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

Effects of vegetation cover on soil moisture sensing with x-band radar

Zhao Changling, Hao Weixing, Li Shengping, Li Xiaohong
Institute of Remote Sensing Applications, Academia Sinica
Beijing, China

Ji Jiankang, Li Guikui
China Research Institute of Radiowave Prodagation
Xingxiang, Henan, China


Abstract
In this paper, the relationship between backscattering coefficient and soil moisture is studies for bare and vegetation (sweet potato, soyabean and peanut) - covered fields. Backscattering measurement carried out by means of sensing soil moisture with X-band radar image is affirmed. The results estimated by a simple model indicate that the sensitivity and accuracy of X-band are lower than that of C-band for vegetation-covered fields, but there will still be certain sensitivity and accuracy wit proper parameters. And the prediction error is about ±25% at X-band. As for bare soil fields, C-and C-band are quite similar accuracy.

Introduction
The use of microwave remote sensing in soil moisture measurements has caused very close attention over the past years. Significant progress in finding out about the relationship between backscattering coefficient and soil moisture has been achieved (1) - (10). Theoretical model and experiments have shown that, backscattering coefficient and soil moisture are closely related. In fact, the relationship is quire complicated. Ti s concerned in the factor as random roughness, soil texture (type), and vegetation cover as well as microwave frequency, incidence angel and polarization. Most of the research in this area was done at C-band in the past. Ulaby, et al. (9), think that C-band (freq. 4.25-4.75 GHz), HH polarization and small incidence angle (q = 10°) are optimum parameters in sensing soil moisture by radar.

In Asia, SAR and SLAR and X-band have been widely used. Someone is attending to study soil moisture by X-band radar image. Then whether or not could the X-band imaging radar be used to predict the soil moisture under vegetation cover? In this study, experimental data were summarized. We have analyzed the relationship between backscattering coefficient at X-band and bare soil moisture, and especially the effects of vegetation cover in detail. The ability of sensing soil moisture with X-band radar image is affirmed.

General Situation of Experiment
Bare soil field measurements were carried out at Xingxiang, Henan province of China, during May 10- June 8, 1989. And some vegetation covered field measurements were carried out at Pandian, near Xingxiang, China, during August 4 - September 27, 1989, crops observed were sweet potato, soyabean and peanut. The measurement system used in the experiments is X-band microwave scatterometer (fre: 9375 MHz) made by Shanghai institute of electron physics. The incidence angel can be changed (0° - 48°, step 6°). The azimuth can be changed continuously.

In the experiments we have noted the effects of soil texture composition, random roughness of the soil surface to microwave response of soil moisture. To minimize the influence of soil texture we used mf the percent of field capacity to represent the soil moisture. Considering that microwave penetration of soil is limited, thus we only collected the soil samples of a fixed layer, 5 cm depth, to measure soil moisture.

Some researchers reported that when the incidence angle is between 5° to 7°, the effect of the roughness of soil surface is smaller to the measurements. The optimum incidence angle is different to various bands. Generally the smaller angle is better. In this study, we mainly discuss the case of 60 the incidence angle.

Bare Soil
Fig.1 shows the response of backscattering coefficient s° for bare soil as a function of mf the soil water content of the top 5-cm soil layer expressed in percent of field capacity at three incidence angles (q=6°, 12° and 18°).

mj = (100mf / FCg)% ----------------------(1)

Where mf is the weight soil water content of top 5-cm layer in g water/g drysoil, and FCg is the weight soil water content at 1/3 bar tension (commonly known as field capacity). Through the analysis of soil texture composition we can get the sand an clay content. And FCg can be calculated using the empirical relation derived by Schmugge (11).

FCg = 25.1 - 0.21S + 0.22C ---------------------------(2)

Where S and C are percent of sand and clay content, respectively. Application of a least squares linear regression fit yields.

s° g(dB) = 0.261 mmj - 25.83 dB, q= 6, r = 0.81 --------------------(3)

where r is the correlation coefficient.
Now we can further express (3) in natural units of m2 m-2

s°g= 0.003exp(0.06mj)m2. m-2, q = 6°, r = 0.81 -----------------(4)

