Positional Accuracy

From Wikicap - European Commission

Go up to the main MLL page

In GIS, among other technical issues, positional accuracy of the coordinates is perhaps the most important as it represent the "raison d’etre" of the GIS. The topic covers concerns for absolute and relative accuracy, error, uncertainty, scale, resolution and precision and affects the ways in which data can be used and interpreted. As digital geospatial data always represents an abstraction and approximation of the reality, there is a certain degree of inherent inaccuracy with respect to the real world. This also stems from the fact that when reality is modelled into a digital representation, there is always a process converting the continuous values that characterize any world phenomenon into the discrete set of values used by the computer systems (quantization). The adverse effects of approximation and quantization can be reduced by information systems with a higher numerical level of detail (see Table 2). So the question always is: how well can digital spatial data represent the real world and how well does the tools or sensors compute or capture (measure) the true values?

For LPIS, minimum accuracy requirement is defined in Article 70. of Regulation (EU) 1306/2013 as at least equivalent to that of cartography at a scale of 1:10.000 and, as from 2016, at a scale of 1:5.000. This translates into:

  • a horizontal absolute positional accuracy expressed as RMSE of 1,25m (5.000 x 0,25mm = 1,25m), or the equivalent CE95 value,
  • display range and feature type content compatible with a map with a scale 1:5.000 (i.e. topographic maps rather than urban survey maps),
  • using orthoimagery <= 0,5m GSD.

The values specified above are also the basis of the the Geospatial Positioning Accuracy Standards of the US Federal Geographic Data Committee (from 1999), where the planimetric (X or Y) accuracy for a mapping scale of 1:5.000, expressed in RMSE is 1,25 meters. In GIS environments, measuring length and surface (area) depends on the geometry coordinates inside a system setting. It is therefore important to work in a unique and common coordinate reference system for the national or regional coverages.

Coordinate reference system

To be compatible, spatial data sets must be defined in or transformed into a common spatial canvas: the CRS (coordinate reference system). Each national mapping agency will have defined its own CRS for an optimal cartography of the coordinates, but any coordinates can be transformed from one CRS to another if both CRS are properly defined. For this, the CRS contains a geodetic datum (a reference to the earth’s globe) and a coordinate projection system (transforming the idealised globe to the mapping plane). Improper CRS processing will cause problems when data are integrated.

Shifts

Shifts usually occur as an improper implementation of the unique CRS by one or more data sets. In the figure below there is a systematic displacement between reference parcel layer and orthophoto. It is visible that the selected reference parcel is shifted eastwards and a bit northwards (as indicated by red arrow). Shape and area of the parcel is not different than the feature on the orthophoto image it represents. This kind of error is systematic and, when identified, easy to adjust by re-projecting and translating the affected data set(s).

Typicalshift2.jpg
Figure 10: A typical shift between reference parcel layer and orthophoto caused by different use of CRS

A non-systematic shift is more problematic when it occurs within a system. It is usually a legacy issue within the data sets that were converted from old paper maps into the digital format. Some parts within the coverage had a different coordinate system adjusted to the local needs in the past. So, in the process of merging all areas and creating a single seamless digital spatial coverage, some of the feature boundaries are not following the same line on the newest imagery. This kind of error makes a risky quantification of the agricultural areas within the parcels, and have to be addressed on a case by case basis by the responsible authorities.

NonSystematicShift2.jpg
Figure 11: A non-systematic shift between reference parcel layer and orthophoto

The figure above is showing a non-systematic shift. Some of the parcel boundaries are perfectly following the lines of the hedges, while others are displaced unevenly. Spatial discrepancies are up to 15 meters. A localised, parcel based correction is not a solution, as it would not be applied in the same direction for all surrounding parcels. Some parcels would grow into their neighbour. In such a case, a complete revision in a form of mapping adjustment of the data set should be carried out.

Edge-matching

Edge matching is a spatial processing procedure of connecting vector data digitized from separate paper map sheets. Each paper map sheet has been subject to an individual stretch and shrink and separate maintenance processes. Objects on the edge of the map sheet are usually cut in two parts due to the map sheet margin. To assure the continuity over neighbouring map sheets, the object must share common coordinates on both map sheets. In GIS, this operation is called merging within the same feature class (and the same object ID number). Edge-matching is the process to determine which edges (lines) should be linked among candidates. Another common case to encounter happens when it comes to joining different administrative units. The boundaries of the neighbouring administrative units might not coincide. Administrative boundaries are often of a scale much smaller than 1:5.000. Usually, the land inventory is made in such a way that inside of one administrative region, parcels have a homogeneous geography. But along the administrative border, neighbouring parcel might be overlapping. On the figure below there is an example of administrative boundary that doesn’t follow the true physical boundaries of the parcels. Note that yellow and red lines are slightly displaced, hence forming gaps and slivers between neighbouring parcels if not topologically fixed.

