Maps and spatial data

Plots › Map (or in the explorer, under the project): the observations on a map. The first time, choose the project's coordinates: two of its variables, X (east, or the longitude) and Y (north, or the latitude), and their coordinate system: longitudes and latitudes in degrees (WGS 84), Lambert-93, the CC zones, Lambert II étendu (NTF), a UTM zone (another EPSG code: 326xx north, 327xx south), LAEA Europe, web Mercator, or plane coordinates in metres of another system (distances stay right, without a background map). Nuée proposes them from the variables' names (lon/lat, X_L93/Y_L93, X/Y...). They are saved with the project, and Ctrl+Z takes them back.

The map draws the points with the same scale on both axes. Colour: a variable of the project, or a column of a fitted model's results (scores, residuals, fitted values, clusters...); Size: a quantitative variable or column; Labels: the observations' names. Background: none by default, so nothing leaves the computer; OpenStreetMap, the IGN map or IGN's aerial photographs (France) are downloaded when chosen, and their servers then receive the area shown. A box, lasso or click selects observations, shared with the project's models as in the other plots.

Conversions between systems are written in Nuée (Lambert, transverse Mercator, LAEA formulas; NTF moved to WGS 84 by IGN's three translations, to about a metre; RGF93 and ETRS89 taken as WGS 84): they match PROJ, the library QGIS uses, to the micrometre.

GeoPackage (.gpkg), the format of QGIS: File › Open reads one of its layers (Nuée asks which when there are several) as a table; points give two coordinate columns (X and Y, or longitude and latitude), lines and polygons their centre (the centroid of a polygon's area), and the project's coordinates are set. File › Export data and a model's Export write a GeoPackage when the project has coordinates: one point per observation with all its columns, a model's other result tables beside it; QGIS opens it with its coordinate system.

Raster values (Edit › Add raster values…, or the button on the map): an elevation, a slope, a land cover, a climate layer… read from raster files at each observation's position, and added to the project as new variables, one per band. Choose one or several files: GeoTIFF (.tif, as QGIS, IGN's BD ALTI and RGE ALTI, Copernicus, CORINE, WorldClim or CHELSA write it) or ESRI ASCII grids (.asc, with their .prj). In the browser version, type each file's full path (several: separated by ";"): the engine reads the files where they are. Before adding, each file shows its size, format, compression, cells and their size, bands and their type, coordinate system, no-data value, and how many observations fall in it; a file that can't be read says why. For each file, choose the value: the cell holding the observation (classes, land cover; proposed for 8-bit integers and palettes), the bilinear interpolation of the four nearest cells (continuous values; proposed for reals), or the mean, or the majority class, of the cells whose centre is within a radius in metres (a tie goes to the class of the observation's own cell, then to the smallest; with no centre within the radius, the observation's own cell). The values are those of terra::extract. An observation without coordinates, outside the raster or on a cell without a value gets a missing value, and the result counts each case. Each band gets a name (its description in the file, or the file's name), which you can change; a name already taken becomes "name (2)". Files side by side with the same bands, type, system and cells (the tiles of a DEM) are proposed as one raster: one variable per band, each observation taking the value of the tile that holds it (a circle takes the cells of every tile). Edit › Undo (Ctrl+Z) removes the variables. They are saved with the project like any other, and remember where they come from: the Methods text of a model that uses them says which files, cell size, system and method gave them. The R console shows the call, ajouter_valeurs_rasters(...), and what it gave.

The files are read in Nuée, without GDAL, and only the strips or tiles that hold observations: a national DEM stays quick (a fraction of a second for a hundred observations in an uncompressed or Deflate file, one to three seconds with LZW, decoded in R). Read: classic TIFF and BigTIFF, in strips or tiles, bands interleaved or separate; uncompressed, LZW, Deflate, PackBits, LZMA and ZSTD (ZSTD with R 4.4 and later), with predictors 2 and 3; integers of 8, 16 and 32 bits, reals of 32 and 64 bits; GDAL's no-data value, scale, offset and band names; the georeferencing (ModelPixelScale and ModelTiepoint, ModelTransformation, or a .tfw file), pixels as areas or as points. The raster's coordinate system comes from its EPSG code, or its parameters, and must be one Nuée converts (WGS 84, ETRS89, RGF93, Lambert-93, CC42 to CC50, Lambert II étendu, UTM, LAEA Europe, web Mercator): another one is refused with its code; reproject the raster in QGIS (Raster › Projections › Warp) or with gdalwarp. A raster without a coordinate system is taken in the project's. Refused, with a message: JPEG, WebP, LERC, JPEG 2000 and CCITT compression, values of 1, 2, 4, 12 or 24 bits, 16-bit reals, complex numbers, YCbCr colours, ground control points (GCPs). Read without their colours: palettes (the class numbers are read). Left aside: overviews (the values come from the full resolution), the other images of a file, masks and .aux.xml files.

Plots › Spatial autocorrelation (from the map, or for a regression from its model): Moran's I of a variable or of a regression's residuals (or, for counts, presence-absence and positive values, its quantile residuals), with its expectation without autocorrelation, z and the one-sided p (positive autocorrelation) under randomisation, and p by permutations (999 by default). For a linear model without random effects, a second line gives the test that accounts for the fit (Cliff and Ord 1981, as spdep::lm.morantest). Neighbourhood: the k nearest neighbours of each observation (made mutual), the observations within a distance (empty: the smallest distance that gives each observation a neighbour), or the inverse of the distance; rows summing to 1 (spdep's "W" style) or the weights as they are. Under it, the correlogram (Moran's I in classes of distance up to half the largest one, two-sided p corrected by Holm over the classes, in red at 0.05) and the semivariogram (half the mean squared difference between the values of the pairs of each class, as gstat::variogram, up to a third of the area's diagonal), with the variance as a dashed line. Distances are along great circles for longitudes and latitudes, straight lines for plane coordinates. Observations at the same place (a plot surveyed several years) are each other's nearest neighbours: "Mean by place" replaces their values by their mean.