def main():
# get specifications for run to read
model_ids = ['811','811','911','911']
assim_type = '_p25adapted'
ens_size = 500
# get pixels, ids and number of iterations to read
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf'+assim_type+'/' + model_ids[0] + '/'
pixels = ['3809','3524','2224','4170','1945','3813','4054','3264','1271','3457']
mcmc_ids = ['119','3','3','119']
n_iters = ['40000000','1000000','1000000','40000000']
nbe_mae, lai_mae, abgb_mae, gpp_mae = [], [], [], []
# run through pixels
for pixel in pixels:
# get that pixel's outputs for each MCMCID
nbe_pred, lai_pred, abgb_pred, gpp_pred = [], [], [], []
for model_id, mcmc_id, n_iter in zip(model_ids, mcmc_ids, n_iters):
# set directories
cur_dir = os.getcwd() + '/'
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf'+assim_type+'/' + model_id + '/'
cbr_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbr'+assim_type+'/' + model_id + '/'
output_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/output'+assim_type+'/' + model_id + '/'
plot_dir = '../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames('../../misc/', model_id)
# read cbf file for that pixel
cbf_pixel = rwb.read_cbf_file(glob.glob(cbf_dir + '*_' + pixel+'.cbf')[0])
# read obs and obs unc for that pixel
nbe_obs, lai_obs, abgb_obs, sif_obs = cbf_pixel['OBS']['NBE'], cbf_pixel['OBS']['LAI'], cbf_pixel['OBS']['ABGB'], cbf_pixel['OBS']['GPP']
nbe_an_unc, nbe_seas_unc, lai_unc, abgb_unc = cbf_pixel['OBSUNC']['NBE']['annual_unc'], cbf_pixel['OBSUNC']['NBE']['seasonal_unc'], cbf_pixel['OTHER_OBS']['MLAI']['unc'], cbf_pixel['OBSUNC']['ABGB']['unc']
conv_chains_pkl = read_pickle(glob.glob(cbr_dir + model_id + assim_type + '*_MCMC'+mcmc_id + '_'+n_iter+'_best_subset.pkl')[0])
conv_chains_pkl.columns = ['pixel','bestchains','conv']# if model_id!='911' else ['pixel','bestchains'] #rename columns for easier access
# grab cbrs corresponding to that pixel, MCMCID and number of iterations
files = glob.glob(cbr_dir + '*MCMC'+mcmc_id+'_' + n_iter + '_'+ pixel+'*.cbr')
files.sort()
best_chains = conv_chains_pkl.loc[conv_chains_pkl['pixel']==pixel]['bestchains'].values[0][1:]
# run through cbrs
cbr_chain_list = []
for chain in best_chains:
print(chain)
# read cbr for one file and transform Bday, Fday
file = [i for i in files if pixel+'_'+chain+'.cbr' in i][0]
cbr_chain = autil.modulus_Bday_Fday(rwb.read_cbr_file(file, {'nopars': len(parnames)}), parnames)
print(cbr_chain.shape)
# read forward run for that cbr
flux_chain = rwb.readbinarymat(output_dir + 'fluxfile_' + file.partition(cbr_dir)[-1][:-3]+'bin', [cbf_pixel['nodays'], autil.get_nofluxes_nopools_lma(model_id)[0]])
pool_chain = rwb.readbinarymat(output_dir + 'poolfile_' + file.partition(cbr_dir)[-1][:-3]+'bin', [cbf_pixel['nodays']+1, autil.get_nofluxes_nopools_lma(model_id)[1]])
# add chain to list for GR calculation
if np.shape(cbr_chain)[0]==ens_size:
cbr_chain_list.append(cbr_chain)
# add forward run chain to aggregated matrix
flux_pixel = np.copy(flux_chain) if best_chains.index(chain)==0 else np.concatenate((flux_pixel, flux_chain), axis=0)
pool_pixel = np.copy(pool_chain) if best_chains.index(chain)==0 else np.concatenate((pool_pixel, pool_chain), axis=0)
# compute gelman rubin
if len(cbr_chain_list)>1:
gr = autil.gelman_rubin(cbr_chain_list)
print('%i of %i parameters converged' % (sum(gr<1.2), len(parnames)))
else:
gr = np.nan
cbr_pixel = np.vstack(cbr_chain_list)
print(pool_pixel.shape)
print(cbr_pixel.shape)
# nbe, lai, and abgb predictions at pixel
# list with elements corresponding to MCMCIDs considered (e.g. first element is MCMCID 119)
nbe_pred.append(autil.get_output('NBE', model_id, flux_pixel, pool_pixel, cbr_pixel, autil.get_nofluxes_nopools_lma(model_id)[2]))
lai_pred.append(autil.get_output('LAI', model_id, flux_pixel, pool_pixel, cbr_pixel, autil.get_nofluxes_nopools_lma(model_id)[2])[:,:-1])
abgb_pred.append(autil.get_output('ABGB', model_id, flux_pixel, pool_pixel, cbr_pixel, autil.get_nofluxes_nopools_lma(model_id)[2])[:,:-1])
gpp_pred.append(autil.get_output('GPP', model_id, flux_pixel, pool_pixel, cbr_pixel, autil.get_nofluxes_nopools_lma(model_id)[2]))
# plot time series
lbls = [model_id+'_MCMC'+mcmc_id for model_id, mcmc_id in zip(model_ids, mcmc_ids)]
plot_output_ts(cbf_pixel, nbe_pred, nbe_obs, nbe_an_unc, lbls=lbls, var='NBE', savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models'+'_NBE_'+pixel+'.png')
plot_output_ts(cbf_pixel, lai_pred, lai_obs, lai_unc, lbls=lbls, var='LAI', savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models'+'_LAI_'+pixel+'.png')
plot_output_ts(cbf_pixel, gpp_pred, sif_obs, 0, lbls=lbls, var='GPP', savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models'+'_GPP_'+pixel+'.png')
# plot box plots
plot_dist_compare(nbe_pred, nbe_obs, [nbe_an_unc, nbe_seas_unc], lbls=lbls, var='NBE', savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models'+'_NBE_'+pixel+'_dist_')
plot_dist_compare(lai_pred, lai_obs, lai_unc, lbls=lbls, var='LAI', savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models'+'_LAI_'+pixel+'_dist_')
plot_dist_compare(abgb_pred, abgb_obs, abgb_unc, lbls=lbls, var='ABGB', savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models'+'_ABGB_'+pixel+'_dist_')
# plot obs vs median comparison
nbe_mae.append([mae_real_numbers_only(f, nbe_obs)[0] for f in nbe_pred])
lai_mae.append([mae_real_numbers_only(f, lai_obs)[0] for f in lai_pred])
abgb_mae.append([mae_real_numbers_only(f, abgb_obs)[0] for f in abgb_pred])
print(rank_mae(nbe_mae, lbls))
print(rank_mae(lai_mae, lbls))
print(rank_mae(abgb_mae, lbls))
plot_maes(nbe_mae, pixels, savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models_NBE_mae')
plot_maes(lai_mae, pixels, savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models_LAI_mae')
plot_maes(abgb_mae, pixels, savepath=cur_dir+plot_dir+'demcmc_mcmc/', title='all_models_ABGB_mae')
return
def main():
cur_dir = os.getcwd() + '/'
plot_dir = '../../../../../../scratch/users/cfamigli/cardamom/plots/'
os.chdir(plot_dir + 'dists/')
# get list of model ids
models_full = list(set([el.split('_')[0] for el in glob.glob('*.png')]))
# remove 101, temporary until 102-->101
models_full.remove('102')
os.chdir(cur_dir)
# set lists of variables and pixels
vrs = [
'NBE', 'cumNBE', 'LAI', 'GPP', 'Reco', 'Rauto', 'Rhet', 'lit', 'root',
'som', 'wood'
]
pixels = [
'3809', '3524', '2224', '4170', '1945', '3813', '4054', '3264', '1271',
'3457'
]
# set MCMC ID
mcmc_id = sys.argv[1]
n_iter = sys.argv[2]
assim_type = '_longadapted'
nmodels_leave_out = sys.argv[3]
models = random.sample(models_full,
len(models_full) - int(nmodels_leave_out))
print(models)
# dataframe will hold model structural uncertainty (Ms) and model parametric uncertainty (Mp) for each pixel-var combination
# n is number of models that make up the suite
partitioning = DataFrame(columns={'Ms', 'Mp', 'n'})
