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

Digital analysis of salinity of soil using multisource data

Peng Wanglu
Dept. of Geography Beijing Normal University,
Beijing, China

Li Tianjie
Institute of Environmental Sciences Beijing Normal
University, Beijing, China


Abstract
This paper the research work of Stalinization of soil at the YANGGAO region YANBEI China. For the highly precise quantitative analysis of the Stalinization not only remote sensing data Tm or MSS but also two non remote sensing data are needed depth of ground water and mineralization rate of ground water according to the theory of genesis of soil for the analysis of compounded multisource generalized Bays classification is used on the that various information sources are independent global membership function with probability are used to combine various information in order to make direct operation to the pixels and classifications of the salinity of Soil The experiment u order to make direct operation to the pixels and because increased speed of processing it's simplicity and improved precision of classification of the salinity of soil. The experiment indicated that this analysis method is sound of salinity At last MSS data of 1977 and TM data of 1986 after processing are compared for getting change of Stalinization during 10 years This worm indicates that the computer quantitative analysis of compounded multisource is one of effective research mean of salinized soil.

Introuction
The research of soil salinization ad the harnes of land -generation is one of emphasis of pedology geography and environment science. The interpretation of land sat remote sensing images and the computer processing are the important means to make qualitative quantitative and dynamic analyses. But the landsat remote sensing images are synthetic reflection of spectral features of various factors such as type soil combination soil covering structure and soil forming factor Consequently it is very incomplete for analyzing solonetz-solonchak\with only spectral feature because the influence of other factors can not be ignored in the view of genesis of soil it is indispensable to combine remote sensing data with ground in the view of genesis relationship among regional topography hydrology hydrogeology and soil data to study to realize the quantitative analysis of salinized soil it is imperative to improve the accuracy of discrimination and to make macroscopic analysis of the interrelation to improve between soil water motion and other factors of geography environmental conditions.

Recent years ynthetic quantitative mostly has used the methods of step by step with remote sensing data and non-remote sensing data (1) after the discrimination with remote sensing data the fuzzy parts of the different type of targets and same spectrum are found then further analysis will be made according to non remote sensing data 2 after classifying the level with non remote sensing data such as slope ,altitude etc the data in the region of each level are reclassified with remote sensing data so as avoid some indistinct surface features of different levels Geographic information system can be used a tool for these kind of operation.

In order to make quantitative analysis of salinized soil by means of the theory of genesis of soil the experiment use both remote sensing data and non remote sensing data experts experiences and increases processing speed and accuracy After registering land sat data of different times the dynamic change can be compared this work can be considered as a scientific basis for the work of transforming local soil.

The principle of classification with compound multi-information.
According P.H.Swain J.A Richards and T.Lee(1) remote sensing data or non remote data may be regarded as independent sources data their locations must be matched accurately A pixel can be regarded as measurement
X =[x1,x2,......xs....xn]T, s=1,2,3,......n, n is the number of independent sources the information class of a pixel denotes wj,
J =1,2 .......M . M is the number of information classes The Dsi (i=1,2,…. Ms ) indicates the ith c;lass of the sth source the function f (Dsi/xs) indicates the strength of association between xs of sth source. and ith class Dsi the function indicates the strength of association between the ith data class dsi of the data source S (relating to xs) and the information class wj last global membership function

F'j{f[Wj/dsi(xs)] rs | i = 1,2,...m, s = 1,2,..........m}

is used. The rs is the weight (the "quality factor" for the source S. In consequence the discrimination rule the pixel X of all source is

If

F°= Max.Fj(j = 1,2,..M)

Then
X is in class W*
From byes classification theory

(Fj(X) = P(Wj/X)= P(Wj/x1,x2,x3.........xs.......xn)


according to the hypothesis "statistical independence of sources" (ignore the weight of sources) the discrimination can be simplified as


The synthesis processing method extends the Bayes classification for remote sensing data brings the thematic elements of non-remote sensing in to the probability statistics theory for the analysis of classification which can be called generalized Bayes classification.

