def project(year):
print("projecting %d" % year)
# Open the mask shape file
shp_file = 'c:/data/from-adriana/tropicalforests.shp'
shapes = fiona.open(shp_file)
# Read Adriana's abundance model (mainland)
mod = modelr.load('../tropical-bii/simplifiedAbundanceModel.rds')
predicts.predictify(mod)
# Import standard PREDICTS rasters
rasters = predicts.rasterset('1km', '', year, 'medium')
rasters['primary_lightintense'] = SimpleExpr(
'primary_lightintense', 'primary_light + primary_intense')
rasters['cropland_lightintense'] = SimpleExpr(
'cropland_lightintense', 'cropland_light + cropland_intense')
rasters['pasture_lightintense'] = SimpleExpr(
'pasture_lightintense', 'pasture_light + pasture_intense')
rasters['rd_dist'] = Raster('rd_dist', '/out/1km/roads-adrid.tif')
rasters['hpd'] = Raster(
'hpd', '/vsizip//data/from-adriana/HPD01to12.zip/yr%d/hdr.adf' % year)
rs = RasterSet(rasters, shapes=shapes, crop=True, all_touched=True)
what = 'LogAbund'
rs[what] = mod
stime = time.time()
rs.write(what, utils.outfn('1km', 'adrid-%d.tif' % year))
etime = time.time()
print("executed in %6.2fs" % (etime - stime))
def project_year(model, model_dir, what, scenario, year):
print("projecting %s for %d using %s" % (what, year, scenario))
models = select_models(model, model_dir)
# Read Sam's abundance model (forested and non-forested)
mod = modelr.load(models[0])
predicts.predictify(mod)
# Import standard PREDICTS rasters
by_age = 'young_secondary' in mod.syms
print('by_age: %s' % str(by_age))
rasters = predicts.rasterset('luh5', scenario, year, by_age)
rs = RasterSet(rasters)
if what == 'bii':
vname = 'bii'
rs[vname] = SimpleExpr(vname, '%s / %f' % (mod.output, intercept))
else:
vname = mod.output
rs[vname] = mod
if what not in rs:
print('%s not in rasterset' % what)
print(', '.join(sorted(rs.keys())))
sys.exit(1)
stime = time.time()
data, meta = rs.eval(what, quiet=False)
etime = time.time()
print("executed in %6.2fs" % (etime - stime))
oname = os.path.join(os.environ['OUTDIR'],
'luh5/%s-%s-%d.tif' % (scenario, what, year))
#hpd, _ = rs.eval('hpd', quiet=False)
#hpd_max, meta2 = rs.eval('hpd_max', quiet=False)
with rasterio.open(oname, 'w', **meta) as dst:
#bb = dst.bounds
#ul = map(int, ~meta2['affine'] * (bb[0], bb[3]))
#lr = map(int, ~meta2['affine'] * (bb[2], bb[1]))
#cap = ma.where(hpd > hpd_max[ul[1]:lr[1], ul[0]:lr[0]], True, False)
#show(hpd, norm=colors.PowerNorm(gamma=0.2))
#show(cap * 1)
#data.mask = np.logical_or(data.mask, cap)
dst.write(data.filled(meta['nodata']), indexes=1)
if None:
fig = plt.figure(figsize=(8, 6))
ax = plt.gca()
show(data, cmap='viridis', ax=ax)
plt.savefig('luh2-%s-%d.png' % (scenario, year))
return
def project_year(model, model_dir, what, scenario, year):
print("projecting %s for %d using %s" % (what, year, scenario))
models = select_models(model, model_dir)
# Read Sam's abundance model (forested and non-forested)
modf = modelr.load(models[0])
intercept_f = modf.intercept
predicts.predictify(modf)
modn = modelr.load(models[1])
intercept_n = modn.intercept
predicts.predictify(modn)
# Open forested/non-forested mask layer
fstnf = rasterio.open(utils.luh2_static('fstnf'))
# Import standard PREDICTS rasters
rastersf = predicts.rasterset('luh2', scenario, year, 'f')