df_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/processed_df/'
for var in vrs:
print('Variable: ' + var)
Mp_pixels = np.zeros(
len(pixels)) * np.nan # list of Mp for each pixel, for mapping
for pixel in pixels:
print('Pixel: ' + pixel)
nsteps = 228 if assim_type == '_longadapted' else 240
meds, ub, lb = np.zeros((len(models), nsteps)) * np.nan, np.zeros(
(len(models), nsteps)
) * np.nan, np.zeros(
(len(models), nsteps)
) * np.nan # medians, upper bounds, lower bounds of prediction through time
Mp, n = 0, 0
for model in models:
print(model)
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf' + assim_type + '/' + model + '/'
cbr_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbr' + assim_type + '/' + model + '/'
output_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '/' + model + '/'
parnames = autil.get_parnames(cur_dir + '../../misc/', model)
os.chdir(cur_dir + cbr_dir)
#files = set(glob.glob('*.cbr')) - set(glob.glob('*MCMC'+mcmc_id+'*.cbr'))
#files = glob.glob('*MCMC'+mcmc_id+'*.cbr')
files = set(
glob.glob('*MCMC' + mcmc_id + '_' + n_iter + '_*.cbr'))
pixel_chains = autil.find_all_chains(
files, pixel
) # list of files corresponding to each chain at that pixel, e.g. 2224_1, 2224_2, 2224_3, 2222_4
pixel_chains.sort()
n_chains = len(pixel_chains)
if n_chains > 0:
cbf_pixel = rwb.read_cbf_file(
cur_dir + cbf_dir +
pixel_chains[0].partition('_MCMC')[0] + '_' + pixel +
'.cbf')
cbr_chain_list = []
for pixel_chain in pixel_chains:
print(pixel_chain)
cbr_chain = rwb.read_cbr_file(
pixel_chain, {'nopars': len(parnames)
}) # cbr file for one chain
cbr_chain_list.append(
cbr_chain
) # list of separate cbrs for each chain, use for gelman rubin
cbr_pixel = np.copy(cbr_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(cbr_pixel, cbr_chain),
axis=0) # concatenate all chain cbrs
#autil.plot_par_histograms(cbr_chain, parnames=parnames, savepath=cur_dir+plot_dir+'dists/', title=model+'_'+pixel_chain[:-3]+'png')
flux_chain = rwb.readbinarymat(
cur_dir + output_dir + 'fluxfile_' +
pixel_chain[:-3] + 'bin', [
cbf_pixel['nodays'],
autil.get_nofluxes_nopools_lma(model)[0]
])
pool_chain = rwb.readbinarymat(
cur_dir + output_dir + 'poolfile_' +
pixel_chain[:-3] + 'bin', [
cbf_pixel['nodays'] + 1,
autil.get_nofluxes_nopools_lma(model)[1]
])
#autil.plot_flux_pool_timeseries(cbf_pixel, cbr_chain, flux_chain, pool_chain, autil.get_nofluxes_nopools_lma(model)[2], savepath=cur_dir+plot_dir+'timeseries/', title=model+'_'+pixel_chain[:-3]+'png')
flux_pixel = np.copy(flux_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(flux_pixel, flux_chain),
axis=0) # concatenate all chain flux outputs
pool_pixel = np.copy(pool_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(pool_pixel, pool_chain),
axis=0) # concatenate all chain pool outputs
gr = autil.gelman_rubin(
cbr_chain_list) # gelman rubin function from matt
gr_thresh = 1.2 # below this value parameters are assumed to be convergent
print('%i of %i parameters converged with GR<%.1f' %
(sum(gr < gr_thresh), len(parnames), gr_thresh))
#autil.plot_par_histograms(cbr_pixel, parnames=parnames, savepath=cur_dir+plot_dir+'dists/', title=model+'_'+pixel_chain[:-6]+'.png')
#autil.plot_flux_pool_timeseries(cbf_pixel, cbr_pixel, flux_pixel, pool_pixel, autil.get_nofluxes_nopools_lma(model)[2], savepath=cur_dir+plot_dir+'timeseries/', title=model+'_'+pixel_chain[:-6]+'.png')
if (sum(gr < gr_thresh) / len(parnames) < .9
): # don't include nonconvergent runs in analysis
continue
else:
fwd_data = autil.get_output(
var, model, flux_pixel, pool_pixel, cbr_pixel,
autil.get_nofluxes_nopools_lma(
model)[2]) # get forward data for var
if len(fwd_data) > 0:
if fwd_data.shape[1] > nsteps:
fwd_data = fwd_data[:, :-1]
fwd_data = autil.remove_outliers(fwd_data)
# fill medians, upper bounds, and lower bounds
meds[models.index(model), :] = np.nanmedian(
fwd_data, axis=0)
ub[models.index(model), :] = np.nanpercentile(
fwd_data, 75, axis=0)
lb[models.index(model), :] = np.nanpercentile(
fwd_data, 25, axis=0)
fwd_data = autil.remove_below_25_above_75(
fwd_data
) # set values outside of 25th-75th range to nan
Mp += np.nanvar(
fwd_data, axis=0
) # sum of intra-ensemble variance, only compute on 25th-75th
n += 1
Ms = np.nanvar(meds, axis=0) # inter-median variance
Mp = Mp / n if n != 0 else float('nan')
Ms_div_sum = Ms / (Ms + Mp)
Mp_div_sum = Mp / (Ms + Mp)
partitioning.loc[pixel + '_' + var] = {
'Ms': np.nanmean(Ms_div_sum),
'Mp': np.nanmean(Mp_div_sum),
'n': n
}
Mp_pixels[pixels.index(pixel)] = np.nanmean(Mp_div_sum)
print(partitioning.to_string())
partitioning.sort_index(
axis=1).to_pickle(cur_dir + df_dir + 'summary' + assim_type + '_MCMC' +
mcmc_id + '_' + date.today().strftime("%m%d%y") +
'_' + str(len(models)) + '.pkl')
return
def main():
combinations = [['811', '119', '40000000'], ['811', '3', '1000000'],
['911', '119', '40000000']]
assim_type = '_longadapted'
metric = sys.argv[1]
vrs = [
'NBE', 'cumNBE', 'LAI', 'GPP', 'Reco', 'Rauto', 'Rhet', 'lit', 'root',
'som', 'wood'
]
pixels = [
'3809', '3524', '2224', '4170', '1945', '3813', '4054', '3264', '1271',
'3457'
]
ens_spread = np.ones(
(len(pixels), len(vrs), len(combinations))) * float('nan')
conv = np.ones((len(pixels), len(combinations))) * float('nan')
cur_dir = os.getcwd() + '/'
for pixel in pixels:
comb_count = 0
for comb in combinations:
model_id = comb[0]
mcmc_id = comb[1]
it = comb[2]
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf' + assim_type + '/' + model_id + '/'
cbr_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbr' + assim_type + '/' + model_id + '/'
output_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '/' + model_id + '/'
plot_dir = '../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames(cur_dir + '../../misc/', model_id)
os.chdir(cur_dir + cbr_dir)
files = glob.glob('*MCMC' + mcmc_id + '_' + it + '_' + pixel +
'*.cbr')
pixel_chains = autil.find_all_chains(files, pixel)
pixel_chains.sort() # filenames
if model_id == '911': pixel_chains = pixel_chains[-4:]
print(pixel_chains)
cbf_pixel = rwb.read_cbf_file(
cur_dir + cbf_dir + pixel_chains[0].partition('_MCMC')[0] +
'_' + pixel + '.cbf')
cbr_chain_list = []
for pixel_chain in pixel_chains:
print(pixel_chain)
cbr_chain = rwb.read_cbr_file(pixel_chain,
{'nopars': len(parnames)})
cbr_pixel = np.copy(cbr_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(cbr_pixel, cbr_chain), axis=0)
flux_chain = rwb.readbinarymat(
cur_dir + output_dir + 'fluxfile_' + pixel_chain[:-3] +
'bin', [
cbf_pixel['nodays'],
autil.get_nofluxes_nopools_lma(model_id)[0]
])
pool_chain = rwb.readbinarymat(
cur_dir + output_dir + 'poolfile_' + pixel_chain[:-3] +
'bin', [
cbf_pixel['nodays'] + 1,
autil.get_nofluxes_nopools_lma(model_id)[1]
])
flux_pixel = np.copy(flux_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(flux_pixel, flux_chain), axis=0)
