A new approach to radiative transfer and related

atmospheric correction problems

            

 

H. Berger , T. Hay and E. Margerum

 

USA Topographic Engineering Laboratory, Alexandria, VA 22315

 

 

ABSTRACT

 

Most atmospheric correction codes are based in part or in whole on the radiative transfer equation (RTE), which is an integrodifferential equation.   It is well known to be an ad hoc equation which can and has produced incorrect answers.  This paper initiates a program for a new way of exploring where the RTE can produce unphysical answers in parameter-ratio ranges of genuine concern.  The program begins with formulating a new technique for rigorously transforming the scalar RTE, without approximation, into a “pure” partial differential equation (PDE), i.e. one involving only partial derivatives of finite and relative small order.  The virtue of this approach is that there are only a small number of analytical and numerical techniques for dealing with integrodifferential equations compared to the vast array of techniques for PDEs.  We develop a variety of tools more powerful than needed for the particular physical problem to demonstrate the robustness of the technique.  We then consider an atmosphere where Rayleigh scattering is dominant  and derive its PDE (We are unaware of anyone having done this before).  We also transform  a class of nonlinear integrodifferential equations into linear PDEs and solve for a multiplicity of solutions.

 

Keywords:  radiative transfer equation, atmospheric correction, Rayleigh scattering atmosphere, linear integrodifferential equations, linear partial differential equations, nonlinear integrodifferential equations, annihilation operators, Whitham’s method

 

1. INTRODUCTION

 

The motivation for this work is the concern about the very real possibility that most major atmospheric correction codes for optical wavelengths may have parameter-ratio ranges where they give unphysical results.  This is because most are based on the scalar radiative transport equation (RTE) (or one of its generalizations) which has long been known to be an ad hoc equation.1  

 

The RTE is an integrodifferential equation of such a character that the analytical and numerical techniques available to investigate its solutions nature are relatively few compared with the immense array of techniques developed over the years for partial differential equations (PDE).  For this reason this paper demonstrates that the RTE can be rigorously reformulated, without approximation, as a “pure” linear PDE, i.e. one involving only partial derivatives.  There is, however, a different PDE corresponding to each phase function (defined later). 

 

The paper applies a new technique developed to derive the RTE-based PDE for a Rayleigh scattering atmosphere, a fourth order linear PDE.  To obtain a clear sense of what this has accomplished, the reader is invited to inspect the traditional set of  coupled integral equations with variables that are integrals of the ones of interest, for Rayleigh scattering, that are derived from the RTE, given in the Appendix.  Note that Rayleigh scattering occurs even in “clear” air.

  

The ultimate purpose of this program is 1) to broaden the search for such unphysical parameter-ratio ranges in optical-wavelength atmospheric analyses for parameter values of any real concern and 2) to explore the power of the PDE approach to atmospheric problems.  The basic program is to develop solutions of the RTE for different atmospheric phenomena of interest using the new technique and compare them to rigorous solutions, where available, from electromagnetic theory.

 

This should expose any such unphysical parameter ranges.  Rayleigh scattering is a good candidate for such an approach because of the availability of electromagnetic solutions and because a comparison with the traditional coupled integral equations formulation for Rayleigh scattering gives some measure of the simplicity introduced by the PDE approach.

 

There have been a small number of cases in the literature considering special, specific circumstances under which the RTE has been demonstrated to give the correct answer or a good approximation.  This has strengthened confidence that it can often be relied on to produce accurate results. The literature, however, observes that incoherent fields are assumed so that power additivity takes place rather than field additivity or some other relationship.  It is known from the Van Cittert-Zernike theorem of optical coherence that even spatially incoherent sources can generate highly coherent fields over large regions of space.  The literature also relates the solution of the RTE as a statistical average of a randomly fluctuating Poynting vector whose Fourier transform is approximately the mutual coherence function. 1 

 

Ishimiru  also observes that although the description of interference and diffraction effects are included in the scattering and absorption characteristics of a single particle, transport theory itself does not include diffraction effects.1 As a specific example of the problem with using the RTE, we cite a recent work in which a careful field-based analysis of scattering in a dense medium with particle correlations considered produced a governing equation in the same general form as the RTE but differing in some details.2  The RTE has been recently demonstrated to have parameter-ratio ranges with the potential for large errors in microwave frequency non-atmospheric investigations for a few special cases.  In those it was possible to use numerical solutions of the rigorous electromagnetic theory for comparison.  That work was reported in a number of journal articles.

