"""Implementations relating to merging of datasets and arrays."""
import numpy as np
import rasterio
from pyproj import CRS
from .raster import full_from_transform
[docs]
def merge(
data_arrays,
dst_crs=None,
dst_bounds=None,
dst_res=None,
align=True,
mask=None,
merge_method="first",
**kwargs,
):
"""Merge multiple tiles to a single DataArray.
If mismatching grid CRS or resolution, tiles are reprojected
to match the output DataArray grid. If no destination
grid is defined it is based on the first DataArray in the list.
Based on :py:meth:`rasterio.merge.merge`.
Arguments
---------
data_arrays: list of xarray.DataArray
Tiles to merge
dst_crs: pyproj.CRS, int
CRS (or EPSG code) of destination grid
dst_bounds: list of float
Bounding box [xmin, ymin, xmax, ymax] of destination grid
dst_res: float
Resolution of destination grid
align : bool, optional
If True, align grid with dst_res
mask: geopands.GeoDataFrame, optional
Mask of destination area of interest.
Used to determine dst_crs, dst_bounds and dst_res if missing.
merge_method: {'first','last','min','max','new'}, callable
Merge method:
* first: reverse painting
* last: paint valid new on top of existing
* min: pixel-wise min of existing and new
* max: pixel-wise max of existing and new
* new: assert no pixel overlap
**kwargs:
Key-word arguments passed to
:py:meth:`~hydromt.raster.RasterDataArray.reproject`
Returns
-------
da_out: xarray.DataArray
Merged tiles.
"""
# define merge method
if merge_method == "first":
def copyto(old_data, new_data, old_nodata, new_nodata):
dmask = np.logical_and(old_nodata, ~new_nodata)
old_data[dmask] = new_data[dmask]
elif merge_method == "last":
def copyto(old_data, new_data, old_nodata, new_nodata):
dmask = ~new_nodata
old_data[dmask] = new_data[dmask]
elif merge_method == "min":
def copyto(old_data, new_data, old_nodata, new_nodata):
dmask = np.logical_and(~old_nodata, ~new_nodata)
old_data[dmask] = np.minimum(old_data[dmask], new_data[dmask])
dmask = np.logical_and(old_nodata, ~new_nodata)
old_data[dmask] = new_data[dmask]
elif merge_method == "max":
def copyto(old_data, new_data, old_nodata, new_nodata):
dmask = np.logical_and(~old_nodata, ~new_nodata)
old_data[dmask] = np.maximum(old_data[dmask], new_data[dmask])
dmask = np.logical_and(old_nodata, ~new_nodata)
old_data[dmask] = new_data[dmask]
elif merge_method == "new":
def copyto(old_data, new_data, old_nodata, new_nodata):
data_overlap = np.logical_and(~old_nodata, ~new_nodata)
assert not np.any(data_overlap)
mask = ~new_nodata
old_data[mask] = new_data[mask]
elif callable(merge_method):
copyto = merge_method
# resolve arguments
if data_arrays[0].ndim != 2:
raise ValueError("Mosaic is only implemented for 2D DataArrays.")
if dst_crs is None and (dst_bounds is not None or dst_res is not None):
raise ValueError("dst_bounds and dst_res not understood without dst_crs.")
idx0 = 0
if mask is not None and (dst_res is None or dst_crs is None):
# select array with largest overlap of valid cells if mask
areas = []
for da in data_arrays:
da_clipped = da.raster.clip_geom(geom=mask, mask=True)
n = da_clipped.raster.mask_nodata().notnull().sum().load().item()
areas.append(n)
idx0 = np.argmax(areas)
# if single layer with overlap of valid cells: return clip
if np.sum(np.array(areas) > 0) == 1:
return data_arrays[idx0].raster.clip_geom(mask)
da0 = data_arrays[idx0] # used for default dst_crs and dst_res
# dst CRS
if dst_crs is None:
dst_crs = da0.raster.crs
else:
dst_crs = CRS.from_user_input(dst_crs)
# dst res
if dst_res is None:
dst_res = da0.raster.res
if isinstance(dst_res, float):
x_res, y_res = dst_res, -dst_res
elif isinstance(dst_res, (tuple, list)) and len(dst_res) == 2:
x_res, y_res = dst_res
else:
raise ValueError("dst_res not understood.")
assert x_res > 0
# TODO test y_res > 0
dst_res = (x_res, -y_res) # NOTE: y_res is multiplied with -1 in rasterio!