pool_pixel = np.copy(pool_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(pool_pixel, pool_chain), axis=0)
cbr_chain_list.append(cbr_chain)
print(np.shape(cbr_chain))
print(np.shape(cbr_pixel))
gr = autil.gelman_rubin(cbr_chain_list)
print('%i of %i parameters converged' %
(sum(gr < 1.2), len(parnames)))
conv[pixels.index(pixel),
comb_count] = sum(gr < 1.2) / len(parnames) * 100
for var in vrs:
print(var)
try:
obs = cbf_pixel['OBS'][var]
obs[obs == -9999] = float('nan')
except:
obs = np.ones(cbf_pixel['nodays']) * np.nan
n_obs = np.sum(np.isfinite(obs))
fwd_data = autil.get_output(
var, model_id, flux_pixel, pool_pixel, cbr_pixel,
autil.get_nofluxes_nopools_lma(model_id)[2])
if len(fwd_data) > 0:
if fwd_data.shape[1] > cbf_pixel['nodays']:
fwd_data = fwd_data[:, :-1]
fwd_data = autil.remove_outliers(fwd_data)
med = np.nanmedian(fwd_data, axis=0)
ub = np.nanpercentile(fwd_data, 75, axis=0)
lb = np.nanpercentile(fwd_data, 25, axis=0)
ens_spread[pixels.index(pixel),
vrs.index(var), comb_count] = np.nanmean(
abs(ub -
lb)) if metric == 'spread' else np.sqrt(
np.nansum((med - obs)**2) / n_obs)
comb_count += 1
for var in vrs:
autil.plot_spread_v_iter(
ens_spread,
pixels,
vrs.index(var),
var,
it,
metric,
cur_dir + plot_dir + 'spread_v_iter',
'iter_test_compare_' + assim_type + '_' + model_id + '_' + var +
'_' + metric,
single_val=True
) #'iter_test_MCMC'+mcmc_id+'_'+model_id+'_'+var + '_' + metric)
autil.plot_conv_v_iter(conv,
pixels,
it,
cur_dir + plot_dir + 'spread_v_iter',
'iter_test_compare' + assim_type + '_' + model_id +
'_conv',
single_val=True)
return
def main():
# set run information to read
model_id = sys.argv[1]
mcmc_id = sys.argv[2] # 119 for normal, 3 for DEMCMC
n_iter = sys.argv[3]
nbe_optimization = sys.argv[4] # 'OFF' or 'ON'
ens_size = 500
assim_type = sys.argv[5]
# set directories
cur_dir = os.getcwd() + '/'
misc_dir = cur_dir + '../../misc/'
cbf_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/cbf' + assim_type + '/' + model_id + '/'
cbr_opt_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/cbr' + assim_type + '/' + model_id + '/'
cbr_ef_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/cbr' + assim_type + '_ef/' + model_id + '/'
cbr_pft_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/cbr' + assim_type + '_pft/' + model_id + '/'
output_opt_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '/' + model_id + '/'
output_ef_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '_ef/' + model_id + '/'
output_pft_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '_pft/' + model_id + '/'
plot_dir = cur_dir + '../../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames('../../misc/', model_id)
# get list of cbfs
os.chdir(cbf_dir)
cbf_files = glob.glob('*.cbf')
cbf_files.sort()
pixel_lst = []
os.chdir(cur_dir + '/../')
# initialize lists for error maps
card_unc, opt_obs_err, pft_obs_err, ef_obs_err, obs_std = np.zeros(
len(cbf_files)) * np.nan, np.zeros(len(cbf_files)) * np.nan, np.zeros(
len(cbf_files)) * np.nan, np.zeros(
len(cbf_files)) * np.nan, np.zeros(len(cbf_files)) * np.nan
opt_pft_trend, opt_ef_trend, opt_pft_seas, opt_ef_seas, opt_mean, pft_mean, ef_mean = np.zeros(
len(cbf_files)) * np.nan, np.zeros(len(cbf_files)) * np.nan, np.zeros(
len(cbf_files)) * np.nan, np.zeros(
len(cbf_files)) * np.nan, np.zeros(
len(cbf_files)) * np.nan, np.zeros(
len(cbf_files)) * np.nan, np.zeros(
len(cbf_files)) * np.nan
pft_mean_within_opt_unc, ef_mean_within_opt_unc = np.zeros(
len(cbf_files)) * np.nan, np.zeros(len(cbf_files)) * np.nan
################################################## iterate through pixels ##################################################
############################################################################################################################
include_ef = True
include_pft = True
include_opt = True
write_txt_sh_pft_rerun = True
# initialize
n_fluxes = autil.get_nofluxes_nopools_lma(model_id)[0]
n_pools = autil.get_nofluxes_nopools_lma(model_id)[1]
# load list of globcover labels
gl_lbls = list(
read_csv(misc_dir + 'Globcover2009_Legend.csv')['Value'].values)
n_classes = len(gl_lbls)
# load globcover csv for av_fracs determination
gl_fracs = read_csv(misc_dir + 'globcover_fracs.csv', header=0)
# load bestchains for cbr_files
conv_chains = read_pickle(cbr_opt_dir + model_id + assim_type + '_ALL' +
'_MCMC' + mcmc_id + '_' + n_iter +
'_best_subset.pkl')
conv_chains.columns = ['pixel', 'bestchains',
'conv'] #rename columns for easier access
# create csv to track pft reruns
pft_rerun_filename = 'pft_rerun_' + model_id + assim_type + '_MCMC' + mcmc_id + '_' + n_iter + '.csv'
pft_rerun = open(misc_dir + pft_rerun_filename, 'w')
w = csv.writer(pft_rerun)
# run through all pixels
for cbf_file in cbf_files:
ind = cbf_files.index(cbf_file)
pixel = cbf_file[-8:-4]
pixel_lst.append(pixel)
print(pixel)
# read in fracs and types for pixel
if int(pixel) in gl_fracs['pixel'].values:
fracs_at_pixel = gl_fracs.loc[gl_fracs['pixel'] == int(
pixel)].values[0][1:]
types_at_pixel = get_types_at_pixel(gl_fracs, pixel)
else:
fracs_at_pixel = np.zeros(len(gl_lbls))
types_at_pixel = []
# read in cbf
cbf_pixel = rwb.read_cbf_file(cbf_dir + cbf_file)
nsteps = cbf_pixel['nodays']
################################################## get PFT forward runs ##################################################
##########################################################################################################################
can_plot_pft = False
if include_pft:
pixel_rerun = []
pft_spec = '5rp_'
# initialize matrices to hold weighted average of fluxes and pools
flux_pft_pixel = np.zeros((1, nsteps, n_fluxes))
pool_pft_pixel = np.zeros((1, nsteps + 1, n_pools))
#flux_pft_pixel = np.zeros((ens_size, nsteps, n_fluxes))
#pool_pft_pixel = np.zeros((ens_size, nsteps+1, n_pools))
# read all forward runs (each pft's run) for a given pixel
print(types_at_pixel)
for pft in types_at_pixel:
suffix = cbf_file[:-9] + '_MCMC' + mcmc_id + '_' + n_iter + '_PFT' + str(
int(pft)
) + '_forward_' + pixel + '.bin' #cbf_file[:-8]+'MCMC'+mcmc_id+'_'+n_iter+'_PFT'+str(int(pft))+'_'+pixel+'.bin'
if (len(glob.glob(output_pft_dir + 'fluxfile_' + suffix)) > 0
) & (len(glob.glob(output_pft_dir + 'poolfile_' + suffix))
> 0):
print(str(int(pft)))
flux_pft = rwb.readbinarymat(
output_pft_dir + 'fluxfile_' + suffix,
[nsteps, n_fluxes])
pool_pft = rwb.readbinarymat(
output_pft_dir + 'poolfile_' + suffix,
[nsteps + 1, n_pools])
#autil.plot_general_timeseries(autil.get_output('NBE', model_id, flux_pft, pool_pft, cbr_data=[], lma_ind=autil.get_nofluxes_nopools_lma(model_id)[2]), 'NBE', cbf_pixel, plot_dir+'timeseries/pft/', model_id + '_MCMC'+mcmc_id + '_'+n_iter + '_' + pixel + '_'+str(int(pft))+'.png')
# add each flux and pool matrix (corresponding to each pft) according to pft fractions, as weighted average
flux_pft[np.isnan(flux_pft)] = 0.
