a data set: downloading some ground environmental factors of certain areas: longitude and latitude, normalized vegetation index, temperature, rainfall, land cover type, slope, slope surface, elevation and shadow, and dividing all the data into a training set, a confirmation set and a test set for model construction according to a proportion;

[0006] Model construction: taking the 10 ground environmental factors as inputs of the NSGA-BP neural network model, the input layer of the NSGA-BP neural network has 10 neurons, the output layer has only one neuron, and the output is soil moisture;

[0007] Model training: the NSGA-BP neural network model is trained by using the divided training set and confirmation set, during the model training, the soil moisture measured in the field is taken as a reference, the parameters of the NSGA-BP neural network model are continuously adjusted, so that the output of the NSGA-BP neural network model is aligned with the soil moisture measured in the field; and the test set is used to test the accuracy of the trained model, and the NSGA-BP neural network model with an accuracy reaching a threshold value is saved;

[0008] Accuracy verification: first, the reliability of the NSGA-BP neural network model reaching the threshold is verified by ten-fold cross-validation method; second, the verification data set is input into the trained NSGA-BP neural network model, and the NSGA-BP inversion soil moisture data set is output, and the NSGA-BP inversion soil moisture data set is compared with the measured soil moisture; if the accuracy meets the requirements, it means that the NSGA-BP neural network model trained to reach the threshold is reliable and effective; if the accuracy does not meet the requirements, the NSGA-BP neural network model is trained again until the accuracy meets the requirements.

[0009] Actual monitoring: the surface environmental factors of the soil moisture area to be monitored are input into the trained NSGA-BP neural network model, and the NSGA-BP neural network model outputs the soil moisture of the area, and the monitored soil moisture has high precision.

[0010] The application proposes a multi-data fusion learning method based on machine learning, constructs a multi-data fusion soil moisture model, and uses surface environmental data including (latitude and longitude information, NDVI, rainfall, air temperature, land cover type and four terrain factors (slope, slope, elevation and shadow)). Since the relationship between the above-mentioned surface environmental data and soil moisture is a complex nonlinear way, it is difficult to fuse multiple data types and draw a soil moisture map using traditional linear statistical regression algorithm. Compared with traditional algorithms, machine learning technology performs well in dealing with complex nonlinear problems. In particular, the genetic algorithm (NSGA) optimized genetic algorithm back propagation neural network model has high stability and good fitting. The method of the application can obtain high-precision soil moisture data without field monitoring, which greatly reduces the cost of obtaining high-precision soil moisture data. BRIEF DESCRIPTION OF DRAWINGS

[0011] Figure 1 The flowchart of the application. DETAILED DESCRIPTION

[0012] The soil moisture monitoring method based on the NSGA-BP neural network model includes the following parts:

[0013] 1. Establishing a data set

[0014] Download some surface environmental factors of certain areas: latitude and longitude, normalized vegetation index (NDVI), temperature, rainfall, land cover type, slope, slope, elevation and shadow. The surface environmental factors can be obtained in Google Earth Engine (GEE), and all data are divided into 70%, 15% and 15% as the training set, the validation set and the test set for model construction.

[0015] 2. Model construction

[0016] The present application adopts a genetic algorithm back propagation (NSGA-BP) neural network model, and takes 10 ground environmental factors as inputs of the NSGA-BP neural network model, so that the input layer of the NSGA-BP neural network has 10 neurons, and the output layer has only one neuron.

[0017] 3. Model training

[0018] The NSGA-BP neural network model is trained using the divided training set and validation set, and the test set is used to test the accuracy of the trained model. The NSGA-BP neural network model with a threshold accuracy is saved (a qualified neural network after training). During model training, the soil moisture measured in the field is used as a reference to align the output of the NSGA-BP neural network model with the soil moisture measured in the field.

[0019] 4. Accuracy verification

[0020] First, the reliability of the neural network model reaching the threshold is verified by the ten-fold cross-validation method. Second, the verification data set is input into the trained neural network model, and the NSGA-BP inverted soil moisture data set is output. The NSGA-BP inverted soil moisture data set is compared and analyzed with the soil moisture measured in the field. If the accuracy meets the requirements, it means that the neural network model trained to reach the threshold is reliable and effective.

[0021] 5. Making soil moisture map

[0022] Each 500m square in the experimental area corresponds to a latitude and longitude coordinate. The 10 ground environmental factors corresponding to all latitude and longitude coordinates in the experimental area are extracted through the latitude and longitude coordinates to form a map input data set.

[0023] The mapped input data set is input into the NSGA-BP neural network model reaching the threshold, and the NSGA-BP inverted soil moisture data set for mapping is obtained by training output. The NSGA-BP inverted soil moisture data set for mapping is imported into ArcGIS, and the "point to raster" function of ArcGIS is used to convert all NSGA-BP inverted soil moisture data sets for mapping into raster images (soil moisture maps).

[0024] 6. Verification of soil moisture map

[0025] The soil moisture map (500x500m) and NASAUSDA (0.25x0.25) product are analyzed and compared, and the data of the two are close. The NASAUSDA (0.25x0.25) product is the official standard map of the soil moisture map.

Claims

1. The soil moisture monitoring method based on the NSGA-BP neural network model is characterized by: The following steps are involved: Establish a dataset: Download the surface environmental elements of certain areas: latitude and longitude, normalized vegetation index, temperature, rainfall, land cover type, slope, slope surface, elevation and shadow, and divide all the data into training set, confirmation set and test set for model construction in proportion; Model construction: 10 surface environmental factors are used as inputs to the NSGA-BP neural network model. The input layer of the NSGA-BP neural network has 10 neurons, and the output layer has only one neuron, which outputs soil moisture. Model training: Use the divided training set and validation set to train the NSGA-BP neural network model. During model training, use the field-measured soil moisture as a reference and continuously adjust the parameters of the NSGA-BP neural network model to align the output of the NSGA-BP neural network model with the field-measured soil moisture. Use the test set to test the accuracy of the trained model and save the NSGA-BP neural network model whose accuracy reaches the threshold. Accuracy Verification: First, the reliability of the NSGA-BP neural network model that reached the threshold was tested using the ten-fold cross-validation method. Second, the validation dataset was input into the trained NSGA-BP neural network model, and the NSGA-BP inverted soil moisture dataset was output. This NSGA-BP inverted soil moisture dataset was then compared and analyzed with the field-measured soil moisture. If the accuracy met the requirements, it indicated that the NSGA-BP neural network model that had been trained to reach the threshold was reliable and effective. If the accuracy did not meet the requirements, the NSGA-BP neural network model was trained again until the accuracy met the requirements. Actual monitoring: The surface environmental factors of the area where soil moisture is to be monitored are input into the trained NSGA-BP neural network model, and the NSGA-BP neural network model outputs the soil moisture of the area.

Images