 

To illustrate how the transformation technique works, Section 2 discusses in considerable detail an example of an integrodifferential equation is solved so that the new method will be transparent to almost any reader.  This example can be followed easily, but nevertheless contains the essence of the difficulties to be surmounted.

 

The use of the technique is illustrated again in Section 5 to solve an example of a nonlinear integrodifferential equation, which satisfies the defining equation and the boundary conditions.  If [f(y)] appears in the equation raised to the Nth power (N = integer), then N boundary conditions will be needed.  Two boundary conditions are required here.

 

Withham discussed a technique for converting some integrodifferential equations into PDEs, but works for more limited circumstances than are of interest here.3  Ochs and Kristensson considered a problem which involved the propagation of electromagnetic waves in two types of dispersive dielectrics.4 The annihilation operators required were second order constant coefficients involving ordinary derivatives.  The functions to be annihilated in the integrands were separable.  There was one single integral in each partial differential equation. 

 

Those authors expressed an awareness that the annihilation operator method they used, which is a valuable one and clearly useful here, could be generalized but gave no sense this could include nonlinear equations.4  Unlike the technique used in reference 4, the method developed here uses the defining equation directly to determine the additional relationships needed for a unique solution.  Furthermore, it can be used whether ordinary or partial derivatives are present.

 

Previously one of the authors (H. Berger) presented, in an unpublished 30 March 1998 memorandum, a method for converting the scalar RTE into a second order linear partial differential equation, free of any integrals, for the special case of  isotropic, and possibly lossy,  scatterers.  The method discussed here  represent a method capable of attacking much more general scattering problems and can be seen to include the technique this author used earlier.

 

There are few, if any, universities in the USA which offer full length courses on integrodifferential equations.  There is only a tiny community of people skilled in dealing with them.  Yet almost all universities offer multiple full-length courses on PDEs.  Most physics and mathematics majors are required to take multiple full length courses on them in their education and most electrical, aeronautical and mechanical engineering students are required to take at least one such course.  Thus transforming the RTE into PDEs opens the area to exploration by technical professionals from many backgrounds including some we have not listed. 

 

Further, the solution of PDEs is currently a very active area with a great many practitioners, and has been for decades, in many engineering disciplines.  The three IEEE Transactions, Microwave Theory and Techniques, Antennas and Propagation and  Electromagnetic Compatibility issue research and exploratory reports that are dominated by articles authored by people that apply PDEs in new and innovative ways to solve engineering problems.  Of course the various mathematics societies offer many journals devoted exclusively to PDEs.  This brief description in no way comes close to exhausting the list of journals that include many articles in this area.  Finally the modern monograph and textbook literature on PDEs so completely dwarfs what is available on integrodifferential equations that comparison is difficult.

 

2.  A  RELATIVELY  SIMPLE  EXAMPLE

 

While we are interested in solving the radiative transfer equation, to illustrate how the basic technique works, a much simpler example is first offered here involving a linear integrodifferential equation with an ordinary derivative.  The method has been used when partial derivatives are present and the more complex RTE is considered later.  We will consider the integrodifferential equation for f(x) given in eq.(1).  The basic idea is to try to convert this equation into an ordinary linear differential equation and to use the defining equation to constrain the relation between what appears to be an excess of independent constants to insure that the correct and unique solution is obtained.  The equation and boundary condition are

 

                                                        2           

                                    df/dx  =  ò g(x) f(y) dy,     g(x)  =  x,   f(0) = A = constant                                                                     (1) 

                                                        0           

 

For this specific function g(x) = x  we can find a particular differential operator (D2) that does the following.           

 

                                                     (D2) g(x)  = ( d2  / dx2 ) g(x)   =  0                                                                                         (2)

 

This equation also defines (D2) that is said to annihilate g(x).