pool_pft[np.isnan(pool_pft)] = 0.
if (flux_pft.shape[0] > 0) & (pool_pft.shape[0] > 0):
lbl_ind = gl_lbls.index(int(pft))
flux_pft_pixel += flux_pft * fracs_at_pixel[lbl_ind]
pool_pft_pixel += pool_pft * fracs_at_pixel[lbl_ind]
can_plot_pft = True
else:
pixel_rerun.append(pft)
else:
pixel_rerun.append(pft)
if len(pixel_rerun) > 0:
w.writerow([pixel] + pixel_rerun)
################################################ get optimal forward runs ################################################
##########################################################################################################################
can_plot_opt = False
if include_opt:
# get pixel's convergent chain numbers
if pixel in conv_chains['pixel'].values:
best_chains = conv_chains.loc[
conv_chains['pixel'] == pixel]['bestchains'].values[0][1:]
flux_opt, pool_opt = [], []
# aggregate best chain outputs into one list
for chain in best_chains:
suffix = cbf_file[:
-8] + 'MCMC' + mcmc_id + '_' + n_iter + '_' + pixel + '_' + chain + '.bin'
if (len(glob.glob(output_opt_dir + 'fluxfile_' + suffix)) >
0) & (len(
glob.glob(output_opt_dir + 'poolfile_' +
suffix)) > 0):
flux_opt.append(
rwb.readbinarymat(
output_opt_dir + 'fluxfile_' + suffix,
[nsteps, n_fluxes]))
pool_opt.append(
rwb.readbinarymat(
output_opt_dir + 'poolfile_' + suffix,
[nsteps + 1, n_pools]))
can_plot_opt = True
# stack list elements for plotting
flux_opt = np.vstack(flux_opt)
pool_opt = np.vstack(pool_opt)
################################################### get EF forward runs ###################################################
###########################################################################################################################
can_plot_ef = False
if include_ef:
ef_spec = 'clipped_PLS_soilgrids_poolobs_rescaled_forward_'
# if 'wpolys' in ef_spec: use '_MCMC'
# else: use 'MCMC'
suffix = cbf_file[:
-9] + '_MCMC' + mcmc_id + '_' + n_iter + '_' + ef_spec + pixel + '.bin' #cbf_file[:-8]+'MCMC'+mcmc_id+'_'+n_iter+'_EF_'+pixel+'.bin'
if (len(glob.glob(output_ef_dir + 'fluxfile_' + suffix)) > 0) & (
len(glob.glob(output_ef_dir + 'poolfile_' + suffix)) > 0):
flux_ef = rwb.readbinarymat(
output_ef_dir + 'fluxfile_' + suffix, [nsteps, n_fluxes])
pool_ef = rwb.readbinarymat(
output_ef_dir + 'poolfile_' + suffix,
[nsteps + 1, n_pools])
can_plot_ef = True
##################################################### plot and compare ####################################################
###########################################################################################################################
can_decompose = True if (can_plot_opt) & (can_plot_pft) & (
can_plot_ef) else False
# plot optimal and pft predictions together
output_opt = autil.get_output(
'NBE',
model_id,
flux_opt,
pool_opt,
cbr_data=[],
lma_ind=autil.get_nofluxes_nopools_lma(model_id)[2]) if (
include_opt) & (can_plot_opt) else np.ones(nsteps) * np.nan
output_pft = autil.get_output(
'NBE',
model_id,
flux_pft_pixel,
pool_pft_pixel,
cbr_data=[],
lma_ind=autil.get_nofluxes_nopools_lma(model_id)[2]) if (
include_pft) & (can_plot_pft) else np.ones(nsteps) * np.nan
output_ef = autil.get_output(
'NBE',
model_id,
flux_ef,
pool_ef,
cbr_data=[],
lma_ind=autil.get_nofluxes_nopools_lma(model_id)
[2]) if (include_ef) & (can_plot_ef) else np.ones(nsteps) * np.nan
card_unc[ind], opt_obs_err[ind], pft_obs_err[ind], ef_obs_err[
ind], obs_std[ind] = autil.plot_opt_pft_ef_timeseries(
output_opt,
output_pft,
output_ef,
'NBE',
pixel,
autil.rowcol_to_latlon([pixel]),
cbf_pixel,
err_v_obs=False,
savepath=plot_dir + 'forward_compare/timeseries/' + model_id +
'/',
title=model_id + '_MCMC' + mcmc_id + '_' + n_iter + '_' +
pft_spec + ef_spec + pixel + '.png')
if can_decompose:
opt_pft_trend[ind], opt_ef_trend[ind], opt_pft_seas[
ind], opt_ef_seas[ind], opt_mean[ind], pft_mean[ind], ef_mean[
ind], pft_mean_within_opt_unc[ind], ef_mean_within_opt_unc[
ind] = timeseries_decompose(
output_opt,
output_pft,
output_ef,
pixel,
savepath=plot_dir + 'forward_compare/decomp/' +
model_id + '/',
savename=model_id + '_MCMC' + mcmc_id + '_' +
n_iter + '_' + pft_spec + ef_spec + pixel)
# close csv for rerun tracking
pft_rerun.close()
# plot decomposition results
plot_decomposed(
[opt_pft_trend, opt_ef_trend], [opt_pft_seas, opt_ef_seas],
[opt_mean, pft_mean, ef_mean],
[pft_mean_within_opt_unc, ef_mean_within_opt_unc],
savepath=plot_dir + 'forward_compare/decomp/' + model_id + '/',
savename=model_id + '_MCMC' + mcmc_id + '_' + n_iter + '_' + pft_spec +
ef_spec)
# plot error maps
for data, plot_title, vmin, vmax in zip(
[
card_unc, opt_obs_err, pft_obs_err, ef_obs_err, obs_std,
opt_obs_err / obs_std, pft_obs_err / obs_std, ef_obs_err / obs_std,
pft_obs_err / obs_std - opt_obs_err / obs_std,
ef_obs_err / obs_std - opt_obs_err / obs_std,
pft_obs_err / obs_std - ef_obs_err / obs_std
], [
'opt_unc', 'opt_err', 'pft_err', 'ef_err', 'obs_std',
'norm_opt_err', 'norm_pft_err', 'norm_ef_err',
'norm_pft_minus_norm_opt_err', 'norm_ef_minus_norm_opt_err',
'norm_pft_minus_norm_ef_err'
], [0., 0., 0., 0., 0., 0., 0., 0., -1., -1., -1.],
[0.7, 0.7, 0.7, 0.7, 0., 2., 2., 2., 1., 1., 1.]):
data_nonan, pixel_lst_nonan = remove_nan(data, pixel_lst)
stipple = card_unc if (plot_title == 'ef_err') | (
plot_title == 'pft_err') else None
autil.plot_map(
nrows=46,
ncols=73,
land_pixel_list=[file[-8:-4] for file in cbf_files],
pixel_value_list=pixel_lst_nonan,
value_list=data_nonan,
vmin=vmin,
vmax=vmax,
cmap='bwr',
savepath=plot_dir + 'forward_compare/maps/' + model_id + '/',
savename=model_id + '_MCMC' + mcmc_id + '_' + n_iter + '_' +
pft_spec + ef_spec + plot_title,
stipple=stipple) #vmax=np.nanpercentile(data_nonan, 90)
# save errors for comparison analysis
DataFrame(list(zip(pixel_lst, list(ef_obs_err / obs_std))),
columns=[
'pixels', 'norm_mae'
]).to_pickle(misc_dir + 'mae_pkls/' + model_id + '_MCMC' +
