The difference between scipy.spatial.cKDTree and pygeos.STRtree #889
Description
In the opendrift/readers/basereader/unstructured.py ，I found the method build_ckdtree and nearest_ckdtree take a long time when running the openoil, so I made the following changes in the method which find the nearest point:
# ----------------------------unstructured.py----------------------------
# def _build_ckdtree_(self, x, y):
# from scipy.spatial import cKDTree
# P = np.vstack((x, y)).T
# return cKDTree(P)
# Replace _build_ckdtree_ with _build_strtree_
def _build_strtree_(self, x, y):
from pygeos import STRtree, points
return STRtree(points(x,y))
#def __nearest_ckdtree__(self, idx, x, y):
# q = np.vstack((x, y)).T
# return idx.query(q, k = 1, workers = self.PARALLEL_WORKERS)[1]
# Replace __nearest_ckdtree__ with __nearest_strtree__
def __nearest_strtree__(self, idx, x, y):
from pygeos import points
return idx.nearest(pygeos.points(x,y))[1]
# line 180
# return self.__nearest_ckdtree__(self.nodes_idx, x, y)
return self.__nearest_strtree__(self.nodes_idx, x, y)
# line 186
# return self.__nearest_ckdtree__(self.nodes_idx, x, y)
return self.__nearest_strtree__(self.nodes_idx, x, y)# ----------------------------reader_netCDF_CF_unstructured.py----------------------------
# line 173
# self.nodes_idx = self._build_ckdtree_(self.x, self.y)
self.nodes_idx = self._build_strtree_(self.x, self.y)
# line 176
# self.nodes_idx = self._build_ckdtree_(self.x, self.y)
self.faces_idx = self._build_strtree_(self.x, self.y)Here I use the current data from FVCOM and the constant wind to drive openoil, so I only modified the reader_netCDF_CF_unstructured.py and unstructured.py, but the graph of tracks is different from before the modification. I wonder if this modification is correct? If it's correct, STRtree may be a better choice than cKDTree, beacauce it can increase the operating speed of opendrift by more than 10%.
