Luận án tiến sĩ: Remote Sensing Techniques of Geospatial Geotechnical Site Characterization Applied to Competence Studies of Mine Tailings Impoundments and Slope Stability Investigations

This dissertation, ‘Remote Sensing Techniques of Geospatial Geotechnical Site Characterization Applied to Competence Studies of Mine Tailings Impoundments and Slope Stability Analysis’,

SITECHARACTERIZATION USING REMOTE SENSING ca 14

This dissertation examines the viability of determining the engineering parameters of slopes, using remote sensing and Geographic Information System (GIS) data. Principal inputs for slope stability models include the soil density (p), cohesion (c), the angle of internal friction ($), the slope angle (a), and the depth to the groundwater table (GWT), which governs the driving pore pressures.

The theory behind the remote sensing approach for site characterization is based on research in the fields of geophysics and soil dynamics Early research of using active wave response to measure soil properties was already underway in the early 1960’s(Hardin & Richart, 1963) During this time, correlations between visible groundcover and soil classes were published (Eyre, 1963) In the USSR, coordinated research established by proxy soil characterization standards between features and groundcover visible in aerial photographs, and the subsurface (Moskalenko, 1961) During this decade, the derivation of soil properties, particularly the relative dielectric permittivity ¢,, from passive radiant energy was also studied (Bowers & Hanks, 1965) Towards the end of that decade, a relationship between the density p, porosity rị, and the shear modulus G was established (Hardin & Black, 1968; Richart et al., 1970) Research of this topic continued throughout the 1970’s, where a relationship between the shear modulus and the shear strength of soils was defined, from which the soil cohesion c and the angle of internal friction $ are determined (Hara, 1974; Athanasopoulos & Richart, 1981) From the mid-1970’s through the early 1980’s, changes in soil electric properties and how they correspond with changes with the void ratio (e) at lower electromagnetic frequencies were studied (Hoekstra & Delaney, 1974; Delaney & Arcone, 1982) Soil temperature changes related to changes in soil moisture and void ratios were examined using infrared radiation through the early 1980’s (Hanks & Ashcroft, 1980) This became the basis of satellite surveys and monitoring of soil moisture contents Also at this time, the relationship between the relative permittivity ¢, of a soil, and it’s volumetric moisture content was defined (Topp et al., 1980) This research continued in Finland, where relationships between a soil’s moisture content and e; was extended to include the soil’s mineral composition (Hallikainen et al., 1985) This ultimately contributed to the definition of the relationship between soil density p and e; (Ulaby, 1988).

In the USSR, as early as the late 1960’s, and resuming in the 1980’s, dielectric properties of soil to determine the re-emission (or penetration) depth 6 of an active wave signal, and how the re-emission depth and soil moisture content are related to the depth of the groundwater table (GWT), were examined (Golovanov, 1968; Shutko

& Reutov, 1982, 1990) The relationships between surface moisture and GWT depth were applied to aerial remote sensing studies at several sites throughout the USSR.

The site characterization efforts thus were intended primarily for on-site or laboratory geophysical applications Soil classifications by remote sensing, at this point, still relied heavily on using by proxy methods, such as geobotanical indicators During

1980, Hunt published the benefits of using electromagnetic band sensors for remote sensing applications, noting such benefits as the greater ability of electromagnetic (particularly microwave and radar) wave signals in discerning environmental conditions on the ground, as opposed to the previously more common monochromatic or pan-chromatic images which satellites were collecting These electromagnetic bands are the same frequencies used in the dielectric studies of soils.

Beginning in the mid-1980’s, groundcover was classified from remote sensing images, based on the “spectral signature”, or spectral response of the material across several light bands (Clark & Roush, 1984) Through the mid-1990’s, the spectral signatures of different minerals and earth materials were catalogued in a database, and computer algorithms to match the spectral signatures from image pixels to a catalogued signature were applied, as a means of identifying the earth material (Clark et al., 1991, 1993; Clark & Swayze, 1995) This directly led to computer-generated geologic and soil mapping, with satellite imagery as the primary input (Kruse, 1990). However, no known attempt had been made to directly compute the engineering parameters of earth material from imagery Any parameters extracted from any geologic maps generated by imagery, would have to be by proxy.

A purpose of this dissertation is to utilize the remote sensing techniques for extracting engineering properties of soils, with the abundance of data available from remote sensing imagery, to perform a geotechnical site characterization This site characterization is then used as input for computerized slope stability analysis The slope stability analysis is performed by extending common slope engineering practice to accommodate geospatial data.

The approach covered in this dissertation uses digital satellite imagery and terrain data, in the form of a Digital Elevation Model (DEM), as the source data Prior to any slope analysis, a geotechnical site characterization must be performed to determine the values of p, 6, c, a, and GWT This site characterization is covered in great detail in this dissertation Though these parameters may be assigned values by proxy (or by association with certain visible or spectral features and characteristics of the surface), this dissertation will focus on calculation these parameters from the relative dielectric permittivity, or €.

OBJECTIVES chú Hà HH HH HH Tà T11 TT 114111481 111 01111 1111111110710 17

The ultimate goal of the research presented in this dissertation is to reduce the cost and effort of landslide hazard analysis through novel techniques of disseminating geotechnical data from geospatial multispectral data in the form of remote sensing imagery The desired output are maps illustrating the geospatial distribution of landslide risk indices, derived directly from multispectral imagery by quantified deterministic methods This ultimate goal may satisfy several objectives.

1 Geospatial site characterization of geotechnical parameters derived from imagery This dissertation will demonstrate methods of approximating geotechnical parameters through imagery analysis.

2 Comparison of geotechnical parameters approximated through imagery analysis with field-collected data Site-specific field-collected values of key engineering parameters are compared with values computed for each site through imagery, to ascertain the accuracy and validity of imagery analysis.

3 Landslide hazard assessment through imagery analysis Using key parameters approximated through imagery analysis as input, a prescribed series of slope stability computations will transform an aerial or satellite image into a patch map or contour map which numerically designates each pixel by an index which indicates the risk of slope.

4 Related Problems and Applications In addition to slope stability analysis, methods of site characterization presented in this dissertation are also applied to problems and issues relating to ground moisture and bearing capacity.

The sum of these objectives results in a method of geotechnical reconnaissance and identification of slopes and areas which are potentially problematic, thus demonstrating that image analysis is a useful tool for investigating sites in very remote, inaccessible, or dangerous areas.

SITE CHARACTERIZATION OF GROUND PROPERTIES; A REMOTE

REQUIRED INPUT PARAMETERS FOR SLOPE STABILITY ANALYSIS

As discussed in Chapter One, remote sensing imagery has been employed for slope failure risk assessment, but up to recent times has been limited to aerial photography. This practice does not allow for monitoring whicha satellite’s orbital coverage of the area can provide Satellite imagery is also less expensive, and easier to obtain, than commissioning an over-flight, plus the associated costs and efforts of processing and mosaicking aerial photography each time the slope is imaged.

However, current remote sensing analysis does not provide definitive results for slope stability analysis; rather, results are dependent on the analyst’s judgment Limited engineering parameters are derived which allow for a quantitative analysis That is, insufficient parameters are derived to perform customary engineering methods for slope analysis.

Techniques can be applied to multispectral satellite imagery with image analysis software, which permit the measurement of dielectric properties of the earth material. The dielectric permittivity of a material is, “a measure of the extent to which the electric charge distribution in a material can be distorted or polarized by application of an electric field” (Berthelot et al., 2001; Hasted, 1973) This distortion or polarization potential is dependent on the composition of the material Dielectric permittivity is a unique and an inherent property from which certain engineering parameters of the soil can be calculated These engineering properties are related to the earth material’s structure and composition, to which the permittivity is directly related.

REMOTE SENSING APPROACH FOR SLOPE ENGINEERING

With global satellite coverage and commercial availability of imagery, engineering analysis of slopes can be condensed from the previously mentioned traditional approaches, towards more rapid and cost-effective analysis using remote sensing.Using recently acquired high-resolution satellite imagery, image analysis software,and GIS, quantitative deterministic slope analysis can be performed Traditional analysis of imagery, in conjunction with techniques for determining the variables for stability calculations from imagery, provides a more comprehensive method for examining slopes, which have limited accessibility for field work, or for preliminary analysis of the slope prior to investing time and effort into doing actual field monitoring.

