| 1 |
<h2>DESCRIPTION</h2> |
| 2 |
|
| 3 |
<em>v.surf.bspline</em> performs a bilinear/bicubic spline |
| 4 |
interpolation with Tykhonov regularization. The <b>input</b> is a 2D |
| 5 |
or 3D vector <em>point</em> layer. Values to interpolate can be the z |
| 6 |
values of 3D points or the values in a user-specified attribute column |
| 7 |
in a 2D or 3D vector layer. Output can be a raster |
| 8 |
(<b>raster_output</b>) or vector (<b>output</b>) layer. Optionally, a |
| 9 |
"sparse point" vector layer can be input which indicates the location |
| 10 |
of <b>output</b> vector points. |
| 11 |
|
| 12 |
<h2>NOTES</h2> |
| 13 |
|
| 14 |
<p>From a theoretical perspective, the interpolating procedure takes |
| 15 |
place in two parts: the first is an estimate of the linear coefficients |
| 16 |
of a spline function, which is derived from the observation points using a |
| 17 |
least squares regression. The second is the computation of the |
| 18 |
interpolated surface or interpolated vector points. As used here, the |
| 19 |
splines are 2D piece-wise non-zero polynomial functions calculated |
| 20 |
within a limited, 2D area. The length, in mapping units, of each spline |
| 21 |
step is defined by <b>ew_step</b> for the east-west direction and |
| 22 |
<b>ns_step</b> for the north-south direction. For optimal performance, |
| 23 |
the length of spline step should be no less than the distance between |
| 24 |
observation points. Each vector point observation is modeled as a |
| 25 |
linear function of the non-zero splines in the area around the |
| 26 |
observation. The least squares regression predicts the coefficients |
| 27 |
of these linear functions. Regularization avoids the need to have one |
| 28 |
observation and one coefficient for each spline (in order to avoid |
| 29 |
instability). |
| 30 |
|
| 31 |
<p>With regularly distributed data points, a spline step corresponding |
| 32 |
to the maximum distance between two points in both the east and north |
| 33 |
directions is sufficient. However, data points are often not regularly |
| 34 |
distributed and require statistical regularization or estimation. In |
| 35 |
such cases, v.surf.bspline will attempt to minimize the gradient of |
| 36 |
bilinear splines or the curvature of bicubic splines in areas lacking |
| 37 |
point observations. As a general rule, the spline step length should be |
| 38 |
greater than the mean distance between observation points (twice the |
| 39 |
distance between points is a good starting point). Separate east-west |
| 40 |
and north-south spline step length arguments allows the user to |
| 41 |
account for some degree of anisotropy in the distribution of |
| 42 |
observation points. Short spline step lengths, especially spline step |
| 43 |
lengths that are less than the distance between observation points, |
| 44 |
can greatly increase the processing time. |
| 45 |
|
| 46 |
<p>The maximum number of splines for each direction at each |
| 47 |
time is fixed, regardless of the spline step length. As the total |
| 48 |
number of splines increases (i.e., with small spline step |
| 49 |
lengths), the region is automatically split into subregions for |
| 50 |
interpolation. Each subregion can contain no more than 150x150 |
| 51 |
splines. To avoid subregion boundary problems, subregions are created |
| 52 |
to partially overlap each other. A weighted mean of observations, |
| 53 |
based on point locations, is calculated within each subregion. |
| 54 |
|
| 55 |
<p>The Tykhonov regularization parameter, <b>lambda_i</b>, acts to |
| 56 |
smooth the interpolation. With a small <b>lambda_i</b>, the |
| 57 |
interpolated surface closely follows observation points; a larger |
| 58 |
value will produce a smoother interpolation. |
| 59 |
|
| 60 |
<p>The input can be a 2D or 3D point vector layer. If input is 3D |
| 61 |
and <b>column</b> is not given than z-coordinates are used for |
| 62 |
interpolation. Parameter <b>column</b> is required when input is 2D |
| 63 |
vector layer. |
| 64 |
|
| 65 |
<p><em>v.surf.bspline</em> can produce raster (<b>raster_output</b>) OR |
| 66 |
vector <b>output</b> but NOT simultaneously. Note that topology is not |
| 67 |
built for output point vector layer. If required, the topology can be built |
| 68 |
using <em><a href="v.build.html">v.build</a></em>. |
| 69 |
|
| 70 |
<p>If output is a point vector layer and <b>sparse</b> is not specified, |
| 71 |
the output vector layer will contain points at the |
| 72 |
same locations as observation points in the input layer but the values |
| 73 |
of the output points will be interpolated values. If a <b>sparse</b> |
| 74 |
point vector layer is specified, the output vector layer will contain points |
| 75 |
at the same locations as the sparse vector layer points. The values will be |
| 76 |
those of the interpolated raster surface at those points. |
| 77 |
|
| 78 |
<p>A cross validation "leave-one-out" analysis is available to help to |
| 79 |