Mismatch.jpg
Figure 12: A mismatch of administrative boundaries between two bordering municipalities

The geographic homogeneity requirement for the data sets should not be seen too restricted as long as the minimum quality is fulfilled. Best practice to deal with the spatial merger is:

  • define homogenous patches,
  • bring into target CRS,
  • perform edge matching where possible,
  • resurvey where needed.

An overall solution for dealing with spatial displacement, regardless of its nature, might be to subdivide the system into smaller homogeneous parts and go from a national to regional based system. The decision should be followed by a careful analysis of all parameters to guarantee consistency in IACS.

Units and precision

With reference to Article 33(1) of Regulation (EU) 1307/2013 farmer applications shall contain declarations of the parcels corresponding to the eligible hectares accompanying payment entitlement. Hence, hectares are the basic units needed in the system for the payment. Declarations require an area precision of 0,01 hectare. System shall operate on the basis of 1:5.000 map scale, and there are many constraints for the maximum widths for the landscape features and EFA elements expressed in meters. So for technical purposes, it is recommended to express all area values in the system in the square meters with at least 2 decimal places, or hectares with 6 decimal places. All length and distances should be expressed in meters with 2 decimal places. As a rule of thumb, to exclude rounding issues, calculations and operations will be done with a 2 digits higher decimal precision compared to the same recorded in the system (see Table below).

m2 Ha m
Parcel area stored in the system 51048,25 5,104825
Parcel area displayed on the screen 51048 5,1048
Length of a feature stored in the system 7802,48
Length of a feature displayed on the screen 7802


Table 2: Example of unit precision of areas and lengths in IACS

Minimum mapping units

In technical terms, the minimum mapping unit (MMU) at a given mapping scale is the size (expressed in area units) below which a feature cannot be reliably identified and reasonably represented by a polygon, or its individual area properly quantified. In the given context, it refers to the size of the smallest feature captured in a vector mode from a raster background. Since LPIS uses 0,5 m resolution orthophotos for all spatial IACS processes (LPIS update, LPIS QA, CAPI in OTSC), setting up the minimum mapping unit in the data capturing process is of key importance. Digitisation in vector mode will be in the function of the display scale of the source orthophoto map. In the figure below there is an example of a pond present on an infrared 0,5 m resolution orthoimagery. Using the GIS tool, the pond is zoomed in and out at a different map scale on the map viewer (screen) for the purpose of photointerpretation and a possible mapping.

PondFound.jpg
Figure 13: A pond found on an orthoimage represented on the screen at a three different display scales

At a zoom scale of 1: 5.000, the feature is barely visible, but at least its presence can be detected. The zoom at scale 1:1.000 provides sufficient information on the nature of the feature in order to be at least identified as a pond. However, quantification of its area through mapping as polygon is not possible. The zoom at scale 1: 100 makes the feature bigger enough on screen to allow digitization. However, due to the size of the pixel, the pixelisation (quantization) effect prevents the reliable representation of the features as polygon, and consequently correct area calculation. Since neither the identification of the nature of the feature, nor its area quantification is possible at the scale of 1: 5.000, this feature is considered below the MMU and thus should not be represented in the LPIS as individual mapped feature. The example above clearly shows the link between the mapping scale, the spatial resolution (pixel size) and the MMU. In order to be considered suitable for a giving cartographic scale, the pixel size of the orthoimagery should allow both a proper identification of the feature type and correct area quantification. Opposite is also valid: for an image with a given pixel size, the MMU is the size of the smallest object of interest that can be both identified and quantified. As a consequence it can be said, that the size of the minimum mapping unit is a function largely of:

  • the cartographic scale of the product – defining the minimum size of the mappable feature
  • the purpose of the product – defining the features of interest the dataset should contain

For example, MMUs of topographic maps and wetland maps at the same cartographic scale might not be the same. Even within a dataset (for example, land cover map) there might be different MMUs for different land cover classes. There are currently two different functional minimum mapping units in operation in IACS. The 2004 LPIS creation guidance allowed for an MMU on ineligible inclusions up to 0,1 ha or 1.000 m2 for calculating the MEA. This provision directly applicable to all production blocks, whose exterior perimeters were expected to follow agricultural land borders. There was no MMU at production block perimeters. The OTSC specifications required individual accountability for all combined ineligible features larger than 0,01 ha or 100 m2, resulting in an area measured and area determined. To have an idea how large is the 100 m2 compared to previous example with a pond, there is a 10 x 10 meters box drawn (or 20 x 20 pixels; 1 pixel = 0,5 m) around the pond in the figure below.

