Reproject Raster Data with Rioxarray
OverviewTeaching: 60 min
Exercises: 20 minQuestions
How do I work with raster data sets that are in different projections?
Reproject a raster in Python using rasterio.
Accomplish the same task with rioxarray and xarray.
Things You’ll Need To Complete This Episode
See the lesson homepage for detailed information about the software, data, and other prerequisites you will need to work through the examples in this episode.
Sometimes we encounter raster datasets that do not “line up” when plotted or analyzed. Rasters that don’t line up are most often in different Coordinate Reference Systems (CRS), otherwise known as “projections”. This episode explains how to line up rasters in different, known CRSs.
Raster Projection in R
If you loaded two rasters with different projections in QGIS 3 or ArcMap/ArcPro, you’d see that they would align since these software reproject “on-the-fly”. But with R or Python, you’ll need to reproject your data yourself in order to plot or use these rasters together in calculations.
For this episode, we will be working with the Harvard Forest Digital Terrain Model (DTM). This differs from the surface model data we’ve been working with so far in that the digital terrain model (DTM) includes the tops of trees, while the digital surface model (DSM) shows the ground level beneath the tree canopy.
Our goal is to get these data into the same projection with the
rioxarray.reproject() function so that
we can use both rasters to calculate tree canopy height, also called a Canopy Height Model (CHM).
First, we need to read in the DSM and DTM rasters.
Reading in the data with xarray looks similar to using
rasterio directly, but the output is a xarray object called a
DataArray. You can use a
xarray.DataArray in calculations just like a numpy array. Calling the variable name of the
DataArray also prints out all of its metadata information. Geospatial information is not read in if you don’t import rioxarray before calling the
import rioxarray surface_HARV = rioxarray.open_rasterio("data/NEON-DS-Airborne-Remote-Sensing/HARV/DSM/HARV_dsmCrop.tif") terrain_HARV = rioxarray.open_rasterio("data/NEON-DS-Airborne-Remote-Sensing/HARV/DTM/HARV_dtmCrop_WGS84.tif") surface_HARV
<xarray.DataArray (band: 1, y: 1367, x: 1697)> [2319799 values with dtype=float64] Coordinates: * band (band) int64 1 * y (y) float64 4.714e+06 4.714e+06 4.714e+06 ... 4.712e+06 4.712e+06 * x (x) float64 7.315e+05 7.315e+05 7.315e+05 ... 7.331e+05 7.331e+05 Attributes: transform: (1.0, 0.0, 731453.0, 0.0, -1.0, 4713838.0) crs: +init=epsg:32618 res: (1.0, 1.0) is_tiled: 0 nodatavals: (-3.4e+38,) scales: (1.0,) offsets: (0.0,) AREA_OR_POINT: Area
We can use the CRS attribute from one of our datasets to reproject the other dataset so that they are both in the same projection. The only argument that is required is the
dst_crs argument, which takes the CRS of the result of the reprojection.
terrain_HARV_UTM18 = terrain_HARV.rio.reproject(dst_crs=surface_HARV.rio.crs) terrain_HARV_UTM18
<xarray.DataArray (band: 1, y: 1493, x: 1796)> array([[[-9999., -9999., -9999., ..., -9999., -9999., -9999.], [-9999., -9999., -9999., ..., -9999., -9999., -9999.], [-9999., -9999., -9999., ..., -9999., -9999., -9999.], ..., [-9999., -9999., -9999., ..., -9999., -9999., -9999.], [-9999., -9999., -9999., ..., -9999., -9999., -9999.], [-9999., -9999., -9999., ..., -9999., -9999., -9999.]]]) Coordinates: * x (x) float64 7.314e+05 7.314e+05 ... 7.332e+05 7.332e+05 * y (y) float64 4.714e+06 4.714e+06 ... 4.712e+06 4.712e+06 * band (band) int64 1 spatial_ref int64 0 Attributes: transform: (1.001061424448915, 0.0, 731402.3156760389, 0.0, -1.00106... scales: (1.0,) offsets: (0.0,) AREA_OR_POINT: Area _FillValue: -9999.0 grid_mapping: spatial_ref
You might wonder why the result of
-9999at the edges whereas when we read in the data,
surface_HARVdid not show the
-9999values. This is because xarray by default will wait until the last necessary moment before actually running the computations on an xarray DataArray. This form of evaluation is called lazy, as opposed to eager, where functions are always computed when they are called. If you ever want a lazy DataArray to reveal it’s underlying values, you can use the
rioxarraywill only show the values in the corners of the array.