mcmc_id + '_' + n_iter + '_' + ef_spec + '.pkl')
DataFrame(list(zip(pixel_lst, list(pft_obs_err / obs_std))),
columns=[
'pixels', 'norm_mae'
]).to_pickle(misc_dir + 'mae_pkls/' + model_id + '_MCMC' +
mcmc_id + '_' + n_iter + '_' + pft_spec + '.pkl')
# plot discrete map showing best parameterization (lowest error) for each pixel
'''best_param_nonan, pixel_lst_nonan = best_param_nonancol([opt_obs_err, pft_obs_err, ef_obs_err], pixel_lst)
autil.plot_map(nrows=46, ncols=73, land_pixel_list=[file[-8:-4] for file in cbf_files], pixel_value_list=pixel_lst_nonan, value_list=best_param_nonan, cmap=LinearSegmentedColormap.from_list('mycmap', [(0, 'dodgerblue'), (0.5, 'orangered'), (1., 'limegreen')]),savepath=plot_dir+'forward_compare/maps/'+model_id+'/', savename=model_id+'_MCMC'+mcmc_id+'_'+n_iter+'_'+ef_spec+'best_param')'''
best_param_nonan, pixel_lst_nonan = best_param_nonancol(
[pft_obs_err, ef_obs_err], pixel_lst)
autil.plot_map(
nrows=46,
ncols=73,
land_pixel_list=[file[-8:-4] for file in cbf_files],
pixel_value_list=pixel_lst_nonan,
value_list=best_param_nonan,
cmap=LinearSegmentedColormap.from_list('mycmap', [(0, 'orangered'),
(1., 'limegreen')]),
savepath=plot_dir + 'forward_compare/maps/' + model_id + '/',
savename=model_id + '_MCMC' + mcmc_id + '_' + n_iter + '_' + pft_spec +
ef_spec + 'best_param')
rgb_triplets = err_rgb_triplets([opt_obs_err, pft_obs_err, ef_obs_err],
pixel_lst)
autil.plot_map_rgb(
nrows=46,
ncols=73,
land_pixel_list=[file[-8:-4] for file in cbf_files],
pixel_value_list=pixel_lst,
value_list=rgb_triplets,
savepath=plot_dir + 'forward_compare/maps/' + model_id + '/',
savename=model_id + '_MCMC' + mcmc_id + '_' + n_iter + '_' + pft_spec +
ef_spec + 'rgb')
############################################### create resubmission for pft ###############################################
###########################################################################################################################
if write_txt_sh_pft_rerun:
# set additional directories
mdf_dir = '../code/CARDAMOM_2.1.6c/C/projects/CARDAMOM_MDF/' if nbe_optimization == 'OFF' else '../code/CARDAMOM_Uma_2.1.6c-master/C/projects/CARDAMOM_MDF/'
runmodel_dir = '../code/CARDAMOM_2.1.6c/C/projects/CARDAMOM_GENERAL/' if nbe_optimization == 'OFF' else '../code/CARDAMOM_Uma_2.1.6c-master/C/projects/CARDAMOM_GENERAL/'
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf' + assim_type + '/' + model_id + '/'
cbf_pft_ic_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf' + assim_type + '_pft_ic/' + model_id + '/'
cbr_pft_dir = '../../../../../scratch/users/cfamigli/cardamom/files/cbr' + assim_type + '_pft/' + model_id + '/'
output_dir = '../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '/' + model_id + '/'
output_pft_dir = '../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '_pft/' + model_id + '/'
if mcmc_id == '119':
frac_save_out = str(int(int(n_iter) / 500))
elif mcmc_id == '3':
frac_save_out = str(
int(int(n_iter) / 500 * 100)
) # n_iterations/ frac_save_out * 100 will be ensemble size
# set up which files to rerun
pft_rerun = read_csv(misc_dir + pft_rerun_filename,
header=None,
sep=',',
names=['pixel'] + gl_lbls)
txt_filename = 'pft_ic_combined_list_' + model_id + assim_type + '_MCMC' + mcmc_id + '_' + n_iter + '_rerun.txt'
txt_file = open(txt_filename, 'w')
cl_count, row_count = 1, 0
for cbf_file in cbf_files:
pixel = cbf_file[-8:-4]
if int(pixel) in pft_rerun['pixel'].values:
pixel_classes = pft_rerun.loc[pft_rerun['pixel'] == int(
pixel)].values[0][1:]
for cl in pixel_classes:
if ~np.isnan(cl):
f = cbf_file[:-9] + '_PFT' + str(int(cl)) + '_' + pixel
txt_file.write(
'%sCARDAMOM_MDF.exe %s%s %s%s %s 0 %s 0.001 %s 1000'
% (mdf_dir, cbf_pft_ic_dir[3:], f + '.cbf',
cbr_pft_dir, f + '.cbr', n_iter, frac_save_out,
mcmc_id))
txt_file.write(
' && %sCARDAMOM_RUN_MODEL.exe %s%s %s%s %s%s %s%s %s%s %s%s'
% (runmodel_dir, cbf_pft_ic_dir[3:], f + '.cbf',
cbr_pft_dir, f + '.cbr', output_pft_dir,
'fluxfile_' + f + '.bin', output_pft_dir,
'poolfile_' + f + '.bin', output_pft_dir,
'edcdfile_' + f + '.bin', output_pft_dir,
'probfile_' + f + '.bin'))
cl_count += 1
if np.mod(cl_count, 5) == 0:
txt_file.write('\n')
row_count += 1
else:
txt_file.write(' && ')
txt_file.close()
sh_file = open(txt_filename[:-3] + 'sh', 'w')
autil.fill_in_sh(sh_file,
array_size=row_count,
n_hours=10,
txt_file=txt_filename,
combined=True)
return
def main():
### set specifications
model_id = sys.argv[1]
run_type = 'ALL'
mcmc_id = '119'
n_iter = '40000000'
ens_size = 500
assim_type = '_longadapted'
### set directories
cur_dir = os.getcwd() + '/'
cbf_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/cbf'+assim_type+'/' + model_id + '/'
cbr_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/cbr'+assim_type+'/' + model_id + '/'
output_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/files/output'+assim_type+'/' + model_id + '/'
plot_dir = cur_dir + '../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames(cur_dir + '../../misc/', model_id)
# get list of cbfs
os.chdir(cbf_dir)
cbf_files = glob.glob('*.cbf')
cbf_files.sort()
os.chdir(cur_dir)
# initialize lists of pixel names and rmses
pixels_plot = []
nbe_rmse, lai_rmse = [], []
for cbf_file in cbf_files:
print(cbf_file, cbf_files.index(cbf_file))
cbf_pixel = rwb.read_cbf_file(cbf_dir + cbf_file)
pixel = cbf_file[-8:-4]
cbr_files = glob.glob(cbr_dir + '*MCMC'+mcmc_id+'_'+n_iter+'_' + pixel + '_*.cbr')
cbr_files = sorted(cbr_files, key=lambda x: int(x.partition(pixel+'_')[-1].partition('.cbr')[0]))
# get all possible XX member combinations of cbr files
n_chains_to_converge = 4
cbr_files_all_subsets = [list(i) for i in itertools.combinations(cbr_files, n_chains_to_converge)]
continue_check = True
for subset in cbr_files_all_subsets:
if continue_check:
# read parameters and compute gelman rubin
cbr_chain_list = []
