You are on page 1of 11
34 __EARSel. ADVANCES IN REMOTE SENSING Vol. 2, No.3 - XI, 1993 Application of Satellite Remote Sensing to Soil and Land Use Mapping in the Rolling Hilly Areas of Nanjing, Eastern China ‘Zhang Ming, Rudi Goosens & L.Daels Lab. of regional geography and landscape study, The State University of Gent, Krijgslaan 281, S8-Al, B-9000 Gent, BELGIUM ABSTRACT ‘The main purpose of the study is to make a maximum use of remote sensing data and GIS techniques to assess land use and soil classification in the rolling hilly area around Nanjing, eastern part of China (Sub-tropical region). Landsat MSS data perform a classification using the ILWIS software (Integrated Land and Watershed Informa- tion System developed by ITC, Ensched, The Nether- lands). ‘The supervised classification was based upon a multispectral analysis and on “ground truth” (maps and existing reports). The minimum distance to mean classi- fier, the maximum likelihood classifier and the box classi- flectance rial, the soil classifica to 4 units and 10 ins 5 classes. It has been fication were used. According to the characteristics of the surface n tion indicated 7 classes that belo subunits, The land use map conta proved, that, within limitations, classification algorithms and threshold parameters have an important influence on the classification result and should be selected carefully based on the training area, INTRODUCTION Study aim ‘The main purpose of the study is to examine the possi bilities of using remote sensing data for a general survey of the soil conditions and land use in the study area This examination consists of 4 parts 1. The establishment of the relationships between the ground truth and the images, in this case based upon reports and maps, and the satellite images. 2. The establishment of the correlation between the soil conditions and the present day landuse, vegetation and other landscape factors 3. The establishment of the most appropriate enhiance- ‘ment techniques and the best classification methods for the detection of the different soil conditions. 4. The mapping of the different soils using satellite image classification. Study Region ‘The hilly area around the Nanjing region is located at 31°08"-33°02°N and 118°41"-119°56°E (Fig, 1a). ‘This re- gion extends from the Huai-He River in the North to the west boundary with the An-Hui province, Mainly it con- sists of undulated rolling hills and Jow mountains with some plains formed in the delta of the rivers; especially along the Yangtze River. It covers an area of about 13388.81 km’ of which about 76.5% is formed by undu- lating rolling hills and low mou 3.5% is plains and polder. Geologically, the low mountain area consists of sandstone, quartzes, conglomerate or limestone; the hilly area is covered by Quaternary deposits of which the Xiashu loess of the Late Pleistocene Epoch is predominant, some mag- matic outerops are observed; the plain and polder along the Yangtze River and the inter mountain basins are com- posed of alluvium on which most of the paddy soil is found. The Paddy soils are formed under the farming conditions of alternating dryness and irrigation. Oxidation and reduction play a dominant role in the soil-forming process. The redoxic layer resulting of submerging the profile is the most important diagnostic horizon of the Paddy soil The vegetation consists of a large number of plant species with a high rate of growth, and the range in variety of vegetation resources is great, Because of the influence of man over many years, the typical natural vegetation char acterised by broad-leaved evergreen and deciduous mixed forest is found only in some remote parts of the hills and low mountains where a nature reserve was established by Ming er al: Application of Sutellite Remote Sensing to Soil and Land Use Mapping... 35 Fig. 1a Location map of the study area, the hilly area af Nanjing region, in the central eastern part of China. b. Schematic drawing of the training area located in the central part ofthe study region. the government. In the area where people are engaged in productive activities the land use consists of field crops (rice, wheat, maize etc), Mason’s Pine (Pinus masso- niana) forest, Chinese fir (Cunninghamia lanceolata) stands, and some economic forest, shrubs and grasses. ‘The training area, Jianglin county. isa part of the hilly area of the Nanjing region in the middle east of China (Fig. 1b) ‘The climate is a transitional zone between the north sub- tropical and the temperate region with rainfall between ‘900 mnv/year and 1200 mm/year. Ibis located on the south bank of the Yangtze River. Another main river, Qing-hua river, goes through the study area from south-east to north-west. There are several low mountains and hills located on both sides of the Qing-hua river. They are mainly consisting of limestone, sandstone, basalt and are also covered by loess material in some places (especially in the lower part of the hills and plain area). Along the Qing-hua river, alluvial materials are dominant. 