Watershed calculation

[adolmajian]

Any idea on how to use richdem for watershed delineation?

[KMarkert]

I wrote a function on how to do this with Python which takes the results from rd.FlowProportions() as an array and a gauge location as list of [y,x](should be index based on image dims) as arguments and returns an array where 1 is the catchment. This is really slow in Python so probably not the best approach and would need to be optimized for large scale application....

import numpy as np

def catchment(flowDir, initPt):
    # dictionary to lookup the neighbors based on flow direction array
    dir_map = {
        1: {
            'dir': 5, # flattened index of the value to reference when searching for a grid
            'idx': np.unravel_index(5,[3,3]) # index in reference to the center pixel to adjust for
        },
        2: {
            'dir': 8,
            'idx': np.unravel_index(8,[3,3])
        },
        3: {
            'dir': 7,
            'idx': np.unravel_index(7,[3,3])
        },
        4: {
            'dir': 6,
            'idx': np.unravel_index(6,[3,3])
        },
        5: {
            'dir': 4,
            'idx': np.unravel_index(3,[3,3])
        },
        6: {
            'dir': 1,
            'idx': np.unravel_index(0,[3,3])
        },
        7: {
            'dir': 2,
            'idx': np.unravel_index(1,[3,3])
        },
        8: {
            'dir': 3,
            'idx': np.unravel_index(2,[3,3])
        },
    }

    catchIdx = [] # list of pixels that contribute to starting point
    pts = [initPt] # list of points to search which neighboring cells flow through

    searching = True
    while searching:
        nextIdxes = [] # blank list to append for next iter step
        # loop through all of the points and find which neighbors contribute
        for p in pts:
            flows = []
            # get neighbors
            for k,v in dir_map.items():
                flows.append(flowDir[p[0]-1:p[0]+2,p[1]-1:p[1]+2,k].ravel()[v['dir']])

            # find the direction of neighbors and append
            for j in np.where(np.array(flows)>0)[0]:
                dy,dx = [d-1 for d in dir_map[j+1]['idx']]
                ni = [p[0]+dy,p[1]+dx]
                nextIdxes.append(ni)

        # append the contributing cells to the total list
        catchIdx.extend(nextIdxes)

        # reset the pts variable for next iter
        pts = nextIdxes

        if len(nextIdxes) <= 0 :
            searching = False

    # transform the catchment indexes to 2-d array
    catchOut = np.zeros_like(flowDir[:,:,-1])
    catchOut[tuple(zip(*catchIdx))] = 1

    return catchOut

[r-barnes]

@adolmajian - What do you mean by watershed delineation? If you describe, very specifically, what you want, I can probably make it happen.