for cbr_file in subset:
cbr_chain = rwb.read_cbr_file(cbr_file, {'nopars': len(parnames)})
cbr_chain = autil.modulus_Bday_Fday(cbr_chain, parnames)
if np.shape(cbr_chain)[0]==ens_size:
cbr_chain_list.append(cbr_chain)
if len(cbr_chain_list)>1:
gr = autil.gelman_rubin(cbr_chain_list)
if sum(gr<1.2)/len(parnames)>=0.9:
continue_check = False
cbr_agg = np.vstack(cbr_chain_list)
pixels_plot.append(pixel)
best_subset = subset.copy()
else:
gr = np.nan
# if there is a convergent subset, read fluxes and pools
if not continue_check:
convergent_chain_nums = [el.partition('.cbr')[0].partition(pixel)[-1][1:] for el in best_subset]
convergent_files = [el.partition('.cbr')[0].partition(model_id+'/')[-1] for el in best_subset]
flux_pixel = []
pool_pixel = []
for filename in convergent_files:
flux_chain = rwb.readbinarymat(output_dir + 'fluxfile_' + filename+'.bin', [cbf_pixel['nodays'], autil.get_nofluxes_nopools_lma(model_id)[0]])
pool_chain = rwb.readbinarymat(output_dir + 'poolfile_' + filename+'.bin', [cbf_pixel['nodays']+1, autil.get_nofluxes_nopools_lma(model_id)[1]])
if (flux_chain.shape[0]==ens_size) & (pool_chain.shape[0]==ens_size):
flux_pixel.append(flux_chain)
pool_pixel.append(pool_chain)
nbe_pred = autil.get_output('NBE', model_id, np.vstack(flux_pixel), np.vstack(pool_pixel), cbr_agg, autil.get_nofluxes_nopools_lma(model_id)[2])
lai_pred = autil.get_output('LAI', model_id, np.vstack(flux_pixel), np.vstack(pool_pixel), cbr_agg, autil.get_nofluxes_nopools_lma(model_id)[2])
nbe_obs, lai_obs = cbf_pixel['OBS']['NBE'], cbf_pixel['OBS']['LAI']
nbe_rmse.append(rmse_real_numbers_only(nbe_pred, nbe_obs))
lai_rmse.append(rmse_real_numbers_only(lai_pred, lai_obs))
print(rmse_real_numbers_only(nbe_pred, nbe_obs), rmse_real_numbers_only(lai_pred, lai_obs))
autil.plot_map(nrows=46, ncols=73, land_pixel_list=[file[-8:-4] for file in glob.glob(cbf_dir + '*.cbf')], pixel_value_list=pixels_plot, value_list=nbe_rmse, savepath=plot_dir+'maps/', savename='rmse_nbe_' + model_id + assim_type+ '_MCMC' + mcmc_id + '_' + n_iter)
autil.plot_map(nrows=46, ncols=73, land_pixel_list=[file[-8:-4] for file in glob.glob(cbf_dir + '*.cbf')], pixel_value_list=pixels_plot, value_list=lai_rmse, savepath=plot_dir+'maps/', savename='rmse_lai_' + model_id + assim_type+ '_MCMC' + mcmc_id + '_' + n_iter)
rmse_df = DataFrame(list(zip(pixels_plot, nbe_rmse, lai_rmse)))
rmse_df.columns = ['pixel','nbe_rmse','lai_rmse']
rmse_df.to_pickle(cur_dir + '../../misc/rmse_' + model_id + assim_type+ '_MCMC' + mcmc_id + '_' + n_iter + '.pkl')
#################################################################################################################################################################
# analyze regionally
'''region_mask = Dataset(cur_dir + '../../misc/fourregion_maskarrays.nc')
region_mask.set_auto_mask(False)
regionmat, lat, lon = region_mask['4region'][:], region_mask['lat'][:], region_mask['lon'][:]
lat[0] = -90
lat[-1] = 90
model_ids = ['811', '911']
rmse_dfs = []
for model_id in model_ids:
rmse_df = read_pickle(cur_dir + '../../misc/rmse_' + model_id + assim_type+ '_MCMC' + mcmc_id + '_' + n_iter + '.pkl')
rmse_df.columns = ['pixel','nbe_rmse','lai_rmse']
regions = []
for pixel in rmse_df[rmse_df.columns[0]].tolist():
pixlat, pixlon = rwb.rowcol_to_latlon(pixel)
regions.append(regionmat[np.argwhere(lat==pixlat)[0][0], np.argwhere(lon==pixlon)[0][0]])
rmse_df.insert(loc=1, column='region', value=regions)
rmse_dfs.append(rmse_df)
print(rmse_dfs[0].groupby('region')['nbe_rmse'].mean(), rmse_dfs[0].groupby('region')['lai_rmse'].mean())
print(rmse_dfs[1].groupby('region')['nbe_rmse'].mean(), rmse_dfs[1].groupby('region')['lai_rmse'].mean())'''
return
def main():
model_id = sys.argv[1]
run_type = sys.argv[2] # ALL or SUBSET
mcmc_id = sys.argv[3] # 119 for normal, 3 for DEMCMC
n_iter = sys.argv[4]
var_to_plot = sys.argv[5] # GR, a flux or pool, or PARXX
ens_size = 500
assim_type = '_longadapted'
cur_dir = os.getcwd() + '/'
if 'scripts' not in cur_dir:
cur_dir = cur_dir + 'scripts/'
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf'+assim_type+'/' + model_id + '/'
cbr_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbr'+assim_type+'/' + model_id + '/'
output_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/output'+assim_type+'/' + model_id + '/'
plot_dir = '../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames(cur_dir + '../../misc/', model_id)
os.chdir(cbr_dir)
files = glob.glob('*MCMC'+mcmc_id+'_'+n_iter+'_*.cbr')
pixel = sys.argv[6]
print(pixel)
pixel_chains = autil.find_all_chains(files, pixel)
pixel_chains.sort() # filenames
print(pixel_chains)
cbf_pixel = rwb.read_cbf_file(cur_dir + cbf_dir + pixel_chains[0].partition('_MCMC')[0]+'_'+pixel+'.cbf')
cbr_chain_list = []
for pixel_chain in pixel_chains:
print(pixel_chain)
cbr_chain = rwb.read_cbr_file(pixel_chain, {'nopars': len(parnames)})
cbr_pixel = np.copy(cbr_chain) if pixel_chains.index(pixel_chain)==0 else np.concatenate((cbr_pixel, cbr_chain), axis=0)
flux_chain = rwb.readbinarymat(cur_dir + output_dir + 'fluxfile_' + pixel_chain[:-3]+'bin', [cbf_pixel['nodays'], autil.get_nofluxes_nopools_lma(model_id)[0]])
pool_chain = rwb.readbinarymat(cur_dir + output_dir + 'poolfile_' + pixel_chain[:-3]+'bin', [cbf_pixel['nodays']+1, autil.get_nofluxes_nopools_lma(model_id)[1]])
flux_pixel = np.copy(flux_chain) if pixel_chains.index(pixel_chain)==0 else np.concatenate((flux_pixel, flux_chain), axis=0)
pool_pixel = np.copy(pool_chain) if pixel_chains.index(pixel_chain)==0 else np.concatenate((pool_pixel, pool_chain), axis=0)
if np.shape(cbr_chain)[0]==ens_size:
cbr_chain_list.append(cbr_chain)
print(np.shape(cbr_chain))
### COMPUTE GELMAN RUBIN
if len(cbr_chain_list)>1:
gr = autil.gelman_rubin(cbr_chain_list)
gr_pixel = sum(gr<1.2)/len(parnames)
else:
gr_pixel = -9999.