SPECTRAL REQUIREMENTS cành HH HH Hà tinh 20

The image analysis approach to slope engineering employs passive remote sensing data These types of images record the re-emitted solar radiation from geographic locations, that is, waves across all bands of the light spectrum from the sun, which are either directly reflected in the visible light bands, or in the case of ultraviolet and infrared light, are absorbed by the ground and subsequently slowly emitted from the earth material The degree of re-emission of these light waves reflects the character and properties of the ground conditions The electromagnetic wave spectrum is shown in Figure 2.1.

For this type of analysis, infrared light re-emitted by the ground is chosen, due to it’s lower frequency which provides a greater penetration of the ground Signal attenuation depth is inversely proportional to frequency Ultraviolet (UV) light has such a high frequency that there is, in most mineral material, virtually no absorption of UV light Penetration depths of light in the infrared bands range from less than one micron in the near infrared to perhaps tens of centimeters, for loose materials, in the mid to far-infrared range (Hunt, 1980).

Since the measured reflectance is primarily in the infrared ranges, it minimizes whether the image was acquired during day or night, as radiation is re-emitted at a fairly constant rate and recharged with daily exposure to sunlight Temporal factors to consider are the degree of cloud cover (which should be at an absolute minimum), as well as snow coverage (if applicable), and associated air temperature, moisture and precipitation, and seasonal climate The presence of snow cover on the ground will interfere with the characteristic re-emission of infrared light by the earth material It is also important that vegetative cover is minimal In the mid-infrared and far- infrared ranges, infrared re-emission provides a degree of information about plant moisture contents, heat stresses, and mechanical stresses These vegetation factors may interfere with the infrared re-emission which characterizes the earth materials underlying the vegetation For slope analysis at mining or civil engineering sites, however, heavy vegetation can often be minimal due to mining operations and early or no reclamation.

In the case of the presence of vegetation upon the slope which is to be examined, the influence of the vegetation upon reflection of light from the visible red band and the re-emission of infrared radiation can generally be filtered out, leaving an index which numerically gives the re-emission of solar radiation from the soil below This is known as the Perpendicular Vegetation Index (PVI), developed by Richardson and

PVI = {[Sa - Val’ + [Sm - Vir}}°° (2.1)

Where S is the soil reflectance, V is the vegetation reflectance, and the subscripts “R” and “IR” represent reflected light in the red visible band and the infrared band, respectively Other approaches for filtering out vegetation will be discussed in

The required spectral resolution for approximating the density and the strength properties of the earth material need to, at the very least, cover the near-infrared to infrared bands (0.76 to 12.5 pm wavelength), but also the lower-frequency microwave bands Re-emission of infrared light by the earth material is the basis of determining the dielectric permittivity The dielectric permittivity is then used to find the ground density p and porosity n These parameters, in turn, are used to calculated the strength values of the ground material.

A band ra km 300 kHz radio

Radar Bands UHF: A=5-6m f= 50-100MHs iim | 700 Mie pond: 4=1-3m

S-band: A=8-15 an j mm —— Chand: A“4-8 an

— Pisibie Light blue: A= F552 pm ulfraviolet (f= 576-662 Tis

,P0-576 THz ] mm x-ray 300 PHz ved: A=.63.69 pm fr4I476 THs

—T——— near-ẽR: 1=.76 90 pm £7 355-399 THs gamma ray ¥

Figure 2.1: The Electromagnetic Wave Spectrum

Dielectric permittivity can also be applied to finding the degree of saturation of the soil This is achieved through the determination of the soil’s porosity, as well as the volumetric moisture content of the soil For surface geophysics, moisture surveys have traditionally been executed in microwave bands, which have a much lower frequency (and conversely a higher penetration depth) than the infrared light bands possess (Greaves et al., 1996) From the active microwave response, dielectric permittivity is calculated Upon satellite platforms, soil moisture surveys involve infrared response of passive solar radiation, since infrared signals can discern the presence of moisture According to Walrafen (1972), the “bending and stretching motions of bonds between the hydrogen and oxygen atoms in the water molecule produce distinct and unique identifying features in the infrared spectrum”.

DETERMINATION OF DIELECTRIC PERMITTIVŨY che 23

The relative dielectric permittivity e; of a material is determined through signal reflectance at varying frequencies Traditionally, subsurface geophysical methods use wave signals in the microwave or radar bands for measuring the permittivity, however, physical laws which govern the behavior of waves in these frequencies, also apply to waves in the infrared band.

By observing that the earth material to be studied is a dielectric material, and assuming that there is negligible refraction losses of waves hitting earth material, one can ideally assume that reflectance is complementary to absorbance (Pepper, 1995), or:

Af) +R=1 (2.2) where A(f) = frequency dependant absorption, and R is the mean reflection of electromagnetic radiation, and is comprised of the following wave components:

Where R; and R, are the reflection of s- and p-polarized electrical components of an electromagnetic (EM) wave The values of A(f) and R can be found, in terms of percent reflectance, by examining the digital number of a single picture element(pixel) of an image, which represents the strength of the reflected or absorbed signal.Passive microwave or infrared imagery is defined by the amount of electromagnetic solar radiation which is absorbed, then re-emitted by the earth’s surface and surface features This is the law of the conservation of energy, which states that thermal energy cannot be created nor destroyed, only transferred So absorbed radiation

(Promes et al., 1988) Therefore, the electromagnetic radiation which the ground is emitting directly governs the pixel value of that ground area Most imaging systems use 8-bit binary data formats, which means that pixel values range from 0 (no signal remission) to 255 (having a range of 2Ÿ or 256, values) The pixel’s brightness value, or digital number (DN) will fall within this range for 8-bit data, 255 will be the brightest possible value of any pixel in the image.

Solar radiation is un-polarized, meaning that at least several types of waves are active in several different planes Radiation from electromagnetic bands are composed of an electrical wave, with a transverse magnetic wave which is not coplanar to the electrical wave; rather, it travels in an orthogonal plane Therefore, EM radiation is un-polarized until it’s interaction with a common (non-metallic) surface, this contact between the wave and the surface leading to polarization.

Infrared imagery cumulatively shows both reflected and re-emitted infrared radiation However, most detected infrared energy is re-emitted from a material as radiation flux, in a direction perpendicular to the material’s surface, when that surface is modeled as an infinite plane and an infinite depth of the material (or a semi-infinite medium) is assumed The reflected component of the radiation scatters at the incident angle, and is usually not detected by a sensor directly overhead.

Electromagnetic energy which has penetrated the ground’s surface to stimulate heat through particle vibration also gives a greater indication of the electrical properties of the ground material, this heat is what is primarily detected by infrared sensors, far besides that is the heat directly reflecting from the ground’s surface.

Although it is at times a difficult assumption to make, it can be assumed that there are negligible losses of light due to backscattering, atmospheric interferences, or surface roughness If so, one can take the re-emission R as a compliment to A(f) If the re- emission of the s- and p-polarized electrical components of the EM wave (R; and R;) can be determined, the dielectric permittivity of the earth material can be found through the amplitudes of the propagation vectors of the electric fields of the earth material and of free air The re-emission components of R are described as follows:

Where j is the examined medium (the earth material) and & is amplitude of the propagation vector of polarization of the electric field j and ¢ is the dielectric permittivity of the earth material j (farads/meter), and is equal to €,€9 (Kraus, 1984; Palik, 1985) The relative dielectric permittivity e; (which is known as the dielectric constant) is a ratio of the permittivity of the material j, to the permittivity of free space (Bowers & Hanks, 1965) The dielectric constant is represented by the term £ọ, and is equal to 8.854 x 10”? farads/meter Amplitudes are computed through the dielectric permittivities of the materials at the interface through these equations

Where Šo is the amplitude of the propagation vector of free air or space (differences between air and space are assumed negligible) and is determined as follows:

Equations 2.4 and 2.5 are referred to as Fresnel Intensity Coefficients of the s- polarized electrical wave energy (perpendicular to the surface of a medium) and the p-polarized energy (parallel to the surface), from the electrical component of an electromagnetic wave Fresnel equations are a ratio of reflected or re-emitted electric field amplitudes to the initial electric field amplitudes (the initial electric field is indicated by the summed quantities in the denominators of Equations 2.4 and 2.5) for electromagnetic radiation which is incident on a dielectric material When these amplitude coefficients are squared, they become intensity coefficients, defined by includes the dielectric permittivities of the materials at the interface (in this case the permittivity of free space, or air, and the permittivity of the medium), since the p- polarized waves oscillate parallel to this interface so Equation 2.5 provides the vertical reflection coefficient across this interface, reflection being perpendicular to the plane of oscillation.