1, DOCUMENTS AND MAIN SOIL TYPE, DESCRIPTIONS 1.1 Data used Landsat MSS Data (CCT with path/row 129/38) dated 13 June 1979 was analysed using the ILWIS image process- ing software package (Integrated Land and Watershed Information System developed by ITC. Enschede, The Netherlands) on a PC/386 computer system in conjunction with published soil survey reports/maps, supported by limited field investigations. The image size is 512 * 480 pixels, representing an area of about 1100 km”, 1.2 Documents During the study, the following materials have provided the primary information: © Soil maps 1:125000 scale for sample area (1979), 1; 1000000 for the whole region (1986): = Land use maps, 1:25000 scale (1986). Vegetation maps, 1:1000000 scale (1986). Climate maps, 1:1000000 scale (1986); - Soil survey report which was section of Jianglin county in 1979; - Vegetation, geology, climate, land use survey and soil resource management report, which have been published by the science committee of Jiangsu province. + Topographic maps. 1:50000 scale (1970), Jianglin county and Jianglin town have been used during the field surveys. wade by the soil survey 1.3 Soil descriptions The local soil classification system in the study region is based on the processes of soil formation which are mainly influenced by the climate, parent material and. human activity (Din Rui-Xin, 1985). 36 ____FARSel. ADVANCES IN REMOTE SENSING Vol. 2, No. 3- X1, 1993 First order soil units are defined according to common The soil subunit in the second category reflects the vari- properties and soil forming processes. This means that the ation of diagnostic horizons and pattern of soil profiles in same units must have the same diagnostic horizons, pat- a main unit-There are mainly 4 soil units and 10 subunits terns of profile etc... The paddy soil, for example, formed in the study area which are described in Table 1 (S. RX, under the farming conditions must consist of an oxidation- M. Zhang et al, 1987). reduction diagnostic horizon. ‘Table 1: Major Type Of Soll Deseription In The Study Area. wnir soso | ~~ ciIARACTER PATTERN Joam to Toamy clay texture, homogencous (due to intensive human activities); ton and manganese iMtuviaion and elluviaton; ground water table is laround0.7- 1.0m terraced fields of hilly area, plowed horizon - light in texture erosion on the land surface; distinct whitish bleached horizon or side-beached horizon formed by the side seepage flow ofthe water. the clay: iluvated = topiand is elluviated - lower part; The ground water tableso8 1.5 m, Paddy soil bottom of the valley or polder area under poor crainage conditions: silt clay to loamy clay: the ative ‘Waterlogged soil [Fe-MIn begins to accumulate in percosubmergic| A-P-W-GIBg-C horizon following the separiton ofthe soi fom the swamping; The groundwater table < 50 em. developed on the alluvial deposits. Me parent material | js mosily calcareous and interayered by sands and clay; the clay coating: inthe discontinuous state and Jess rusty mottings in the solum. The soil forming processes: initial stage; The ground water table| ‘variation 0: 1.5 m Xashu loess parent material and distributed on the Duripan yellow brown strongly eroded hilly area; loam to loamy clay: earth subangular blocky structure; the pH value around 6.0 65; tere isa duripan within 1 m of profile. ash loess materials and distributed around the top) ‘Yellow brown earth | whiush yellow earth |and middle slope oF hilly area. The texture is silt loam ABC. 1 loan; subangular blocky structure: | weathered residual deposited material of andesite or andesite basalt. pH: 6.0 - 6.5; silt loam to loam;| A-BIC-R or A-CR or subangular blocky structure; well developed argilie even AR, horizon; containing afew small stones; “Ando sit Ochi andoso) weathered materials of basalt or volcanic ash loan ABC clayey loam; granular structure ‘weathered material of limestone in the Tow mountain | ea, Dark brownish colour; sit clay to loamy clay; sulbangular blocky trture;pH:7.5-8.0; the orga { matter content > 2.5%. weathered material of