GeographicLib  2.6
TransverseMercatorExact.cpp
Go to the documentation of this file.
1 /**
2  * \file TransverseMercatorExact.cpp
3  * \brief Implementation for GeographicLib::TransverseMercatorExact class
4  *
5  * Copyright (c) Charles Karney (2008-2022) <karney@alum.mit.edu> and licensed
6  * under the MIT/X11 License. For more information, see
7  * https://geographiclib.sourceforge.io/
8  *
9  * The relevant section of Lee's paper is part V, pp 67--101,
10  * <a href="https://doi.org/10.3138/X687-1574-4325-WM62">Conformal
11  * Projections Based On Jacobian Elliptic Functions</a>;
12  * <a href="https://archive.org/details/conformalproject0000leel/page/92">
13  * borrow from archive.org</a>.
14  *
15  * The method entails using the Thompson Transverse Mercator as an
16  * intermediate projection. The projections from the intermediate
17  * coordinates to [\e phi, \e lam] and [\e x, \e y] are given by elliptic
18  * functions. The inverse of these projections are found by Newton's method
19  * with a suitable starting guess.
20  *
21  * This implementation and notation closely follows Lee, with the following
22  * exceptions:
23  * <center><table>
24  * <tr><th>Lee <th>here <th>Description
25  * <tr><td>x/a <td>xi <td>Northing (unit Earth)
26  * <tr><td>y/a <td>eta <td>Easting (unit Earth)
27  * <tr><td>s/a <td>sigma <td>xi + i * eta
28  * <tr><td>y <td>x <td>Easting
29  * <tr><td>x <td>y <td>Northing
30  * <tr><td>k <td>e <td>eccentricity
31  * <tr><td>k^2 <td>mu <td>elliptic function parameter
32  * <tr><td>k'^2 <td>mv <td>elliptic function complementary parameter
33  * <tr><td>m <td>k <td>scale
34  * <tr><td>zeta <td>zeta <td>complex longitude = Mercator = chi in paper
35  * <tr><td>s <td>sigma <td>complex GK = zeta in paper
36  * </table></center>
37  *
38  * Minor alterations have been made in some of Lee's expressions in an
39  * attempt to control round-off. For example atanh(sin(phi)) is replaced by
40  * asinh(tan(phi)) which maintains accuracy near phi = pi/2. Such changes
41  * are noted in the code.
42  **********************************************************************/
43 
45 
46 namespace GeographicLib {
47 
48  using namespace std;
49 
50  TransverseMercatorExact::TransverseMercatorExact(real a, real f, real k0,
51  bool extendp)
52  : tol_(numeric_limits<real>::epsilon())
53  , tol2_(real(0.1) * tol_)
54  , taytol_(pow(tol_, real(0.6)))
55  , _a(a)
56  , _f(f)
57  , _k0(k0)
58  , _mu(_f * (2 - _f)) // e^2
59  , _mv(1 - _mu) // 1 - e^2
60  , _e(sqrt(_mu))
61  , _extendp(extendp)
62  , _eEu(_mu)
63  , _eEv(_mv)
64  {
65  if (!(isfinite(_a) && _a > 0))
66  throw GeographicErr("Equatorial radius is not positive");
67  if (!(_f > 0))
68  throw GeographicErr("Flattening is not positive");
69  if (!(_f < 1))
70  throw GeographicErr("Polar semi-axis is not positive");
71  if (!(isfinite(_k0) && _k0 > 0))
72  throw GeographicErr("Scale is not positive");
73  }
74 
79  return utm;
80  }
81 
82  void TransverseMercatorExact::zeta(real /*u*/, real snu, real cnu, real dnu,
83  real /*v*/, real snv, real cnv, real dnv,
84  real& taup, real& lam) const {
85  // Lee 54.17 but write
86  // atanh(snu * dnv) = asinh(snu * dnv / sqrt(cnu^2 + _mv * snu^2 * snv^2))
87  // atanh(_e * snu / dnv) =
88  // asinh(_e * snu / sqrt(_mu * cnu^2 + _mv * cnv^2))
89  // Overflow value s.t. atan(overflow) = pi/2
90  static const real
91  overflow = 1 / Math::sq(numeric_limits<real>::epsilon());
92  real
93  d1 = sqrt(Math::sq(cnu) + _mv * Math::sq(snu * snv)),
94  d2 = sqrt(_mu * Math::sq(cnu) + _mv * Math::sq(cnv)),
95  t1 = (d1 != 0 ? snu * dnv / d1 : (signbit(snu) ? -overflow : overflow)),
96  t2 = (d2 != 0 ? sinh( _e * asinh(_e * snu / d2) ) :
97  (signbit(snu) ? -overflow : overflow));
98  // psi = asinh(t1) - asinh(t2)
99  // taup = sinh(psi)
100  taup = t1 * hypot(real(1), t2) - t2 * hypot(real(1), t1);
101  lam = (d1 != 0 && d2 != 0) ?