Oscillation of the electrical and magnetic components of an electromagnetic wave propagates electromagnetic radiation in the direction of the wave When an electromagnetic wave hits a surface at an angle, both the electrical and the magnetic components of the electromagnetic wave are subdivided into wave components which act parallel and perpendicular to the material’s surface The electric field’s component of oscillation, perpendicular to the target surface, is the polarized electrical energy in the s-plane (s-polarized electrical component) The electric field’s parallel component of oscillation is the polarized electrical energy in the p- plane Equations 2.4 and 2.5 refer to reflection of the “s” (perpendicular to the reflecting surface) and “p” (parallel to that surface) components of the polarized electrical component of an electromagnetic wave.

Image acquisition platforms characteristically detect the R; component of electromagnetic radiation; the R, component tends to obscure the appearance of ground features This is why subaquatic features may be observed from the sky directly above or at certain angles of view, but from some skewed perspectives not directly overhead these subaquatic features are obscured by the reflection of sunlight from the water’s surface Considering the influence of viewing perspective, where © is the inflection angle, or the angle of the incident wave signal on the camera from the Earth's surface; this term is geometrically dependent on the satellite's altitude and it's nadir's proximity to the study site This inflection angle can be defined by these terms: ©=sin! {y/x} (2.8)

Where “x” is the distance from the nadir (center axis) of the image to the observed location or object and “y” is the estimated altitude of the satellite (these estimates are based on orbital velocity and extent of image coverage) Because the width of a satellite’s path, or swath, results in images of rather narrow areas on the ground in relation to the satellite’s altitude, © is usually less than 5°.

Using an active remote sensing approach, where a signal (such as radar) is emitted from a satellite, more or less directly overhead from the target, this angle is virtually negligible With a passive remote sensing approach, where one is primarily concerned about radiation being re-emitted from the ground and not directly reflecting from the ground at it’s incident angle, © is still neglected because radiation re-emitted through the earth’s surface travels mostly upward towards the satellite- mounted sensors, at a very small ©.

Since © is assumed to be such a small angle, the "sin? ©" term can be omitted because {sin7(~5°)} is a negligible value, therefore Equations 2.6 and 2.7 simplify to: gj =a (2.9) Éo=o (2.10)

Subsequently, Equations 2.4 and 2.5 can be expressed as:

Rs = {[eg”Š - ứg°'1/ [eo 7 + si”? (2.11)

Rp = {[ei(€o'”) -eo(ei**)] / [seo '”)+eo(j°9]}” — (2.12)

To solve for gj for any known DN, the function A(f), which was expressed as the ratio of emissivity to the maximum pixel value accommodated by the data format, is used.

For 8-bit data for example, the maximum value is 2Ÿ, or 256, which results in the following expressions:

Setting Equation 2.3 equal to Equation 2.14, R = {256-DN}/ 256 = 0.5{Rs + Rp} Inserting Equations 2.11 and 2.12 into Equation 2.3 yields this solution:

PENETRATION DEPTH OF WAVE SIGNAL, chu Hà tiệt 29

The “skin depth” 8, or the distance which a wave can travel through a medium before losing it’s energy to electromagnetic impedance of the material, can be determined when the frequency of the applied wave signal and the relative dielectric permittivities are known The absorption of the wave signal’s energy by a material is known as attenuation, and the depth of penetration 6 is a reciprocal of the attenuation factor a’ This energy loss is due to the wave’s distortion, or straining, of the molecular packing order of the material through which it passes (in the instant that it passes through; with enough applied wave energy materials will display plastic behaviour and melt) The heat generated by this distortion reduces the total elastic energy of the wave This loss of wave energy can be modeled with this function(Burger, 1992):

Where E(z,t) is the wave energy at a given time t and depth z, Eo = energy penetrating medium at the surface, and the first natural base logarithmic term is the attenuation term This term decreases with depth z The second natural base logarithmic term is an imaginary term, and represents wave propagation through the medium The propagation factor is B’, and the attenuation factor is a’.

The attenuation constant is defined as: o = {o/c} {0.5[(e,? + 7)? -er]?? (2.20) where œ is the angular signal frequency and equals 2rƒ where ƒ is the wave transmission frequency in Hertz, ¢,’ is the real component of the complex relative dielectric permittivity, ¢,” is the imaginary component (or the dielectric loss factor; and ð = 2.17 Once the coefficient A is solved for the entire image, it is saved as “A.grd” The normal stress oayg must also be known at these arbitrary depths They are computed using Equations 3.3, 3.4, and3.5, through p (or Densitypci.grd) and the given depth increments, and assuming v is0.5 (see Section 2.8, Chapter Two) This is saved as “Normstress***.grd”.

Figure 8.9: Cohesion Computed from *.grd Files for p & yn This image results from also applying a 6 pixel x 6 pixel moving window average to the image illustrating the computed cohesion, as an image “smoothing” technique This is saved as “c6x6.grd”’.

The formulae for shear modulus G (Equation 3.2) are entered into the MapCalculator, using A.grd for A, Porosity.grd for n, and using the filesNormstress***.grd for the average normal stresses o,,, For Equation 3.8, the K value specific to each depth increment is computed with Equation 3.13, keeping in mind that the overconsolidation ratio OCR is set to 1.0 (see Section 3.1, ChapterThree) This means that in the Map Calculator, the shear modulus G is set up for solution with the various files entered in a configuration mandated by Equation 3.2,then divided by depth-corresponding files mandated by Equation 3.13 This yields the shear strength for that particular depth in pounds per square inch (psi) With to computed for each depth, the files for to are saved as “Shear***.grd”, where the asterisks represent the depths in inches.

DATA INDEPENDENCCE .- HH Hà 2 Hà Hà Hà nh ni 111 r0 tr 130

The techniques presented in this dissertation for the determination of the principle input parameters of slope stability analysis demonstrate that the principle parameters p, >, c, and / are each directly computed from DN, the “brightness value” of the pixel representing the ground respective to the principle input values This is a cause of concern, as potential errors or irregularities encountered within the computational procedures of any one parameter will also affect the accuracy of other computed parameters It is also arguable that determining four of the five principle parameters

(the fifth being slope inclination œ, from a separate DEM data source) from a single source, the image and it’s pixel values, is not advisable Errors within the preliminary steps of image processing or errors in the computation of fundamental values, such as

€, Will ultimately compound the error of the end product, which is the margin of safety against slope failure.

The idea of obtaining the four seemingly unrelated parameters p, 9, c, and / from a single source, the DN, may also fuel skepticism This dissertation, particularly material presented in Chapters Two, Three, and Six, demonstrates the theory behind the computational steps outlined by Figure 8.14 In Figure 8.14, it is shown that the computational procedures of the bivariates and c are dependent on p, as well as the parameters Ô and 7 from the intermediate computations which lead to these shear strength properties, as well as the projected groundwater level / Furthermore, / can be directly calculated from ¢, through 8 and n only, though the shear strength parameters are also dependent on p The groundwater level / is independent of p, yet the bulk density p is also directly dependent on £y, like / is.

