Open AccessAssessment of possibility for using remote sensing data and Markov chains for prediction of vegetation cover development
Author(s) -
Tamara Myslyva,
Vera Ivanovna Bushueva,
Victoria Volyntseva
Publication year - 2020
Publication title -
vescì nacyânalʹnaj akadèmìì navuk belarusì. seryâ agrarnyh navuk
Language(s) - English
Resource type - Journals
eISSN - 1817-7239
pISSN - 1817-7204
DOI - 10.29235/1817-7204-2020-58-2-176-184
Subject(s) - remote sensing , raster graphics , markov chain , vegetation (pathology) , normalized difference vegetation index , raster data , environmental science , computer science , leaf area index , geography , machine learning , artificial intelligence , medicine , ecology , pathology , biology
In conditions of global climate change, it is important to develop reliable models allowing to reliably predict plant development based on combination of the Earth remote sensing data and statistical modeling. Modeling by means of Markov chains is an efficient and at the same time simple way to predict random events, which include prediction of performance of phytomass of agricultural crops. The Earth remote sensing data obtained from the Sentinel-2 satellite with spatial resolution of 10 m were used to calculate the value of vegetation index NDVI and obtain different time rasters (2017-2019) with different degrees of vegetation cover development. To construct the matrix of probability of transition from one state to another for different levels of vegetation cover development, functionality of geoinformation systems (GIS) were used allowing to classify raster images, transform them into vector layers, and establish intersection areas. The probability matrix was later used to predict vegetation cover development using the Markov model as a predictor. The developed prediction model was tested for feasibility of the χ 2 test. The results obtained showed that both the modeled values and the actual area of vegetation distribution with different degrees of development, determined from the available raster image of 2019, correlated well with each other. The research results can be useful both in developing forecasting methods and in directly predicting the crop yield of primarily dense-cover agricultural crops, as well as for estimating performance of pastures and creating efficient pasture rotations.