Tenmeters2.jpg
Figure 14: 10x10 meter box around the pond (zoomed at ~ 1:500)

The following table provides a suggestion of the smallest mappable objects, which could be integrated in a 1:10.000 and 1:5.000 coverage as derived from relevant orthophoto, without risk of exaggerating their area or displacing their boundaries.

1 m pixel size orthophoto 0,5 m pixel size orthophoto
Area Linear object Area Linear object
In the field 0,1 ha 10 m width 0,03 5 m width
On the map 3x3 mm 1 mm 3x4 mm 1 mm


Table 3: indicative size of the smallest mappable objects

The table given above, provide information the size of the identifiable objects in relation to the pixel size. Taking also into account some previous studies of JRC, it can be concluded that:

  • to detect the presence of a feature, it must be at least 3 x pixel size,
  • to identify the nature of a feature, it must be at least 20 x pixel size,
  • to map (measure the area of) a feature, it must be 35 x pixel size.

Thus, OTSC can assess presence of a feature with much lower resolution than was required for the original capture in the LPIS. It should be noted also that the nature of the object determines the resolution required.

Minimum parcel sizes

The Regulations and DG Agri guidance specify minimum area dimensions for particular elements that have to be respected in any case.

Error propagation and final accuracy

Spatial overlay

Information from two different spatial data sets are comparable and can be used together for spatial analysis if their specification is well-know and if their quality (spatial accuracy/precision, temporal accuracy) is well recorded. Quality issues present in the sources layers will certainly affect any results from their integration. Often the result can be of even of worse quality than the worst of the two. Although GIS provides the technology for overlaying any different data sets, as far it respect certain format, combination inconsistencies and errors can occur and jeopardize a meaningful and final information of the result. Different datasets sharing same specification with respect to information content and spatial accuracy can be considered compatible for their integration into any spatial data processing without specific considerations. Otherwise, a careful analysis must be carried out in advance to assess at what extent the datasets can be integrated and what the impact on the final product will be.

Data quality

Error is introduced at almost every step of LPIS database creation. For example, the data may contain measurement inaccuracies, if an inappropriate map source was used. Measurement inaccuracies can occur during data capture, i.e. digitising data from a printed map (very often in conversion of cadastre sheets into the digital format), or errors arising from measurement instruments that is not validated. Further errors may be introduced during data processing or even using inappropriate tools for data analysis. In this paragraph the focus is only on positional error propagation in LPIS, since other error types (related to data content) are a matter of concern in LPIS QA ETS guidance. The usual accumulation of errors in the process of creating the LPIS or derived information is adding up the following:

  • data acquisition error,
  • data processing error,
  • data conversion error,
  • data analysis error.

Testing of the final LPIS (positional) accuracy is possible with an independent source of higher accuracy. It can be done on a map of a larger scale and with the use of GNNS and survey data. An estimation of overall accuracy can be achieved by summing up the squares of the computed accuracy of each individual component and taking the square root of the sum. In short, the necessary quality of any spatial data can and should be measured by such tests and computations. It should be documented in the metadata of the data set and consulted before every attempted use.

Interaction of spatial datasets with different specifications

Use of data sets shall be homogeneous with reference to the Regulation. Maps, land registry documents or other cartographic references, as well as aerial or satellite imagery have to be homogeneous within the set in order to assure a required level of accuracy. This doesn’t mean that all data sets need to be at the same accuracy level, as long as the combined and resulting information meets the minimum requirement. I.e. some lower resolution imagery could be used to identify the crop pattern (to assign the true crop class), but the actual measurement of the area for that class should be done against the data set that provide the required minimum accuracy. Some third party spatial data sets could have an added value in identifying issues on the parcel level. In such cases and if the level of accuracy isn’t the same, no high level nor accurate positioning or matching of feature boundaries can be carried out between two data sets. The same feature represented on two or more map scales can look slightly different due to simplification and reduction of feature shape complexity by the so called cartographic generalization process. In this process a map operator moves the simplified shape of the features’ geometry in order to retain presence of the feature at the expense of the features’ true position. In certain cases, the semantic relationship between the thematic information stored in the third party spatial dataset (wetland map, protected areas, and national land cover information) and the LPIS, needs to be assessed prior to any integration. This is especially valid when the difference in the thematic scope between the datasets is of such magnitude that they might be hardly related or compatible. In the figure below is the example of “Natura2000” zone overlaid with reference parcel layer and orthophoto imagery at the 1:5.000 scale. Since “Natura2000” data has been digitized from a lower precision map 1:100.000 (see generalized lines along the road), one cannot simply do the cutting to exclude the small southern part of the reference parcel concluding it is not within the protected zone. Hence, it is used only to identify the whole parcel within the protected zone (see that the majority of the land belonging to RP is within the zone).