### DETERMINE DATA TO WRITE TO FILE
if var_to_plot == 'GR':
data = np.copy(gr_pixel)
elif 'PAR' in var_to_plot:
parnum = int(var_to_plot.partition('PAR')[-1])
if gr_pixel>0.9:
data = np.nanmedian(cbr_pixel[:,parnum-1])
else:
data = -9999.
else:
if gr_pixel>0.9:
data = np.nanmean(np.nanmedian(autil.get_output(var_to_plot, model_id, flux_pixel, pool_pixel, cbr_pixel, autil.get_nofluxes_nopools_lma(model_id)[2]), axis=0))
else:
data = -9999.
with open(cur_dir + '../../misc/' + model_id + '_' + pixel_chains[0].partition('_MCMC')[0] + '_MCMC' + mcmc_id + '_' + n_iter + '_' + var_to_plot + '.csv','a') as f:
writer = csv.writer(f)
new_row = [pixel, data]
assert len(new_row)==2
writer.writerow(new_row)
return
def main():
model_id_start = sys.argv[1]
run_type = sys.argv[2] # ALL or SUBSET
metric = sys.argv[3] # spread or RMSE
assim_type = '_p25adapted'
compare_between = sys.argv[4] # MCMCID or MODEL or NBEUNC
n_iters = [
['40000000'], ['40000000']
] #['500000','1000000','2500000','5000000','10000000'],['40000000']]#[['100000', '250000', '500000', '1000000', '1750000', '2500000', '5000000'], ['100000', '250000', '500000', '1000000', '5000000', '10000000', '25000000','50000000']]
vrs = [
'NBE', 'cumNBE', 'LAI', 'GPP', 'Reco', 'Rauto', 'Rhet', 'lit', 'root',
'som', 'wood'
]
pixels = [
'3809', '3524', '2224', '4170', '1945', '3813', '4054', '3264', '1271',
'3457'
]
cur_dir = os.getcwd() + '/'
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf' + assim_type + '/' + model_id_start + '/'
cbr_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbr' + assim_type + '/' + model_id_start + '/'
output_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '/' + model_id_start + '/'
plot_dir = '../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames('../../misc/', model_id_start)
if compare_between == 'MCMCID':
comps = ['3', '119']
elif compare_between == 'MODEL':
comps = [model_id_start, '911']
mcmc_id = '119'
elif compare_between == 'NBEUNC':
comps = [assim_type, '_p25adapted_NBEuncreduced']
mcmc_id = '119'
ens_spread = [
np.ones((len(pixels), len(vrs), len(n_iters[0]))) * float('nan'),
np.ones((len(pixels), len(vrs), len(n_iters[1]))) * float('nan')
]
conv = [
np.ones((len(pixels), len(n_iters[0]))) * float('nan'),
np.ones((len(pixels), len(n_iters[1]))) * float('nan')
]
for pixel in pixels:
for comp in comps:
if compare_between == 'MCMCID':
mcmc_id = comp
elif compare_between == 'MODEL':
model_id_start = comp
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf' + assim_type + '/' + comp + '/'
cbr_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbr' + assim_type + '/' + comp + '/'
output_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/output' + assim_type + '/' + comp + '/'
plot_dir = '../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames(cur_dir + '../../misc/', comp)
elif compare_between == 'NBEUNC':
assim_type = comp
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf' + comp + '/' + model_id_start + '/'
cbr_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbr' + comp + '/' + model_id_start + '/'
output_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/output' + comp + '/' + model_id_start + '/'
plot_dir = '../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames(cur_dir + '../../misc/',
model_id_start)
os.chdir(cur_dir + cbr_dir)
for it in n_iters[comps.index(comp)]:
files = glob.glob('*MCMC' + mcmc_id + '_' + it + '_' + pixel +
'*.cbr')
pixel_chains = autil.find_all_chains(files, pixel)
pixel_chains.sort() # filenames
#if ((comp=='911') & (pixel_chains[0][-5]=='1')): pixel_chains.pop(0)
#if ((comp=='911') & (pixel_chains[0][-5]=='2')): pixel_chains.pop(0)
print(pixel_chains)
cbf_pixel = rwb.read_cbf_file(
cur_dir + cbf_dir + pixel_chains[0].partition('_MCMC')[0] +
'_' + pixel + '.cbf')
cbr_chain_list = []
for pixel_chain in pixel_chains[:4]:
print(pixel_chain)
cbr_chain = rwb.read_cbr_file(pixel_chain,
{'nopars': len(parnames)})
cbr_pixel = np.copy(cbr_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(cbr_pixel, cbr_chain), axis=0)
flux_chain = rwb.readbinarymat(
cur_dir + output_dir + 'fluxfile_' + pixel_chain[:-3] +
'bin', [
cbf_pixel['nodays'],
autil.get_nofluxes_nopools_lma(model_id_start)[0]
])
pool_chain = rwb.readbinarymat(
cur_dir + output_dir + 'poolfile_' + pixel_chain[:-3] +
'bin', [
cbf_pixel['nodays'] + 1,
autil.get_nofluxes_nopools_lma(model_id_start)[1]
])
flux_pixel = np.copy(flux_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(flux_pixel, flux_chain), axis=0)
pool_pixel = np.copy(pool_chain) if pixel_chains.index(
pixel_chain) == 0 else np.concatenate(
(pool_pixel, pool_chain), axis=0)
cbr_chain_list.append(cbr_chain)
print(np.shape(cbr_chain))
print(np.shape(cbr_pixel))
gr = autil.gelman_rubin(cbr_chain_list)
print('%i of %i parameters converged' %
(sum(gr < 1.2), len(parnames)))
conv[comps.index(
comp)][pixels.index(pixel),
n_iters[comps.index(comp)].index(it)] = sum(
gr < 1.2) / len(parnames) * 100
for var in vrs:
print(var)
try:
obs = cbf_pixel['OBS'][var]
obs[obs == -9999] = float('nan')
except:
obs = np.ones(cbf_pixel['nodays']) * np.nan
n_obs = np.sum(np.isfinite(obs))
fwd_data = autil.get_output(
var, model_id_start, flux_pixel, pool_pixel, cbr_pixel,
autil.get_nofluxes_nopools_lma(model_id_start)[2])
if len(fwd_data) > 0:
if fwd_data.shape[1] > cbf_pixel['nodays']:
fwd_data = fwd_data[:, :-1]
fwd_data = autil.remove_outliers(fwd_data)
med = np.nanmedian(fwd_data, axis=0)
ub = np.nanpercentile(fwd_data, 75, axis=0)
lb = np.nanpercentile(fwd_data, 25, axis=0)
ens_spread[comps.index(comp)][
pixels.index(pixel),
vrs.index(var),
n_iters[comps.index(comp)].index(it)] = np.nanmean(
abs(ub -
lb)) if metric == 'spread' else np.sqrt(
np.nansum((med - obs)**2) / n_obs)
print(ens_spread[comps.index(comp)]
[pixels.index(pixel),
vrs.index(var),
n_iters[comps.index(comp)].index(it)])
print(ens_spread)
for var in vrs:
