GISdevelopment.net ---> AARS ---> ACRS 1990 ---> Poster Session

The estimation of cotton-growing areas by Remote Sensing

Li Zewen, Jiang Dong, Lu Denghuai
China Institute of Land Survey & Planning

Zhang Cuizhi
Beijing Municipal Academy of Agriculture & Forestry Science


Abstract
From 1986 to 1990, using remote sensing, the authors estimated the cotton-growing areas in Henan province, China, surveyed the farming calendar of the main crops, studied the development features, and the spectral reflectance, selected the Landsat TM digital image of cotton's heading initial and middle periods of ripening, adopted the maximum likelihood method to conduct the supervised classification of TM images. We gained data and distribution of cotton-growing areas. The research proves that it is feasible to use remote sensing technology to estimate the cotton-growing areas. The methodology has features of strong objectivity, high speed and low cost. It provides a reliable technological method to monitor cotton-growing areas.

Preface
The cotton-growing areas in china are obtained by the way of presenting the itemized report to the higher level. This way will be affected by man-made factors and limited by the presenting time and statistic method. It is impossible to obtain the accurate data of cotton-growing areas in time. But using remote sensing to estimate the cotton-growing areas before harvest, so that it can provide important basis for decision-making departments to estimate the cotton yield in advance.

General situation of test area
Huojia county lies on Northern Henan plain, with 437.15 km2 in area. It possesses abundant resources of water, soil, climate and organisms. The terrain is smooth, the traffic convenient, the condition of social economic technology better and the level of productivity higher. The dominant crops are wheat, rice, maize, cotton and soybean, etc.

The main contents of the research
The particular contents are as follow:
  • To research the crop growth and development features of cotton, wheat, maize, rice and soybean, etc. to seek the development characteristics of other crops; and to seek the maximum difference of growing period between cotton and other crops;
  • To determine the spectral reflectance of cotton in different growing periods so as to provide basis to select the best band in TM data to identify cotton;
  • To select the types and acquired data of remote sensing data;
  • To select proper classification method in order to improve the accuracy of estimation of the cotton-growing areas.
The development features of cotton and farming calender of the test area
Cotton is perennial woody plant, originated in tropical and subtropical grassland with features of high temperature, drought and short time of sunshine. it has been changed into annual plant after having introduced a fine variety, cultivated and domesticated by human beings for a long term the life of cotton begins with the germination of seeds, the develops into the vegetative growth i.e. growing roots, increasing leaves, growing stems, branching, etc., on this basis, conduct the development of reproduction of squaring, heading, flowering, yield formation and ripening, etc. and finishes its life till the seeds are ripen. According to the order of the formation of its organs and the process of development, the development of cotton can be divided into five growing periods: germination and emerge period vegetative period (including establishment and head development periods), flowering period, yield formation period and ripening period. [2]

In order to seek the maximum difference of growing periods between cotton and other crops, we have surveyed the farming calender of dominant crops of wheat, maize, soybean and cotton in the test area. The development of these crops is as Table 1.

Table 1. Table of growth and development of dominant crops


From Table 1. we can see that in the periods of flowering and ripening the difference of growing situation between cotton and other crops is the largest, so the difference shown on the satellite image should be the largest. it is most suitable to choose the satellite image of cotton's flowering or ripening period to conduct classification.

Spectrum determination of cotton and the corresponding crops
Different growth and development steps of crop-coenosium have different tones. the way that the crops absorb and reflect spectrum are different. it is the basis of estimation of cotton-growing areas to choose the phonological periods which the colony's tones and compositions are most different from the other crops. In the test area, the crops have the same periods of growth and development as cotton are soybean, wheat, peanut, sweat potato and maize, etc. (called the corresponding crops). Therefore, we should pay special attention to seek maximum differences of spectrum of growth and development periods between cotton and other crops.

In 1987, the research group used the objects spectrometer to determine the spectral reflectance of cotton, the corresponding crops i.e. soybean, peanut, sweat potato and maize, and the exposed soil in periods of cotton's heading (July 4), end of budding and beginning of blossom (July 23), flowering and bulling (August 1), yield formation (August 25), initial and middle periods of ripening (Sept. 29) middle and late periods of ripening (October 10). We have found that, in the periods of cotton's heading, initial and middle reopening, the difference of characteristic curve of spectrum between the cotton fields and the other crops' is evident (Figure 1.)


Figure 1. The characteristics curves of spectrum of cotton and the
corresponding crops in each of the growing and developing period in 1987


In 1988, the research group, separately determined the spectral reflectance of cotton and the corresponding crops in the periods of cotton's establishment (July 23), heading (July 10), yield formation (August 29), initial and middle periods of ripening (Sept. 28). parts of their characteristic curves of spectrum are as Figure 2.