rsf = RasterSet(rastersf, mask=fstnf, maskval=0.0)
rastersn = predicts.rasterset('luh2', scenario, year, 'n')
rsn = RasterSet(rastersn, mask=fstnf, maskval=1.0)
#rsn = RasterSet(rastersn)
if what == 'bii':
vname = 'bii'
rsf[vname] = SimpleExpr(
vname, 'exp(%s) / exp(%f)' % (modf.output, intercept_f))
rsn[vname] = SimpleExpr(
vname, 'exp(%s) / exp(%f)' % (modn.output, intercept_n))
rsf[modf.output] = modf
rsn[modn.output] = modn
else:
vname = modf.output
assert modf.output == modn.output
rsf[vname] = modf
rsn[vname] = modn
if what not in rsf:
print('%s not in rasterset' % what)
print(', '.join(sorted(rsf.keys())))
sys.exit(1)
stime = time.time()
datan, meta = rsn.eval(what, quiet=False)
dataf, _ = rsf.eval(what, quiet=True)
data_vals = dataf.filled(0) + datan.filled(0)
data = data_vals.view(ma.MaskedArray)
data.mask = np.logical_and(dataf.mask, datan.mask)
#data = datan
etime = time.time()
print("executed in %6.2fs" % (etime - stime))
oname = '%s/luh2/%s-%s-%d.tif' % (utils.outdir(), scenario, what, year)
with rasterio.open(oname, 'w', **meta) as dst:
dst.write(data.filled(meta['nodata']), indexes=1)
if None:
fig = plt.figure(figsize=(8, 6))
ax = plt.gca()
show(data, cmap='viridis', ax=ax)
plt.savefig('luh2-%s-%d.png' % (scenario, year))
return
def project_year(model, model_dir, scenario, year):
"""Run a projection for a single year. Can be called in parallel when
projecting a range of years.
"""
print("projecting %s for %d using %s" % (model, year, scenario))
# Import standard PREDICTS rasters
rasters = predicts.rasterset('luh2', scenario, year)
rs = RasterSet(rasters)
what, model = select_model(model, model_dir)
# Read Sam's models
if model:
mod = modelr.load(model)
predicts.predictify(mod)
rs[mod.output] = mod
if what in ('CompSimAb', 'CompSimSR', 'Abundance', 'Richness'):
if what in ('CompSimAb', 'CompSimSR'):
expr = '(inv_logit(%s) - 0.01) / (inv_logit(%f) - 0.01)'
else:
expr = '(exp(%s) / exp(%f))'
rs[what] = SimpleExpr(what, expr % (mod.output, mod.intercept))
if what not in rs:
print('%s not in rasterset' % what)
print(', '.join(sorted(rs.keys())))
sys.exit(1)
stime = time.time()
data, meta = rs.eval(what, quiet=True)
etime = time.time()
print("executed in %6.2fs" % (etime - stime))
oname = '%s/luh2/%s-%s-%04d.tif' % (utils.outdir(), scenario, what, year)
with rasterio.open(oname, 'w', **meta) as dst:
dst.write(data.filled(meta['nodata']), indexes=1)
if None:
fig = plt.figure(figsize=(8, 6))
show(data, cmap='viridis', ax=plt.gca())
fig.savefig('luh2-%s-%d.png' % (scenario, year))
return
def luh2_rasterset(scenario, year):
rset = luh2_types('historical', 1999)
rset_add(rset, 'perennial', 'c3per + c4per')
rset_add(rset, 'annual', 'c3ann + c4ann')
rset_add(rset, 'nitrogen', 'c3nfx')
rset_add(rset, 'rangelands', 'range')
rset_add(rset, 'secondaryf', 'secdyf + secdif + secdmf')
rset_add(rset, 'secondaryn', 'secdyn + secdin + secdmn')
rset_add(rset, 'secondary',
'secdyf + secdif + secdmf + secdyn + secdin + secdmn')
return RasterSet(rset)
'1km/mainland-from-igor-edited.tif')
mask_ds = rasterio.open(mask_file)
# Richness compositional similarity model with updated PriMin-Urb coefficient
mod = modelr.load('/home/vagrant/katia/models/updated_compsim_rich.rds')