Section 3.2, Chapter Three, explains the shear strength parameters @ and c, which characterize how shear strength changes with the increase or decrease in average normal stress Shear strength is a bivariate computation dependant on these parameters ¢ and c, which define the Mohr-Coulomb Failure Envelope (see Equation 3.15, Chapter Three) Chapter Three shows how the Mohr-Coulomb Failure Envelope is solved for through p and n, then Equation 3.18 solves for , and is used in conjunction with Equation 3.17 to solve for c This approach makes the solution for c directly dependent on ¿, as well as p and n (see Equations 3.2, 3.8, & 3.13, Chapter Three) Using the approach favoured in this dissertation, the groundwater level / is dependent on 9 and n (see Equations 6.7 & 6.8, Chapter Six), leaving n as a common factor in the computation of c and /.

This illustrates how two very different parameters, such as soil cohesion and phreatic surface levels, are defined by common characteristics, in this case the porosity. phreatic surface depth, where saturation is 100% Porosity also defines soil structure. Differing values of porosity are a clue of the degree of contact points or interlock between individual soil particles The nature of inter-particle contact of soil grains is a defining trait of the soil’s shear strength (see Section 3.1, Chapter Three) Shear strength is due primarily to the soil grain’s structure and composition Irregular soil grains have a greater surface area than smooth grains of comparable size, therefore tend to have less points of contact per unit area with adjacent grains, or less opportunity to interlock with adjacent grains Irregular grain shapes (resulting in a less effective packing order of the soil) as well as the greater surface areas, lead to higher void ratios Equation 3.14 shows that as the void ratio increases, the shear strength decreases Void ratios affect the shear strength parameters of soil, as well as the soil’s potential capacity to accommodate groundwater, the soil’s drainage and hydraulic conductivity, and the changes in effective stress induced by groundwater (Cosby et al., 1984).

Soil grains composed of materials tending to have rough or jagged surfaces will also have higher shear strength parameter values of @ and c than grains with smooth surfaces, due to the interlocking of particles Such grains are typically composed of strong, dense, non-soluble minerals Softer, less dense earth materials, such as clays, are composed of particles which do not have rough surfaces and do not interlock as easily, hence the shear strength parameters are not enhanced by any particle surface roughness In the case of soils with high clay contents, the introduction of moisture also decreases the value of $ and c This is due in part of the swelling of the soil with the addition of water, due to the strong adhesive properties of water to clay particles. This swelling increases the void ratio, which in turn decreases the shear strength parameters.

These considerations indicate the relationship of such seemingly different parameters such as c and / Shear strength parameters are influenced by n, and even 9, which also define / Ordinarily, 6 is a factor to consider when computing bulk density p (seeEquations 2.31 & 2.33, Chapter Two), but the preferred method presented in this dissertation bypasses this step and p is computed directly from e; (see Equation 2.36,

Chapter Two) All parameters are ultimately dependent on ¢, Recall from Chapter Two that this is “a measure of the extent to which the electric charge distribution in a material can be distorted or polarized by application of an electric field” (Berthelot et al., 2001) The polarization of electromagnetic energy to which a material is subjected is dependent on the properties of the material, governed by the material’s composition, which is characterized by p, n, and 9 In this way, the determination of £; from a pixel’s DN is imperative for finding p, n, and 9, which paves the way for finding the additional principal parameters 9, c, and /.

Figure 8.14: Computational Sequence of Principal Parameters The most direct approach towards determining the principal geotechnical parameters needed for slope stability analysis is diagrammed It is clearly shown that the principal parameters p, ý c, and Ì are derived from a single source, DN Since c and | are the ultimate products of this sequence of calculations, their strength of association and degree of independence must be tested.

Since all principal parameters are ultimately derived from ¢,, the statistical correlation between the end results of the principal parameters’ computational sequence should be determined Figure 8.14 shows that this sequence ultimately yields the seemingly unrelated principal parameters c and / The statistical correlation is not to be confused with the correlation coefficients presented in Figures 8.4, 8.5, 8.6, & 8.7, Section 8.4, which are derived from root mean square errors The statistical correlation cor(Xj, Yj) is the strength of association between the sequential variables

Xj and Yj: cor(Xi, Yi) = {cov(Xj, Yi) }/oxoy (8.9)

Where cov(Xi,Vị) is the covariance of the sequential input values X; and Yj. Covariance is a statistical measure of the correlation of fluctuations of the two different quantities X; and Y; The quantities in the denominator of Equation 8.9, ox and oy, are the standard deviations of the sets of variates X; and Y; The covariance of these variates is determined by this equation: cov(Xi,¥i) = {12" [Xi —X][Yi - VỊ} {1/7} (8.10)

Where X and Y are the mean values of the sets of ứ variates X; and Yj Figure 8.14 shows that the ultimate results of the preferred computational sequence of principal input parameters used in this case study are c and / Each pixel location/ground truth site is listed in Tables 8.1 and 8.2, along with the values of c and / measured at the study site and the computed values of c For cross-validation of the phreatic surface depth, discussed in Section 8.4, off-site well data was used, for temporal reasons and to minimize the effects of depth restrictions Thus in order to find the covariance of c and / of a common location, / must first be computed for each pixel location Table 8.4 summarizes the phreatic surface depths at given sample locations, computed through the capillarity and matric potential methods discussed in Section 6.3, Chapter Six Using this data set for statistical correlation computations ensures that each set of principal parameters has a common data source, the pixel covering the sample location in the imagery.

Using Equations 8.9 & 8.10 in conjunction with the principal parameters c and / from Tables 8.2 & 8.4, the absolute statistical correlation of c and / computed from imagery is found to be 0.576 For the field data shown in Table 9.1, the statistical correlation is 0.229,

Initial inspection of Figure 8.14 indicates that the statistical correlation between the principal parameters c and / should be strong, due to their common source ¢,, though a weak correlation between c and / is desirable With a statistical correlation of0.576, the independence of these ultimate principal parameters is greater than expected, since an absolute statistical correlation of 1.0 would denote total mutual dependence Weaker correlation may be due in part to p being computed directly and independently of 9 and n, rather than using Equations 2.31, 2.32, & 2.33 to find p. These equations require the particle density p, of the soil’s mineral component, which is unknown, so Equation 2.36 is used in lieu of Equations 2.31, 2.32, & 2.33 One must also consider the void ratio-dependent step-wise nature of calculating the amplitude function of the wave form through the earth medium F(e) (see Equation 3.6, Chapter Three) when computing the shear modulus, from which the principal parameter c is ultimately derived from Equations 3.2 & 3.8 This function may change with n, and will affect the relationship between / and c, since the outcome of c is dependent on F(e), though / is not influenced by F(e).

Table 8.4: Computed Phreatic Surface Depths

These computed values of the principal parameter Ì are used in conjunction with calculated values of principal parameter c from Table 9.2 to determine the degree of independence of the data used in an image analysis approach to slope stability analysis.

CPT 1 1.80CPT 2 1.10CPT 3 1.20CPT 4 1.05CPT 5 4.15CPT 6 3.32CPT 7 4.54CPT 8 4.32CPT 9 0.75CPT 10 3.95CPT 11 0.55CPT 12 4.87CPT 13 2.16CPT 14 3.59CPT 15 5.75CPT 16 4.81CPT 17 5.15CPT 18 4.02

DISCUSSION co cuc HH nền Y4 0n 0800004140000 0000001610000040800009000008 137 9.1: ANALYSIS OF TAILINGS DAM FAILURE chua 137 9,22 CASESTUDY - ng, 0 0114 080111 1 tt 1114010111 201100111 140 10.3: NARROWING THE GAP BETWEEN ENGINEERING AND SCIENCE

9.1: ANALYSIS OF TAILINGS DAM FAILURE

The tailings dam and impoundment selected for the case study presented in Chapter Eight covers a large area The downstream dam itself has a length of about three miles (see Figure 8.13, Chapter Eight) The geographic extent of this site already draws the appeal of using aerial or satellite imagery to examine the stability of this large dam structure The dimensions of the downstream dam will make regular on- site inspections of the entire dam too tedious to be effectively implemented.