limestone io the Tow mountain afea, Reddish colour, silt clay to foamy clay granular structure; pH: around 8.0; there are some calcie nods; organic matter < 2.5%, Permeable soils APWBECe | side eating sis APWSLBEC Percolating soil APWE. APW.BEC. CChestnutish soit Brownish rendzinas RENDZINA SOILS, Reddish rendzina * Symbol used inthe profile pattern: A: Plowed layer, P:Plowpan; —_W: Percosubmergic layer B: Illuvium; G:Gley layer; _C: Parent material layer; SL: Side bleaching layer; R: Parent rock. Ming et al: Application of Satelite Remot 2. METHODS AND PROCEDURES A classification of the land use and the soil types based on, reflectance characteristics of the earth surface materials, was performed on the four bands of the landsat MSS data. ‘The ground truth was provided by means of local survey reports, local maps and site visits. The ILWIS software package was used for the multispectral classification, Supervised classification was used in this study. It rests upon using suitable algorithms to label the pixels in the image as representing particular ground cover type, or classes (John A, Richards, 1986), TRAINING PROCESS: The overall objective of this, step was t0 assemble a set of statistics that describe the spectral response pattern for each land cover type to be classified in an image of the study area, 1) Locate and define several relatively small blocks of data, dispersed over the entire study site, each contain- ing a few cover types of interest. These areas are known as candidate training sets, In this study, the training sets are chosen in a representative way so that they covered all the reflectance variations in the scene, Twelve train- ing sets were created in the false colour composite image (FCC) of the study area and represented 10.4% of the total image. ed to determine the ge is resolved. This 2) Clusters in the feature space were spectral classes into which the im ‘Table 2: The Classes Description, sasing 1 Soil and Land Use Mapping... 37 is performed on a representative subset of data by sampling procedure to obtain sets of spectral data that ca be used to determine decision rules for the classifi: cation of each pixel in the whole image data set. 3) According to the sample pixels of each training set in the different colour clusters with the help of a separ- ability algorithm, twelve classes were defined such as dark blue colour for class 1, deep blue colour for class 2; the bright blue colour for class 3; the deep green colour for class 4; yellow brown colour for class 8; the red colour for 12 and so on. During the study, a mean of DN was calculated for each spectral class and the standard deviation indicated the distribution of samp- ling pixel’s value to mean. The class statistics were used to check the overlap between each spectral class in the feature space (Fig.2). 4) Using available ground truth reference information, the spectral classes were associated with ground cover {ypes mainly based upon the soils and land use types. In this way, the spectral classes were defined for the ‘whole training area. Evaluation of the sampling pixels as part of the training set refinement process was per- formed to see if all features need to be retained for a reliable classification or not. On the basis of the ground truth and the statistical variables of the reflectance values, twelve spectral classes were distinguished Each spectral class is described in table 2. (vo. ‘MEAN VALUE" CLASS NAME _DESCRIPTION L_| 35521-s/01.3/14.2048) Water 1 [Pure water with lowest reflection (dark colour in FCC) 2_| 33.2/29.9720 8/11 2163) Water 2 Muddy water in smaller rivers and lake (biue colour in FCC) 3 arvs.acy Paddy soil with a poor drainage condition and without crop: ground 3. | 28.429.7132.4725.4(30) Pally PL nee able Som nas > ‘Paddy soil with poor drainage condition and with crops, most of them | io eee Podiy-p? ae rice, cotton and maize: ground water table < 50 em - > - addy soil with good drainage condition and without erop eovers| S| een es SC) eddy characterised by alluvial horizon (Fe) altstesanrsetien re Pay sl wit goed ring cones end wih ry Sover l. Pad Paddy soils formed from loess material; characterised by clay illuvial 36.9148.1/55,9148 0023) 7 ‘ Bade ed eal horizon and contain duripan, — | ‘Yellowish brown earth developed from loess material or weathered 8 | 36.8749.1162.5/57.6(47) Y-BEI ‘material of andesite; and mainly distributed around rolling bill rea; | | 7 tis normally dry land without iigaton, 29.4726 653.955.8025) |____Bas-7 Yellowish brown earth developed from weathered material of basal pal concen teel asl? Ando evlped re were muri of snl (ith msi | 27.8198.5/45.2146.9(20) |_____Lime-1____| Rendzinas developed from limestone and O.M% > 2.5% 12_| 23.9119 5143.7146.6(25) Lin —|Renazina developed from limestone and OM. < 2.5% * Mean value: Mean value of brightness in FOC: band band 2/band 3/band 4 (number of sample pixels). 