 

If we apply the operator (D2) to both sides of eq.(1) we quickly  find

 

                                                                                         2

                                                     d [(D2) f] / dx  =  ò (D2) g(x) f(y) dy                                                                                      (3)

                                                                                         0

 

Since the operator, (D2), operates only on functions of x and (D2) g(x)  =  0.  then the integral in eq.(3) becomes zero, and the equation reduces to the linear  ordinary differential equation

 

                                                      d [(D2) f] / dx  =  d3f / dx3   =  0.                                                                                         (4) 

 

At first glance, this is a simpler equation to solve than eq.(1).  Of course more boundary conditions are needed with eq.(4) than with eq.(1).  The general solution to eq.(4) is obviously

 

                                                            f(x)  =  B1x2  +  B2x  +  B3                                                                                                  (5)

 

              B1, B2 = undetermined constants,         B3  =  A = boundary condition  constant

 

This process ends, incomplete, unless we can generate more boundary conditions.  To do this eq.(1) is rewritten as

 

                                                                    df / dx  =  C1 x                                                                                                        (6)

 

where C1 is the integral in eq.(1) after x is factored outside of it.  Eq.(6) integrates to

 

                                              f(x)  =  (C1/2) x2  +  C2,        C2  =  constant                                                                                (7)

 

Comparing eqs.(5) and (7) it is clear that B2 = 0 and A =  B3 = C2.  Now we are not free to generate a new boundary condition because the relation between C1 and A are constrained by eq.(1).  Substituting Eq.(7) into Eq.(1) produces   

 

                                                         2           

                                          C1  =  ò [ (C1/2) y2  +  C2 ] dy  =  ( 8/6) C1  +  2 C2                                                                                                                  (8)

                                                         0

 

Carrying out the remaining simple algebra produces the solution

 

                                                            f(x)  =  - 6A [ (x2/2)  -  (1/6) ]                                                                                        (9)

 

This is a valid solution because it is seen to satisfy the original equation, eq. (1), and the boundary condition.  The RTE certainly appears more challenging than this example but the same basic approach will allow its transformation into a PDE whose form will depend on the form of the phase function (to be defined later).  Integrodifferential equations involving partial derivatives have been solved by this method as well.

 

3.  GENERALIZED  ANNIHILATION  OPERATOR  TECHNIQUE

 

The annihilation operator technique employed in Reference 4 uses linear constant-coefficient differential operators of first and second order and the integrands can be expressed as a sum of separable polynomial functions.  We will generalize that here in a fairly obvious way, and to an extent beyond what is needed to consider the Rayleigh scattering problem that we focus on in later sections.  The reason we do this is to make clear that this is a robust method capable of ultimately handling much more general scattering problems.  But first we try to make clear what the terms used in the first sentence of this paragraph mean.

 

The second order linear constant-coefficient differential operator has the form

 

                                                   DA2  =  A0  +  A1 (d  /dx)  +  A2 (d2  /dx2)                                                                             (10)

 

where  the A coefficients are constants.  The integrands in Reference 3 have the form

 

                                                             I22(x,y)  =  S g2(x) f2(y)                                                                                            (11)

 

where S is the summation sign and g2(x) and f2(y) are both quadratic (second-order) polynomials. 

 

It is clear from the relatively simple example in the Introduction, eqs. (1) – (9), that the same kind of results are obtainable for an arbitrary function g(x), so long as there exists an N th order constant-coefficient operator

                    

                                                                          N

                                                             DAN  =  S   An (dn  / dxn)                                                                                           (12)

                                                                          n=0

 

such that (DAN) g(x) = 0.  If we could find such an operator, i.e. an annihilation operator for g(x), then its use, when we have an arbitrary g(x) in eq.(1) instead of just x, would, in place of eq.(4), lead to

 

                                                                   dNf / dxN  =  0.                                                                                                    (13)

 

It is probably also reasonably clear that if instead of having g(x) in the integrand there appeared some g(x,y) the method would still work so long as (DAN) g(x,y) = 0.  This is in fact the case for Rayleigh scatters as we show and make use of later.  The next step in generalizing the annihilation operator method is  to consider the operator in eq.(10) when the An are not constant but rather functions of x, i.e. An  = An(x).  This can be demonstrated in a straightforward manner. 

 

More generally if the operator in eq. (12) has coefficients An which are functions of x then the arguments in the paragraph immediately above, supporting eqs. (7–8), still apply.  We then see that the annihilation operator technique applies for all g(x,y) and An(x) for which

 

                                                                    (DAN) g(x,y)  =  0                                                                                                 (14)

 

  and for which integrals like those in the equations exist.