autil.plot_spread_v_iter(
ens_spread,
pixels,
vrs.index(var),
var,
n_iters,
metric,
cur_dir + plot_dir + 'spread_v_iter',
'iter_test' + assim_type + '_' + compare_between + '_' +
model_id_start + '_' + var + '_' + metric,
single_val=True
) #'iter_test_MCMC'+mcmc_id+'_'+model_id_start+'_'+var + '_' + metric)
autil.plot_conv_v_iter(conv,
pixels,
n_iters,
cur_dir + plot_dir + 'spread_v_iter',
'iter_test' + assim_type + '_' + compare_between +
'_' + model_id_start + '_conv',
single_val=True)
return
def main():
model_id = sys.argv[1]
run_type = sys.argv[2] # ALL or SUBSET
mcmc_id = sys.argv[3] # 119 for normal, 3 for DEMCMC
n_iter = sys.argv[4]
ens_size = 500
assim_type = '_p25adapted'
use_bestchains_pkl = False
cur_dir = os.getcwd() + '/'
cbf_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbf'+assim_type+'_ef_ic/' + model_id + '/'
cbr_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/cbr'+assim_type+'_ef/' + model_id + '/'
output_dir = '../../../../../../scratch/users/cfamigli/cardamom/files/output'+assim_type+'_ef/' + model_id + '/'
plot_dir = '../../../../../../scratch/users/cfamigli/cardamom/plots/'
parnames = autil.get_parnames('../../misc/', model_id)
# load list of land pixels
pixels = list(set([file[-8:-4] for file in glob.glob(cbf_dir + '*.cbf')])) if run_type=='ALL' else ['3809','3524','2224','4170','1945','3813','4054','3264','1271','3457']
pixels.sort()
# load list of cbrs
cbr_files = glob.glob(cbr_dir+'*MCMC'+mcmc_id+'_'+n_iter+'_*PLS*forward*.cbr')
# for loop over pixels
gr_pixels = np.zeros(len(pixels))*np.nan # list of GR for each pixel, for mapping
par_pixels = np.zeros((len(pixels), len(parnames)))*np.nan
for pixel in pixels:
print(pixel, pixels.index(pixel))
pixel_chains = autil.find_all_chains(cbr_files, pixel)
pixel_chains.sort() # filenames
if use_bestchains_pkl:
conv_chains_pkl = read_pickle(glob.glob(cbr_dir + model_id + assim_type + '*_MCMC'+mcmc_id + '_'+n_iter+'_best_subset.pkl')[0])
conv_chains_pkl.columns = ['pixel','bestchains','conv'] #rename columns for easier access
if pixel in conv_chains_pkl['pixel'].values:
bestchains = conv_chains_pkl.loc[conv_chains_pkl['pixel']==pixel]['bestchains'].values[0][1:]
print(bestchains)
pixel_chains = [pixel_chain for pixel_chain in pixel_chains if pixel_chain.partition(pixel+'_')[-1][:-4] in bestchains]
else:
continue
#cbf_pixel = rwb.read_cbf_file(cur_dir + cbf_dir + pixel_chains[0].partition('_MCMC')[0]+'_'+pixel+'.cbf')
cbf_filename = glob.glob(cur_dir + cbf_dir + '*'+pixel+'.cbf')[0]
cbf_pixel = rwb.read_cbf_file(cbf_filename)
cbr_chain_list = []
for pixel_chain in pixel_chains:
print(pixel_chain)
cbr_chain = rwb.read_cbr_file(pixel_chain, {'nopars': len(parnames)})
cbr_chain = autil.modulus_Bday_Fday(cbr_chain, parnames)
cbr_pixel = np.copy(cbr_chain) if pixel_chains.index(pixel_chain)==0 else np.concatenate((cbr_pixel, cbr_chain), axis=0)
#autil.plot_par_histograms(cbr_chain, parnames=parnames, savepath=cur_dir+plot_dir+'dists/', title=model_id+'_'+pixel_chain[:-3]+'png')
try:
flux_chain = rwb.readbinarymat(cur_dir + output_dir + 'fluxfile_' + pixel_chain.partition(cbr_dir)[-1][:-3]+'bin', [cbf_pixel['nodays'], autil.get_nofluxes_nopools_lma(model_id)[0]])
pool_chain = rwb.readbinarymat(cur_dir + output_dir + 'poolfile_' + pixel_chain.partition(cbr_dir)[-1][:-3]+'bin', [cbf_pixel['nodays']+1, autil.get_nofluxes_nopools_lma(model_id)[1]])
#autil.plot_flux_pool_timeseries(cbf_pixel, cbr_chain, flux_chain, pool_chain, autil.get_nofluxes_nopools_lma(model_id)[2], savepath=cur_dir+plot_dir+'timeseries/', title=model_id+'_'+pixel_chain[:-3]+'png')
flux_pixel = np.copy(flux_chain) if pixel_chains.index(pixel_chain)==0 else np.concatenate((flux_pixel, flux_chain), axis=0)
pool_pixel = np.copy(pool_chain) if pixel_chains.index(pixel_chain)==0 else np.concatenate((pool_pixel, pool_chain), axis=0)
except Exception as e:
pass
if np.shape(cbr_chain)[0]==ens_size:
cbr_chain_list.append(cbr_chain)
#print(np.shape(cbr_chain))
if len(cbr_chain_list)>1:
gr = autil.gelman_rubin(cbr_chain_list)
#print(gr)
print('%i of %i parameters converged' % (sum(gr<1.2), len(parnames)))
gr_pixels[pixels.index(pixel)] = sum(gr<1.2)/len(parnames)
else:
gr = np.nan
par_pixels[pixels.index(pixel),:] = np.nanmedian(cbr_pixel, axis=0)
#autil.plot_par_histograms(cbr_pixel, parnames=parnames, savepath=cur_dir+plot_dir+'dists/', title=model_id+assim_type+'_MCMC'+mcmc_id+'_'+cbf_filename.partition(cbf_dir)[-1][:-4]+'.png')
#autil.plot_flux_pool_timeseries(cbf_pixel, cbr_pixel, flux_pixel, pool_pixel, autil.get_nofluxes_nopools_lma(model_id)[2], savepath=cur_dir+plot_dir+'timeseries/', title=model_id+assim_type+'_MCMC'+mcmc_id+'_'+cbf_filename.partition(cbf_dir)[-1][:-4]+'.png')
#vmax = [None,None,None,None,None,None,None,None,None,None,None,None,None,None,None,None,None,210,200,215,6600,195,24000,None,None,None,900,None,None,None,None,None,None,None] #np.nanpercentile(par_pixels[:,par], 90)
for par in range(len(parnames)): autil.plot_map(nrows=46, ncols=73, land_pixel_list=[file[-8:-4] for file in glob.glob(cur_dir + cbf_dir + '*.cbf')], pixel_value_list=pixels, value_list=par_pixels[:,par], vmax=np.nanpercentile(par_pixels[:,par], 90), savepath=cur_dir+plot_dir+'maps/', savename='par'+str(par)+'_' + model_id +assim_type+ '_MCMC' + mcmc_id +'_'+ n_iter+'_EF_clipped_PLS_soilgrids_poolobs_rescaled_forward')
#autil.plot_map(nrows=46, ncols=73, land_pixel_list=[file[-8:-4] for file in glob.glob(cur_dir + cbf_dir + '*.cbf')], pixel_value_list=pixels, value_list=np.ones(len(pixels)), savepath=cur_dir+plot_dir+'maps/', title='test_pixels.png')
#autil.plot_map(nrows=46, ncols=73, land_pixel_list=[file[-8:-4] for file in glob.glob(cur_dir + cbf_dir + '*.cbf')], pixel_value_list=pixels, value_list=gr_pixels*100, savepath=cur_dir+plot_dir+'maps/', savename='gr_' + model_id + assim_type+ '_' +run_type+ '_MCMC' + mcmc_id + '_' + n_iter)
return