38 EARScL ADVANCES IN REMOTE SENSING Vol. 2, No.3 BANDI het “td al ott STANDARD DEVIATION s ren be BAND 2 i He bet He fey EE STANDARD DEVIATION ta oH 15 22 z i 20 “ eee =e Gaajeaty eva ae ciate ” | Bano s Bann 4 | - ; t i “ a HE ag; ||,° y EE Eso : ge z EE eo Bo rt i Ba a Zs Zz Zo i 2” z mo He 1234567690012 Lass 12345678 9 10N12 cass Fig. 2 Means and Standard Deviations Distribution (Overlapping between the different classes) IFICATION CHECK: On the basis of these classes, the maximum likelihood algorithm, box classifier train. ing set that is used for testing pixel wise the classification, and k-nearest neighbour were used to classify before the ste image is classified. They use these train- ing sets to descriptions of the category spectral response patterns as interpretation keys for the pixels of the unident- ified cover type and categorized them into their appropri ate classes. To obtain the best results of classification, the different classification methods with different parameters for each method were used. The results were visually inspected according to the available ground truth data and based on the amount of unclassified pixels in the sampling windows CLASSIFICATION: The box classifier, the maximum likelihood classifier and k-nearest neighbour classifier and appropriate parameters were selected by the “Classifica tion Check” step. As a factor to multiply the standard deviation defining the size of the box (MF) for Box classi- fier we choose MF = 3, 5, 7, and 10, The results were named as B1, B2, B3 and B4 respx threshold for the maximum search distany ively in table 3. As TD) form Ming er al: Application of Satellite Remote Sensing to Soil and Land Use Mapping. mum likelihood classifier we used TD = 10, 20 and 30, named as M1,-M2 and.MB separately in table 3. Since it, was taken too much computer time, we only choose search, radius(SR), the maximum distance from a datum in feature space to a specified number of surrounding pixels (neigh- bours), SR = 5; and the number of neighbours (NB), the number of pixels in feature space considered within the search radius, NB =3 for the k-nearest neighbour classifier (K1 in the table 3), The table 3 also shows the percentage of each class from digitised ground truth maps in the training area, The results from each method were in- spected by statistical correlation analysis with them and the best classification algorithm and parameters were se- lected according (0 the highest correlation coefficient (Table 4 and table 5). ‘Table 3: Classification Results from Different Algorithms and Parameters in the Training Area (%). (Wais mH M2 M3 KI MET NBS Soa o im 1 oa [0 | 6s | 2 1s] 020 I 02 | 3 | 64 | 217 4 | 137_| 74 | 4 5 | ms | a9 219 | 204 6 | 282 | ns 263237 [1 [af o6s 196 | oa [3 20 zn [ as ° [er Tas os 1009 | 6? 18 as u 02 07 7 pi} oo a 02 a ToraL | 100) 100 0 100 100 ° The digitised soil map from the raining area Regrouping Level ‘Table d: The Significance of Correlation Coefficient (SCC) Between the Digitised Soil Map and The Clas jed Imagen Different a | I 4 —— eeonours| yy gy | gs ome | |e) KT nam | Ran | LEVEL.1 | es7as7/ 0.111863) 0.1373, 00017] -0.00050| 0870s) 0523055, 0Si3077| 130553] 0.684 LEVEL 2 | 08668| 0.36981 0.11365] 0.36131, 0.05603) 09536 0383576) 0s0Mi6|_9 | 0.656) 0.798 LEVEL 3 | 0.85731) 0.45457, 0.02659] -0.33574 -0.17765| 0.9748 0.83501| 0.82695, 6 | O.8I1| 0.917 ‘able §: Conusion Matrs for Classification in the Training Are (To subunits level Ground Cores pixels rom lassie imagery in st set a: [8 | 9a0 | 42 TOTAL | Paw WATER | 185087) 30960 igo) 270 77] 73 i PADDYL | 30.966, 165612 as) 2.212246 240 PADDY2 | 11050) 45957 30228) 5.998) 1475] Pappy3 | —_14300 10412 8.186, 8.110 a YER xs) 36 1744). 