Virtual Gridding Operation

On this page, you will find general help for the ProRaster product family, including links to documentation, instructional videos, and training videos.
Previous Topic: Point Import Operation
Next Topic: Tensioned Minimum Curvature
Back to: ProRaster Scientific Help
Back to: ProRaster Help
Back to: ProRaster
Go to: ProRaster Essential help page, ProRaster Premium help page, ProRaster Scientific help page.
The ProRaster User Guide is available for download as a PDF.

The Virtual Gridding Operation creates a virtual raster (MVR) that implements a virtual gridding operation.

Like any other virtual raster, raster cell values are not computed until you consume the raster by either rendering it or using it in some other processing operation.

To start gridding, select “Virtual Gridding > Virtual Grid…” on the Point Processing menu button.

The operation is designed to operate on 2D LiDAR point clouds as well as general 2D multi-banded point data. You must supply point cloud data in an MRP file format and it follows that you must first master and execute the Point Import operations for LiDAR and Multi-banded point clouds.

If you want to grid multiple MRP files at once, use the Raster Source Editor to create a point raster source containing all the MRP files. All MRP files must be the same flavour and have the same band structure. Virtual gridding supports multi-banded point clouds and well as LiDAR pulse-return point clouds. In the latter case, you can filter pulses to acquire specific returns using a LAS Query.

The virtual gridding operation is repeatable, and editable. The MVR file that is produced by the operation will thereafter be listed on the Virtual Gridding menu. To edit an existing MVR, select the MVR in the menu to reopen the property dialog populated with the settings for that virtual gridding operation. If the MVR is already being displayed it will be updated with any changes you make, allowing you to make iterative edits to evaluate changes to gridding properties.

This operation is complicated and resource intensive! Keep the documentation to hand and, if you don’t already have an AMD Threadripper based data processing workstation, go get one!

Getting started

To get started and to set the minimum number of properties –

Define the input MRP file and output MVR file.
Turn on “Clip to the bounding polygon union of the input MRP files” and turn off all other clipping options.
Select the “Triangulation with Linear Interpolation” gridding method and set the Pad Size to Small.
Choose the band to grid. If the MRP contains LiDAR data choose the “Z” band (and optionally select the “Ground” LAS Query).
Hit OK and the MVR will display in a new map window.

Tiled Gridding

The virtual gridding operation generates the grid tile by tile as it is needed. Each tile is populated independently of other tiles (except when feathering), drawing from a subset of the point cloud data. Multithreading is employed at tile scope – each tile is generated and populated by a thread working in parallel.

To grid a tile, the processing thread gathers the point data that intersects that tile. When we pad a tile, the size of the tile is expanded so that points outside of the tile, lying in the pad region, are also loaded. By using points outside the tile, we can help the system generate grid data that matches adjacent tiles better along the tile edges. The larger the pad is, the better the edges match. However, the system must grid both the pad and the tile, and this adds processing overhead that increases as the pad size increases.

With all gridding methods, you will see artefacts and effects from this tile-based approach. When using triangulation, you may see odd triangles near the edges of the dataset, and it may not generate a perfect Delauney triangulation along the tile edges. Tile edge effects are particularly noticeable for minimum curvature because it generates very smooth grids and any change in smoothness is very noticeable. To counter this, feathering is used to blend tiles at the edges.

The gridding operation takes notice of the resolution level. It grids the base resolution level, and it grids overview resolution levels, but it does not grid underview resolution levels. Underview resolution levels are generated by interpolation from the base resolution level (just like a standard raster). The interpolation method will be determined by how you are consuming the grid. If you are rendering the grid, for example, then you can control the interpolation from the rendering algorithm property pages.

Gridding methods

You can select from three gridding methods.

“Nearest Point to Cell Assignment” is a Nearest Neighbour method that you will only use is specific circumstances. It is the only method that will generate RGB grids from red, green, and blue input data bands.

“Triangulation with Linear Interpolation” is a Delauney Triangulation method that is useful for LiDAR data and as a general workhorse for data exploration.

“Tensioned Minimum Curvature” is a more advanced technique that produces smoothly interpolated grids. It is used to grid smoothly varying phenomena and commonly used in geophysical applications to grid potential fields like the local magnetic field of the Earth. The “Automated Trend Reinforcement” option applies real-time trend analysis to reinforce local trends and improve trend continuity.