Chapter Eight demonstrates one of the means by which image analysis will indicate potentially problematic areas throughout the dam Figure 8.13 identifies the locations of earth columns which are possibly unstable, given a projected slip surface depth of 18’ Unstable columns are not a cause for great concern if they are isolated; they will be contained by surrounding stable earth columns However, concern may arise where there is a heavier concentration of unstable earth columns, as shown on the downstream side of the dam facing the northeast in Figure 8.13 As discussed in Section 8.8, Chapter Eight, the greatest hydraulic thrust on the dam acts downstream of the greatest concentration of retained water or slurries This hydraulic thrust generates the greatest fluid pressure upon the upstream (submerged) face of the dam, which in turn may lead to an increased rate of seepage through the dam Figure 8.13 identifies this area as being potentially problematic, based on a higher concentration of pixels indicating potential instability Therefore, this area can be singled out for further and more in-depth observation and study, whereas the competence of other sections comprising the majority of the dam structure need not be examined as closely.

Figure 8.1, Chapter Eight, shows a sand core installed within the dam structure, for the drainage of impoundment water which has infiltrated the dam structure. sediments carried by the tailings slurries This may particularly be a problem in the event that the grain size distribution of the core material coincides with the grain size distribution of the material of which the dam is comprised in such a way that the core, over time, meets Terzaghi and Peck’s criteria of filter material selection (Holtz & Kovacs, 1981) This problem with the dam’s outer constituent material may evolve over time and through the infiltration of tailings slurries, in which the solids in turn settle out of the slurry as it seeps through the dam This addition of solid fines to the dam’s outer structure changes the grain size distribution of the outer structure’s material In all likelihood, the introduction of additional fines will reduce the overall cumulative grain size distribution in such a way that the core will inadvertently meet the filter criteria (Holtz & Kovacs, 1981) and subsequently trap additional fines from the outer structure transported through seepage If these conditions force the drainage core to act as filter instead, then over time the core could become congested and a planned and orderly discharge of seepage from the dam will not be possible Instead, excessive seepage may lead to the wash out or sliding of the dam’s structurally imperative material on the downstream side, or even the fissuring or rupture of the inner dam structure, which facilitates the development of higher pore pressures on the downstream side, which could eventually lead to sliding.

The identification of locations throughout the dam which may be problematic due to excessive pore pressures also helps with problems related to run-off from the site. Figure 8.2, Chapter Eight, indicates which streams originating at the toe of dam could be tested and monitored for sediments, based on the stream’s proximity to potential troublesome or unstable parts of the dam.

Through the employment of remote sensing imagery, the preliminary identification of problematic parts of the dam can be done as frequently as new imagery becomes available, using the techniques covered in this dissertation Using these deterministic methods for preliminary failure hazard assessments limits the need to employ or implement ground control points or to use GPS to monitor existing creep (Alloway et al., 1998) The dam analyzed in Chapter Eight of this dissertation is part of a defunct mining site Often in such cases the property owners or management may lack the necessary technical equipment and time for monitoring mass movement within dam structure, through specialized surveying, GPS, soil testing, or geophysical equipment.

A federal or state environmental protection authority with jurisdiction over such a site may also lack the necessary time and resources for comprehensive and continuous on- the-ground site monitoring In such instances, imagery analysis as presented in this dissertation may be adopted as a broad, quick, and low-budget approach for preliminary stability monitoring Imagery contains a wealth of coded information which is more current and site-specific some of the more common methods of qualitative analysis discussed in Chapter One and repeated throughout this dissertation.

A pivotal theme of this dissertation is the vast amount of inter-dependent albeit unique data contained within a single image With so much information contained within each and all pixels and at the disposal of the analyst, imagery analysis can be applied to a wide variety of environmental concerns of any site, this case study focusing on a mine tailings dam One example is fugitive dust control within the tailings basin With a high volume of fine-grained material distributed over the large area covered by such a tailings basin, such a tailings basin is a potential source of man-made dust storms which can affect a large part of the surrounding area, given the right combination of climatic conditions Imagery can be used to pinpoint dust sources throughout the basin Of course, given that imagery is acquired during the genesis of a local dust storm, a dust cloud will be seen on the image However, some of the analysis techniques presented in this dissertation can be applied to this problem Equations 2.26 & 2.27, Chapter Two, are used to find the moisture content of the material Dry areas are potential dust sources With these spots identified through imagery analysis, extra remediation efforts can be made at these locations.Keep in mind that such a tailings basin covers such a vast area that it would be impractical to comb the basin over and sample the tailings at a large number of locations to test the moisture content and find a current moisture content distribution.Fugitive dust issues are not restricted to the area’s dry season, either During the winter, the ground in the very moist (particularly saturated) areas of the basin freezes,leaving tailings particles suspended within the ice Over time, exposure to solar radiation causes the disappearance of a large portion of the ice through sublimation; as this ice transpires, tailings particles within the ground, not bound in place by the ice, are liberated and dust is the result of this (Price et al., 1998) Imagery can identify areas within the basin where this freeze/thaw/dust cycle may occur by indicating where the ground is frozen Lower frequency band signals (particularly with radar bands) attenuate almost immediately in ice A low frequency image with very little or no frequency response over certain areas will indicate the presence of ice within those areas, thus potential sources of winter dust are identified.

In the case study is presented in Chapter Eight of this dissertation, of a tailings impoundment located in White Pine, Michigan Methods discussed in this dissertation are performed on imagery within a GIS system With the geospatial data representing ground properties and parameters, slope stability analysis utilizes these data to assess the stability of a tailings dam This tailings dam is composed of earth materials (primarily clays), as well as tailings materials (mine wastes which in this case were produced throughout the processing of copper ore) The dam structure is not of a uniform composition; the composition is actually quite heterogeneous when one considers the varying composition of the tailings materials employed in it’s construction, over time This is due to different qualities of ore mined over time, as well as the varying quality of tailings coming from the concentrators and smelters, due to such factors as temporal changes in metal prices and mineral processing becoming more efficient over time.

The tailings dam structure, built up over these three decades, reflects these changes in material the tailings composition Therefore, it is not sufficient to assume uniform ground conditions throughout the entire dam structure, and it certainly is not sufficient to apply customary slope stability models, focusing on trial cross sections,for such a vast structure, since these models are mostly applied to smaller and more uniform embankments It is wiser to use an area slope stability model, not limited to specific slope profiles, with comprehensive geospatial geotechnical data as input.

Remote sensing provides the raw data to compute site-specific geotechnical data throughout the area outlined in the imagery, and GIS provides the capability to compute safety indices against sliding for slopes throughout this area The relatively remote location of this site impedes frequent comprehensive field inspections and analysis; examination of updated imagery, as it becomes available, allows site inspectors to monitor overall stability of the roughly two mile long dam.

10.3: NARROWING THE GAP BETWEEN ENGINEERING AND SCIENCE

This dissertation has been designed with a target audience of engineering professionals in mind, focusing more on obtaining a practical solution through remote sensing imagery and explaining some theoretical aspects, well known to scientists, in greater detail for the benefit of a reader who has a stronger engineering background.

All material discussed in this dissertation is based on established theory and practice in the fields of physics, geophysics, soil physics, soil dynamics, geobotany, hydrology, remote sensing, GIS, and geotechnical engineering Innovations which this dissertation introduces to the reader are the computation of the dielectric permittivity of the ground from remote sensing imagery (Section 2.4, Chapter Two), computation of the unsaturated shear strength properties through the dielectric permittivity (Chapter Three), using surface moisture to compute the depth to the phreatic surface (Section 6.3, Chapter Six), and modifying the DRASTIC model for slope stability forecasting (Section 6.4, Chapter Six) This dissertation also demonstrates the ease and reliability of using imagery to perform desk-top preliminary analysis of geotechnical problems with common GIS tools This may prove to become an effective means of saving time and cutting costs for preliminary geotechnical analysis and site reconnaissance (see Section 1.6, Chapter One) The presented techniques and methods are at this time established yet not extensively researched Geotechnical engineering is a relatively young field of study, and the remote sensing and GIS fields have evolved even more recently.

