update
parent
fa41272395
commit
1b0ca75d66
@ -0,0 +1,232 @@
|
|||||||
|
"""This module contains functions to analyze the shoreline data along transects'
|
||||||
|
|
||||||
|
Author: Kilian Vos, Water Research Laboratory, University of New South Wales
|
||||||
|
"""
|
||||||
|
|
||||||
|
# load modules
|
||||||
|
import os
|
||||||
|
import numpy as np
|
||||||
|
import matplotlib.pyplot as plt
|
||||||
|
import pdb
|
||||||
|
|
||||||
|
# other modules
|
||||||
|
import skimage.transform as transform
|
||||||
|
from pylab import ginput
|
||||||
|
import pickle
|
||||||
|
import simplekml
|
||||||
|
|
||||||
|
def find_indices(lst, condition):
|
||||||
|
"imitation of MATLAB find function"
|
||||||
|
return [i for i, elem in enumerate(lst) if condition(elem)]
|
||||||
|
|
||||||
|
|
||||||
|
def create_transect(origin, orientation, length):
|
||||||
|
"""
|
||||||
|
Create a 2D transect of points with 1m interval.
|
||||||
|
|
||||||
|
Arguments:
|
||||||
|
-----------
|
||||||
|
origin: np.array
|
||||||
|
contains the X and Y coordinates of the origin of the transect
|
||||||
|
orientation: int
|
||||||
|
angle of the transect (anti-clockwise from North) in degrees
|
||||||
|
length: int
|
||||||
|
length of the transect in metres
|
||||||
|
|
||||||
|
Returns:
|
||||||
|
-----------
|
||||||
|
transect: np.array
|
||||||
|
contains the X and Y coordinates of the transect
|
||||||
|
|
||||||
|
"""
|
||||||
|
x0 = origin[0]
|
||||||
|
y0 = origin[1]
|
||||||
|
# orientation of the transect
|
||||||
|
phi = (90 - orientation)*np.pi/180
|
||||||
|
# create a vector with points at 1 m intervals
|
||||||
|
x = np.linspace(0,length,length+1)
|
||||||
|
y = np.zeros(len(x))
|
||||||
|
coords = np.zeros((len(x),2))
|
||||||
|
coords[:,0] = x
|
||||||
|
coords[:,1] = y
|
||||||
|
# translate and rotate the vector using the origin and orientation
|
||||||
|
tf = transform.EuclideanTransform(rotation=phi, translation=(x0,y0))
|
||||||
|
transect = tf(coords)
|
||||||
|
|
||||||
|
return transect
|
||||||
|
|
||||||
|
def draw_transects(output, settings):
|
||||||
|
"""
|
||||||
|
Allows the user to draw shore-normal transects over the mapped shorelines.
|
||||||
|
|
||||||
|
Arguments:
|
||||||
|
-----------
|
||||||
|
output: dict
|
||||||
|
contains the extracted shorelines and corresponding dates.
