High-spatial-resolution multilayer soil moisture simulation method based on machine learning

Abstract

The invention relates to the field of soil moisture data, and discloses a machine learning-based high-spatial-resolution multilayer soil moisture simulation method, which comprises the following steps of S1, performing large-scale multi-source remote sensing data fusion by utilizing an ESTARFM model based on a GEE platform, so as to obtain high-spatial-temporal-resolution NDVI data and LST data; step S2, on the basis of an Extreme Gradient Boosting model, in combination with reanalysis and actual measurement data, estimating moisture information of soil at the depth of 0 to 100 cm; and S3, evaluating the precision of the soil moisture model by using indexes such as a Pearson correlation coefficient, a root-mean-square error, a skew value and a mean absolute error. An enhanced space-time adaptive reflection fusion model is realized on a Google Earth Engine platform, NDVI and LST derived from MODIS and NDVI and LST derived from Landsat are fused, NDVI and LST data with a spatial resolution of 30m and other soil moisture background field data are generated every 8 days, and the precision of the soil moisture model is evaluated by using the indexes such as the Pearson correlation coefficient, the root-mean-square error, the skew value and the mean absolute error. Comprising rainfall and soil texture variables, and soil humidity data is actually measured by combining an extreme gradient enhancement model with the variables.

Description

Technical Field

[0001] The present invention relates to the field of soil moisture data, and in particular to a high spatial resolution multi-layer soil moisture simulation method based on machine learning. Background Art

[0002] At present, the research and development of soil moisture data has received extensive attention. Whether it is using visible light bands, infrared bands or microwave bands, the theoretical methods for inverting soil moisture have basically matured. The main methods include: water deficit index, soil thermal inertia, temperature vegetation drought index, and microwave band-based soil moisture inversion. The relationship between backscattering coefficient and soil moisture is established to further invert soil moisture. However, although the soil moisture data obtained through visible light and infrared bands have high spatial resolution, it is difficult to solve the problem of obtaining soil moisture information under the influence of clouds. The obvious disadvantage of using microwave data to obtain soil moisture is that the spatial resolution is very low, and most of the soil moisture information is obtained at a spatial resolution of tens of kilometers.

[0003] Although some progress has been made in the research on calculating soil moisture data using multi-source remote sensing data, there is still a lot of room for improvement. It is difficult to combine the characteristics of long time series, high spatiotemporal resolution, and multiple layers. Most data sets are limited to improving data characteristics in one aspect. In particular, due to the limitations of the data, the lack of understanding of the multi-layer soil moisture characteristics in some areas has brought uncertainty and difficulties to soil moisture prediction. Therefore, it is necessary to further develop soil moisture monitoring technology suitable for large scales to obtain more comprehensive, continuous and refined soil moisture data. To this end, the present invention proposes a high spatial resolution multi-layer soil moisture simulation method based on machine learning to solve the above problems. Summary of the invention

[0004] The purpose of the present invention is to provide a high spatial resolution multi-layer soil moisture simulation method based on machine learning to solve the problems raised in the above background technology.

[0005] To achieve the above object, the present invention provides the following technical solution: a high spatial resolution multi-layer soil moisture simulation method based on machine learning, comprising the following steps:

[0006] Step S1, based on the GEE platform, the ESTARFM model is used to fuse large-scale multi-source remote sensing data, thereby obtaining high temporal and spatial resolution NDVI data and LST data;

[0007] Step S2, based on the Extreme Gradient Boosting model, the soil moisture information at a depth of 0-100 cm is estimated by combining the reanalysis and measured data;

[0008] Step S3, using Pearson correlation coefficient, root mean square error, deviation value and mean absolute error to evaluate the accuracy of soil moisture model.

[0009] Preferably, the step S2 estimates soil moisture information in ten layers, one layer at a time of 10 cm.

[0010] Preferably: the calculation of the correlation coefficient R in step S3 is as follows,

[0011]

[0012] Preferably: the root mean square error RMSE of step S3 is calculated as follows:

[0013]

[0014] Preferably, the calculation process of the deviation value bisa in step S3 is as follows:

[0015]

[0016] Preferably, the mean absolute error RMSE calculation process of step S3 is as follows:

[0017]

[0018] Preferably, the step S3 further calculates ubRMSE to characterize random errors and eliminate bias. The calculation process is as follows:

[0019]

[0020] Compared with the prior art, the present invention has the following beneficial effects:

[0021] The present invention implements an enhanced spatiotemporal adaptive reflectance fusion model (ESTARFM) on the Google Earth Engine (GEE) platform, fuses the NDVI and LST derived from MODIS with the NDVI and LST derived from Landsat, generates NDVI and LST data with a spatial resolution of 30 m every 8 days, and other soil moisture background field data, including precipitation and soil texture variables; and uses the extreme gradient boosting (XGBoost) model to combine the above variables with the measured soil moisture data. BRIEF DESCRIPTION OF THE DRAWINGS

[0022] Figure 1 is a flow chart of the present invention;

[0023] Figure 2 This is a result diagram of an embodiment of the present invention. DETAILED DESCRIPTION

[0024] 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.