 

Now consider an integrodifferential equation involving two integrals.  For clarity we use the relatively following simple case as an example.

 

                                                                   b                               d

                                                    df/dx  =  ò g(x,y) f(y) dy   +   ò h(x,z) f(z) dz                                                                                   (15)

                                                                   a                               c

If

 

                                                     (DAN) g(x,y)  =  0  and  (DBM) h(x,z)  =  0                                                                                     (16)

 

                                                               (DAN) (DBM)  =  (DBM) (DAN)                        

 

then  applying the two operators to eq. (15) we get the ordinary differential equation

 

                                                              d[(DAN) (DBM) f(x)] / dx  =  0                                                                                                (17)

 

which is free of integrals, the desired result. 

 

This idea is easily generalized in a trivial, but important, way to the more complex circumstance where the integrand of any one integral in an integrodifferential equation involves a kernel function that is the sum of terms.  Each of which are the products of functions and for one of each of these the corresponding annihilator operators are known.  This frequently comes up when the phase function in the RTE is expanded in a Legendre polynomial series.

 

 Finally these ideas are applied to integrodifferential equations involving one or more multiple integrals.  We illustrate the idea through another relatively simple example.  Consider

 

                                  f(x,y)/x  =  òò g(x,y,w,z) f(w,z,y) dw dz  +  òò h(x,y,w,z) f(w,z,y) dw dz,                                           (18)

 

where the limits of the integrations are fixed numbers.   Then if (DAN) g(x,y,z)  =  0,  and (DBM) h(x,y,z) = 0 and are communtaive operators and both of these operators are applied to eq. (18)  we now get a partial differential equation, free of integrals,  that appears as

 

                                                          [ (DAN) (DBM) f(x,y) ] / x  =  0.                                                                                         (19)

                

Clearly the operators in eq. (19) could be extended to include partial derivatives with respect to x and y and the method would still work. 

 

Most of the generalizations of the annihilation operator technique in this section can be further generalized and are more advanced than what is needed to investigate the Rayleigh scattering problem.  However the intent was to demonstrate that the method could be generalized to the point where a wide variety of atmospheric scattering problems, possibly some much more complex than Rayleigh scattering, could be investigated with this technique.

 

4. THE PDE FOR THE RTE AND A RAYLEIGH SCATTERING ATMOSPHERE

 

In this section we briefly consider the application of the annihilation operator technique to the scalar radiative transfer equation.  That equation, for azimuth-independent plane-parallel media, is the integrodifferential equation (Reference 5, pg. 15)

 

                                                                                                1

                                          m I(t, m) / t   =  I(t,m)  +  (1/2) ò p(m,m’) I(t,m’) dm                                                                             (20)

                                                                                               -1

 

where I = specific intensity, m  =  cosJ , J  =  zenith angle and t  =  optical thickness.    If p(m,m’) is annihilated by some operator, (Dm), then it follows from the previous discussion that

 

                                                              [(Dm) mI] / t  =  (Dm) I                                                                                                         (21)

 

In this section some of the potential virtues of using the annihilation operator technique may become clearer.  The atmospheric phase function for Rayleigh scattering is well known and is given on page 121 in reference 6 (after integration over the azimuth angles j and j’) as

 

                                      p(m,m’)  = (3/4) a [ 1  +   m2m2   + (1/2) (1 - m 2)(1 - m2) ]                                                                          (22)

 

Then straightforward differentiation yields

 

                                                                     3p/m3   =   0                                                                                                                       [23]                                                                                   

 

From this we see that the annihilation operator for the Rayleigh phase function is 3  /m3.  Applying this annihilation operator to the RTE in eq. (20) with p( m,m’ ) given in eq. (22), we find that

 

                                              33I / m 2t    +  (m) 4I / m3t    =   3I / m3                                                                                    (24)

 

This fourth order, linear, variable coefficient  PDE is then equivalent to the radiative transfer equation for Rayleigh scattering.   Generating an additional two boundary conditions here may not be straightforward but this may not be necessary when the method in the elementary example is used.. 