198820, 12780 ANDO) L720) 6 N35 S110 5.898) 212.001 [enp| 17s] 93] 19666361208) 10322 : TOTAL 248.760) 248.760) 218.760) 248-140) 248,760] 248.7001 Pe(%) | 7830, 6740) 61a) 73.0) 8090) 80.301 i 7580] 40 MAPPING: The 12 spectral classes that were classified and were added together according to the spectral charac: ter and the units in the ground truth that they belong to (able 2). Same units were labelled to same colour by regrouping different classes as one colour set. The area for each unit was calculated and the result was show in table 6. The output images were named as Map! for soil map and Map? for land use respectively. EARSeL ADVANCES IN Rl MOTE SENSING Vol. 2, No. 3- XI, 1993 Jand also but not on the data of the image recording; the crop 3 was regrouped from class 8 and 9, itis non-irrigated Jand and the main crops are wheat, soybean or sweet potato; the classes 10 to 12 belong to the forest and grass land (F & G) and mainly consists of pine, shrubs and ‘Table 6: The Classes Description for Soil Map and Land Use Map. ‘No. | Class Name | _ ‘SOIL MAP (Km2) ___TAND USE MAP (Km2} nit Possible subunit Unit__|_Land use| Vegetation | Water |_| Water | Fishing and Water (45.1) pearls cael os (45.1) | eultivation 3__| Paddy-pl P| ‘Waterlogged paddy Rice, wheat, | 4 | Paday-p2 a 1 soils (451) Crp), cotton, maize ; Goes) itated and | 5S | Paddy-wi D 2 Permeable paddy ‘greenmanure soils(s203) |__| _| [6 Paday-w2 | ee ae . ‘Side bleaching paddy soils ae Irrigated land | “Nest cotton 7 Paday-t1 y 3 iia 268) maize rape 3.2) Yellow brown earth wheat, | 132) __1___Yellow brown earth! Grey's | Non-itrigated | | spp? | Dark yellow brown earth soybean, 9 | 87 | anposon | YH? | Res (76) land goed 10 | Bas-12 ‘Andosol ‘Andosols 222) | | Forest & Pine, shrubs ime- jombre rendzina (1.1) F (4.4) 11_/__Lime:|__ RENDZINA i Sombre rendsina .ty | F&GOA) | acs | and grasses 2 | 1 = rendzinaaa (1.1) J For the soil map (Map 1), the class | and 2 were regrouped, to one unit for water. The class 3 and 4 were regrouped to ‘one unit as paddy 1, mainly waterlogged soils. The class, 5 and 6 were regrouped to paddy 2 and mainly permeable soils. The class 7 is representing side bleaching soils. The class 8 (YBE 1) is yellow brown earth developed from loess material which may belong to the whitish yellow earth subunit, The class 9 (YBE 2) is dark yellow brown, earth developed from weathered material of basalt which may belong to the Andosol subunit. The class 10 belongs to the Andosol subunit that is developed from weathered ‘material of basalt and with an umbric epipedon. The class 11 (Rendzina 1) are rendzinas developed upon limestone and contain organic matter more than 2.5%: the class 12 are rendzinas 2 developed upon limestone and contain ‘organic matter less than 2.5% For the landuse map (Map 2.), the water area was re- grouped from class 1 and 2, it consists of rivers and lake Which are mainly used for fish and pearls; the erop 1 was regrouped from class 3,4 and 5, itis irrigated land and the main crops ate rice, cotton, maize and rape or/and green ‘manure; the crop 2 came from class 6 and 7, itis irrigated 3. RESULTS AND DISCUSSION 3.1 Class ication methods Four classified images from the Box classification by using different multiplication factors (MF=3; 5; 7; 10), ives different results (Table 3). The MF=3 gives the best result because the location and the percentage of each class are comparable withthe ground truth, the correlation coef- ficient with the digitised ground truth snap was 0.578 that is higher than 0.553 (n=13) at 5% significance level. The results with MF=5, MF=7 and MP=10 were much lower and it was decrease when the value of multiplication factors increases (table 3). These because of their multi- plication factor are too large that enlarges the size of classifier “box” so that caused them overlap each other that happens for the class 5 and 6, for instance, can separate with class 10 and 11 at band 1 and band 4 (Fig.2), but it will be overlap each other after enlarged multiplication factors. According to Bayes’ Rule, if feature vector is on the overlap of several boxes, the class with the smallest box is selected as the “most likely” one (Ben Gorte, et al, ‘Ming er al: Application of Satellite Remote Sensing to Soil and Land Use Mapping. 41 1988). It was a reason why the class 5 class 6 and class 7 contained highest percentage of classified pixels in BL ‘with MF=3 that has high degree of correlation with ground truth but they were decrease when MF increase from 5 to 10. The high percentages were shifted to class 9, 10 and 11 after enhanced MF value (B2 to B4 in table 3 and table 4) because of their mean value and standard dev lower at some band as a “small box” (Fig. 2). We have to carefully selecting sample pixels for each class and try to create “bespoke” boxes by tailoring the “off-the-peg” boxes until they fit (Curran P.J, 1984), ‘Three classified images were obtained from the Maximum Likelihood classifier where different thresholds were used (ED=10; 