[0025] Example

[0026] See also Figure 1 The high spatial resolution multi-layer soil moisture simulation method based on machine learning in the figure includes the following steps: Step S1, based on the GEE platform, the ESTARFM model is used to fuse large-scale multi-source remote sensing data to obtain high temporal and spatial resolution NDVI data and LST data; Step S2, based on the Extreme Gradient Boosting model, the soil moisture information at a depth of 0-100 cm is estimated in combination with reanalysis and measured data; Step S3, the accuracy of the soil moisture model is evaluated using indicators such as the Pearson correlation coefficient, root mean square error, deviation value and mean absolute error.

[0027] In this embodiment, the ESTARFM model is used to solve the problem of parameter downscaling. By integrating multi-source data, the XGBoost model is applied to generate 30m×30m soil moisture every 8 days. This model is implemented on the GEE platform for the first time to downscale NDVI and LST data. Based on the XGBoost model, a large-scale, high-spatial-resolution, long-term series and multi-layer soil moisture simulation is realized. The ESTARFM model is used to fuse the NDVI data obtained by MODIS and Landsat. After the quality control procedure, the soil moisture measured data observations from 53 stations every ten days in the Northeast region of the China Meteorological Administration from May to September 2000 to 2004 are used as soil moisture modeling.

[0028] The main goal of the ESTARFM model is to generate synthetic images with high spatial and temporal resolution by fusing information from multiple remote sensing image sources (such as MODIS and Landsat). It is based on the following principles: Spatial adaptive fusion: ESTARFM uses the spatial information of high-resolution images (such as Landsat) to guide the synthesis process to obtain higher spatial resolution. It refines low-resolution images (such as MODIS) into high resolution by spatial diffusion and correction. Temporal adaptive fusion: ESTARFM takes into account the temporal changes of remote sensing images and fuses images at adjacent moments to obtain more accurate synthetic images.

[0029] The input of the ESTARFM model includes low-resolution image sequences (such as MODIS image sequences) and high-resolution images (such as Landsat images). It generates high-resolution synthetic images by analyzing the temporal and spatial characteristics of low-resolution images and corresponding them with high-resolution images using appropriate fusion algorithms. In this study, the ESTARFM model was implemented in GEE to produce NDVI and LST with a spatial resolution of 30M for 8 days in a large-scale range of the study area.

[0030] When using the Xgboost model to calculate multi-layer soil moisture, multiple parameters are used as model input parameters and the measured soil moisture data of 0-100cm every 10cm layer to calculate the multi-layer soil moisture covering the study area. The surface soil moisture is calculated first, and then the soil moisture data of the second layer and deeper layers are calculated using the soil moisture data of the previous layer as the main parameter.

[0031] To comprehensively evaluate the simulated soil moisture performance, five indicators were used, including correlation coefficient (R), bias, and root mean square error (RMSE). In addition, ubRMSE was calculated to characterize random errors and provide a more reliable RMSE estimate by eliminating bias. The statistical indicators were calculated as follows:

[0032]

[0033]

[0034] In this embodiment, refer to Figure 2 Verification test for this embodiment: In order to evaluate the simulation effect of the XGBoost model on the soil moisture of 10 layers (0-100cm) in the typical black soil area in Northeast China, the meteorological data sharing network was used to obtain the soil moisture data of 27 meteorological observation stations in the region to verify the XGBoost model. The accuracy analysis of each soil layer was performed, and the measured data were divided into training set and validation set in a ratio of 7:3. The model analysis was performed on the 10 layers of soil moisture data in the range of 0-100cm. The results showed that the worst values ​​of R, RMSE, ubRMSE and Bias of the training set were 0.86, 1.49, 1.49 and -0.039, respectively. For the validation set, the lowest R value was 0.83.

[0035] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0036] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A high spatial resolution multi-layer soil moisture simulation method based on machine learning, characterized in that: The steps include: Step S1, based on the GEE platform, the ESTARFM model is used to fuse large-scale multi-source remote sensing data, thereby obtaining high temporal and spatial resolution NDVI data and LST data; Step S2, based on the Extreme Gradient Boosting model, the soil moisture information at a depth of 0-100 cm is estimated by combining the reanalysis and measured data; Step S3, using Pearson correlation coefficient, root mean square error, deviation value and mean absolute error to evaluate the accuracy of soil moisture model.

2. The high spatial resolution multi-layer soil moisture simulation method based on machine learning according to claim 1 is characterized in that: The step S2 estimates soil moisture information for ten layers, one layer at a time of 10 cm.

3. The high spatial resolution multi-layer soil moisture simulation method based on machine learning according to claim 2 is characterized in that: The calculation of the correlation coefficient R in step S3 is as follows:

4. The method for simulating multi-layer soil moisture with high spatial resolution based on machine learning according to claim 3, characterized in that: The root mean square error RMSE of step S3 is calculated as follows:

5. The method for simulating multi-layer soil moisture with high spatial resolution based on machine learning according to claim 4, characterized in that: The calculation process of the deviation value bisa in step S3 is as follows:

6. The method for simulating multi-layer soil moisture with high spatial resolution based on machine learning according to claim 5, characterized in that: The mean absolute error RMSE calculation process of step S3 is as follows:

7. The high spatial resolution multi-layer soil moisture simulation method based on machine learning according to claim 6 is characterized in that: The step S3 also calculates ubRMSE to characterize random errors and eliminate bias. The calculation process is as follows:

Images