 

5.  SOLVING NONLINEAR INTEGRODIFFERENTAL EQUATIONS

 

As mentioned in the Introduction, the scalar radiative transport equation has been derived by some authors as an approximation, indirectly from some forms of the Boltzmann transport theory and directly from some random media theories. Some of these theories take nonlinear forms that have remained largely intractable.  Thus it is of some interest that we have unexpectedly found that the generalized method of annihilation operators can be used to solve certain kinds of  nonlinear integrodifferential equations that it converts to linear ordinary or linear partial differential equations. 

 

We discuss the basic idea by outlining the solution method of a particular example.  The accuracy of the solution is verified.  It also seems as though an additional boundary condition is needed for uniqueness that is not required in the linear case.  Additionally the method seems like it may be applicable to a nonlinear form of the radiative transport equation.  

 

The integrodifferential equation and boundary condition we consider  is

 

                                                                          2

                                                           df/dx  =  ò x [ f(y) ]2 dy.      f(0)  =  A                                                                                         [25)

                                                                          0

The method of solution runs parallel to that given for eq.(1) but with one key aspect different.  Following the kind of analysis given for eq.(1) we find with some messier algebra that eq. (25) yields

 

 

                                                                 f(x)  =   - A { [x2 / K1,2 ]  - 1 }                                                                                            (26)

 

where

 

                                                                 K1,2  =  - { (4/3)  ±  Ö(26/90) } 

 

 

Thus we obtain two solutions from eq. (30), corresponding to the ± signs  that satisfy the equation and the boundary condition.

 

This type of result is not seen for linear problems and strongly suggests that an additional boundary condition is needed for a unique solution.   This needs to be explored further..  The preceding approach seems as though it should also be successful for integrands involving f(x) raised to the Nth power, N = any integer, which should produce N distinct solutions satisfying the defining equation and the  single boundary condition used.  Here it would seem that N boundary conditions are needed to obtain a unique solution.

 

 

                     6:  WHITHAM’S  METHOD  FOR  OBTAINING  A  PDE  FROM  AN           

                                                     INTEGRODIFFERENTIAL  EQUATION

 

There exists in the literature a special case where an equation that is similar to the RTE is converted into a partial differential equation (PDE).  We will briefly comment on this and observe that the approach used is far more limited than the annihilation operator approach.  Whitham considers the  equation

 

                                                                                 µ

                                                            y(x,t)/t  +  ò K(x-z) {y(z,t)/z}dz  =  0                                                                              (27) 

                                                                                -µ

 

where µ represents infinity.2

 

Whitham observes that because this is a linear equation then one would expect elementary solutions of the form

 

                                                                        y  =  exp (kx – wt)                                                                                                       (28)

 

He indicates that this is possible if  the phase velocity, given by the ratio

 

                                                                            c(k)   =   w/k,                                                                                                          (29)

 

is the Fourier transform of the kernal function, K(x).  He further indicates that if c(k) can be represented by a finite length polynomial involving k raised only to even integer powers then one could derive a partial differential equation, which he displays, that is equivalent to the original integrodifferential equation. 

 

One could use the same kind of reasoning for the RTE and arrive at the same kind of conclusion as well as an equivalent partial differential equation under the same very special conditions.  This approach, which is very interesting, is not equivalent in any way to the annihilation operator approach.  It is very limited in its applicability while the annihilator approach will yield a partial differential equation under far more general circumstances, specifically whenever one can find the appropriate annihilation operator.  We infer from the ease of doing so for Rayleigh scattering, above, that it may not be too difficult for many other practical problems. 

 

7.  CONCLUSIONS

 

This paper has reminded the atmospheric correction community of the potential for serious problems in the use of current major atmospheric correction codes and discussed an approach to investigating that problem.  The problem arises from the use of the radiative transfer equation (RTE), which is widely regarded as an ad hoc usage in atmospheric problems.  Recently published work, carried out for microwave frequency non-atmospheric problems, have shown in special cases where rigorous results from electromagnetic theory could be obtained for comparison, parameter ranges in which the use of the RTE produced serious errors.                                                                                                    

                                                                                                                                                                    

 The approach to uncovering similar parameter ranges for optical wavelength atmospheric problems we are following is the same:  find instances where solutions to the RTE could be found and compared with rigorous solutions from electromagnetic theory.  A method has been developed which seems to make it far more likely that this approach could actually be persued on a phenomenon-by-phenomenon basis starting with the Rayleigh scattering problem.