predicts.predictify(mod)
# Import standard PREDICTS rasters
rasters = predicts.rasterset('1km', 'medium', year=2005)
# create an ISL_MAINL raster
# set it to Mainlands this time round (set Mainlands to 1 and Islands to 0)
rasters['ISL_MAINMAINLAND'] = SimpleExpr('ISL_MAINMAINLAND', '2')
rasters['ISL_MAINISLAND'] = SimpleExpr('ISL_MAINISLAND', '0')
rasters['adjGeogDist'] = SimpleExpr('adjGeogDist', '0')
rasters['cubeRtEnvDist'] = SimpleExpr('cubeRtEnvDist', '0')
# set up the rasterset, cropping to mainlands
rs = RasterSet(rasters, mask=mask_ds, maskval=0, crop=True)
# if you're projecting the whole world, use this code instead
#rs = RasterSet(rasters)
# evaluate the model
# model is logit transformed with an adjustment, so back-transformation
rs[mod.output] = mod
rs['output'] = SimpleExpr(
'output', '(inv_logit(%s) - 0.01) / (inv_logit(%f) - 0.01)' %
(mod.output, mod.intercept))
rs.write('output', utils.outfn('katia', 'bii-sr-cs-mainlands.tif'))
from projections.rasterset import RasterSet, Raster
import projections.predicts as predicts
import projections.r2py.modelr as modelr
# Open the mask shape file
shp_file = os.path.join(os.environ['DATA_ROOT'],
'from-adriana/tropicalforests.shp')
shapes = fiona.open(shp_file)
# Read Adriana's abundance model (mainland)
mod = modelr.load(os.path.join(os.environ['MODEL_DIR'], 'ab-model.rds'))
predicts.predictify(mod)
# Import standard PREDICTS rasters
rasters = predicts.rasterset('luh5', 'historical', 1990, True)
rs = RasterSet(rasters, shapes=shapes, all_touched=True)
what = mod.output
rs[mod.output] = mod
stime = time.time()
data1, meta_data1 = rs.eval(what)
etime = time.time()
print("executed in %6.2fs" % (etime - stime))
show(data1)
##
## Compare with good raster
##
out = rasterio.open('adrid-good.tif')
good = out.read(1, masked=True)
diff = np.fabs(data1 - good)
#!/usr/bin/env python
from projections.atlas import atlas
from projections.rasterset import RasterSet, Raster
import projections.predicts as predicts
import projections.rds as rds
# Read Katia's abundance model (mainland)
mod = rds.read('../models/ab-mainland.rds')
predicts.predictify(mod)
# Import standard PREDICTS rasters
rasters = predicts.rasterset('rcp', 'aim', 2020, 'medium')
rs = RasterSet(rasters)
rs[mod.output()] = mod
data = rs.eval(mod.output())
# Display the raster
atlas(data, mod.output(), 'viridis')
# get the differences
# diff = cuberoot(s1) - cuberoot(s2)
# we want cuberoot(s1+1) = 0
# so the difference is 0 - cuberoot(s2)
# then center and scale (subtract mean and divide by sd)
mean_rd50_diff = float(
df[df['Value'] == 'CRootDens_50km_diff']['mean'])
mean_rd1_diff = float(df[df['Value'] == 'CRootDens_1km_diff']['mean'])
sd_rd50_diff = float(df[df['Value'] == 'CRootDens_50km_diff']['sd'])
sd_rd1_diff = float(df[df['Value'] == 'CRootDens_1km_diff']['sd'])
rasters['CRootDens_50km_diff_cs'] = SimpleExpr(
'CRootDens_50km_diff_cs',
'((0 - clip_rd50) - %f) / %f' % (mean_rd50_diff, sd_rd50_diff))
rasters['CRootDens_1km_diff_cs'] = SimpleExpr(
'CRootDens_1km_diff_cs',
'((0 - clip_rd1) - %f) / %f' % (mean_rd1_diff, sd_rd1_diff))
# set up the rasterset, cropping by tropical forests
rs = RasterSet(rasters, shapes=shapes, crop=True, all_touched=False)