Soil Covered by Vegetation
When the soil is vegetation covered, the microwave power suffers a two-way attenuation due to the propagation through the vegetation layer. In addition, the vegetation layer contributes a backscatter component of its own due to scattering. If the soil and vegetation backscattering components are assumed to add incoherently at the radar and if multiple scattering contributions are ignored, the backscattering coefficient of the total s°c may be written as

s°c(q) = sv (q) + [ s°g(q) / L2(q) ] m2 . m-2 --------------------(5)

Where
s°c(q) - total backscattering coefficient
s°v(q) - vegetation backscattering coefficient, given above by (4),
Lg(q) - One way loss factor of vegetation layer,
q(°) - angle of incidence

From (5), if we find out the functional relationship between s°c(q)and s°g(q), we will get the backscattering coefficient s°v(q) of some vegetation and the two way attenuation L2(q) due to the vegetation cover. We should emphasize that L2(q) is significant parameter which represents the capability of microwave penetration through vegetation cover (sweet potato, soyabean and peanut)we can see the effects of canopy and soil to s°c(q).

Below we will have a further discussion about the case of individual crop type an three crop types combines.
  1. Individual Crop Types


    2. Fig 2 shows the measured values of s°gp plotted versus mf for five fields of sweet potato. A least square linear regression results in

    s°gp(dB) - 0.118mf - 13.20 dB, q = 6°, r = 0.55 -----------------------(6)

    In order to analyse more clearly the effect of vegetation cover to the microwave transmission , (6) may be expressed in natural units of m2. m-2 s 0gp is the measured backscattering coefficient for the sweet potato fields shown in fig 2 and s 0g is calculated from the measured values of mf That leads to

    s°gp = 0.146 + 0.786, s°g m2 . m-2, q = 6°, r = 0.54 -------------------(7)

    as shown in fig 3 comparison of (7) to (5) leads to


    A similar analysis for soyabeans and peanuts leads to the values given in table I Using (4), (5) and the values given in Table I, plots ofs°c(dB) versus mf were obtained for sweet potato, soyabean and peanut and are shown in Fig. 4

    Table 1 Results Estimated at X-band
    N Crop Type s°v (m2 . m-2) L2(dB) r
    15 Sweet potato 0.146 1.05 0.54
    14 Soyabeans 0.105 1.71 0.50
    16 Peanuts 0.123 13.28 0.35
    45 Three crops 0.226 4.93 0.49

  2. All Crops combined


    3. if we were to ignore differences between crop type and, therefore, treat all available vegetation covered data as a single class, shown in Fig 5 and Fig. 6, would be obtained. Fig. 5 shows the actual measured data with s°c (dB) plotted versus mf The lest squares linear regression is given by

    Table IL Results estimated at C-band
    N Crop Type s°gv (m2 . m-2) L2(dB) R
    34 Corn 0.164 0.97 0.81
    22 Milo 0.070 2.15 0.82
    54 Soyabeans 0.054 0.51 0.81
    33 Wheat 0.110 1.31 0.94
    143 Four corps 0.066 1.25 0.91

    s°c(dB) = 0.057 mf - 10.90 (all corps), q=6°, r - 0.49 -----------------------------(8)

    (8) may be expressed in natural units of m2 m-2

    sc = 0.226 + 0.321 sg , m2m-2(all crops), q=6°, r - 0.36 ----------------------------(9)

    Thus, we may obtain for all crops combined.


    Now we have obtained s°c and s°g as a function of mf based on (9) and (4) respectively, as shown in Fig . 7 It is observed that when vegetation is present, the effect of vegetation to the estimation of soil moisture is bigger for lower values of mf while a high values of mf the effect becomes negligible. As for the vegetation covered soil, the level of s°c is lower than that of s°g by 4.39 dB. It should be emphasized that the 4.39 dB is an average value. In fully grown, the canopy contain a lot of water content, so during which s0v and L would be higher. And to the grain - (i). ling stage, canopy loses most of its water content and, therefore, exhibits lower values of s°v and L.