Padding is available for all methods, and feathering of adjacent tiles is available for minimum curvature.

Input – Output

For first time use, open the “Virtual Gridding” dialog by hitting “Virtual Gridding > Virtual Grid…” on the Point Processing menu button. Specify the MRP inputs, MVR output, and all gridding properties. Hit OK to save the MVR and to display it in ProRaster. If all goes well, you will see your point cloud gridded and rendered in real time. As you zoom and pan, the MVR will respond by generating different tiles of the grid at different scales.

You can then edit the gridding operation to modify the grid in real time. The MVR you first produced will be saved to the Virtual Gridding most-recently used list and will be listed on the Virtual Gridding menu. Select the MVR, and the gridding dialog will open, populated with all your grid settings. Make any changes you want and hit OK again. If the MVR is currently being displayed, it will automatically be updated.

If your MVR file is not visible on the menu button list, you can still edit a virtual gridding raster. Just drag & drop the MVR file onto ProRaster Scientific and the gridding dialog will open reflecting the properties recorded in the MVR file. If you just want to display the MVR and not edit, simply hit the Cancel button on the dialog and the MVR will be displayed unmodified.

As input, either select or browse to an MRP file, or choose a raster source that contains one or more MRP files of the same flavour, and band structure. The MRP files do not have to have the same coordinate system as the gridding operation will reproject all coordinates on-the-fly as required.

Clipping

The clipping options give you three methods that clip the grid spatially, and one method to filter input points.

You can enable none, one, or any number of these methods simultaneously. To pass the clipping tests, all enabled methods must return a valid result.

Clip to the bounding polygon union of the input MRP tiles

The MRP datasets contain tiles only when points are in those tiles. Consequently, the complex polygon that represents the union of those tiles can be used to clip the virtual grid. It is cheap to acquire the polygon and is a good place to start your clipping experimentation.

Clip to a polygon

The clipping gold standard is to generate a detailed, complex polygon and to clip the gridding to that polygon. Select a polygon from the recently used list or browse to a polygon file. Polygon files must be in MapInfo Pro table format. All polygons found in the table will be used to clip the gridding.

Clip by distance

This test will clip an output grid cell if the distance to the nearest input data point is greater than the minimum distance. It works slightly differently based on the gridding method. For Nearest Neighbour, it is a precise test. For Triangulation, it is used to clip long thin triangles that exceed the specified distance. For Minimum Curvature the distance test is approximate and not precise.

This clipping method is scale dependent. When you are zoomed in and looking at the grid at its highest resolution, the minimum distance parameter will be used. As you zoom out, the distance must also be increased otherwise all cells would quickly be clipped. The distance in steadily increased until it reaches the maximum distance parameter. As you zoom right out even this distance may be overridden by the system to try to ensure that content is not all clipped away.

If you do not see any content from a virtual gridding operation, the usual culprit is distance-based clipping. Turn it off to verify that this is causing the issue.

Clip by filtering

This method filters the input data stream to remove unwanted points. Choose the data band on which to filter, then select a Data Conditioning Filter that will operate on the input data band. If the input data value fails the Data Conditioning test, then the point will be rejected. Use the Data Conditioning Editor to define your filter in advance.

For example, if you are gridding an aeromagnetic survey you may wish to grid only the primary survey lines and reject all the secondary tie lines. You might achieve this by filtering on the Line data band. In the Data Conditioning Editor, you would create a filter that will reject line values over a particular range (assuming the survey and tie lines are numbered quite differently, which they always are).

Point Data Source

In this section you will choose the data band that you want to grid. Select a data band from the drop list. The output grid will always be a 32-bit floating point value.

If you are gridding LiDAR data and you are using the Nearest Neighbour gridding method, then you can choose to output an RGB color. In this case, you need to select the Red, Green, and Blue bands.

If you are gridding LiDAR data, then you can select a LAS Query to filter the pulse-return data to select returns that meet a set of criteria. ProRaster ships with a set of standard Las Queries and you can use the LAS Query Editor to design your own queries.

Resolution and Detail

The virtual grid has a base resolution cell size that will match the cell size of the first MRP point cloud file used as input. The default cell size will be ¼ the average separation between points (assuming you have designed the MRP so that the average point separation is double the MRP cell size).

