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

An approach for estimating forest stock volume by using space Remote Sensing data

Zhao Xianwen Yuan Kaixian Bao Yingzhi
Zhao Xianwen Yuan Kaixian Bao Yingzhi Chinese Academy of Forestry,
Beijing, China

Cao Faji
Jilin Surveying and Designing Institute of Forestry


Abstract
The study is a new attempt to estimate forest stock volume by using Landsat TM image and some ground sampling plots with multi-analysis method. Independent variables were used including qualitative and quantitative factors. The quantitative factors are colour hue and group of tree species, while the quantitative factors are density value and ratio of bands. In this way, the potentiality of remote sensing data can be brought into better ply. The result shows that the accuracy of estimation of forest stock volume is more than 80%. It is a convenient and economical method. To estimate forest stock volume with remote sensing data especially satellite data has been an interesting topic for foresters. In recent years, many internal and external reports discussed an estimation of forest stock volume directly using remote sensing data. Strahler, A.H., Tang S.Z., Xu G.H., Zhao X. W. studied this topic from different aspects. This paper described the procedures of an estimation approach on the basis of previous studies and analyzed the results in comparison with the actual measured values.

Materials
One hundred of group sample plots were systematically distributed over the whole experiment area-Pingquan County, each with five sub-plots. The arrangement pattern is shown in Figure 1. Tally measurement and angle gauge measurement were performed on an area of 0.01 hectare and 0.02 hectare for each sub-plot, respectively. The sub-plot was of square shape. When these measured data were used as base variables, they had two different forms; one was from the average of the five sub-plots of a, sample plot and was referred to a group averaged in the tables, the other was only from the central sub-plot and was referred to as central. At the same time, density values of TM images of the scale 1:1,000,000 were measured with point densitometer (aperture=0.02um) on each sample plot for bands 1,2,3,4, and band ratio values were also caculated for e. g. 4.3, (4-3)/ (4-3). The chromaticity an categories of tree species groups were visually interpreted with the color composites of MSS images of the scale 1:100,000.

Method and Scheme
  1. Method

    The method of multi-variable estimation was used. The mathematic axpession is as follows

    Y = A1X1+A3 X3 +…………+An Xn

    In this study, the base variable Y was substituted for by different sorts of values of stock volume of the sample plots, which were obtained by various ways of design of the sample plots (0.01 ha., 0.02 ha. And angle gauge measurement----changeable circular standard plots), and by various ways of point choice (group-averaged, central). The independent variables were respectivgely the density values measured on the satellite images and their ratios, chromaticity and categories of three species groups. Among them, the density values and their ratios were the quantitative factors, while the chromaticity and categories of three species groups were the qualitative factors. Therefore, the method adopted in this study is a a special case of multivariate estimation---a quantitative method, which is a mixed type of problem characteristic of both quantitative and qualitative factors.

  2. Scheme

    In this study, the variables were decided according to the following scheme:

    Central angle gauge sample plot, group-averaged square sample plot of 0.01 ha., and group-averaged sample plot of angle gauge measurement. The square sample plot of 0.01 ha. Was not used in te method of multivariateestimation due to its unstability. The quantitative factors and qualitative factors were selected as independent variables. The quantitative factors wee the measured values of TM images in bands 1,2,3,4 and band ratio values for 4/3, (4-3)/(4-3) , respective . While the qualitative valies were the 16 colors were: dark red, midium red, light red; dark yellow, midium yellow, light ytllow; dark blue, midium blue, light blue; dark and the categories of tree species groups were the four: confers, broadleaves, bush and the rest. The qualitative factors were doded with the binary code 0/1 Beside the establishment and calculation of the estimate equations, multivariate equations were also established according to the major basins- the southern and northern drainage system of pu River and Lyan River. And the statistics were calculated accordingly.