                                                                                                                                                        .         

 This paper has considerably extended and augmented the annihilation operator technique demonstrated in Reference 4 for the equations of electromagnetic wave propagation in dispersive dielectrics.  It makes use of the original defining equation as a constraint and can be applied to both integrodifferential equations with ordinary and partial derivatives present.  It has applied this to the scalar radiative transfer equation usually used in atmospheric correction codes.  Although the techniques developed here should be useful for many different kinds of scatterers in the atmosphere we have applied it here to Rayleigh scattering which occurrs even in clear air.                                                                                                                                                               

                                                                                                                                                                      

 A pure partial differential equation (PDE) has been rigorously derived without approximation, i.e. one involving only partial derivatives.  It is the first such RTE-based equation for Rayleigh scattering to our knowledge.  The virtue of having transformed the problem into this form is that there is an immense solution technique literature for PDEs and a very small one for the RTE beyond numerical methods.  Although the PDE for Rayleigh scattering is a fourth-order PDE with variable coefficients some progress toward solving it has already been made and will be reported elsewhere.  Other RTEs with different phase functions will transform into different PDEs. 

 

The work reported here includes the extension of the above methods to certain types of non-linear integrodifferential equations.  This has been done by way of a relatively elementary example. This example suggests, if generalizations are valid, it may be useful in dealing with certain more complex theories regarded by some as more accurate expressions of transport theory than the RTE for atmospheric problems.

 

REFERENCES

 

1.   A. Ishimaru, “Wave Propagation and Scattering in Random Media,” IEEE Press, N.Y. and Oxford Univ. Press, Oxford, Chap. 7, 1997.

 

2.  B. Wen, L Tsang, D.P. Winebrenner and A. Ishimaru, “Dense Medium Radiative Tranfer Theory:  Comparison with Experiment and Application to Microwave Remote Sensing and Polarimetry, IEEE Trans. Geosci, Remote Sensing , 28. pp. 46-59, Jan, 1990.

 

3.  G. B. Whitham, “Linear and Nonlinear Waves”, John Wiley, New York, p. 368, 1974.

 

4.  R. L. Ochs, Jr., and G. Kristensson, “Using Local Differential Operators to Model Dispersion in Dielectric Media,” J. Opt. Soc. Am., Part A,  15, pp. 2208-2215, Aug. !998.

 

5.  S. Chandrasekhar,  “Radiative Transfer”, Dover Pubs., New York, 1960.

 

6.  M. E. Thomas and D. D. Duncan, “Atmospheric Transmission”, Chap. 1 in “Atmospheric Propagation of Radiation”, F. G. Smith, Ed., Vol. 2 of “The Infrared and Electro-Optical Systems Handbook,  J. S. Aceetta, et. al., Ed., SPIE Optical Engineering Press, 1993.

 

               APPENDIX: THE  TRADITIONAL  COUPLED  INTEGRAL  EQUATIONS  FOR         

                                  RAYLEIGH  SCATTERING

 

Chandrasekhar considers Rayleigh scattering in a plane-parallel azimuth-averaged homogeneous atmosphere and derives the integral equation for it as5

 

                                                                                                1                                        1

                       m I(t, m) / t   =  I(t,m)  -  (3/16)[ (3- u2) ò I(t,m’) dm  +  (3u2 – 1) ò (m’)2  I(t,m’) dm]                                  (30)

                                                                                          -1                                       -1

 

He then observes that this integral equation with multiple integrals can be reformulated as a fairly complex set of coupled linear integral equations which we give below.5  But first we must define the key symbols and how they relate to I.  These are

 

                                                  J(t, m )   =  1/(4Õ) ò  I(t,m ) dW                                                                                                         (31) 

 

where dW is a differential element of solid angle and the axial-symmetric K-integral is

 

 

                                                                      1

                                                K(t)  =  (1/2) ò  I(t , m , j)  [m]2 dm                                                                                                     (32)

                                                                     -1

 

It is also necessary to define a type of exponential integral with an integer index, n, which takes on the values 1,3 and 5 in the coupled integral equations.  It is

 

                                                     µ

                                       En(y)  =  ò  {[ exp(-xy) ] / xn } dx ,     y  =  ½t -