102  atan2(dnu * snv, cnu * cnv) - _e * atan2(_e * cnu * snv, dnu * cnv) :
103  0;
104  }
105 
106  void TransverseMercatorExact::dwdzeta(real /*u*/,
107  real snu, real cnu, real dnu,
108  real /*v*/,
109  real snv, real cnv, real dnv,
110  real& du, real& dv) const {
111  // Lee 54.21 but write (1 - dnu^2 * snv^2) = (cnv^2 + _mu * snu^2 * snv^2)
112  // (see A+S 16.21.4)
113  real d = _mv * Math::sq(Math::sq(cnv) + _mu * Math::sq(snu * snv));
114  du = cnu * dnu * dnv * (Math::sq(cnv) - _mu * Math::sq(snu * snv)) / d;
115  dv = -snu * snv * cnv * (Math::sq(dnu * dnv) + _mu * Math::sq(cnu)) / d;
116  }
117 
118  // Starting point for zetainv
119  bool TransverseMercatorExact::zetainv0(real psi, real lam,
120  real& u, real& v) const {
121  bool retval = false;
122  if (psi < -_e * Math::pi()/4 &&
123  lam > (1 - 2 * _e) * Math::pi()/2 &&
124  psi < lam - (1 - _e) * Math::pi()/2) {
125  // N.B. this branch is normally not taken because psi < 0 is converted
126  // psi > 0 by Forward.
127  //
128  // There's a log singularity at w = w0 = Eu.K() + i * Ev.K(),
129  // corresponding to the south pole, where we have, approximately
130  //
131  // psi = _e + i * pi/2 - _e * atanh(cos(i * (w - w0)/(1 + _mu/2)))
132  //
133  // Inverting this gives:
134  real
135  psix = 1 - psi / _e,
136  lamx = (Math::pi()/2 - lam) / _e;
137  u = asinh(sin(lamx) / hypot(cos(lamx), sinh(psix))) *
138  (1 + _mu/2);
139  v = atan2(cos(lamx), sinh(psix)) * (1 + _mu/2);
140  u = _eEu.K() - u;
141  v = _eEv.K() - v;
142  } else if (psi < _e * Math::pi()/2 &&
143  lam > (1 - 2 * _e) * Math::pi()/2) {
144  // At w = w0 = i * Ev.K(), we have
145  //
146  // zeta = zeta0 = i * (1 - _e) * pi/2
147  // zeta' = zeta'' = 0
148  //
149  // including the next term in the Taylor series gives:
150  //
151  // zeta = zeta0 - (_mv * _e) / 3 * (w - w0)^3
152  //
153  // When inverting this, we map arg(w - w0) = [-90, 0] to
154  // arg(zeta - zeta0) = [-90, 180]
155  real
156  dlam = lam - (1 - _e) * Math::pi()/2,
157  rad = hypot(psi, dlam),
158  // atan2(dlam-psi, psi+dlam) + 45d gives arg(zeta - zeta0) in range
159  // [-135, 225). Subtracting 180 (since multiplier is negative) makes
160  // range [-315, 45). Multiplying by 1/3 (for cube root) gives range
161  // [-105, 15). In particular the range [-90, 180] in zeta space maps
162  // to [-90, 0] in w space as required.