    In this paper, except the density values of various bands and their ratios were used, the factors of chromaticity and category groups were also employed. In this way, for one thing, the restrictions of using merely the density values were prevented (some bands or band ratios are correlated), for another, the advantages of multi -source information of remote sensing was brought into a full play.
Results and Analysis
  1. Four groups of equations from four different base variable were obtained in accordance with the above described ways of sample plot design and selection. The results demonstracted no significant difference for the various ways of sample selection.

    The county comprises mainly two drainage areas of Pu river and Luan River, To make the estimate equations tally properly with the actual situation, they were calculated separately in accordance with the two major river systems. The correlation coefficients were obviously raised after the two major drainage areas had neen treated separately, and the estimates thus achieved represented properly the distribution of stock volume.

    At the same time, the effects of stock volume estimation by merely using density values were compared with that of stock volume estimation by addition of the ratio terms. The comple x correlation coefficient for the regression equations established between stock volume and density values of TM bands 1,2,3,4 was only 0.4818, however, after the addition of ratios 4/3 and (4-3)/(4+3), the complex correlation coefficient increased to 0.7262, which indicated the important influence of the ratio terms. When the qualitative factors were added, an approximate deter mination of the boundary lines between forest and non-forests could be made, though the estimation precision was not further improved.

  2. Tanking the group angle gauge measurement as an example, there are two kinds of methods to compute the stock volume for the whole county:

    1. The stock volume of forested land (19,72, M3/ha.) was multiplied by the total area of the county (329,832 ha.) and then by the coverage (30.65%), and the total stack volume of the county (1,993, 557.934 M3) was obtained.

      Table 1 comparision of the sesults actually measured of Ping Quan county with those estimated by various mehtods (figures were of relative errors)
      Methods size/form
      of plots
      choice of
      plots
      compared vith
      non-strtified
      Sampling
      compared with
      stratified
      butnon-mappd smplng
      Note
      Multivar
      Regress.
      Estimate
      Angle gauge
      0.02 /sq.
      Group-av.
      Group-av.
      +2.6
      +17.7
      -10.7
      +1.8
       
      Stratif.
      2-stage
      sampling
        0.01
      0.01
      -4.4
      14.4
      -8.0
      -14.9
      3 selected from
      5 plots per group
      0.02/sq.
      angle gauge
      Centeral
      Centeral
      +18.8
      +5.4
      +15.2
      +1.7
      Sampling according
      To Air photos
      Double
      Sampling
      Estimation
      angle gauge
      0.01/sq
      Centeral
      Group-av.
      +22.88
      -11.73
      +10.6
      -24.6
       


    2. The stock volume estimates of all the points (including the points actually measured and those interpreted) were substituted for with the regression equation, and the general average of stock volume for the whole county was computed (6.9475 m3/ha.). The total stock volume of the whole county was estimated to be 2,291,500 M3) in this method. The two methods had 6.9% difference, and the second method of calculation seems to be more reasonable.

    It can be seen from Table 1 that the accuracy of various procedures usually reaches 80%. Each procedure has its different advantages and applicability: the double sampling procedure with air photos and ground plots (method 3) can increase estimation precision by doing an amount of indoor work but limited field check, while the two-stage sampling procedure by combining satellite data and aerial photographic data can increse work efficiency. The direct application, however, of satellite images in estimation of stock volume is a new test and experiment. It is worth of notation that the combined procedure of qualitative and quantitative facotrs proves to be simple and convenient, and easy to spread It can also satisfy the precision requirements.
References
  1. Strahler, A.H., timber inventory using LANSAT, The Eighth Candian symposium on Remote Sensing, pp 665-673.

  2. Tang S.Z., Xu G. H., A Study on the method of estimation of forest stock volume by LANDSAT digital image data-principle and Method, Remote Sensing Research and Application Materials, Science and Technology Document Press, pp 142-147, 1984.

  3. Zhao X. W., A preliminary study of the estimation of forest distribution and stock volume by directly using satellite images, Gaungdong Forest Newslatter, ()2; 18-20, 1894.

  4. Dong W.q.etal. quantitativeapproachesandtheir application, Jiling Peoples' Press, 22, 1979