{"id":35,"date":"2020-04-04T07:00:00","date_gmt":"2020-04-04T07:00:00","guid":{"rendered":"http:\/\/wcwp.rf.gd\/2020\/04\/04\/https-landscapearchaeology-org-2020-wgs\/"},"modified":"2020-04-04T07:00:00","modified_gmt":"2020-04-04T07:00:00","slug":"wgs","status":"publish","type":"post","link":"https:\/\/landscapeanalysis.org\/?p=35","title":{"rendered":"Name the enemy: geographic projections in GIS"},"content":{"rendered":"

So, you have your data, you\u2019ve installed a GIS program \u2026 and there we go: slope gradient \u2013 check, solar exposition \u2013 check, water drainage \u2013 check, visibility range \u2013 check \u2026<\/p>\n

But somehow, the outputs don\u2019t seem right: things don\u2019t fall where they should, values go off by hundreds and your maps just look strange.<\/p>\n

Nine out of ten cases when QGIS plugins that I maintain \u201cdo not work\u201d are due to geographic projection errors. Based on this experience, and on what I see around, I would go as far as to estimate that the majority of raster based GIS analyses being done today is flawed<\/strong>. The main reason is the lack of proper projection management. Don\u2019t believe me? Take a look at your own models, after having read this.<\/p>\n

Simply put, modern GIS works with just about anything that can be thrown inside. So, let\u2019s try a visibility analysis with a piece of freely available SRTM elevation model (provided by NASA<\/a>). I\u2019m using SAGA GIS<\/a> software.<\/p>\n

\"viewshed1\"
\nViewshed analysis made over a raw SRTM dataset (observer point is marked by an orange dot).<\/em><\/p>\n

Whoa ! How come the visible range has a shape of an ellipse ? Well, we can make it round, but then the background topography gets unrealistically stretched out.<\/p>\n

\"viewshed2\"<\/p>\n

What is happening here? Let\u2019s take a look at the information associated with our elevation data.<\/p>\n

\"WGS84\"<\/p>\n

We can see that our data is in degrees, and that our pixel size is 0.000-something, which cannot be meters (in that case, a pixel would be smaller than a millimetre!). What does a degree unit look like ? Well, we can only represent it in 3D because it is curved in 3 dimensions: figure below.<\/p>\n

\"grids\"
\nOne angular unit in a latitude\/longitude grid vs one pixel unit in an image grid.<\/em><\/p>\n

Latitude\/longitude grid, such as the one above, can be seen in any GIS manual and is commonly featured on maps. How come, then, that it causes problems in GIS? Well, the truth is that the majority of raster GIS modules borrow their design from image analysis algorithms<\/strong>. These feed on pixel grids and don\u2019t care about geography. The problem with our visibility analysis is that it interpreted elongated, curved units as square pixels.<\/p>\n

What your GIS program sees is in most cases just an image composed of supposedly square pixels. Deformed angular units will totally skew the result. This is valid for almost any raster based analysis.<\/p>\n