# evaluate the model
# model is logit transformed with an adjustment, so back-transformation
rs[mod.output] = mod
rs['output'] = SimpleExpr(
'output', '(inv_logit(%s) - 0.01) / (inv_logit(%f) - 0.01)' %
(mod.output, ref))
rs.write(
'output',
utils.outfn('temporal-bii/' + version, 'bii-cs-%d.tif' % year))
# clip the cs rasters so they can't go below 0 but no upper limit
inname = 'C:/ds/temporal-bii/' + version + '/bii-cs-' + str(
year) + '.tif'
outname = 'C:/ds/temporal-bii/' + version + '/bii-cs-low-bound-' + str(
year) + '.tif'
ru.clip(inname, outname, 0, None)
# calculate bii (with only a lower bound)
# pull in all the rasters for computing bii
bii_rs = RasterSet({
'abundance':
Raster(
'abundance', 'C:/ds/temporal-bii/' + version +
'/bii-ab-low-bound-%d.tif' % year),
'comp_sim':
Raster(
'comp_sim', 'C:/ds/temporal-bii/' + version +
'/bii-cs-low-bound-%d.tif' % year),
'bii_ab':
SimpleExpr('bii_ab', 'abundance * comp_sim')
})
# write out bii raster
bii_rs.write(
'bii_ab',
utils.outfn('temporal-bii/' + version, 'bii-%d.tif' % year))
# calculate bounded bii
# clip the abundance rasters
inname = 'C:/ds/temporal-bii/' + version + '/bii-ab-' + str(
year) + '.tif'
mod = modelr.load('/home/vagrant/katia/models/updated_compsim_abund.rds')
predicts.predictify(mod)
# Import standard PREDICTS rasters
rasters = predicts.rasterset('1km', 'medium', year=2005)
# create an ISL_MAINL raster
# set it to Mainlands this time round (set Mainlands to 1 and Islands to 0)
rasters['ISL_MAINMAINLAND'] = SimpleExpr('ISL_MAINMAINLAND', ISLMAIN)
rasters['ISL_MAINISLAND'] = SimpleExpr('ISL_MAINISLAND', 0)
rasters['adjGeogDist'] = SimpleExpr('adjGeogDist', 0)
rasters['cubeRtEnvDist'] = SimpleExpr('cubeRtEnvDist', 0)
# set up the rasterset, cropping to mainlands
rs = RasterSet(rasters, mask=mask_ds, maskval=0, crop=True)
# if you're projecting the whole world, use this code instead
#rs = RasterSet(rasters)
# evaluate the model
# model is logit transformed with an adjustment, so back-transformation
rs[mod.output] = mod
intercept = mod.partial({'ISL_MAINMAINLAND': ISLMAIN})
print("intercept: %.5f" % intercept)
if args.mainland:
assert math.isclose(intercept, 1.939964, rel_tol=0.001)
else:
assert math.isclose(intercept, 1.439107176, rel_tol=0.001)
rs['output'] = SimpleExpr(
'output', '(inv_logit(%s) - 0.01) / (inv_logit(%f) - 0.01)' %
ab_max = 1.655183
sr_max = 1.636021
else:
suffix = 'islands'
ab_max = 1.443549
sr_max = 1.413479
folder = 'clip' if args.clip else 'no-clip'
# pull in all the rasters for computing bii
bii_rs = RasterSet({'abundance': Raster('abundance',
utils.outfn('katia',
folder,
'ab-%s.tif' % suffix)),
'comp_sim': Raster('comp_sim',
utils.outfn('katia',
'ab-cs-%s.tif' % suffix)),
'clip_ab': SimpleExpr('clip_ab',
'clip(abundance, 0, %f)' % ab_max),
'bii_ab': SimpleExpr('bii_ab', 'abundance * comp_sim'),
'bii_ab2': SimpleExpr('bii_ab2', 'clip_ab * comp_sim'),
})
# write out bii raster
bii_rs.write('bii_ab' if args.clip else 'bii_ab2',
utils.outfn('katia', folder,
'abundance-based-bii-%s.tif' % suffix))
# do the same for species richness
# pull in all the rasters for computing bii
bii_rs = RasterSet({'sp_rich': Raster('sp_rich',
utils.outfn('katia',
'pow(RdDens_50km, 1/3.)')