Geotechnical engineering and remote sensing sciences have extremely different allow sufficient intermingling of these subjects The differences in professional disciplines of the foremost researchers in geotechnical engineering and remote sensing are so vast that there are presently not very many researchers who are crossed-trained between these disciplines, explaining why the issue of using remote sensing sciences for geotechnical problem solving has never been adequately addressed It can be safely stated that most civil or mining engineers have never taken a course in remote sensing, GIS, or even geophysics Conversely, it is most unlikely that very many remote sensing scientists have ever studied soil mechanics. Comprehensive knowledge of all techniques involved in this type of research stimulates new ideas or the exploration of new approaches towards geotechnical problem solving with imagery Another explanation of the lack of interaction between these fields is that using remote sensing for geotechnical problem solving is a very new idea, therefore not much discussion has taken course to generate sufficient interest or demand for this type of research It is believed that demand for this type of research will increase as worldwide urban sprawl is being forced onto the hillsides It is also hoped that as urban development becomes more regulated throughout the world, urban planners will take note of these techniques as a quick and cost-effective method of preliminary zoning and planning.

CONCLUSIONS & RECOMMENDATIONS ằ 143

ADVANTAGES OF REMOTE SENSING APPROACH àc he 143 10.2: PROBLEMS WITH REMOTE SENSING APPROACH che 145 10.3: AUTOMATED LANDSLIDE HAZARD WARNING SYSTEM eo 149 10.4: POTENTIAL USERS & BENEFICIARIES OF THESE TECHNIQUES

Landslides are a common occurrence in many parts of the world, especially after seismic events Catastrophic landslides are often induced by earthquakes, and many earthquake victims are not caught in collapsing buildings, but are buried by the ensuing mudslides This is particularly true in developing countries, where urban sprawl pushes settlements onto more unstable hillsides, and construction is subject to little or no regulation As an example, in January 2001 in El Salvador, an earthquake which measured 7.6 on the Richter Scale killed up to 1600 people (Associated Press, 2003) The majority of these people succumbed to a mudslide which cut through the town of Santa Tecla, which was triggered by the quake Creeping failures are often catalyzed by seismic tremors In Southern California, urban expansion has forced developers to build residences in the hills surrounding the Los Angeles basin. Frequent seismic tremors on the San Andreas Fault induce gradual creeping failures to occur within these hillside developments As a result, residential property damage, due to slope failures, is a major problem in that area.

Such examples outline the need for a comprehensive system for landslide hazard analysis, over a wide geographic range Traditional field sampling, monitoring, and testing of soils and wells, for each location within the area, is just not possible.Sampling and testing of soils is time consuming and labour intensive Monitoring wells or peizometers for pore water pressures are costly in terms of equipment and time The same applies for satellite surveillance of ground control points with global positioning system (GPS) units emplaced on the ground This type of monitoring also requires the installation of specialized equipment at key locations thought to represent landslide activity characteristic to that site, and amounts to little more than forecasting an acceleration of an existing slide based on the monitoring of existing ground conditions and movement (Duffy & Whitaker, 1999) With the techniques presented in this dissertation, imagery analysis handles the same tasks almost instantaneously, inexpensively, covering the entire geographic extent of the image, without using an observation over time approach.

The easiest sources of ground data, which also cover broad areas in a continuous manner, are digital satellite images Imagery is currently often used for landslide studies, albeit as a tool of qualitatively observing characteristic warning signs or monitoring any existing movement, rather than in a predictive capacity Active remote sensing approaches are favoured here, including radar, LIDAR (light detection and ranging), or laser imagery, which provide data representing surface terrain, at a very fine resolution (Bitelli et al., 2004) However, satellite surveillance of slopes with lasers or LIDAR are restricted to observing and monitoring existing creep or other small precursory mass movements leading up to a landslide.

In terms of obtaining data for landslide forecasting, the traditional field-based methods are time consuming and can only be limited to a few locations This does not allow the development of site-specific continuous, or rasterized, geospatial data sets Most analysis is currently done in a qualitative manner, where troublesome spots are already identified through past occurrences or existing creep In the absence of GPS or LIDAR surveillance, existing creep requires a team of surveyors to take regular measurements, or creep is monitored through placing strain gauges in boreholes or trenches, which first need to be excavated Topographic maps of the site are examined, rainfall data is monitored, wells and peizometers may or may not be installed, and similar sites are identified The risk of a sliding failure at these sites is assessed by their similarities to the known trouble spots The extent of satellite image analysis is to estimate the presence of groundwater by qualitatively determining the degree of moisture of the slope material Areas which appear dark on the image are thought to have higher moisture contents, and a closer proximity to the phreatic surface, than lighter areas would indicate If any geotechnical sampling of the soil or lab testing occurs, to determine the soil strength properties, it is mostly limited to locations where a road or structure is to be built, or where a steep slope may ominously loom over an existing road or structure.

This type of traditional landslide hazard mitigation requires the efforts of surveyors, well diggers, geologists, geotechnical field technicians, lab technicians, and engineers Analysis is limited to perhaps a handful of sites at a time, is not updated frequently, and is expensive in terms of labour costs Computerized engineering analysis of slopes through digital imagery can be performed within a very short period of time, by a single trained user, at a fraction of the cost of using traditional methods A major additional benefit is that the extent of the area for which forecasting is performed is only limited by the geographic boundaries of the spatial datasets used as input In terms of monitoring slopes, updates in landslide forecasting can be performed as often as new geospatial data becomes available This method provides instant area-wide geotechnical sampling, as each pixel in the image yields estimates of the properties of the ground material.

The computerized digital image analysis approach towards landslide prediction not only saves time, efforts, and resources, it also opens up slope areas which are inaccessible, remote, or not open to intrusive testing and monitoring methods Slope analysis can be performed almost entirely in a desktop manner Field efforts can be limited to perhaps having one or a few ground control sites within each image for cross-validation of the calculated parameters, which could be used to account for any potential error which may result from slight corruptions of the geospatial datasets. Arguing the convenience, ease, and low cost of imagery analysis may be detracted by questions of the accuracy of the methods and case study presented in this dissertation, but additional research and development of this topic will guarantee the improvement of this imagery analysis approach With imagery representing a myriad of ground conditions, compressed and coded into each pixel’s value, digital satellite images emerge as a convenient, yet overlooked, source of data from which geotechnical and groundwater properties can be computed.

10.2: PROBLEMS WITH REMOTE SENSING APPROACH

The most fundamental problem is that techniques discussed in this dissertation extract remote sensing data are often presented as “soft” data, that is, it is determined by proxy or through the interpretations of analysts Geophysical formulas are sometimes indirect, and sometimes even empirical, means of calculating the parameters of the soil However, improvements through research, in the field of geophysics will lead to greater accuracy and reliability of these calculated parameters.

The formulae outlined within this treatise and used in the example cases outlined in Chapter Eight were developed with the intent of using infrared and microwave band signals Infrared signals are used to determine the dielectric permittivity, from which the soil density and the strength properties are computed To find the phreatic surface, the most direct approach involves ground emissivity (Section 6.2, Chapter Six) However, these methods are either intended or have only be used with low frequency microwave band signals, and may not be compliant with higher frequency infrared band electromagnetic wave response used to determine the relative dielectric permittivity (Section 2.4, Chapter Two).

Considering the transitive nature of wave theory, and the similar behaviour of wave signals throughout the entire spectrum, all computations were made with the infrared band of the image The problem with this approach is that a signal from this band has a penetration depth of less than one centimeter, so all computed properties and parameters represent values to be found at what is essentially the surface With this constraint, it becomes necessary to make the assumption that the ground is homogeneous (in terms of density and porosity) and isotropic All subsurface values are extrapolated values In most cases however, density increases as porosity decreases with depth Ideally, one should use very low frequency signals (such as P- band RADAR), which have higher depths of penetration, and which are also able to punch through groundcover (if any) with great ease With higher penetration depths,direct readings of the dielectric permittivities at different depths become available,which eliminates the need for provisions needed to infer the values of these permittivites based on the assumption of homogeneity Dielectric permittivity changes with soil depth, as the porosity and moisture content vary This is another source of potential error, not being able to take direct readings of the soil’s permittivity at changing depths, due to the unavailability of low-frequency data.