|
||||||
|
settings: dict
|
||||||
|
contains parameters defining :
|
||||||
|
transect_length: length of the transect in metres
|
||||||
|
|
||||||
|
Returns:
|
||||||
|
-----------
|
||||||
|
transects: dict
|
||||||
|
contains the X and Y coordinates of all the transects drawn. These are also saved
|
||||||
|
as a .pkl and .kml (+ a .jpg figure showing the location of the transects)
|
||||||
|
|
||||||
|
"""
|
||||||
|
sitename = settings['inputs']['sitename']
|
||||||
|
length = settings['transect_length']
|
||||||
|
filepath = os.path.join(os.getcwd(), 'data', sitename)
|
||||||
|
|
||||||
|
# plot all shorelines
|
||||||
|
fig1 = plt.figure()
|
||||||
|
ax1 = fig1.add_subplot(111)
|
||||||
|
ax1.axis('equal')
|
||||||
|
ax1.set_xlabel('Eastings [m]')
|
||||||
|
ax1.set_ylabel('Northings [m]')
|
||||||
|
ax1.grid(linestyle=':', color='0.5')
|
||||||
|
for i in range(len(output['shorelines'])):
|
||||||
|
sl = output['shorelines'][i]
|
||||||
|
date = output['dates'][i]
|
||||||
|
ax1.plot(sl[:, 0], sl[:, 1], '.', markersize=3, label=date.strftime('%d-%m-%Y'))
|
||||||
|
# ax1.legend()
|
||||||
|
fig1.set_tight_layout(True)
|
||||||
|
mng = plt.get_current_fig_manager()
|
||||||
|
mng.window.showMaximized()
|
||||||
|
ax1.set_title('Click two points to define each transect (first point is the origin of the transect).\n'+
|
||||||
|
'When all transects have been defined, click on <ENTER>', fontsize=16)
|
||||||
|
|
||||||
|
# initialise variable
|
||||||
|
transects = dict([])
|
||||||
|
counter = 0
|
||||||
|
# loop until user breaks it by click <enter>
|
||||||
|
while 1:
|
||||||
|
try:
|
||||||
|
pts = ginput(n=2, timeout=1e9)
|
||||||
|
origin = pts[0]
|
||||||
|
except:
|
||||||
|
fig1.gca().set_title('Transect locations', fontsize=16)
|
||||||
|
fig1.savefig(os.path.join(filepath, sitename + 'transects.jpg'), dpi=200)
|
||||||
|
break
|
||||||
|
counter = counter + 1
|
||||||
|
# create the transect using the origin, orientation and length
|
||||||
|
temp = np.array(pts[1]) - np.array(origin)
|
||||||
|
phi = np.arctan2(temp[1], temp[0])
|
||||||
|
orientation = -(phi*180/np.pi - 90)
|
||||||
|
transect = create_transect(origin, orientation, length)
|
||||||
|
transects[str(counter)] = transect
|
||||||
|
|
||||||
|
# plot the transects on the figure
|
||||||
|
ax1.plot(transect[:,0], transect[:,1], 'b.', markersize=4)
|
||||||
|
ax1.plot(transect[0,0], transect[0,1], 'rx', markersize=10)
|
||||||
|
ax1.text(transect[-1,0], transect[-1,1], str(counter), size=16,
|
||||||
|
bbox=dict(boxstyle="square", ec='k',fc='w'))
|
||||||
|
plt.draw()
|
||||||
|
|
||||||
|
# save as transects.pkl
|
||||||
|
with open(os.path.join(filepath, sitename + '_transects.pkl'), 'wb') as f:
|
||||||
|
pickle.dump(transects, f)
|
||||||
|
|
||||||
|
# save as transects.kml (for GIS)
|
||||||
|
kml = simplekml.Kml()
|
||||||
|
for key in transects.keys():
|
||||||
|
newline = kml.newlinestring(name=key)
|
||||||
|
newline.coords = transects[key]
|
||||||
|
newline.description = 'user-defined cross-shore transect'
|
||||||
|
kml.save(os.path.join(filepath, sitename + '_transects.kml'))
|
||||||
|
|
||||||
|
return transects
|
||||||
|
|
||||||
|
def compute_intersection(output, transects, settings):
|
||||||
|
"""
|
||||||
|
Computes the intersection between the 2D mapped shorelines and the transects, to generate
|
||||||
|
time-series of cross-shore distance along each transect.
|
||||||
|
|
||||||
|
Arguments:
|
||||||
|
-----------
|
||||||
|
output: dict
|
||||||
|
contains the extracted shorelines and corresponding dates.
|
||||||
|
settings: dict
|
||||||
|
contains parameters defining :
|
||||||
|
along_dist: alongshore distance to caluclate the intersection (median of points
|
||||||
|
within this distance).
|
||||||
|
|
||||||
|
Returns:
|
||||||
|
-----------
|
||||||
|
cross_dist: dict
|
||||||
|
time-series of cross-shore distance along each of the transects. These are not tidally
|
||||||
|
corrected.
