def krigingmodel(DF, dic_polia, dic_latlon):
#start_time_of_gmodel = time.time()
# MODELWI is model without interpolation of shape, it is an array of shape (142,271)
MODELWI = regresia.model_reg(DF, dic_polia, dic_latlon)[0]
# ERROR is caculated as pollutant minus PREDICT, it has shape same as measured pollutant (32)
ERROR_stanice = regresia.reg_funkcia(DF, dic_polia, dic_latlon)[4]
x, y, z = list(DF['lon_x']), list(DF['lat_x']), ERROR_stanice
#xi,yi are new grid coordinates
xi = np.linspace(np.min(dic_latlon['LON'].flatten()),
np.max(dic_latlon['LON'].flatten()), 271)
yi = np.linspace(np.min(dic_latlon['LAT'].flatten()),
np.max(dic_latlon['LAT'].flatten()), 142)
# OK is kriging funkction
OK = OrdinaryKriging(x, y, z, variogram_model='spherical')
"""
OK: is kriging funkction form pykrige package
x, y: are latitude and lontitude of measured concentrations from DF(dataframe)
z : error concentrations of pollutant, from linear regresion model
variogram model: is tzpe of theoretical variogram
"""
# z1 is kriging interpolation array
z1, ss1 = OK.execute('grid', xi, yi)
"""
z1: is interpolated array realised on grid
xi, yi: are new grid coordinates
"""
KRIGING_ERROR = z1
#KOMPLET_MODEL is linear reggresion predict MODELWI plus kriging of residuals
KOMPLET_MODEL = KRIGING_ERROR + MODELWI
return KOMPLET_MODEL, KRIGING_ERROR, np.exp(KOMPLET_MODEL), np.exp(
KRIGING_ERROR)
def validation(DF, dic_polia, dic_latlon, model_type):
"""
leaving-one-out validation of linear reggresion model,
linear reggresion model + idw interpolation of residulas,
linear gergession model + kriging interpolation if residuals
"""
print('###########################')
print('leaving one validation for: ' + model_type)
predicted_value = [
] # this is the list of arrays, each array has 1 station dropped.
for i, row in DF.iterrows():
DF1 = DF.drop([i])
ix, iy = regresia.getclosest_ij(dic_latlon['LAT'], dic_latlon['LON'],
row['lat_x'], row['lon_x'])
if model_type == 'regresia + IDW':
value = model_idw.idwmodel(DF1, dic_polia, dic_latlon)[2][ix, iy]
elif model_type == 'regresia':
value = regresia.model_reg(DF1, dic_polia, dic_latlon)[1][ix, iy]
elif model_type == 'regresia + kriging':
value = model_kriging.krigingmodel(DF1, dic_polia,
dic_latlon)[2][ix, iy]
predicted_value.append(value)
#print('EOI = {}, model = {}, measured = {} '.format(row['EOI'],value,row['pollutant']))
#print('{} {} {} {}'.format(row['name'], row['EOI'], value, row['pollutant']))
predicted_value = np.array(predicted_value)
RMSE = ((np.sum((DF['pollutant'] - predicted_value)**2)) * (1 / len(
(DF['pollutant'] - predicted_value))))**0.5
BIAS = np.sum(
(predicted_value - DF['pollutant'])) / predicted_value.shape[0]
r = np.corrcoef(predicted_value, DF['pollutant'])[0, 1]
print('RMSE={}, BIAS={}, r={}'.format(RMSE, BIAS, r))
def idwmodel(DF, dic_polia, dic_latlon):
#start_time_of_gmodel = time.time()
# MODELWI is model without interpolation of shape, it is an array of shape (142,271)
MODELWI = regresia.model_reg(DF, dic_polia, dic_latlon)[0]
# ERROR is caculated as pollutant minus PREDICT, it has shape same as measured pollutant (32)
ERROR_stanice = regresia.reg_funkcia(DF, dic_polia, dic_latlon)[4]