20; 30). Also the K-nearest neighbour (KNN) classifier was used with a search radius (SR) of S-and the number of neighbour's =3, The best result obtained with TD=20. The correlation coefficient with the digitised soil ‘map was 0.571 (n=13 at 5% significance level), The Maxi- ‘mum Likelihood classification is performed assuming equal probability of occurrence and cost of misclassifica- tion for all classes. In cause of the class 1 and class 2 defined as the class “water” was over-represented in the iraining area that shows to high percentage and lower correlatio with ground truth (M1 in the table 3). We may enlarge the value of thresholds as propose to interac- Lively weight each class to reduce the probability of class- ifying “water” pixels in the whole data set by reducing the “water” weighting (M2 and M3 in the table 3) (Curran P. 1, 1984; Lillesand T.M. et al 1987). Since the errors of commission were enlarge by increase of values of thre- sholds also, it was eased some classes misclassified at M3 (TD=30) which shows the lower correlation coefficient with ground truth tion was 3.2 Classification accuracy to demonstrate the During the mapping stage, we tri relationship between the spectral classes and the mapping unit, The image classification, grouped in 12 spectral classes, has been compared with the digitised soil map. ‘The results in table 4 show the significance of the correla tion coefficient (SCC) with the digitised soil map will be increase after regrouping to different level, The level | is in original 12 spectral classes, The level 2 and level 3 are in the soil subunit and unit level respectively (see also table 7). There are four maps (B1,M2, M3 and K1) with the correlation coefficient are higher than 0.811 (n=6) at ‘5% significance level; one map from TD=20 (M2) with the correlation coefficient of 0.975 that is higher than 0.917 (n=6) at 1% significance level. It is clearly that the correlation coefficient is increasing by regrouping the subunits to one unit. The Maximum Likelihood classifier with TD=20 was selected to classify the satellite image to produce the soil map and landuse map according to their highest significance of correlation coefficient with the seround truth data. Table 5 shows the confusion matrix for the training area in the test site resulting from cross-tabu- lation of the polygon rasterized soil map with the classified MSS image (TD=20). The classification accuracy for the main soil types varied between 61.3 and 86.6 per cent. The Paddyl, Paddy2 and Paddy 3 shows lower accuracy that was partly confused with water. The overall classification accuracy was 75.87 per cent. During the calculation of the overall classification accuracy, it was increased to 84.21 percent when the Paddyl classified as Paddy? or Paddy3 Or inverse was considered correct and added them to one unit (table 7). This means there were some misclassifica- tion between the subunit that was characterised by water capacity and the hydrological processes etc. 3.3 Soil maps Map | indicates 7 classes of soil that belongs to 4 units and 10 subunits. The soil units, can be detected and de- tached by RS operations, but the classification up to sub- units with RS data is more difficult. The reflectance values ‘on the satellite imagery (influenced by the upper layer and vegetation cover) are not always related to variations of the different soils. We have to use the relationship between the spectral characteristics and the “underlying” charac- teristics (geology, drainage condition, soil material) ‘Table 7: Confusion Matrix for Classification in the Training Area (To unit level). Ground Correct pels from classified imagers Truth 2 37 5 3.10 1B TOTAL Pot) WATER 185057 __56,525 23708 737] 737 PADDY 56525] 639,959] 26.296) 9.830 6565) YER | _983)— 20684) 198,820 i27s0[ 12.534 | ANDO. 1.20 11,059 5.898 212.091, 14.991 REND 1075) 9,098 1002 10,322) 212828 TOTAL 245,760) 737,280, 245,760 245,760 245,760| 1,720,320 a Pei‘) 1530 868 509) 863 $06) 1448755 | BAL 2 EARScl. ADVANCES IN REMOTE SENSING Vol. 2, No. 3- XI, 1993 Map I (TD=20) from landvat image (ILWIS ouput) Some of them are coinciding very well between ground truth and the satelite image, such as Rendzinas, Andosol and yellow-brown earth (classification accuracies of 36.60%, 86.30% and 80.90% respectively, see table 5). For some others, the coincidence is lower, especially for the Paddy soil (67.4%, 61.3% and 73.3), the subunit is distinguished mainly according to the hydrological pro- cesses (e.g. drainage conditions), but some Paddy soils