def export_special(roi, description):
fc = ee.FeatureCollection(roi)
roi_mask = fc.geometry().bounds().getInfo()['coordinates']
image_list = list_assets('users/dgketchum/IrrMapper/version_2')
# years = [str(x) for x in range(1986, 1991)]
# target_images = [x for x in image_list if x.endswith(years[0])]
# target = ee.ImageCollection(image_list)
# target = target.mosaic().select('classification').remap([0, 1, 2, 3], [1, 0, 0, 0])
# range_images = [x for x in image_list if x.endswith(tuple(years))]
coll = ee.ImageCollection(image_list)
sum = ee.ImageCollection(coll.mosaic().select('classification').remap([0, 1, 2, 3], [1, 0, 0, 0])).sum().toDouble()
sum_mask = sum.lt(3)
img = sum.mask(sum_mask).toDouble()
task = ee.batch.Export.image.toDrive(
sum,
description='IrrMapper_V2_sum_years',
# folder='Irrigation',
region=roi_mask,
scale=30,
maxPixels=1e13,
# fileNamePrefix='IrrMapper_V2_{}_{}'.format(description, period)
)
task.start()
def first_detection():
# this doesn't work, but it works in Code Editor
for state in TARGET_STATES:
bounds = os.path.join(BOUNDARIES, state)
roi = ee.FeatureCollection(bounds)
mask = roi.geometry().bounds().getInfo()['coordinates']
image_list = list_assets('users/dgketchum/classy')
out_images = []
for yr in range(1986, 2017):
yr_img = [x for x in image_list if x.endswith(str(yr))]
coll = ee.ImageCollection(yr_img)
classed = coll.mosaic().select('classification').remap(
[0, 1, 2, 3], [yr, 0, 0, 0]).rename('{}_min'.format(yr))
out_images.append(classed)
coll = ee.ImageCollection(out_images)
img = coll.reduce(ee.Reducer.minMax()).rename('min', 'max')
pprint(img.getInfo())
task = ee.batch.Export.image.toAsset(image=img,
description='{}'.format(state),
assetId=os.path.join(
ASSET_ROOT,
'{}'.format(state)),
fileNamePrefix='{}'.format(state),
region=mask,
scale=30,
maxPixels=1e10)
print(state)
task.start()
break
def export_raster():
target_bn = 'projects/ee-dgketchum/assets/IrrMapper/IrrMapper_RF'
image_list = list_assets('users/dgketchum/IrrMapper/version_2')
for yr in range(1987, 2022):
images = [x for x in image_list if x.endswith(str(yr))]
coll = ee.ImageCollection(images)
_properties = {
'image_id': 'IrrMapper_RF_{}'.format(yr),
'system:time_start': ee.Date.fromYMD(yr, 1, 1),
'system:time_end': ee.Date.fromYMD(yr, 12, 31)
}
img = coll.select('classification').remap([0, 1, 2, 3], [1, 0, 0, 0])
img = img.updateMask(
img.neq(0)).rename('classification').set(_properties)
id_ = os.path.join(target_bn, '{}'.format(yr))
task = ee.batch.Export.image.toAsset(
image=img,
description='IrrMapper_RF_{}'.format(yr),
assetId=id_,
pyramidingPolicy={'.default': 'mode'},
scale=30,
maxPixels=1e13)
task.start()
print(yr)
def reduce_classification(tables, years=None, description=None, cdl_mask=False, min_years=0):
"""
Reduce Regions, i.e. zonal stats: takes a statistic from a raster within the bounds of a vector.
Use this to get e.g. irrigated area within a county, HUC, or state. This can mask based on Crop Data Layer,
and can mask data where the sum of irrigated years is less than min_years. This will output a .csv to
GCS wudr bucket.