163  ang = atan2(dlam-psi, psi+dlam) - real(0.75) * Math::pi();
164  // Error using this guess is about 0.21 * (rad/e)^(5/3)
165  retval = rad < _e * taytol_;
166  rad = cbrt(3 / (_mv * _e) * rad);
167  ang /= 3;
168  u = rad * cos(ang);
169  v = rad * sin(ang) + _eEv.K();
170  } else {
171  // Use spherical TM, Lee 12.6 -- writing atanh(sin(lam) / cosh(psi)) =
172  // asinh(sin(lam) / hypot(cos(lam), sinh(psi))). This takes care of the
173  // log singularity at zeta = Eu.K() (corresponding to the north pole)
174  v = asinh(sin(lam) / hypot(cos(lam), sinh(psi)));
175  u = atan2(sinh(psi), cos(lam));
176  // But scale to put 90,0 on the right place
177  u *= _eEu.K() / (Math::pi()/2);
178  v *= _eEu.K() / (Math::pi()/2);
179  }
180  return retval;
181  }
182 
183  // Invert zeta using Newton's method
184  void TransverseMercatorExact::zetainv(real taup, real lam,
185  real& u, real& v) const {
186  real
187  psi = asinh(taup),
188  scal = 1/hypot(real(1), taup);
189  if (zetainv0(psi, lam, u, v))
190  return;
191  real stol2 = tol2_ / Math::sq(fmax(psi, real(1)));
192  // min iterations = 2, max iterations = 6; mean = 4.0
193  for (int i = 0, trip = 0;
194  i < numit_ ||
196  ("Convergence failure in TransverseMercatorExact");
197  ++i) {
198  real snu, cnu, dnu, snv, cnv, dnv;
199  _eEu.am(u, snu, cnu, dnu);
200  _eEv.am(v, snv, cnv, dnv);
201  real tau1, lam1, du1, dv1;
202  zeta(u, snu, cnu, dnu, v, snv, cnv, dnv, tau1, lam1);
203  dwdzeta(u, snu, cnu, dnu, v, snv, cnv, dnv, du1, dv1);
204  tau1 -= taup;
205  lam1 -= lam;
206  tau1 *= scal;
207  real
208  delu = tau1 * du1 - lam1 * dv1,
209  delv = tau1 * dv1 + lam1 * du1;
210  u -= delu;
211  v -= delv;
212  if (trip)
213  break;
214  real delw2 = Math::sq(delu) + Math::sq(delv);
215  if (!(delw2 >= stol2))
216  ++trip;
217  }
218  }
219 
220  void TransverseMercatorExact::sigma(real /*u*/, real snu, real cnu, real dnu,
221  real v, real snv, real cnv, real dnv,
222  real& xi, real& eta) const {
223  // Lee 55.4 writing
224  // dnu^2 + dnv^2 - 1 = _mu * cnu^2 + _mv * cnv^2
225  real d = _mu * Math::sq(cnu) + _mv * Math::sq(cnv);
226  xi = _eEu.E(snu, cnu, dnu) - _mu * snu * cnu * dnu / d;
227  eta = v - _eEv.E(snv, cnv, dnv) + _mv * snv * cnv * dnv / d;
228  }
229 
230  void TransverseMercatorExact::dwdsigma(real /*u*/,
231  real snu, real cnu, real dnu,
232  real /*v*/,
233  real snv, real cnv, real dnv,
234  real& du, real& dv) const {
235  // Reciprocal of 55.9: dw/ds = dn(w)^2/_mv, expanding complex dn(w) using
236  // A+S 16.21.4
237  real d = _mv * Math::sq(Math::sq(cnv) + _mu * Math::sq(snu * snv));
238  real
239  dnr = dnu * cnv * dnv,
240  dni = - _mu * snu * cnu * snv;
241  du = (Math::sq(dnr) - Math::sq(dni)) / d;
242  dv = 2 * dnr * dni / d;
243  }
244 
245  // Starting point for sigmainv
246  bool TransverseMercatorExact::sigmainv0(real xi, real eta,
247  real& u, real& v) const {
248  bool retval = false;
249  if (eta > real(1.25) * _eEv.KE() ||
250  (xi < -real(0.25) * _eEu.E() && xi < eta - _eEv.KE())) {
251  // sigma as a simple pole at w = w0 = Eu.K() + i * Ev.K() and sigma is
252  // approximated by
253  //
254  // sigma = (Eu.E() + i * Ev.KE()) + 1/(w - w0)
255  real
256  x = xi - _eEu.E(),
257  y = eta - _eEv.KE(),
258  r2 = Math::sq(x) + Math::sq(y);
259  u = _eEu.K() + x/r2;
260  v = _eEv.K() - y/r2;
261  } else if ((eta > real(0.75) * _eEv.KE() && xi < real(0.25) * _eEu.E())
262  || eta > _eEv.KE()) {
263  // At w = w0 = i * Ev.K(), we have
264  //
265  // sigma = sigma0 = i * Ev.KE()
266  // sigma' = sigma'' = 0
267  //
268  // including the next term in the Taylor series gives:
269  //
270  // sigma = sigma0 - _mv / 3 * (w - w0)^3
271  //
272  // When inverting this, we map arg(w - w0) = [-pi/2, -pi/6] to
273  // arg(sigma - sigma0) = [-pi/2, pi/2]
274  // mapping arg = [-pi/2, -pi/6] to [-pi/2, pi/2]
275  real
276  deta = eta - _eEv.KE(),
277  rad = hypot(xi, deta),
278  // Map the range [-90, 180] in sigma space to [-90, 0] in w space. See
279  // discussion in zetainv0 on the cut for ang.