were covered by irrigation water at the time of the image recording (13 of June). This may cause the misclassifica- tion between the subunits of the paddy soils. This could be solved by a multitemporal analysis. TM and SPOT data may be helpful to map soil materials on a small seale. A. ‘multitemporal analysis could be improve the mapping accuracy, VDR CPHoOW ARGO AION ZHPOZDHG ZH LEGEND em ueTER Wm Prooy 1 ™ propy 2 iB peo 3 me me vaE 2 ANDOSOL mm FENSZINA 1 = RENDZINA 2 Soil map of a part of the Jianglin county, central eastem part of China, produced by maximum likelihood elassifieation 3.4 Land use map The land use map contains 5 units (Map 2.). The forest is, located in the top of the lower mountain area. The crop 1 is located in the plain and the polder area along the Qing- hua river. The crop 2 is distributed between the plain and, middle slope of the hilly area. It is irrigated land also but, not at the moment of the image recording. The crop 3 is, distributed in the middie slope of the lower mountains and, at the top of the hilly area. It is the non-irrigated agricul tural land, The river and lakes are used for fishing and, pearl cultivation. Table 6 shows the relationship between the spectral classes, soil units and Janduse. The soil distribution has a close relationship with the land cover types because the different land use types influenced the soil developing, process. The remote sensing and GIS technique may pro- more detail information about land covertypes. Using these techniques, we have more advantage to study the soil conditions. Ming et al: Application of Satllt ‘Remote Sensing to Soil and Land Use Mappin; a3 MaAcIazZpr mo UD3 ZHOMZDRG ZH Map 2- The landuse map of apartof the ianglin county, central eastern part of China, produced by maximum likelihood classification (TD=20) from landsat image (ILWIS output) CONCLUSIONS, On the basis of the research described in this study, the following conclusions concerning the use of satellite im- ages (Landsat MSS data) for the study of soil conditions, can be made 1. Classification algorithms and threshold parameters have an important influence on the classification results and should be selected carefully based on the training area. GIS was a powerful tool to improve the accuracy of soil and land use mapping based upon the remote sensing data, 2. The landsat MSS data are very suitable for soil mapping, upto the level of soil units, but errors occur on the level of subunits. The relation between the spectral charac teristics and the soil subunits should be checked, based fon ground truth to obtain an idea about the possibility of automatic spectral mapping. 3. By using image processing and GIS, we could found the close relationship between the distribution of soils and land use types. ACKNOWLEDGEMENTS ‘The author would like to thank Dr.Beata M. De Vliegher, Ms. T. Ongena Mr. C. Marius Mr.P-Brackman and other colleagues in our lab. for their kind advices, BIBLIOGRAPHY Barrett, EC. (1976). Introduction to envionmental remote sens- ing. Chapman and Hall Let, 1976. Ben Gorte, Ruud Liem and Jelle Wind, (1988). The ILWIS software kernel, 1988-1, pplS - 22. Carlos R. Valenzuela, (1988). ILWIS Overview. ITC Journal 1988-1, ppd - 14. Curran Paul J., (1985). Principles of remote sensing. Longman, London and New York, 1985, Din Rui Xin, (1985). Lecture note of “The soil genesis and ‘lassiication". Department of soil science, Nanjing Agricultural University, 1985, Gerardo Bocco and Catios R Valenzuela, 1988. Integration of GIS and image processing in soil erosion studies using ILWIS, TTC Journal 1988-4, pp309.319, 4 ___EARSeL ADVANCES IN REMOTE SENSING Vol. 2. No. 3X1. 1993 Haralick, RM. and Fu, K.S. (1983). Pattern recognition and classification: In Colwell, RN. (ed), Manual of Remote Sensing (Gnd edn). American society of photogrammetry, Falls Church, Virginia, pp. 793 - 808. Lillesand TM. et al, (1987). Remote Sensing and Image Inter- pretation, (second ecition).John Wiley & Sons, Ine. 1987 Mather Paul M., (1987). Computer Processing of Remotely- Sensed Images: An introduction. John Wiley & Sons, Inc. 1987, Richards John A. 1986. An introduction remote sensing digital image analysis. Springer-Verlag, Berlin, Heidelberg, New York, London, Paris, Tokyo. 1986. ‘Townshend, 1G. (1981). Terrain analysis and remote sensing, Allen and Unwin, London, Boston, Xu Sheng-Rong, Zhang ming, tal, (1987), Surveys and planned ‘of the soil resources in hilly areas around Nanjing region, Countryside economic in Jiangsu province, pp152-163.

You might also like