Figure 2. The characteristics curves of spectrum of cotton and the
corresponding crops in each of the growing and developing period in 1988


The spectral data acquired in 1987 and 1988 showed that the heading period and initial and middle periods of ripening are the most favorable time to distinguish the cotton fields from other objects. it is unanimous with that, shown in table 1 of the development period of cotton and the corresponding crops, the difference between cotton and other crops in this two periods is the greatest.

The selection of Remote Sensing data
In order to get the higher accuracy of estimation, the TM digital image data of high resolution power should be selected n the cotton's heading period and initial and middle periods of ripening.

We selected the July 12,1986 TM digital images of Huojia country which was the heading period of local cotton. Band 2,3,4 of TM digital images which are the most favorable to the distinguishment of soil and vegetation are mainly used.

Classification
We used the maximum likelihood method to conduct the classification. The steps are as follows:
  1. Preparation of Classification
    Before classifying, the digitizer was used to load the administrative boundary of Huojia country mapped in the topographic maps at a scale of 1:50,000 into computer so as to form a 2-value image called boundary file. The image presented a rectangle with 512X1024 pixels. Each pixel was assigned the value 0 or 1, e.g. the pixels; value of outside country-boundary was equal to 0, and inside equal to 1.

    At first, by using 3 X 3 pixels sampling method, the spectral information of bands 2,3,4, was extracted and the edge enhancement was conducted. Then, based on hydrographic net and network of communication lines, five well-identifiable and correspondent points on the image and topographic maps were selected as control points to dealt with location, enlargement and whirling, etc. it makes the homologous points of the image and topographic maps overlap each other; Finally, we selected images of bands 2,3,4 that were unanimously as the boundary file form bands 2,3,4 and treated them as the objects of classification. The objects multiplied by the boundary file was the result that all the values of pixels outside the Huojia country were 0, it was analogous to a special type.

  2. Determination of the number of types
    In the classification, cotton was regarded as a major type, named RED2, its corresponding cropsi.e. soybean, wheat, and sweat potato were named RED 1. RED3, and RED4 respectively; water area and residential area named WATER and VI; the other objects were merged into one type, named OTHER. So we get seven types.

  3. Encircling of Training Sites In the TM false colour composite images of bands 2,3,4 it was very easy to distinguish the water area and residential area, we used a tracking boll to encircle these two types of training sites. In the TM false colour composite images, the crops of cotton, wheat sweat potato and soybean, etc. display different thick or thin red tones. Based on the result of colour enhancement and cluster analysis of bands 2,3,4 colour composite image and complied with the ground truth data and the result of spectral analysis, we encircled the training sites one by one.

    After having compared and revised repeatedly and determined six types of training sites, we conducted clustering process once again and counted the emergence probability of this six types in order to provide parameters for normal classification.

  4. Classification based on the Maximum Likelihood Method
    The result of supervised classification for the first time was not all of them unanimous as the ground truth. After having revised the limits of classification, we conducted the maximum likelihood method nice again, and got the numbers of pixels of seven types (Table 2) and the special distribution of each type (Fig. 3) finally.

    Table 2. The result of TM Classification by
    maximum likelihood method
    Type order Type name Pixel Percentage
    1
    2
    3
    4
    5
    6
    7    		 Water
    Red1
    Red2
    Red3
    Red4
    VI
    Other    		 88
    4856
    6696
    23034
    8489
    18319
    0    		 0.14
    7.90
    10.90
    37.80
    13.80
    29.80
    0.00
    Total 61482 100.00



    Fig. 3 The map of classification


  5. Analysis of Accuracy
    According to the figures of ground survey in Huojia country, the cotton fields were 11% of the total area of the country in 1986. In this classification, we got 6696 pixels of cotton fields, 10.90% of the total area of the country. If the data of ground survey were true values, the classification accuracy was 97.43%.

    Houjia country lies on the Northern Henan Plain and suitable for cotton planting. In 1986, cotton was mainly planted on its areas of mound land of Xunfengling in Midwest, low-lying land linked with Taihang mountain in the north, the side dyke depression of ancient Yellow river in the central part and the lands around the residential areas of the side dyke depression of ancient Yellow river in the east. The result of the classification was identical with this distribution situation.
References
  • Institute of cotton of Chinese Academy of Agricultural Science, Cultivation Science of Cotton in China, page 54. Shanghai publishing house of science and technology, Dec. 1983.