x, y, z = list(DF['lat_x']), list(DF['lon_x']), ERROR_stanice
#xi,yi are new grid coordinates
xi = dic_latlon['LAT'].flatten()
yi = dic_latlon['LON'].flatten()
IDW_ERROR = np.asarray(idwr(y, x, z, yi, xi))
IDW_ERROR = np.reshape(IDW_ERROR, (142, 271))
# adding pollutant values to interpolated errror array (IDW_ERROR) which shape is (142,271)
for i in range(0, len(x)):
ix, iy = getclosest_ij(dic_latlon['LAT'], dic_latlon['LON'], x[i],
y[i])
IDW_ERROR[ix, iy] = ERROR_stanice[i]
#KOMPLET_MODEL is linear reggresion predict MODELWI plus IDW of residuals
KOMPLET_MODEL = IDW_ERROR + MODELWI
#print('gmodel {0:.3f}'.format(time.time() - start_time_of_gmodel))
return KOMPLET_MODEL, IDW_ERROR, np.exp(KOMPLET_MODEL), np.exp(IDW_ERROR)
def mapy(DF, dic_polia, dic_latlon, name):
"""
mapy is funkction which produce prediction maps transformed from ln(pollutant) to pollutant of linear reggresion in 3 forms:
1) only prediction of linear reggresion
2) prediction of linear reggresion + kriging interpolation of residuals
3) prediction of linear reggresion + idw interpolation of residuals
"""
################################
for i in (regresia.model_reg(DF, dic_polia, dic_latlon)[1],
model_idw.idwmodel(DF, dic_polia, dic_latlon)[2],
model_kriging.krigingmodel(DF, dic_polia, dic_latlon)[2]):
mapb.drawcountries()
mapb.pcolormesh(dic_latlon['LON'],
dic_latlon['LAT'],
i,
cmap=plt.cm.jet,
latlon=True)
mapb.readshapefile('C:/Users/kocok/Desktop/Shapefile0/slovensko',
'slovensko',
drawbounds=True,
linewidth=4)
if i in regresia.model_reg(DF, dic_polia, dic_latlon)[1]:
plt.colorbar(label=name + ' [$\mu$g.$m^{-3}$]', extend='max')
plt.title('Koncentrácia ' + name +
' za rok 2017 (lineárna regresia)',
fontsize=15)
plt.clim(0, 40)
mapb.scatter(DF['lon_x'].values,
DF['lat_x'].values,
c=DF['pollutant'].values,
s=30,
latlon=True,
cmap='jet',
alpha=1,
edgecolors='black')
plt.clim(0, 40)
#plt.show()
plt.savefig('lnNO2 expregresia {}.png'.format(len(
dic_polia.keys())),
bbox_inches='tight',
dpi=600)
plt.clf()
elif i in model_kriging.krigingmodel(DF, dic_polia, dic_latlon)[2]:
plt.colorbar(label=name + ' [$\mu$g.$m^{-3}$]', extend='max')
plt.clim(0, 40)
plt.title(
'Koncentrácia ' + name +
' za rok 2017 (lineárna regresia + OK interpolácia rezíduí)',
fontsize=15)
mapb.scatter(DF['lon_x'].values,
DF['lat_x'].values,
c=DF['pollutant'].values,
s=30,
latlon=True,
cmap='jet',
alpha=1,
edgecolors='black')
plt.clim(0, 40)
#plt.show()
plt.savefig('lnNO2 expregresia+ok pocet poli {}.png'.format(
len(dic_polia.keys())),
bbox_inches='tight',
dpi=600)
plt.clf()
else:
plt.colorbar(label=name + ' [$\mu$g.$m^{-3}$]', extend='max')
plt.clim(0, 40)
plt.title(
'Koncentrácia ' + name +
' za rok 2017 (lineárna regresia + IDW interpolácia rezíduí)',
fontsize=15)
mapb.scatter(DF['lon_x'].values,
DF['lat_x'].values,
c=DF['pollutant'].values,
s=30,
latlon=True,
cmap='jet',
alpha=1,
edgecolors='black')
plt.clim(0, 40)
#plt.show()
plt.savefig('lnNO2 expregresia+idw pocet poli {}.png'.format(
len(dic_polia.keys())),
bbox_inches='tight',
dpi=600)
plt.clf()