# extract the maximum value
max_rd = float(df[df['Value'] == 'CRootRdDens_50km']['max'])
# clip to the maximum value
rasters['clip_rd'] = SimpleExpr(
'clip_rd', 'clip(Croot_RdDens_50km, 0, %f)' % max_rd)
# scale and center
mean_rd = float(df[df['Value'] == 'CRootRdDens_50km']['mean'])
sd_rd = float(df[df['Value'] == 'CRootRdDens_50km']['sd'])
rasters['CRootRdDens_50km_cs'] = SimpleExpr(
'CRootRdDens_50km_cs', '(clip_rd - %f) / %f' % (mean_rd, sd_rd))
# set up the rasterset, cropping by tropical forests
# set all_touched = False as there are some boundary issues with very small countries on the edge
# like Macao
rs = RasterSet(rasters, shapes=shapes, crop=True, all_touched=False)
# evaluate the model
# model is square root abundance so square it
# note that the intercept value has been calculated for the baseline land use when all other variables are held at 0
rs[mod.output] = mod
rs['output'] = SimpleExpr(
'output', '(pow(%s, 2) / pow(%f, 2))' % (mod.output, ref))
rs.write(
'output',
utils.outfn('temporal-bii/' + version, 'bii-ab-%d.tif' % year))
# if you're on the final version (i.e. the last time you're going round)
# write out the pressure data
if version == 'v3':
from projections.rasterset import RasterSet, Raster
from projections.simpleexpr import SimpleExpr
import projections.predicts as predicts
import projections.r2py.modelr as modelr
import projections.utils as utils
import projections.raster_utils as ru
CLIP = 'no-clip'
# pull in all the rasters for computing bii
bii_rs = RasterSet({
'abundance':
Raster('abundance', utils.outfn('katia', CLIP, 'bii-ab-mainlands.tif')),
'comp_sim':
Raster('comp_sim', utils.outfn('katia', 'bii-ab-cs-mainlands.tif')),
'clip_ab':
SimpleExpr('clip_ab', 'clip(abundance, 0, 1.655183)'),
'bii_ab':
SimpleExpr('bii_ab', 'abundance * comp_sim'),
'bii_ab2':
SimpleExpr('bii_ab2', 'clip_ab * comp_sim'),
})
# write out bii raster
bii_rs.write('bii_ab' if CLIP == 'clip' else 'bii_ab2',
utils.outfn('katia', CLIP, 'abundance-based-bii-mainlands.tif'))
# do the same for species richness
# pull in all the rasters for computing bii
bii_rs = RasterSet({
'sp_rich':
Raster('sp_rich', utils.outfn('katia', CLIP, 'bii-sr-mainlands.tif')),
'DistRd', os.path.join(utils.data_root(),
'1km/rddistwgs.tif')) ###Use new raster
## If args.clip is true, limit the predictor variable values to the max seen
## when fitting the model
if args.clip:
rasters['clipDistRd'] = SimpleExpr(
'clipDistRd', 'clip(DistRd, %f, %f)' % (RD_DIST_MIN, RD_DIST_MAX))
else:
rasters['clipDistRd'] = SimpleExpr('clipDistRd', 'DistRd')
rasters['logDistRd_rs'] = SimpleExpr(
'logDistRd_rs', 'scale(log(clipDistRd + 100),'
'0.0, 1.0, -1.120966, 12.18216)')
###Added +100 to DistRd to deal with zero values
# set up the rasterset, cropping to mainlands
rs = RasterSet(rasters, mask=mask_ds, maskval=0, crop=True)