The Procedure of Experiment

  1. Investigating the Environment of Salinized Soil in the experimental Area

    The region studied in this paper is located at the south west part of YANGGAObasin in SHANXI province nearly 113°50" E, 40°20" N this area in the temperate semi arid steppe belt where soil types are mostly meadow and salinized soil so the differentiation Is the distinct and the type of salinized soil are mostly mixed type there exists various degree of salinization.

    It is saline soil and highly salinized soil that the salinity of topsoil is greater than 1.0 percent and 0.6 percent to 1.0 percent respectively which mainly distributes at the alluvial plan along BAIDENG river the soil salinity of which is from 0.4 percent to 0.6 percent namely moderately salinized soil is at the north of the highly salinized soil region the soil salinity of which is from 0.1 percent to 0.4 percent namely weakly salinized soil and or potentially salinized soil is located at a big sides of high weekly salinized soil and or distributes crisscross with moderately salinized soil.it is non salinized soil that the salinity of topsoil is less than 0.1 percent A long which drainages free is of lower mineralization rate to the north of alluvial inclined plan the soil that is weakly salinity and sufficient moisture is non salinized soil.

    According to the field investigation in late June the variety of vegetations and it's growths are related with salinity the wheat with high coverage that is growing well is mainly planted in the area of non salinized soil. Other plants which are low coverage such as chestnut sweet potato watermelon soybean etc. are all sediments in the various salinized area the highly salinized soil areas appear a stretch of white flat land,.

  2. The Preprocessing of the Tm image.

    On the Tm image 1986 the highly salinized soil area in which the soil is high reflect rate for each band appears white the mon salinized spoil area in which the plants are growing well appears red only in the moderately and weakly and salinized soil areas there is little chromatic difference and the soil is difficult to be distinguished .

    In order to get the understanding features of soil and to find the relation between brightness of pixels an the corresponding natural scences three images are compared and classified the false color image of first color image three components after K-L principal component transformation the false color image first three components after K-T Tasseled Cap transformation and the false color image of band 4,3,2, of Tm image the result of comparison is that the image after K-T has the most evident of soil characteristics and the best classification result with only three bands and become the main remote sensing sources of the multi information classification.

  3. The Selecting of Non Remote Sensing Sources.

    The salinized soil in the YANGGAO regions is principally affected by ground water The salinized soil of the ground water type ids formed with the water circulation within ground water soil and atmosphere the buried depth and the mineralization rate of ground water is definitely related to the salinity of soil in general the salinity of soil is weaker in the area with deep depth an low high mineralization rate of ground water the period of the peak of salification is in June and July Comparing the depth and mineralization rate with other factors such as topography surface water buried topography etc they are most important for the salinity research and as the non remote sensing sources in this paper .

    From the soil experiment by expects the two probability tables are listed for the statistical algorithm.

    Table 1. Probability for depth of ground water
    Level Depth of ground water (m) Saline soil and highly salinized soil(%) mid-salinized soil(%) weakly salinized soil(%) non-salinized soil(%)
    1
    2
    3
    4
    5     		>2.4
    2.0-2.4
    1.8-2.0
    1.4-1.8
    1.0-1.4    		 0
    0
    5
    10
    60-70    		 0
    5-10
    10-20
    60-70
    20-30    		 20-30
    30-40
    60-70
    20-30
    10    		 70-80
    50-60
    15
    0
    0


    Table 2 probability for mineralization rate of ground water
    Level mineralization rate of ground water
    (mg./1.)    		 Saline soil and highly salinized soil
    (%)    		 mid-salinized soil
    (%)    		 weakly salinized soil
    (%)    		 non-salinized soil
    (%)
    1
    2
    3
    4
    5
    6     		<600
    600-800
    800-1000
    1000-3000
    3000-6000
    >6000    		 0
    0
    0
    20-30
    70-80
    80-90    		 0
    0
    10-20
    70-80
    20-30
    10-20    		 10-20
    60-70
    70-80
    0
    0
    0    		 80-90
    30-40
    5-10
    0
    0
    0