Another source of error stems from the preprocessing of the imagery, before it is made available to the analyst Preprocessing typically alters pixel values through smoothing algorithms, de-speckling, filters, DN distribution stretches (for contrast enhancement), etc For most pixels, an un-altered image has accurate DN’s which represent the true values of the soil’s properties and characteristics Elimination of outliers (“noise”) involves moving window averages, which alter all pixel values, thereby also altering the dielectric permittivity values computed from each DN. Errors in preprocessing and in the computation of the dielectric permittivity compound the errors in calculated values for density, strength parameters, and the phreatic surface depth.

Some sources of error which can easily be curtailed with updated imagery are the topographic models The most common rasterized data sets which represent terrain are the USGS 7.5’ Digital Elevation Models (DEM’s) As mentioned in ChapterTwo, DEM’s are merely topographic data, digitized from USGS 7.5’ topographic maps These maps are not frequently updated, and the original data is derived from public surveys Often is the case with DEM’s that mining properties, such as the case study site, are blacked out These are often very large tracts which are inaccessible to publicly employed surveyors And if they were shown, the data would be outdated due to the changing nature of excavation dimensions in a mining operation USGSDEM's have also a horizontal resolution of 30 meters, which is too great for an accurate assessment of the geometry of most highway cuts or man-made slopes.Error due to inadequate topographic models can be minimized at some expense,however High resolution (< one meter) DEM’s are available from several satellite systems, and are updated frequently These platforms typically use high frequencyRADAR, laser scanning, or LIDAR (light direction and ranging) techniques to fabricate a digital model of the earth’s surface These types of DEM’s are, at this time, very expensive however Image resolution should be equal to DEM resolution.

One meter, or even sub-meter, resolution multispectral imagery is increasingly common and affordable This dissertation featured a very low budget demonstration of the imagery analysis techniques discussed Due primarily to fiscal constraints, The case study presented in Chapter Eight employed a free one meter resolution near- infrared image in conjunction with a DEM of the tailings dam constructed from digitized survey data With the absence of geotechnical data of the dam structure, engineering parameter validation between ground data and imagery were performed with the tailings material within the impoundment, for which data were available. This could be done, since the purpose of this was to demonstrate the accuracy of the image analysis techniques compared to field data from anywhere within the image; site calibration was restricted to some of the fundamental assumptions about the site’s properties.

At this stage, using only remote sensing data, as a basis for slope stability analysis,does leave some of the ground’s properties in question With or without any ancillary field data, some critical assumptions of each specific site need to be made, which indeed influence the output of a slope stability study The first assumption governs the computation of ¢,, this being based upon whether the aerial or satellite-mounted sensor only detects the s-polarized component of solar radiation, perpendicular to the ground (Equation 2.18, Chapter Two) Additionally, the possible absence of additional ancillary site-specific data pertaining to the local soil leads to critical assumptions of soil properties based mostly on mineralogical factors These assumptions include values of the relative permittivity of the soil’s mineral component em (Equation 2.25, Chapter Two), the saturation matric potential \m` and the water release curve exponent b (Table 6.1, Chapter Six), as well as the radiation attenuation coefficient A, and the local geothermal heat flux F (Equations 5.2 & 5.3,Chapter Five) These assumptions can be guided by consulting geologic maps or preferably SSURGO GIS data (Section 6.4, Chapter Six) Assumptions affecting the values of the shear strength properties ¿ and c are the values of the overconsolidation ratio OCR (Equation 3.13, Chapter Three) and Poisson’s Ratio (Section 2.8, ChapterTwo, and Section 8.2, Chapter Eight).

With the constrained data resources available for the Chapter Eight case study, no alternative was left but to make some improvisations in order to have all the necessary information Specifically, the temporal characteristics of field’s moisture and groundwater data (collected during the autumn of 1997) would differ with the imagery (acquired during the spring thaw of 1998) To compensate for this, meteorological data had to be used to compensate for estimated differences in static groundwater levels between the fall and the spring due to aquifer recharge of snowmelt and precipitation (see Section 7.1, Chapter Seven, and Section 8.2, Chapter Eight) To verify the employed methods of computing the phreatic surface depth, additional imagery for the entire county was acquired and used in conjunction with county well data The county well data was sifted through, with only wells being examined for which the water level data was relatively contemporary to (or within three weeks of) the imagery These provisions yielded only a handful of locations with usable well data The same was true with the cross-examination of imagery and field data for the strength properties of the ground; more data would have been much more desirable With the few data used to validate the site characterization techniques featured in the case study, the correlation coefficients presented in Section 8.4, Chapter Eight, are questionable Ideally, field data would be collected at the same time imagery is acquired, and used in conjunction with an accurate and current DEM having the same resolution as the imagery.

103: AUTOMATED LANDSLIDE HAZARD WARNING SYSTEM

The case study in Chapter Eight yielded what seem reasonable, albeit not totally accurate, results That is to be expected These types of geotechnical characterization and analysis techniques, like more customary geophysical survey methods, yield soft data To date, no data collection method is more reliable than field sampling and testing These techniques, while not reliable enough for concise site characterization prior to any project, provide useful techniques for inexpensive preliminary examinations of sites, as well as fast and inexpensive analysis of locations for which regular field sampling and testing is impractical and costly These techniques also provide an outline for an automated warning system of landslide hazards or soil

FUTURE RESEARCH Gì nọ Hà 12 1 41 21 0211111111111 1e 153

Deterministic site characterization is difficult to standardize because to the spatial variability of soil characteristics, particularly in terms of mineral composition (Equation 2.25, Chapter Two, or Equations 6.8 & 6.9, Chapter Six) It is therefore beneficial at this time to have as much ancillary data about the ground as possible. Existing techniques of geospatial soil classification and site characterization are briefly introduced in Chapter Five, and should not be completely disregarded Those techniques are widely accepted, and would be especially useful with fortifying the results of the deterministic image analysis methods which this dissertation features.

Improvements or developments made with ancillary geospatial data will lead to imagery, regardless of the analysis method used Specifically, a great step would include the availability of more comprehensive geospatial data for soil mapping This should fall under the responsibilities of the SSURGO program, which currently is limited to soil taxonomic classifications having to do with soil chemistry, salinity, and mineralogy The principal subscribers to SSURGO data solve problems related to land use, land management, agriculture, and ecology, but not geotechnical problems. The soil parameters catalogued by SSURGO should be extended to include soil densities, void ratios, shear strength properties, static groundwater table elevations, making SSURGO data more comprehensive The geographic coverage provided by SSURGO data does not include the entire United States, either, making it necessary to continue digitizing and compiling digitize all county, state, and federal soil surveys.

As more comprehensive soil data is compiled, including geotechnical properties, more efforts are needed to improve the geobotanical correlations between soils and vegetation types Geobotany can prove useful in areas where vegetation obscures imagery of the ground, in order to verify the effectiveness of vegetation canopy filters (see Equations 5.2 & 5.3, Chapter Five) in isolating the soil’s component of the pixel value.

Efforts to compile soil classes can also be extended to include a spectral signature characteristic of each class of soil, which can then be used for tricorder or spectral matching comparisons with the unknown soils in the image If this is too gargantuan of a task, then subpixel classifiers could at least become a standard attachment for remote sensing software (see Section 5.2, Chapter Five) With the subpixel classifier,soil properties dependent on soil mineralogy will become discernable, since subpixel classifiers determine fractional composition of the material The common approach thus far to infer ground properties from imagery is analysis through tricorder or spectral matching algorithms (McDougal et al., 1999), necessitating that such algorithms should also become a standard feature for remote sensing software, and databases soil types, properties, and spectral data need to be developed and made available One drawback is that such data represents ground conditions during the time field data is collected, whereas a pure remote sensing approach delivers current data, but improving conventional methods will fortify results derived through imagery analysis Certain obstacles exist Tricorder algorithms and comprehensive soil databases may require excessive computing power and data storage which most desktop computers may not be able to accommodate.