Use of data.jpg
Figure 15: use of two data sets with a different level of accuracy for identification purposes

It is important to point out that determination whether a reference parcel belongs fully inside of the protected area should be a result of a visual interpretation on a case by case basis. It might happen that a big block representing a RP is only partly within the protected zone of “Natura 2000”. In principle, authorities should develop a spatial analysis method to come up with the best practice how to combine two data sets with a different precision for the identification purposes. An example of the possible spatial analysis scenario with a fictious overlapping % could be:

  • perform overlap query between the data sets,
  • calculate the geometric area of the RP overlap,
  • select all RP’s where RP overlap is < 20% of the reference area and perform a visual check of each RP to enter a flag if fully inside of “Natura 2000”,
  • select all RP’s where RP overlap is > 20% of the reference area and perform a further delineation of each RP (based on 0,5 m resolution orthoimagery taking into account Natura 2000 zone line) to separate and determine protected RP MEA and unprotected RP MEA.

Spatial interaction between the LPIS layers

The spatial information stored in the LPIS system can be represented by and operated through four “core” layers:

  • layer of the reference parcels (RP), with their unique ID and reference area,
  • layer of the Basic Payment Scheme (BPS) containing the main agriculture land cover types (arable land, permanent crop and permanent grassland),
  • layer of the various Ecological Focus Areas (EFA),
  • layer of Landscape Features (LF) subject to retention according to GAEC standard 7.

The degree of semantic and topologic relationship -overlay rules and spatial hierarchy- between layers is specific for every layer pair and depends on the scope and definition of the feature types represented by each layer. BPS and RP are interdependent as the BPS layer is derived from delineated agricultural land cover within each reference parcel boundary taking into account all necessary exclusions of ineligible features. In some cases (in the farmers block systems and agricultural parcel systems), RP layer and BPS layer will share a single representation, hence only one layer is present in LPIS. EFA layer contains, among others, a subset of the features residing in the LF layer, if such layer exists, thus no features are digitized twice. It should be noted, that for some LPIS systems that are storing the landscape features by alphanumerical recording at RP level, LF layer with individual geometries might not be even feasible to generate. There is no direct topological relation between the EFA and BPS layers, apart from the fact that the presence of certain EFA elements is restricted to the extent of the arable land within the BPS layer. The same is valid for the LF and BPS layers: individual landscape features are not represented in the BPS layer, as they area is dissolved in (or attributed) to the correspondent “BPS polygon”. The Table below aims to provide an overview of the “core” layers from LPIS: their content, basic topological rules applied, as well as the most common interactions between them (see also guidelines on EFA document DSCG/2014/31-FINAL).

Layer Content of the layer Topological rules in the layer A possibility of spatial interactions in function of:
Area calculation at RP level
A possibility of spatial interactions in function of:
Selection and attribution of features to EFA
Reference parcel (RP) Polygon representations of the reference parcels (management unit of agricultural land) No overlap between features is allowed. Controlled gaps are allowed Polygon area - directly from RP polygon N/A
Basic Payment Scheme (BPS)/Single Area Payment Scheme (SAPS) Polygon representations of the maximum eligible agriculture land cover types – arable, permanent crops, and permanent grassland – found within reference parcels No overlap between features is allowed. Controlled gaps are allowed. BPS area (reference area/MEABPS): - through spatial intersection of RP and BPS layers (and the third like ineligible mask) N/A
Ecological Focus Areas (EFA) Polygon/line/point representation of the EFA found on the BPS layer Controlled overlap between features is allowed.- EFA area – through: 1) spatial intersection of EFA layer with BPS and RP layers; 2) spatial join of resulted EFAs within intersection; 3) attribution of the total area of resulted EFAs to the given RP Selection by location of all LFs present on arable land in BPS layer. Copying of geometries and attributes of LFs into EFA layer
Landscape features (LF), if mapped Polygon/line/point representation of the eligible landscape features subject to retention (GAEC 7) found on the agriculture land of the BPS layer Controlled overlap between features is allowed LF area - Through. 1) spatial intersection of LF layer with RP layer; 2) spatial join of resulted LFs within intersection; 3) attribution of the total area of resulted LFs to the given RP N/A


Table 4: basic characteristics of the core LPIS layers

Go up to the main MLL page