  4. Generalized Bayes Analysis of Multisource Data

    For the equation (1) n=3 the multi source data are multi spectral TM image as a remote sensing data the depth and the mineralization rate of ground water as the non remote as the non remote sensing data. The steps of processing are reading TM data into computer digitizing the contour maps of both depth and mineralization rate of ground water then matching them with TM image after transformation of vector format to raster format by the common like hood supervised classification the value P(wj/xTM) can be got from the training areas of the TM data after K-T class area to total area for each level in the training fields or the valuesP(Wj/xs) from geography observation experiment this paper uses the latter one.

    According maximum like hood classification rule in the condition of same prior probability the discriminating function of the probability P(wj/xTM)may be written.

    PTM= In |Ej|-(X-Mj)T Ej-1 (X-Mj)

    Here Mj Ej are mean vector and covariance matrix of class J respectively then equation (1) could be simplified as

    Fj(X) = PTM + In [PD(wj/x)] + In (wj/x)]---------------(2)

    Here TM and D and M for TM remote sensing data depth and mineralization rate of ground water respectively if prior probabilities of the classes are not the same equation (1) should be used.

  5. Observing of the Dynamic change

    With comparison between MSS image in the March of 1977 the changes of salinized soil be Between 1977 and 1986 has been done the analysis methods have two ways one is the classification of multi source from MSS then comparing the result with the result of TM multi source classification the other is the matching between TM and MSS images both after transformation (with another matrix factor for MSSdata then subtracting 1977 image from 1986 image bright parts in the new image show the change area s which Indicate the shrinkage of salinity field the result transforming saline alkali land in tpo farmland capital construction.
Results and Conclusion
The analysis of soil using multisource data combined remote sensing and non remote sensing more scientific and more accurate according to the theory of genesis of soil table 3 shows the areas of classes with different methods the approximately reference data are from a soil salinity map in YANGGAO area mapped from the interpretations of the TM image comparing the classification result of preprocessed TM image and modification after synthetic classification of multi source data . The confidence of the classification of multi source for salinity is increased and this kind of algorithm can improve the implementation of classification in speed and in simplicity also can use for updating GIS database.

Table-3 The result of classification
proportion in total area saline soil
(%)		 highly salinized soil
(%)		 mid-salinized soil
(%)		 weakly salinized soil
(%)		 non-salinized soil
(%)		 water
(%)
map of
salinity
only TM
data modification
of multisource
data

2.5

2.8

0



7.5

11.2

-0.7
30.7

30.4

0.5
52.1

48.0

0.3
4.5

4.0

0
2.7

3.2

0


The dynamic inspecting of soil salinity could be accomplished using the data of different times the true changes can not be recognized completely in this experiment because the MSS image in March of 1977 has a low resolution when it is not strong salification period Nevtheless some tackled field is a still shown in the images for example a big salinized field at the TM image BAIDENG river is decreased and some highly salinized fields are obvious in the TM image salinized indicates the seriousness Stalinization after the comparison it is known a lot of work still remains to do for improving the large area salinized soil.


In brief the analysis of salinity using multi source data not only can improve the quantitative result the data of the same time but also can analyze the dynamic change with the data of different time this method is effective for the quantitative inspecting managing and improving of the salinized soil Note that fig 1,2 white error color is for saline soil red highly salinized soil light red for mid salinized green for weekly light blue for non salinized and blue for water.

Reference
  1. P.H Swain J.A Richards and T.Lee, Machine processing of remotely senses data symposium 1985 pp 211 218

  2. Wanghu Peng Tianjie Li preprint on 40th congress of the international astronautical federation (1989)

  3. R.J Kauth G.S Thmos IEEE Symposium proceedings of machine processing of remotely sensed data (9176 ) pp 4 B 41 4 B 5