Soil Moisture Investigation Utilizing Machine Learning Approach Based Experimental Data and Landsat5-TM Images: A Case Study in the Mega City Beijing

Yue Qu,Jinqiang Tan,Xu Qian,Yi Xing,Hongqing Song,Zhengyi Li

doi:10.3390/w10040423

Abstract

The characteristics of soil moisture content (SMC) distribution in an area are necessarily analyzed for the design and construction of sponge cities. Combining remote sensing data with experimental data, this paper establishes a machine learning model to reveal the characteristics of SMC. Taking Beijing as an example, the SMC distribution was obtained and the characteristics were analyzed after training and validating. When comparing different machine learning methods, it can be concluded that the support vector classifier (SVC) method trained with remote sensing and grayscale data can achieve the highest accuracy (76.69%). The calculation results show that the districts with the highest and lowest SMC value are Xicheng District (19.94%) and Daxing District (11.04%), respectively, in Beijing. The mean SMC value of Beijing is 15.65%. The SMC distribution characteristic in Beijing shows that the soil in the west and north are relatively wet, while the soil in the east and south are relatively dry. Therefore, it is suggested that the timely monitoring of the SMC of vegetation covered areas at the north and west should be carried out. Water conservation facilities also need to be established with the development of city constructions in the south and east areas.

Highlights

Modern cities should have functions of absorbing, purifying, and utilizing rainwater like sponges in order to prevent extreme rainfall, reduce runoff, and improve the ecological environment
With 2500 remote sensing data used as the training set and 502 remote sensing data used as the validation set, the machine learning is carried out on six wave bands of remote sensing data from Landsat4–5 covering the whole Beijing area in terms of the support vector classifier (SVC) algorithm, decision tree algorithm, and k nearest neighbor algorithm, respectively
The reason for this classification is that most of the Soil moisture content (SMC) values are less than 25% according to the above experimental results and because the soil in which the SMC values are more than 25% can be treated as water

Summary

Introduction

Modern cities should have functions of absorbing, purifying, and utilizing rainwater like sponges in order to prevent extreme rainfall, reduce runoff, and improve the ecological environment. The initiative seeks to reduce the intensity of the rainwater runoff by enhancing and distributing the seepage capacities more evenly across targeted areas. This approach reduces flooding, and enhances groundwater replenishment by building wetlands (which will store rainwater) and laying down permeable roads. Under complex weather and climate conditions, soil moisture is widely used for potential runoff and flood control, and for soil erosion, drought warning, water resources management, and other related fields [11,12,13,14]. It makes great sense to obtain the SMC spatial and temporal distribution and analyze its characteristics

Methods

Results

Discussion

Conclusion