280  ang = atan2(deta-xi, xi+deta) - real(0.75) * Math::pi();
281  // Error using this guess is about 0.068 * rad^(5/3)
282  retval = rad < 2 * taytol_;
283  rad = cbrt(3 / _mv * rad);
284  ang /= 3;
285  u = rad * cos(ang);
286  v = rad * sin(ang) + _eEv.K();
287  } else {
288  // Else use w = sigma * Eu.K/Eu.E (which is correct in the limit _e -> 0)
289  u = xi * _eEu.K()/_eEu.E();
290  v = eta * _eEu.K()/_eEu.E();
291  }
292  return retval;
293  }
294 
295  // Invert sigma using Newton's method
296  void TransverseMercatorExact::sigmainv(real xi, real eta,
297  real& u, real& v) const {
298  if (sigmainv0(xi, eta, u, v))
299  return;
300  // min iterations = 2, max iterations = 7; mean = 3.9
301  for (int i = 0, trip = 0;
302  i < numit_ ||
304  ("Convergence failure in TransverseMercatorExact");
305  ++i) {
306  real snu, cnu, dnu, snv, cnv, dnv;
307  _eEu.am(u, snu, cnu, dnu);
308  _eEv.am(v, snv, cnv, dnv);
309  real xi1, eta1, du1, dv1;
310  sigma(u, snu, cnu, dnu, v, snv, cnv, dnv, xi1, eta1);
311  dwdsigma(u, snu, cnu, dnu, v, snv, cnv, dnv, du1, dv1);
312  xi1 -= xi;
313  eta1 -= eta;
314  real
315  delu = xi1 * du1 - eta1 * dv1,
316  delv = xi1 * dv1 + eta1 * du1;
317  u -= delu;
318  v -= delv;
319  if (trip)
320  break;
321  real delw2 = Math::sq(delu) + Math::sq(delv);
322  if (!(delw2 >= tol2_))
323  ++trip;
324  }
325  }
326 
327  void TransverseMercatorExact::Scale(real tau, real /*lam*/,
328  real snu, real cnu, real dnu,
329  real snv, real cnv, real dnv,
330  real& gamma, real& k) const {
331  real sec2 = 1 + Math::sq(tau); // sec(phi)^2
332  // Lee 55.12 -- negated for our sign convention. gamma gives the bearing
333  // (clockwise from true north) of grid north
334  gamma = atan2(_mv * snu * snv * cnv, cnu * dnu * dnv);
335  // Lee 55.13 with nu given by Lee 9.1 -- in sqrt change the numerator
336  // from
337  //
338  // (1 - snu^2 * dnv^2) to (_mv * snv^2 + cnu^2 * dnv^2)
339  //
340  // to maintain accuracy near phi = 90 and change the denomintor from
341  //
342  // (dnu^2 + dnv^2 - 1) to (_mu * cnu^2 + _mv * cnv^2)
343  //
344  // to maintain accuracy near phi = 0, lam = 90 * (1 - e). Similarly
345  // rewrite sqrt term in 9.1 as
346  //
347  // _mv + _mu * c^2 instead of 1 - _mu * sin(phi)^2
348  k = sqrt(_mv + _mu / sec2) * sqrt(sec2) *
349  sqrt( (_mv * Math::sq(snv) + Math::sq(cnu * dnv)) /
350  (_mu * Math::sq(cnu) + _mv * Math::sq(cnv)) );
351  }
352 
353  void TransverseMercatorExact::Forward(real lon0, real lat, real lon,
354  real& x, real& y,
355  real& gamma, real& k) const {
356  lat = Math::LatFix(lat);
357  lon = Math::AngDiff(lon0, lon);
358  // Explicitly enforce the parity
359  int
360  latsign = (!_extendp && signbit(lat)) ? -1 : 1,
361  lonsign = (!_extendp && signbit(lon)) ? -1 : 1;
362  lon *= lonsign;
363  lat *= latsign;
364  bool backside = !_extendp && lon > Math::qd;
365  if (backside) {
366  if (lat == 0)
367  latsign = -1;
368  lon = Math::hd - lon;
369  }
370  real
371  lam = lon * Math::degree(),
372  tau = Math::tand(lat);
373 
374  // u,v = coordinates for the Thompson TM, Lee 54
375  real u, v;
376  if (lat == Math::qd) {
377  u = _eEu.K();
378  v = 0;
379  } else if (lat == 0 && lon == Math::qd * (1 - _e)) {
380  u = 0;
381  v = _eEv.K();
382  } else
383  // tau = tan(phi), taup = sinh(psi)
384  zetainv(Math::taupf(tau, _e), lam, u, v);
385 
386  real snu, cnu, dnu, snv, cnv, dnv;
387  _eEu.am(u, snu, cnu, dnu);
388  _eEv.am(v, snv, cnv, dnv);
389 
390  real xi, eta;
391  sigma(u, snu, cnu, dnu, v, snv, cnv, dnv, xi, eta);
392  if (backside)
393  xi = 2 * _eEu.E() - xi;
394  y = xi * _a * _k0 * latsign;
395  x = eta * _a * _k0 * lonsign;
396 
397  if (lat == Math::qd) {
398  gamma = lon;
399  k = 1;
400  } else {
401  // Recompute (tau, lam) from (u, v) to improve accuracy of Scale
402  zeta(u, snu, cnu, dnu, v, snv, cnv, dnv, tau, lam);
403  tau = Math::tauf(tau, _e);
404  Scale(tau, lam, snu, cnu, dnu, snv, cnv, dnv, gamma, k);
405  gamma /= Math::degree();
406  }
407  if (backside)
408  gamma = Math::hd - gamma;
409  gamma *= latsign * lonsign;
410  k *= _k0;
411  }
412 
413  void TransverseMercatorExact::Reverse(real lon0, real x, real y,
414  real& lat, real& lon,
415  real& gamma, real& k) const {
416  // This undoes the steps in Forward.
