1 Introduction

In recent years, Chinese soybean planting area is gradually reducing, the domestic soybean self-sufficiency ability has been reduced to about 20 %, China mainly imports soybean from United States, Brazil, Argentina [1]. Timely acquisition of the world soybean crop area and its spatial distribution can be used to forecast crop yields, formulate agricultural policies, ensure national food security [2, 3]. Remote sensing has characteristics of wide coverage and low cost, can provided a technical means for fast, accurate and dynamic monitoring of crop area [4].

United States, European Union, China, and other countries have designed crop area survey systems which take advantage of remote sensing and sampling method [5–7], sampling method could solves the contradiction between the accuracy, timeliness and cost. Spot, Rapideye and Landsat images are frequently used as sample image. Considering the free download of landsat8 data, bigger width of Landsat8, this research chooses landsat8 image as data source.

Crop extraction should consider electromagnetic spectrum characteristics, spatial characteristics, time characteristics and auxiliary data [8]. Supervised classification, unsupervised classification, object-oriented method and decision tree classification are frequently used methods. In crop growth cycle, the change of physiological, shape and structure of crop will resulted in seasonal changes of spectral characteristics. Considering to the characteristics of soybean in Brazil, this paper, tried to give transition area of soybean harvest in 1 or 2 times in a year and try to give classify method of image for sampling method.

2 Study Area and Data Source

2.1 Study Area

Most of Brazil belongs to tropical climate, southern parts belongs to subtropical climate. Due to the tropical climate and long growing season, crop cycles are complicated. Below is a month-by-month (Table 1) account of what to expect during growing season. Brazil soybean mainly distribute in central and southern, exists harvest 1 time in a year and 2 times in a year, according to the number of remote sensing images for soybean extraction, three research areas are selected (Fig. 1), the size of each is 40 km × 40 km. Extraction of soybean in study area 1 located in Paraná need 1 time image, which was acquired 2014-01-21; Extraction of soybean in study area 2 located in south Mato Grosso need 2 times images, which were acquired 2013-11-30 and 2014-02-02; Extraction of soybean in study area 3 located in middle Mato Grosso need 3 time images, which were acquired on 2013-11-30/2014-02-02/2014-04-07 (Table 2).

Table 1. Brazil Soybean Month-By-Month Crop Cycle
Full size table
Fig. 1.
figure 1

The sketch map of study area

Full size image
Table 2. Landsat8 Multi-spectral image
Full size table

2.2 Data

landsat8 Multi-spectral image listed as Table 2 are used.

3 Methodology

First classification step is determine the optimum classification images by visually comparison of multi-period images, which need to retrieval same path/row landsat8 images during 2013.09–2014.06. Second classification step is unsupervised classification method and visual correction.

Study area3 harvest 2 times in a year, the extraction of soybean requires three images of different acquired time. The first reason is that same plot harvest 2 times in a year, the second reason is that soybean sowing time of adjacent land vary about 1–2 month. Therefore, classification result of soybean study area3 = time1 + time2 + time3.

Study area2 harvest 1 times in a year, the extraction of soybean requires two images of different acquired time. The reason is that soybean sowing time of adjacent land vary about 1–2 month. Therefore, classification result of soybean study area2 = time1 + time2.

Study area1 harvest 1 times in a year, the extraction of soybean requires one images of acquired time. Classification result of soybean study area1 = time1 (Fig. 2).

Fig. 2.
figure 2

Flow chart of experiment

Full size image

4 Results and Discussion

Remote sensing images and classification results of the study area 3 are shown in Fig. 3. Remote sensing images and classification results of the study area 2 are shown in Fig. 4. Remote sensing images and classification results of the study area 1 are shown in Fig. 5. Image features are in Table 3 summarized from Figs. 3, 4 and 5.

Fig. 3.
figure 3

Study area 3 (3 images)

Full size image
Fig. 4.
figure 4

Study area 2 (2 images)

Full size image
Fig. 5.
figure 5

Study area 1 (1 image)

Full size image
Table 3. Image feature
Full size table

From Fig. 3 and Table 3, we can see that plot-E which harvest soybean 1 time is bare land on 20131130, soybean on 20140202, bare land on 20140407; plot-D which harvest soybean 2 times is soybean on 20131130, bare land on 20140202, soybean on 20140407. The extraction of soybean is: 639.99 km2 Result = 483.49 km2 (Time1) + 145.00 km2 (Time2) +11.50 km2 (Time3).

From Fig. 4 and Table 3, we can see that plot-C which harvest soybean 1 time is soybean on 20131130, bare land on 20140202; plot-B which harvest soybean 1 time is bare land on 20131130, soybean on 20140202. The extraction of soybean is: 495.64 km2 Result = Time1 Union Time2 (in ArcGIS).

From Fig. 5 and Table 3, we can see that plot-A which harvest soybean 1 time is soybean on 20140121. The extraction of soybean is: 398.04 km2 Result = Time1.

In this paper, we do not have accuracy assessment, because the lack of high resolution remote sensing data and field data.

5 Conclusions

Considering the spatial distribution and harvest times of Brazil soybean, using Landsat8 data, using Landsat8 data, this paper chose 3 study area, determine the optimum classification images by visually comparison of multi-period images, extract soybean by ISODATA unsupervised classification method and visual correction. The conclusion is that Landsat8 path/row of 225/75 (between Study area3 and Study area2) are soybean transition area of harvest 1 time in a year and harvest 2 times in a year; Classification result can be used for sampling survey of national scale and the full coverage survey of county scale. Soybean classification method can be used to improve the method of low resolution image and to guide other medium resolution image.