:param tables: vector data over which to take raster statistics
:param years: years over which to run the stats
:param description: export name append str
:param cdl_mask:
:param min_years:
:return:
"""
sum_mask = None
image_list = list_assets('users/dgketchum/IrrMapper/version_2')
fc = ee.FeatureCollection(tables)
if min_years > 0:
coll = ee.ImageCollection(image_list)
sum = ee.ImageCollection(coll.mosaic().select('classification').remap([0, 1, 2, 3], [1, 0, 0, 0])).sum()
sum_mask = sum.lt(min_years)
# first = True
for yr in years:
if yr not in [2002, 2007, 2012]:
yr_img = [x for x in image_list if x.endswith(str(yr))]
coll = ee.ImageCollection(yr_img)
tot = coll.mosaic().select('classification').remap([0, 1, 2, 3], [1, 0, 0, 0])
if cdl_mask and min_years > 0:
# cultivated/uncultivated band only available 2013 to 2017
cdl = ee.Image('USDA/NASS/CDL/2013')
cultivated = cdl.select('cultivated')
cdl_crop_mask = cultivated.eq(2)
tot = tot.mask(cdl_crop_mask).mask(sum_mask)
elif min_years > 0:
tot = tot.mask(sum_mask)
elif cdl_mask:
cdl = ee.Image('USDA/NASS/CDL/2013')
cultivated = cdl.select('cultivated')
cdl_crop_mask = cultivated.eq(2)
tot = tot.mask(cdl_crop_mask)
tot = tot.multiply(ee.Image.pixelArea())
reduce = tot.reduceRegions(collection=fc,
reducer=ee.Reducer.sum(),
scale=30)
task = ee.batch.Export.table.toCloudStorage(
reduce,
description='{}_area_{}_'.format(description, yr),
bucket='wudr',
fileNamePrefix='{}_area_{}_'.format(description, yr),
fileFormat='CSV')
task.start()
print(yr)
def export_raster(roi, description):
fc = ee.FeatureCollection(roi)
mask = fc.geometry().bounds().getInfo()['coordinates']
image_list = list_assets('users/dgketchum/IrrMapper/version_2')
for yr in range(1986, 2019):
yr_img = [x for x in image_list if x.endswith(str(yr))]
coll = ee.ImageCollection(yr_img)
img = ee.ImageCollection(coll.mosaic().select('classification'))
img = img.first()
task = ee.batch.Export.image.toDrive(
img,
description='IrrMapper_V2_{}_{}'.format(description, yr),
folder='Irrigation',
region=mask,
scale=30,
maxPixels=1e13,
fileNamePrefix='IrrMapper_V2_{}_{}'.format(description, yr))
task.start()
print(yr)
def request_validation_extract(file_prefix='validation'):
"""
This takes a sample points set and extracts the classification result. This is a roundabout cross-validation.
Rather than using holdouts in the Random Forest classifier, we just run all the training data to train the
classifier, and come back later with this function and a seperate set of points (with known classes) to
independently test classifier accuracy.
Other options to achieve this is to use out-of-bag cross validation, or set up a sckikit-learn RF classifier and
use k-folds cross validation.
:param file_prefix:
:return:
"""
roi = ee.FeatureCollection(ROI)
plots = ee.FeatureCollection(VALIDATION_POINTS).filterBounds(roi)
image_list = list_assets('users/dgketchum/IrrMapper/version_2')
for yr in YEARS:
yr_img = [x for x in image_list if x.endswith(str(yr))]
coll = ee.ImageCollection(yr_img)
classified = coll.mosaic().select('classification')
start = '{}-01-01'.format(yr)
d = datetime.strptime(start, '%Y-%m-%d')
epoch = datetime.utcfromtimestamp(0)
start_millisec = (d - epoch).total_seconds() * 1000
filtered = plots.filter(ee.Filter.eq('YEAR', ee.Number(start_millisec)))
plot_sample_regions = classified.sampleRegions(
collection=filtered,
properties=['POINT_TYPE', 'YEAR'],
scale=30)
task = ee.batch.Export.table.toCloudStorage(
plot_sample_regions,
description='{}_{}'.format(file_prefix, yr),
bucket='wudr',
fileNamePrefix='{}_{}'.format(file_prefix, yr),
fileFormat='CSV')
task.start()
print(yr)
def get_ndvi_stats(tables, years, out_name):
fc = ee.FeatureCollection(tables)
i = get_ndvi_series(years, fc)
image_list = list_assets('users/dgketchum/IrrMapper/version_2')
for yr in years:
coll = ee.ImageCollection(image_list).filterDate('{}-01-01'.format(yr), '{}-12-31'.format(yr))
remap = coll.mosaic().select('classification').remap([0, 1, 2, 3], [1, 0, 0, 0])
cls = remap.rename('irr_{}'.format(yr))
i = i.addBands(cls)
stats = i.reduceRegions(collection=fc,
reducer=ee.Reducer.mean(),
scale=30)
task = ee.batch.Export.table.toCloudStorage(
stats,
description='{}'.format(out_name),
bucket='wudr',
fileNamePrefix='{}'.format(out_name),
fileFormat='KML')
task.start()