417  real
418  xi = y / (_a * _k0),
419  eta = x / (_a * _k0);
420  // Explicitly enforce the parity
421  int
422  xisign = (!_extendp && signbit(xi)) ? -1 : 1,
423  etasign = (!_extendp && signbit(eta)) ? -1 : 1;
424  xi *= xisign;
425  eta *= etasign;
426  bool backside = !_extendp && xi > _eEu.E();
427  if (backside)
428  xi = 2 * _eEu.E()- xi;
429 
430  // u,v = coordinates for the Thompson TM, Lee 54
431  real u, v;
432  if (xi == 0 && eta == _eEv.KE()) {
433  u = 0;
434  v = _eEv.K();
435  } else
436  sigmainv(xi, eta, u, v);
437 
438  real snu, cnu, dnu, snv, cnv, dnv;
439  _eEu.am(u, snu, cnu, dnu);
440  _eEv.am(v, snv, cnv, dnv);
441  real phi, lam, tau;
442  if (v != 0 || u != _eEu.K()) {
443  zeta(u, snu, cnu, dnu, v, snv, cnv, dnv, tau, lam);
444  tau = Math::tauf(tau, _e);
445  phi = atan(tau);
446  lat = phi / Math::degree();
447  lon = lam / Math::degree();
448  Scale(tau, lam, snu, cnu, dnu, snv, cnv, dnv, gamma, k);
449  gamma /= Math::degree();
450  } else {
451  lat = Math::qd;
452  lon = lam = gamma = 0;
453  k = 1;
454  }
455 
456  if (backside)
457  lon = Math::hd - lon;
458  lon *= etasign;
459  lon = Math::AngNormalize(lon + Math::AngNormalize(lon0));
460  lat *= xisign;
461  if (backside)
462  gamma = Math::hd - gamma;
463  gamma *= xisign * etasign;
464  k *= _k0;
465  }
466 
467 } // namespace GeographicLib
static T tand(T x)
Definition: Math.cpp:203
static T pi()
Definition: Math.hpp:187
static const TransverseMercatorExact & UTM()
An exact implementation of the transverse Mercator projection.
static T LatFix(T x)
Definition: Math.hpp:303
Header for GeographicLib::TransverseMercatorExact class.
void Reverse(real lon0, real x, real y, real &lat, real &lon, real &gamma, real &k) const
static constexpr int hd
degrees per half turn
Definition: Math.hpp:145
static T AngNormalize(T x)
Definition: Math.cpp:69
static T sq(T x)
Definition: Math.hpp:209
static constexpr int qd
degrees per quarter turn
Definition: Math.hpp:142
Namespace for GeographicLib.
Definition: Accumulator.cpp:12
static T AngDiff(T x, T y, T &e)
Definition: Math.cpp:80
static T degree()
Definition: Math.hpp:197
GeographicLib::Math::real real
Definition: Geod3Solve.cpp:25
#define GEOGRAPHICLIB_PANIC(msg)
Definition: Math.hpp:67
void Forward(real lon0, real lat, real lon, real &x, real &y, real &gamma, real &k) const
Exception handling for GeographicLib.
Definition: Constants.hpp:344
static T tauf(T taup, T es)
Definition: Math.cpp:251
static T taupf(T tau, T es)
Definition: Math.cpp:241