Метод мониторинга влажности почвы, покрытой растительным покровом, с использованием нейронной сети, радарных и мультиспектральных оптических данных Sentinel-1,2

Abstract

In this article, a method for the moisture monitoring of vegetation covered soil was proposed using neural network, radar and optical multispectral data of Sentinel-1,2. Test site was chosen in the Volgograd region at an agriculture field. The moisture retrieval algorithm is based on the use of a neural network to predict reflection coefficient of an electromagnetic wave from the soil, followed by inversion into soil moisture using a dielectric model that takes into account the soil texture. The input parameter of the neural network is the ratio of the microwave radar vegetation index (calculated on the basis of Sentinel-1 data) to the multispectral optical index (calculated on 8-11 channels of the Sentinel-2). Such way calculated index reveals a significantly greater dependence on soil moisture than on vegetation height. The retrieved values of soil moisture were compared with the moisture content of in-situ selected soil samples, which were measured under laboratory conditions by the thermostatic-weight method. The proposed method with a determination coefficient of 0.435 and a standard deviation of 2.4 % allows predicting the soil moisture content of a test area covered with vegetation, relative to soil moisture measured in-situ. The conducted research creates the scientific basis for a new all-weather technology for remote sensing the moisture content of agricultural soils as an element of the precision farming system.