Estimation of changes in land area under wheat and soybean cultivation using satellite images classification techniques in west of Golestan province

Estimation of cultivated area is controlled on the basis of some features such as seed rate, fertilizer and other delivered chemical inputs which are costly andtime consuming. In this way, using satellite data as a new solution not only reduces the drawbacks of conventional estimation methods of cultivated area , but also can be helpful for various programming aims in the agriculture scope. Land cover mapping is important for much planning and management activities. Today, satellite images and remote sensing techniques are extensively used in all sectors, including agriculture and natural resources because they provide updated data and high analyzing abilities. It is one of the fastest and most cost-effective methods to map those lands are available for researchers. In recent years, sattelite imagery and different detection methods have helped researches to detect the features at a lower cost and spend less time. Different methods are avialbale for this purpose.. Each method has some advantages and disadvantages. Artificial neural network, fuzzy logic, support vector machine, decision tree, object-oriented classification and intelligent systems can be considered as advanced classification methods (Guo et al., 2012). In this study, we want to detect wheat and soybean-grown fields with two advanced classification methods (i.e. support vector machine (SVM) and artificial neural network (ANN)) in west of Golestan Province, Iran. Cultivation areas were determined from those maps extracted from satellite images to use them as the base layer for other research goals.

The current research was conducted with the aim of estimating the wheat- and soybean-grown areas during a 16-year period from 2000 to 2016 using Landsat satellite imagery. To this purpose, two classification methods, support vector machine and artificial neural network, were used. In order to classify and detect aforementioned two crops, ground control points (GCPs), the normalized difference vegetation index (NDVI) for agricultural lands, and spectral behavior of wheat and soybean training GCPs were involved. In order to validate the results of classification, the generated maps were checked by GCPs (coordinated with GPS).

Wheat and soybeans were at the maximum vegetative growth in May and September, and were well-detected from other crops. So the images of these two months were used for detection in all studied years. According to the results of previous studies, support vector machine and artificial neural network could be used as two reliable image classification methods for detecting the crops vegetation cover (Rahimzadegan & Pourgholam ,2017; Mokhtari & Najafi, 2015). The accuracy of image classification was assessed using kappa coefficient and overall accuracy. Kappa coefficient and overall accuracy showed the support vector machine method was superior than artificial neural network method in classifying agricultural lands and detecting studied wheat- and soybean-grown fields. In the all images, the calculated overall accuracy coefficient was more than 80% (0.84 to 0.92 for wheat; and 0.84 to 0.90 for soybean) and Kappa coefficient was more than 0.8, indicating the reliability of the classification outputs. According to comparisons of satellite image-based estimations and real recorded ststistics, about 93% of the estimated areas of wheat - and soybean grown fields in the 16 consecutive years of study were within the range of 15% confidence level, which indicates that this method is a reliable method for detecting these two crops using images of April (for wheat) and September (for soybean).

The support vector machine method of classification was identified as the superior method. wheat- and soybean-grown fields maps extracted from satellite images can be used as a base layer for regional modeling, providing yield gap layer, calculating the water requirement, designing crop pattern and etc