|
||||||
|
|
||||||
|
"""
|
||||||
|
shorelines = output['shorelines']
|
||||||
|
along_dist = settings['along_dist']
|
||||||
|
|
||||||
|
# initialise variables
|
||||||
|
chainage_mtx = np.zeros((len(shorelines),len(transects),6))
|
||||||
|
idx_points = []
|
||||||
|
|
||||||
|
for i in range(len(shorelines)):
|
||||||
|
|
||||||
|
sl = shorelines[i]
|
||||||
|
idx_points_all = []
|
||||||
|
|
||||||
|
for j,key in enumerate(list(transects.keys())):
|
||||||
|
|
||||||
|
# compute rotation matrix
|
||||||
|
X0 = transects[key][0,0]
|
||||||
|
Y0 = transects[key][0,1]
|
||||||
|
temp = np.array(transects[key][-1,:]) - np.array(transects[key][0,:])
|
||||||
|
phi = np.arctan2(temp[1], temp[0])
|
||||||
|
Mrot = np.array([[np.cos(phi), np.sin(phi)],[-np.sin(phi), np.cos(phi)]])
|
||||||
|
|
||||||
|
# calculate point to line distance between shoreline points and the transect
|
||||||
|
p1 = np.array([X0,Y0])
|
||||||
|
p2 = transects[key][-1,:]
|
||||||
|
d_line = np.abs(np.cross(p2-p1,sl-p1)/np.linalg.norm(p2-p1))
|
||||||
|
# calculate the distance between shoreline points and the origin of the transect
|
||||||
|
d_origin = np.array([np.linalg.norm(sl[k,:] - p1) for k in range(len(sl))])
|
||||||
|
# find the shoreline points that are close to the transects and to the origin
|
||||||
|
# the distance to the origin is hard-coded here to 1 km
|
||||||
|
logic_close = np.logical_and(d_line <= along_dist, d_origin <= 1000)
|
||||||
|
idx_close = find_indices(logic_close, lambda e: e == True)
|
||||||
|
idx_points_all.append(idx_close)
|
||||||
|
|
||||||
|
# in case there are no shoreline points close to the transect
|
||||||
|
if not idx_close:
|
||||||
|
chainage_mtx[i,j,:] = np.tile(np.nan,(1,6))
|
||||||
|
else:
|
||||||
|
# change of base to shore-normal coordinate system
|
||||||
|
xy_close = np.array([sl[idx_close,0],sl[idx_close,1]]) - np.tile(np.array([[X0],
|
||||||
|
[Y0]]), (1,len(sl[idx_close])))
|
||||||
|
xy_rot = np.matmul(Mrot, xy_close)
|
||||||
|
|
||||||
|
# compute mean, median, max, min and std of chainage position
|
||||||
|
n_points = len(xy_rot[0,:])
|
||||||
|
mean_cross = np.nanmean(xy_rot[0,:])
|
||||||
|
median_cross = np.nanmedian(xy_rot[0,:])
|
||||||
|
max_cross = np.nanmax(xy_rot[0,:])
|
||||||
|
min_cross = np.nanmin(xy_rot[0,:])
|
||||||
|
std_cross = np.nanstd(xy_rot[0,:])
|
||||||
|
# store all statistics
|
||||||
|
chainage_mtx[i,j,:] = np.array([mean_cross, median_cross, max_cross,
|
||||||
|
min_cross, n_points, std_cross])
|
||||||
|
|
||||||
|
# store the indices of the shoreline points that were used
|
||||||
|
idx_points.append(idx_points_all)
|
||||||
|
|
||||||
|
# format into dictionnary
|
||||||
|
chainage = dict([])
|
||||||
|
chainage['mean'] = chainage_mtx[:,:,0]
|
||||||
|
chainage['median'] = chainage_mtx[:,:,1]
|
||||||
|
chainage['max'] = chainage_mtx[:,:,2]
|
||||||
|
chainage['min'] = chainage_mtx[:,:,3]
|
||||||
|
chainage['npoints'] = chainage_mtx[:,:,4]
|
||||||
|
chainage['std'] = chainage_mtx[:,:,5]
|
||||||
|
chainage['idx_points'] = idx_points
|
||||||
|
|
||||||
|
# only return the median
|
||||||
|
cross_dist = dict([])
|
||||||
|
for j,key in enumerate(list(transects.keys())):
|
||||||
|
cross_dist[key] = chainage['median'][:,j]
|
||||||
|
|
||||||
|
return cross_dist
|
Loading…
Reference in New Issue