  3. All data combined


    4. For comparison, linear regression lines are shown for each of bare soil, vegetation covered soil and for both combines in fig. 8
Prediction Accuracy
For extracting the information of soil moisture from radar image, we need a relatively simple estimation model that may be applied to both bare and vegetation - covered soils. Here we suggest a generalized estimation model.

s°m = 0.133 + 1.34 x 10-3 exp (0.06mj) ----------------------(10)

Fig 7 shows s°m plotted as a function of mf where s°m lies way in between the plots for s°0(dB) and s°s(dB) application of (10) leads to

m~j = 110.3 + 16.7 In (s°m - 0.133) ---------------------(11)

it is observed from above equation that when the value of s° is fixed from the radar observations, the soil moisture could be estimated by entering s° in (11)

From the points of application, the estimation accuracy is the matter of interest to us. we define the estimative error as

Dmj= m~j - mj -------------------------(12)

Then the estimation error of both bare and vegetation -covered soil may be got. The results are shown in Fig. 9 When mf <75% the prediction underestimates mf for bare soil and overestimates mf for vegetation covered soil and does the reverse for mf >75%

Now we shall evaluate the prediction accuracy of estimation algorithm using the actual data. The prediction accuracy of soil moisture are shown in Fig 10 and 11 for bare and vegetation covered soil respectively. In the 90% confidence interval the prediction error for bare soil is bounded between ±33% and -10% for vegetation covered cases, the prediction error is bounded between ±40% and -25%. Those errors are due to two types of sources: (1) measurement errors of moisture content; (2) measurement errors of the backscattering coefficient, including the effects of soil surface roughness and vegetation cover on s°. We estimate that in absence of these errors the prediction accuracy would have been within the range ±25%

The Comparison of X-Band and C-Band
On sensing bare soil moisture, the results of X-band and C-band are quite well matched in small incidence angles, and the correlation coefficients are 0.81 and 0.85 respectively.

In the case of vegetation - covered soil, the difference between the two bands is great. From tables I and II the correlation coefficients of X-band are relatively small and the two way losses are 5 to 6 times bigger than that of C-band. From the data of soyabean, we can see that the values of s°v and L2 and C-band are bigger than that of soil under the vegetation cover at the shorter wave length. I other words, the shorter the wave length, the weaker the capability of penetrating vegetation cover, and the lower the backscattering and correlation coefficient will be too. Of course with proper parameters, the certain prediction accuracy will be obtained. With the 90% confidence interval the prediction accuracy of C- and X-band are 15% and 25% respectively.

Conclusion
In this paper, the microwave backscatter from bare and vegetation - covered soil has been investigated using experimental data at X-band and the estimate model. We suggest that.
  1. In the accuracy of estimating the bare soil moisture, X-band is almost identical with C-band with the proper parameters.


  2. In the vegetation covered soil, the accuracy of X band is worse than that of C-band. But with the proper parameters, the certain prediction accuracy can be obtained. In 90% confidence interval the average prediction accuracy of crops may be bound within +2%.


To summarize, the results show that, the vegetation cover must have significant effects on the radar remote sensing soil moisture. And it is more serious in the cases of lower soil moisture under vegetation cover and shorter wavelength. In order to provide more basic data for the design and application. In order to provide more basic data for the design and application of imaging radar, we think that the effect of L-band deserves further investigation.

Acknowledge
In completing this task we have been helped by many of our friends and colleagues. In particular thought we would like to acknowledge the generous and unfailing support of Lu Yonghong, Fen Wenfu, Cehng Ruiting and Yue Xiaoping.

Reference
  1. Ulaby F.T. RSL Technical Rep. 177-51, university of Kansas Center for Research, inc. Lawrence, Kansas, 1974.


  2. Ulaby F.T. et. Al journal of Remote Sensing of Environment, Jan., 1975


  3. Reaves R.G. Manual of remote sensing Vol 2 Keuffel and Esser Co., 1975


  4. Ulaby F.T. IEEE Trans. (AP) Vol. AP-23, No. 1 1975


  5. Ulaby F.T. et. al IEEE Trans, (GE) Vol. GE-16, pp. 286-295, 1978


  6. Ulaby F.T. et. al. IEEE Trans. (GE), Vol. GE-19, pp. 51-61, 1981


  7. Ulaby F.T. et. al. Micro wave remote sensing: active and passive, vol.2, Addison-Wesly publishing Company, 1982.


  8. Ulaby F.T. et.al IEEE Trans. (GE), Vol. GE-20, pp-476-481, 1982


  9. Huang Yang, et al, Remote Sensing Information (Chinese), No.1, pp 24-27, 1986


  10. Shmugge T.J. IEEE Trans. (GE), vol. GE-18, pp. 253-261, 1980