Most current satellite monitoring of the earth is of a higher frequency in order to monitor clouds, the atmosphere, vegetation, and the earth surface, not the subsurface. For more direct satellite coverage of subsurface problems, lower-frequency imagery acquisition capabilities and data need to become more common and available Low frequency radar, such as P-band, can yield direct readings of soil properties at variable depths, eliminating the need to extrapolate surface readings to the subsurface Using low frequency sensors, cloud cover is not an issue, since radar bands with a greater wavelength (interferometric synthetic aperture radar particularly) easily penetrate clouds, in addition to providing greater ground penetration Radar is an active remote sensing system Since it is not detecting residual solar energy, it can operate during the day or night, and not be affected by the position of the sun and atmospheric conditions, which in turn may ultimately affect relative permittivities computed from passive remote sensing data.

In this dissertation and case study, Poisson’s Ratio was assumed, based on critical ground conditions when a slide is initiated This is arguably a weak assumption, since using Equation 3.2, which assumes rigidity, seems to contradict using a Poisson’s ratio of 0.5, the assumption of fluid behaviour of the ground At an evolved failure surface, pore pressures nullify the effective stresses of the soil which indeed does compel the soil at the evolved sliding surface to act as a fluid, though whether the shear strength properties governing the resisting forces against sliding should be computed from unstable conditions, where the soil acts as a fluid, is debatable With no means to objectively determine Poisson’s Ratio, a maximum value of 0.5 had to be assumed, but this works out well since this yields a minimum safety index against sliding (see Section 2.8, Chapter Two) It would be beneficial if the possibility to calculate or approximate Poisson’s Ratio, based on other fundamental soil properties, were to be explored Until then it may be wise just to make an educated guess of what Poisson’s Ratio is, based on characteristic values and the identified soil classes in the image For ongoing landslide hazard monitoring, Poisson’s Ratio may be approximated just by installing GPS ground control points along a slope and monitoring changes in elongation accompanied by lateral compression of the array of ground control points.

With certain common equations used in this approach, research should continue to refine their applications As an example, the fundamental Equations 2.26 & 2.35, Chapter Two, and 3.2 & 3.13, Chapter Three, seem almost empirical Efforts should be made to find a broad function to replace the seemingly site-specific numerical coefficients found in these equations It should be known exactly what determines the value of the numerical constants in these equations, since such constants imply site-specificity.

Closing the discussion for further-improvement suggestions on existing and novel geotechnical site reconnaissance methods, it should be reiterated that what sets the featured techniques in this dissertation apart is that this is not tied to classifications which are referred to ground truth, but use an approach is independent of field observations; that is, entirely remote surveillance Featured here is a fast, easy, and fairly reliable low-budget approach for a landslide hazard assessment, which can be performed by a single analyst in a desktop environment With the exception of a few critical assumptions of existing ground conditions, the methods are objective and quantitative This, in addition to a procedural structure, sets the framework for a possible automated landslide hazard warning system, which could use imagery as an input, SSURGO data and spectral catalogues as features of reference, and which would yield current landslide hazard forecasts in a spatial form, such as a map For the time being and until the deterministic procedures can be improved, refined, and endure the test of time, it would be wise to include as much ground data as possible,for site calibration or verification of geotechnical properties of an earth material with regards to the material identified and characterized with data from an imaging spectroscopy system.

The overall research into this matter should not be restricted to one approach, all options should be examined With several approaches developed and employed, the optimum approach will naturally evolve to predominance over time Different approaches which warrant further examination may be as mundane as running the outlined procedures through different software which have map calculator features or which can otherwise handle rasterized geospatial data With the software used, some data was lost during the computational phases; another program which is more inclined to handle the vast amount of data contained in imagery may better accommodate imagery throughout it’s various intermediate computational phases.This approach may have weaknesses, none which will not be improved or corrected over time with proper additional research The underlying theme is that the speed,ease, and cost of landslide hazard analysis are improved by cutting out, or at least limiting, the need for field trips, ground-truthing, and site calibration Imagery is an unbiased visual representation of the actual ground conditions, providing a wealth of usable data available through systematic decoding and interpretation.

Where DN is the digital number (“brightness value”) of the pixel, ọ is the dielectric permittivity of free space (8.854 x 10° Farads/meter), and g¡ is the dielectric permittivity of the earth material To simplify the formula to solve for gị, set eg =

X, 0 =X’, gj" = y, and ¢; = y’ so that x? =x’ and y? =y:

{256-DN}/256 = {[x?~ 2xy + y?]/ [x2 + 2xy + y?]} x? — Ixy + y? = {[256-DN] /256}x? + {[512-2DN] /256}xy + {[256-DN] / 256}y’; set {256-DN}/ 256 = “n”, so: x? — 2xy + y? = nx? + 2nxy + ny”

Subtracting “x2” from each side of the equation yields: y - 2xy = {n-l }x? + 2nxy + ny”

From which “ny“” is subtracted from each side to solve for “y”, or the variable259 representing ¢;”°:

Adding “2xy” to each side yields:

Since “x’” represents a numerical value, or the dielectric permittivity of free space, then for a given value of DN, this equation assumes a quadratic form for which one can solve for “y” The quadratic equation is: y = {-b+ [b’ — 4ac]°} / 2a

For which the terms “a”, “b”, and “c” can be expressed as: a= {1-n} =DN/256 b=- {n+1}2x = {[DN/256] - 2}2eo? Ÿ c=- {n-1}x? = {DN/256}£o

Substituting these values for “a”, “b”, and “c” into the quadratic equation facilitates the solving of ¢; for a measured value of DN.

After substitution of these expressions for “a”, “b”, and “c” into the equation for “y”, the numbers 4 and sọ can be factored out of the radical for the (b’ — 4ac)” * component of the numerator, leaving a coefficient of 2eg” ` for this component This is a common factor with the other component of the numerator, “b” Recall that ¢j = £r Eọ and “y”” is equal to ¢; Therefore, algebraic manipulation of the equation for “y” with the expressions representing “a”, “b”, and “c”, then solving for £;, yields this formula: tr = {[2-(DN/256)] + 2[1-(DN/256)]°°}/(DN/256) (Equation 9.4)

The driving assumption of the computational steps leading to Equation 9.4 is that the inflection angle of EM radiation © is negligible, a common assumption However, without this assumption, the complex conjugate of the relative permittivity is also regarded as a factor of the effective permittivity In this case, it is assumed that the reflection component perpendicular to the reflecting surface R¿ primarily is detected by an imaging platform, so Equation 2.16 is examined:

R; = {[cos @- (e;— sin©)?”] / [cos © + (e, - sin©)?']}ˆ (Equation 2.16)

Which is re-written as:

R;”Ÿ cos © + Ryo fe, — sin’@}°% = cos ©- {e, — sin’@}°°

Subtracting the term “R,°° (e; — sin’®)°°” from each side, then multiplying each side of the equation by “-1” yields:

-R,°* cos © = -cos @ + Ry” {e, — sin@}®Š + {e;— sin’@}"°

Adding “cos ©” to each side yields: cos @- R,° cos © = R,"'{e, — sin’@}"* + {e,— sin’@}"* cos © {1 - R,°°} = {1 + R:’°} {e,— sin’®)”* cos2@ {[1 - R¿"”]/[1 + R,””]} = e,—sin’®

Appendix B: CONE PENETROMETER TEST RESULTS FOR SHEAR STRENGTH

Cone Penetrometer Tests, borehole data Field sampling executed between 25 September and 18 October, 1997 For each of the 21 Cone Penetrometer Tests (CPT’s 1-21), the parameters cohesion c and internal friction angle $ can be extracted from the shear strength versus depth logs.

For extracting the values of o and c of the materials sampled/tested at sites CPT 1-21, refer to the Cone Penetrometer Test logs (Golder Assoc., 1998) These logs include charts showing the un-drained shear strength S, versus the depth 5 of the cone measuring the shear strength parameter Cohesion is defined as inherent shear strength, or shear strength at zero confining stress Therefore, c = S, at 6 =0 Ất., or at the surface The S, versus 5 charts were computed using an average value of p (104.2