determine the optimal <b>lambda_i</b> value that produces an |
| 80 |
interpolation that best fits the original observation data. The more |
| 81 |
points used for cross-validation, the longer the time needed for |
| 82 |
computation. Empirical testing indicates a threshold of a maximum of |
| 83 |
100 points is recommended. Note that cross validation can run very |
| 84 |
slowly if more than 100 observations are used. The cross-validation |
| 85 |
output reports <i>mean</i> and <i>rms</i> of the residuals from the |
| 86 |
true point value and the estimated from the interpolation for a fixed |
| 87 |
series of <b>lambda_i</b> values. No vector nor raster output will be |
| 88 |
created when cross-validation is selected. |
| 89 |
|
| 90 |
<h2>EXAMPLES</h2> |
| 91 |
|
| 92 |
<h3>Basic interpolation</h3> |
| 93 |
|
| 94 |
<div class="code"><pre> |
| 95 |
v.surf.bspline input=point_vector output=interpolate_surface method=bicubic |
| 96 |
</pre></div> |
| 97 |
|
| 98 |
A bicubic spline interpolation will be done and a point vector layer |
| 99 |
with estimated (i.e., interpolated) values will be created. |
| 100 |
|
| 101 |
<h3>Basic interpolation and raster output with a longer spline step</h3> |
| 102 |
|
| 103 |
<div class="code"><pre> |
| 104 |
v.surf.bspline input=point_vector raster=interpolate_surface ew_step=25 ns_step=25 |
| 105 |
</pre></div> |
| 106 |
|
| 107 |
A bilinear spline interpolation will be done with a spline step length |
| 108 |
of 25 map units. An interpolated raster layer will be created at the |
| 109 |
current region resolution. |
| 110 |
|
| 111 |
<h3>Estimation of lambda_i parameter with a cross validation process</h3> |
| 112 |
|
| 113 |
<div class="code"><pre> |
| 114 |
v.surf.bspline -c input=point_vector |
| 115 |
</pre></div> |
| 116 |
|
| 117 |
<h3>Estimation on sparse points</h3> |
| 118 |
|
| 119 |
<div class="code"><pre> |
| 120 |
v.surf.bspline input=point_vector sparse=sparse_points output=interpolate_surface |
| 121 |
</pre></div> |
| 122 |
|
| 123 |
An output layer of vector points will be created, corresponding to the |
| 124 |
sparse vector layer, with interpolated values. |
| 125 |
|
| 126 |
<h3>Using attribute values instead z-coordinates</h3> |
| 127 |
|
| 128 |
<div class="code"><pre> |
| 129 |
v.surf.bspline input=point_vector raster=interpolate_surface layer=1 \ |
| 130 |
column=attrib_column |
| 131 |
</pre></div> |
| 132 |
|
| 133 |
The interpolation will be done using the values |
| 134 |
in <i>attrib_column</i>, in the table associated with layer 1. |
| 135 |
|
| 136 |
<h3>North Carolina dataset example using z-coordinates for interpolation</h3> |
| 137 |
|
| 138 |
<div class="code"><pre> |
| 139 |
g.region region=rural_1m res=2 -p |
| 140 |
v.surf.bspline input=elev_lid792_bepts raster=elev_lid792_rast \ |
| 141 |
ew_step=5 ns_step=5 method=bicubic lambda_i=0.1 |
| 142 |
</pre></div> |
| 143 |
|
| 144 |
<h2>KNOWN ISSUES</h2> |
| 145 |
|
| 146 |
<p> |
| 147 |
In order to avoid RAM memory problems, an auxiliary table is needed |
| 148 |
for recording some intermediate calculations. This requires |
| 149 |
the <i>GROUP BY</i> SQL function is used. This function is not |
| 150 |
supported by the DBF driver. For this reason, vector output |
| 151 |
(<b>output</b>) is not permitted with the DBF driver. There are |
| 152 |
no problems with the raster map output from the DBF driver. |
| 153 |
|
| 154 |
<h2>REFERENCES</h2> |
| 155 |
|
| 156 |
<ul> |
| 157 |
<li>Brovelli M. A., Cannata M., and Longoni U.M., 2004, LIDAR Data |
| 158 |
Filtering and DTM Interpolation Within GRASS, Transactions in GIS, |
| 159 |
April 2004, vol. 8, iss. 2, pp. 155-174(20), Blackwell Publishing Ltd</li> |
| 160 |
<li>Brovelli M. A. and Cannata M., 2004, Digital Terrain model |
| 161 |
reconstruction in urban areas from airborne laser scanning data: the |
| 162 |
method and an example for Pavia (Northern Italy). Computers and |
| 163 |
Geosciences 30, pp.325-331</li> |
| 164 |
<li>Brovelli M. A e Longoni U.M., 2003, Software per il filtraggio di |
| 165 |
dati LIDAR, Rivista dell'Agenzia del Territorio, n. 3-2003, pp. 11-22 |
| 166 |
(ISSN 1593-2192)</li> |
| 167 |
<li>Antolin R. and Brovelli M.A., 2007, LiDAR data Filtering with GRASS GIS for the |
| 168 |
Determination of Digital Terrain Models. Proceedings of Jornadas de SIG Libre, |
| 169 |
Girona, España. CD ISBN: 978-84-690-3886-9</li> |
| 170 |
</ul> |
| 171 |
|
| 172 |
<h2>SEE ALSO</h2> |
| 173 |
|
| 174 |
<em> |
| 175 |
<a href="v.surf.idw.html">v.surf.idw</a>, |
| 176 |
<a href="v.surf.rst.html">v.surf.rst</a> |
| 177 |
</em> |
| 178 |
|
| 179 |
<p> |
| 180 |
Overview: <a href="https://grasswiki.osgeo.org/wiki/Interpolation">Interpolation and Resampling</a> in GRASS |
| 181 |
|
| 182 |
<h2>AUTHORS</h2> |
| 183 |
|
| 184 |
Original version (s.bspline.reg) in GRASS 5.4: |
| 185 |
Maria Antonia Brovelli, Massimiliano Cannata, Ulisse Longoni, Mirko Reguzzoni<br> |
| 186 |
Update for GRASS 6 and improvements: Roberto Antolin |