# dst bounds
if dst_bounds is None and mask is not None:
w, s, e, n = mask.to_crs(dst_crs).buffer(2 * abs(x_res)).total_bounds
# align with da0 grid
w0, s0, e0, n0 = da0.raster.bounds
w = w0 + np.round((w - w0) / abs(x_res)) * abs(x_res)
n = n0 + np.round((n - n0) / abs(y_res)) * abs(y_res)
e = w + int(round((e - w) / abs(x_res))) * abs(x_res)
s = n - int(round((n - s) / abs(y_res))) * abs(y_res)
align = False # don't align with resolution
elif dst_bounds is None:
for i, da in enumerate(data_arrays):
if i == 0:
w, s, e, n = da.raster.transform_bounds(dst_crs)
else:
w1, s1, e1, n1 = da.raster.transform_bounds(dst_crs)
w = min(w, w1)
s = min(s, s1)
e = max(e, e1)
n = max(n, n1)
else:
w, s, e, n = dst_bounds
# check N>S orientation
top, bottom = (n, s) if y_res < 0 else (s, n)
dst_bounds = w, bottom, e, top
# dst transform
width = int(round((e - w) / abs(x_res)))
height = int(round((n - s) / abs(y_res)))
transform = rasterio.transform.from_bounds(*dst_bounds, width, height)
# align with dst_res
if align:
transform, width, height = rasterio.warp.aligned_target(
transform, width, height, dst_res
)
# creat output array
nodata = da0.raster.nodata
isnan = np.isnan(nodata)
dtype = da0.dtype
da_out = full_from_transform(
transform=transform,
shape=(height, width),
nodata=nodata,
dtype=dtype,
name=da0.name,
attrs=da0.attrs,
crs=dst_crs,
)
ys = da_out.raster.ycoords.values
xs = da_out.raster.xcoords.values
dest = da_out.values
kwargs.update(dst_crs=dst_crs, dst_res=dst_res, align=True)
sf_lst = []
for _, da in enumerate(data_arrays):
if not da.raster.aligned_grid(da_out):
# clip with buffer
src_bbox = da_out.raster.transform_bounds(da.raster.crs)
da = da.raster.clip_bbox(src_bbox, buffer=10)
if np.any([da[dim].size == 0 for dim in da.raster.dims]):
continue # out of bounds
# reproject
da = da.raster.reproject(**kwargs)
# clip to dst_bounds
da = da.raster.clip_bbox(dst_bounds)
if np.any([da[dim].size == 0 for dim in da.raster.dims]):
continue # out of bounds
if "source_file" in da.attrs:
sf_lst.append(da.attrs["source_file"])
# merge overlap
w0, s0, e0, n0 = da.raster.bounds
if y_res < 0:
top = np.where(ys <= n0)[0][0] if n0 < ys[0] else 0
bottom = np.where(ys < s0)[0][0] if s0 > ys[-1] else None
else:
top = np.where(ys > n0)[0][0] if n0 < ys[-1] else 0
bottom = np.where(ys >= s0)[0][0] if s0 > ys[0] else None
left = np.where(xs >= w0)[0][0] if w0 > xs[0] else 0
right = np.where(xs > e0)[0][0] if e0 < xs[-1] else None
y_slice = slice(top, bottom)
x_slice = slice(left, right)
region = dest[y_slice, x_slice]
temp = np.ma.masked_equal(da.values.astype(dtype), nodata)
if isnan:
region_nodata = np.isnan(region)
temp_nodata = np.isnan(temp)
else:
region_nodata = region == nodata
temp_nodata = temp.mask
copyto(region, temp, region_nodata, temp_nodata)
# set attrs
da_out.attrs.update(source_file="; ".join(sf_lst))
da_out.raster.set_crs(dst_crs)
return da_out.raster.reset_spatial_dims_attrs()