<xarray.DataArray (band: 1, y: 1367, x: 1697)> array([[[408.76998901, 408.22998047, 406.52999878, ..., 345.05999756, 345.13998413, 344.97000122], [407.04998779, 406.61999512, 404.97998047, ..., 345.20999146, 344.97000122, 345.13998413], [407.05999756, 406.02999878, 403.54998779, ..., 345.07000732, 345.08999634, 345.17999268], ..., [367.91000366, 370.19000244, 370.58999634, ..., 311.38998413, 310.44998169, 309.38998413], [370.75997925, 371.50997925, 363.41000366, ..., 314.70999146, 309.25 , 312.01998901], [369.95999146, 372.6000061 , 372.42999268, ..., 316.38998413, 309.86999512, 311.20999146]]]) Coordinates: * band (band) int64 1 * y (y) float64 4.714e+06 4.714e+06 4.714e+06 ... 4.712e+06 4.712e+06 * x (x) float64 7.315e+05 7.315e+05 7.315e+05 ... 7.331e+05 7.331e+05 Attributes: transform: (1.0, 0.0, 731453.0, 0.0, -1.0, 4713838.0) crs: +init=epsg:32618 res: (1.0, 1.0) is_tiled: 0 nodatavals: (-3.4e+38,) scales: (1.0,) offsets: (0.0,) AREA_OR_POINT: Area
And we can also save our DataArray that we created with
rioxarray to a file.
reprojected_path = "data/NEON-DS-Airborne-Remote-Sensing/HARV/DTM/HARV_dtmCrop_UTM18.tif" terrain_HARV_UTM18.rio.to_raster(reprojected_path)
Inspect the metadata for
surface_HARV. Are the projections the same? What metadata attributes are different? How might this affect calculations we make between arrays?
# view crs for DTM print(terrain_HARV_UTM18.rio.crs) # view crs for DSM print(surface_HARV.rio.crs)
Good, the CRSs are the same. But …
# view noddata value for DTM print(terrain_HARV_UTM18.rio.nodata) # view nodata value for DSM print(surface_HARV.rio.nodata)
The nodata values are different. Before we plot or calculate both of these DataArrays together, we should make sure they have the same nodata value. Furthermore …
# view shape for DTM print(terrain_HARV_UTM18.shape) # view shape for DSM print(surface_HARV.shape)
(1, 1492, 1801) (1, 1367, 1697)
The shapes are not the same which means these data cover slightly different extents and locations. In the next episode we will need to align these DataArrays before running any calculations.
rioxarrayprovides functionality to align multiple geospatial DataArrays.
Let’s plot our handiwork so far! We can use the
xarray.DataArray.plot function to show the DTM. But if we run the following code, something doesn’t look right …
import matplotlib.pyplot as plt plt.figure() terrain_HARV_UTM18.plot(cmap="viridis") plt.title("Harvard Forest Digital Terrain Model")
Whoops! What did we forget to do to the DTM DataArray before plotting?
Our array has a
-9999.0, which causes the color of our plot to be stretched over too wide a range. We’d like to only display valid values, so before plotting we can filter out the nodata values using the
where()function and the
.rio.nodataattribute of our DataArray.
terrain_HARV_UTM18_valid = terrain_HARV_UTM18.where( terrain_HARV_UTM18 != terrain_HARV_UTM18.rio.nodata) plt.figure() terrain_HARV_UTM18_valid.plot(cmap="viridis") plt.title("Harvard Forest Digital Terrain Model")
If we had saved
terrain_HARV_UTM18to a file and then read it in with
masked=Trueargument the raster’s
nodatavalue would be masked and we would not need to use the
where()function to do the masking before plotting.
Challenge: Reproject, then Plot a Digital Terrain Model
Create 2 maps in a UTM projection of the San Joaquin Experimental Range field site, using the
matplotlib.pyplot(to add a title). Reproject the data as necessary to make sure each map is in the same UTM projection and save the reprojected file with the file name “data/NEON-DS-Airborne-Remote-Sensing/SJER/DSM/SJER_dsmCrop_WGS84.tif”.
If we read in these files with the argument
masked=True, then the nodata values will be masked automatically and set to
numpy.nan, or Not a Number. This can make plotting easier since only valid raster values will be shown. However, it’s important to remember that
numpy.nanvalues still take up space in our raster just like
nodatavalues, and thus they still affect the shape of the raster. Rasters need to be the same shape for raster math to work in python. In the next lesson, we will examine how to prepare rasters of different shapes for calculations.
terrain_HARV_SJER = rioxarray.open_rasterio("data/NEON-DS-Airborne-Remote-Sensing/SJER/DTM/SJER_dtmCrop.tif", masked=True) surface_HARV_SJER = rioxarray.open_rasterio("data/NEON-DS-Airborne-Remote-Sensing/SJER/DSM/SJER_dsmCrop_WGS84.tif", masked=True) reprojected_surface_model = surface_HARV_SJER.rio.reproject(dst_crs=terrain_HARV_SJER.rio.crs) plt.figure() reprojected_surface_model.plot() plt.title("SJER Reprojected Surface Model") reprojected_surface_model.rio.to_raster("data/NEON-DS-Airborne-Remote-Sensing/SJER/DSM/SJER_dsmCrop_WGS84.tif") plt.figure() terrain_HARV_SJER.plot() plt.title("SJER Terrain Model")
In order to plot or do calculations with two raster data sets, they must be in the same CRS.
rioxarray and xarray provide simple syntax for accomplishing fundamental geospatial operations.
rioxarray is built on top of rasterio, and you can use rasterio directly to accomplish fundamental geospatial operations.