You can adjust this up and down by powers of 2 by changing the “Grid Cell Size Adjustment”. Adjust the value up to increase the cell size. Adjust the value down to decrease the cell size. The new cell size is shown on-screen as you make changes. The default grid cell size adjustment of -1 (2^-1= ½) reduces the cell size to ¼ the average point separation.

If you are gridding using minimum curvature, you will want to make sure that the grid cell size is well designed for the station or line spacing. For the other methods the grid cell size is not critical, but you can force the operation to grid at higher or lower resolution, if you wish. If you want to generate a hardcopy of the grid via the Export Raster operation, you may also want to carefully consider the grid cell size.

You can also explicitly define the cell size. If you are using tensioned minimum curvature gridding, you may find this option useful. Check the “Fixed cell size” button and then enter the X and Y cell size as either decimal values or expressed as fractions (integer numerator, integer denominator). You can also exercise control over the position of cells. By default, cell edges align with zero. Check the “Origin aligns cell centre” button to move cell boundaries half a cell south-west so that cell centres align with zero. When gridding line-based data where the line spacing is an integer number of cells, you may find it better to align cell centres along lines.

You can define the tile size of the virtual grid as either Tiny (256×256), Small (512×512), Normal (1024×1024), or Large (2048×2048). If your PC CPU has many cores-threads, then you may find it advantageous to decrease the tile size. Each tile will be generated by a single thread, so as you decrease the tile size you increase the number of tiles being generated and employ more available threads, improving gridding performance. Against this, more tiles means more tile boundaries that need to be feathered, and this may degrade grid quality. Bear in mind that the pad size is a fraction of the tile size, so as you decrease the tile size the pad size also decreases.

For base resolution level (0) and all overview levels (1+), gridding will be performed and point cloud data will be acquired from the MRP files. This is based on the cell size of the MRP files and the system will compute which resolution level in the MRP to target. You may want to change raise or lower this to decrease or increase the number of points used in the gridding. The “Point Source Resolution Level Adjustment” modifies the resolution level that the points are acquired from. Adjust the value up to reduce the number of points and adjust the value down to increase the number of points. Take it one step at a time as each step changes the number of points loaded by a factor of four.

As you change either of these resolution adjustments, the system will respond by reporting what point resolution level will be accessed at base resolution (if there are multiple MRP files it may report a range of resolution levels). Usually, you would aim to use points from level 0 when the grid is at base resolution. If the point level is greater than zero, then you will not be consuming all the available point data. If it is less than zero, then you may be consuming more points than optimal at overview levels.

Padding and Feathering

In all methods, you can use padding. Tensioned Minimum Curvature also allows feathering. These options are always turned on by default.

Padding expands the tile size by a fixed percentage, which increases the number of points fed into the gridding algorithm. Increasing the size of the pad will impact performance, particularly when using minimum curvature. The pay-off is a higher quality grid with fewer tile edge effects.

Feathering is employed when using minimum curvature. In this case, the entire tile and pad region is gridded and then the tile data is finalised by merging data from the pads of adjacent tiles (one along the edges, up to three at the corners). The final tile data is a weighted average of the contributions, and the weight is based on distance. The Taper Order parameter controls how quickly the weight is attenuated by distance.

Nearest Point to Cell Assignment

This is a Nearest Neighbour gridding method that does not interpolate data values. Put simply, it finds the nearest input data point to the centre of each cell and assigns the value of that data point to the cell. This will naturally generate a Voronoi Polygon representation of the data.

This method has no other parameters, but you will certainly want to employ clipping methods to trim the output grid. The cell size does not affect the efficacy of the method.

Triangulation with Linear Interpolation

This method generates a Delauney Triangulation of the input points. Each cell is then populated by finding the triangle that it lies in and performing a linear interpolation from the data values at the corners of the triangle.

This method has no other parameters and, if not clipped, may generate a convex hull (tiling may prevent this). As interpolation is linear, it will not overshoot or undershoot the point values. The cell size does not affect the efficacy of the method.

Tensioned Minimum Curvature

This method generates a smooth surface by interpolation. It aims to generate a surface with continuous second derivatives and minimal total squared curvature.

When employing the Tensioned Minimum Curvature method, you can activate Automated Trend Reinforcement. This is a multi-scale technique that identifies dominant local trends in the source data and then biases the minimum curvature method to reinforce the trend in that direction.