# if you're projecting the whole world, use this code instead
# rs = RasterSet(rasters)
# evaluate the model
# model is square root abundance so square it
# Note that the intercept value has been calculated for the baseline
# land use when all other variables are held at 0
# Therefore I calculate separatedly an intercept where DistRd is set to
# the max value, i.e. logDistRd_RS = 1.
intercept = mod.partial({'ISL_MAINLMAINLAND': ISLMAIN, 'logDistRd_rs': 1.0})
print("intercept: %.5f" % intercept)
if args.mainland:
assert math.isclose(intercept, 0.67184, rel_tol=0.001)
#!/usr/bin/env python
import fiona
import time
from rasterio.plot import show, show_hist
from projections.rasterset import RasterSet, Raster
import projections.predicts as predicts
import projections.modelr as modelr
# Open the mask shape file
shp_file = '../../data/from-adriana/tropicalforests.shp'
shapes = fiona.open(shp_file)
# Read Adriana's abundance model (mainland)
mod = modelr.load('../models/ab-corrected.rds')
predicts.predictify(mod)
# Import standard PREDICTS rasters
rasters = predicts.rasterset('1km', 'foo', 2005, 'historical')
rs = RasterSet(rasters, shapes=shapes, all_touched=True)
rs[mod.output] = mod
what = mod.output
stime = time.time()
data = rs.write(what, 'hires.tif')
etime = time.time()
print("executed in %6.2fs" % (etime - stime))
import fiona
import numpy as np
import time
import matplotlib.pyplot as plt
from rasterio.plot import show, show_hist
from projections.atlas import atlas
from projections.rasterset import RasterSet, Raster
from projections.simpleexpr import SimpleExpr
import projections.predicts as predicts
import projections.rds as rds
# Import standard PREDICTS rasters
rasters = predicts.rasterset('rcp', 'aim', 2020, 'medium')
rs = RasterSet(rasters)
rs['logHPD_rs2'] = SimpleExpr('logHPD_rs', 'scale(log(hpd + 1), 0.0, 1.0)')
data1 = rs.eval('logHPD_rs')
data2 = rs.eval('logHPD_rs2')
data3 = np.where(np.isclose(data1, data2, equal_nan=True), 1, 0)
diff = np.fabs(data1 - data2)
print("max diff %f" % diff.max())
print("max in hpd_ref: %f" % rs['hpd_ref'].data.values.max())
print("max in hpd: %f" % rs['hpd'].data.dropna().values.max())
fig, ((ax1, ax2, ax3), (hx1, hx2, hx3)) = plt.subplots(2, 3, figsize=(21, 7))
show(data1, ax=ax1, cmap='Greens', title='Global max/min')
show(data2, ax=ax2, cmap='Greens', title='Computed max/min')
show(diff, ax=ax3, cmap='viridis', title='Difference', vmin=0, vmax=1.0)
)
rasters['secondary_vegetation_intense'] = SimpleExpr(
'secondary_vegetation_intense',
'young_secondary_intense + intermediate_secondary_intense + mature_secondary_intense'
)
rasters['primary_vegetation_intense'] = SimpleExpr(
'primary_vegetation_intense', 'primary_intense')
rasters['primary_vegetation_light'] = SimpleExpr(
'primary_vegetation_light', 'primary_light')
rasters['plantation_pri'] = SimpleExpr('plantation_pri',
'perennial + timber')
rasters['plantation_pri_intense'] = SimpleExpr(
'plantation_pri_intense', 'perennial_intense + timber_intense')
rasters['plantation_pri_light'] = SimpleExpr(
'plantation_pri_light', 'perennial_light + timber_light')
rasters['plantation_pri_minimal'] = SimpleExpr(
'plantation_pri_minimal', 'perennial_minimal + timber_minimal')
rs = RasterSet(rasters)
# back transform (they were transformed using log(x+1)
rs[mod.output] = mod
rs['output'] = SimpleExpr(
'output', 'clip((exp(%s) - 1) / (exp(%f) - 1), 0, 1e20)' %
(mod.output, mod.intercept))
rs.write(
'output', 'D:/victoria_projections/projections/' + what + '-' +
scenario + '-' + '%d.tif' % year)
import sys
import rasterio
from projections.rasterset import RasterSet
import projections.predicts as predicts
import projections.utils as utils
# Import standard PREDICTS rasters
rasters = predicts.rasterset('1km', 'medium', year = 2005)
for suffix in ('islands', 'mainland'):
# Open the BII raster file
mask_file = 'C:/Users/katis2/Desktop/Final_projections/Clip_variables/abundance-based-bii-%s.tif' % suffix
mask_ds = rasterio.open(mask_file)
# set up the rasterset, cropping to mainlands
rs = RasterSet(rasters, mask=mask_ds, maskval=-9999, crop=True)
# Run through each land-use
for lu in ('cropland', 'pasture', 'primary', 'secondary', 'urban'):
# And every use intensity
for ui in ('minimal', 'light', 'intense'):
name = '%s_%s' % (lu, ui)
print(name)
oname = utils.outfn('katia', '%s-%s.tif' % (name, suffix))
if os.path.isfile(oname) or name in ('secondary_intense',
'urnan_light'):
continue
rs.write(name, oname)
