def get_model_performance_state(model,
yield_type='rainfed',
prediction_type='forward',
state_train=False):
d = load_prediction(model,
yield_type=yield_type,
prediction_type=prediction_type,
state=state_train)
d_r2 = prediction_result_by_state(d, yield_type=yield_type)
d_r2.loc[:, 'rmse'] = d_r2.loc[:, 'rmse'] * 0.0628 # convert to t/ha
return d_r2
def get_national_prediction(model,
yield_type='rainfed',
prediction_type='forward',
area_weight=False,
direct_fn=False):
df = load_prediction(model,
yield_type=yield_type,
prediction_type=prediction_type,
direct_fn=direct_fn)
df = df.dropna()
state_2003 = [
'ILLINOIS', 'INDIANA', 'IOWA', 'MISSOURI', 'NEBRASKA', 'NORTH DAKOTA',
'WISCONSIN'
]
# Only 7 states before 2007
con = df.State.isin(state_2003) & (df.year < 2007)
# con = df.State.isin(state_2003)
# all states after 2007
con = con | (df.year >= 2007)
if area_weight:
df['Predicted_' + yield_type_dict[yield_type] +
'_area'] = (df['Predicted_' + yield_type_dict[yield_type]] *
df[area_type_dict[yield_type]])
df[yield_type_dict[yield_type] +
'_area'] = (df[yield_type_dict[yield_type]] *
df[area_type_dict[yield_type]])
yield_predicted=df[con].groupby('year').sum()['Predicted_' + yield_type_dict[yield_type] + '_area']/ \
df[con].groupby('year').sum()[area_type_dict[yield_type]]
yield_actual=df[con].groupby('year').sum()[yield_type_dict[yield_type] + '_area']/ \
df[con].groupby('year').sum()[area_type_dict[yield_type]]
else:
yield_predicted = df[con].groupby('year').mean()[
'Predicted_' + yield_type_dict[yield_type]]
yield_actual = df[con].groupby('year').mean()[
yield_type_dict[yield_type]]
return yield_predicted, yield_actual
import matplotlib.pyplot as plt
import numpy as np
import pandas as pd
from func_crop_model import prediction_result_global
from plot_model_comparisons import load_prediction
# Constrain to seven states due to CDL availiability
state_2003 = [
'ILLINOIS', 'INDIANA', 'IOWA', 'MISSOURI', 'NEBRASKA', 'NORTH DAKOTA',
'WISCONSIN'
]
# Best climate model
d0 = load_prediction('vpd_spline',
yield_type='rainfed',
prediction_type='leave_one_year_out')
d0_r2 = prediction_result_global(d0[d0.State.isin(state_2003)],
yield_type='rainfed')
d0_r2.loc[:, 'rmse'] = d0_r2.loc[:, 'rmse'] * 0.0628 # convert to t/ha
# Best climate + evi
d = load_prediction('vpd_spline_evi_poly',
yield_type='rainfed',
prediction_type='leave_one_year_out')
d1_r2 = prediction_result_global(d[d.State.isin(state_2003)],
yield_type='rainfed')
d1_r2.loc[:, 'rmse'] = d1_r2.loc[:, 'rmse'] * 0.0628 # convert to t/ha
print(d0_r2)
print(d1_r2)
def make_plot(model='vpd_spline_evi_poly'):
data_12 = load_yield_data()
data_12.loc[:, 'yield_rainfed'] = data_12.loc[:, 'yield_rainfed'] * 0.0628
# Constrain to seven states due to CDL availiability
state_2003 = [
'ILLINOIS', 'INDIANA', 'IOWA', 'MISSOURI', 'NEBRASKA', 'NORTH DAKOTA',
'WISCONSIN'
]
# state_2003 = ['ILLINOIS', 'INDIANA', 'IOWA', 'KANSAS', 'MICHIGAN', 'MINNESOTA',
# 'MISSOURI', 'NEBRASKA', 'NORTH DAKOTA', 'OHIO', 'SOUTH DAKOTA', 'WISCONSIN']
# Load best climate + evi
d = load_prediction(model,
yield_type='rainfed',
prediction_type='leave_one_year_out')
d1_r2 = prediction_result_global(d[d.State.isin(state_2003)],
yield_type='rainfed')
d1_r2.loc[:, 'rmse'] = d1_r2.loc[:, 'rmse'] * 0.0628 # convert to t/ha
d_pre = d1_r2.join(data_12[data_12.State.isin(state_2003)].groupby(
'year').mean()['precip_gs_z'].to_frame())
d_yieldstd = d1_r2.join(data_12[data_12.State.isin(state_2003)].groupby(
'year').std()['yield_rainfed'].to_frame())
fig, axes = plt.subplots(2, 2, figsize=(10, 7.5))
d_pre.plot.scatter(x='precip_gs_z', y='R2', ax=axes[0, 0])
plot_name(d_pre, 'precip_gs_z', 'R2', axes[0, 0])
plot_fitting(d_pre, 'precip_gs_z', 'R2', axes[0, 0], order=2)
d_pre.plot.scatter(x='precip_gs_z', y='rmse', ax=axes[0, 1])
plot_name(d_pre, 'precip_gs_z', 'rmse', axes[0, 1])
plot_fitting(d_pre, 'precip_gs_z', 'rmse', axes[0, 1], order=2)
axes[0, 0].set_title('$R^2$')
axes[0, 1].set_title('RMSE (t/ha)')
axes[0, 0].set_xlabel('Precipitation standard anomaly', fontsize=12)
axes[0, 1].set_xlabel('Precipitation standard anomaly', fontsize=12)
axes[0, 0].set_ylabel('')
axes[0, 1].set_ylabel('')
d_yieldstd.plot.scatter(x='yield_rainfed', y='R2', ax=axes[1, 0])
plot_name(d_yieldstd, 'yield_rainfed', 'R2', axes[1, 0])
plot_fitting(d_yieldstd, 'yield_rainfed', 'R2', axes[1, 0], order=1)
d_yieldstd.plot.scatter(x='yield_rainfed', y='rmse', ax=axes[1, 1])
plot_name(d_yieldstd, 'yield_rainfed', 'rmse', axes[1, 1])
plot_fitting(d_yieldstd, 'yield_rainfed', 'rmse', axes[1, 1], order=1)
axes[1, 0].set_xlabel('Spatial yield variability (t/ha)', fontsize=12)
axes[1, 1].set_xlabel('Spatial yield variability (t/ha)', fontsize=12)
axes[1, 0].set_ylabel('')
axes[1, 1].set_ylabel('')
# Add panel label
for i, s in enumerate([chr(i) for i in range(ord('a'), ord('d') + 1)]):
axes.flatten()[i].text(0.01,
0.95,
s,
fontsize=12,
transform=axes.flatten()[i].transAxes,
fontweight='bold')
plt.subplots_adjust(left=0.05,
right=0.95,
top=0.9,
bottom=0.1,
hspace=0.25)
# plt.savefig('../figure/figure_model_performance_interannual_variability_causes_12states.pdf')
plt.savefig(
'../figure/figure_model_performance_interannual_variability_causes_%s.pdf'
% model)
# plt.savefig('../figure/test.pdf')
print('figure saved')
def make_plot(model='vpd_poly_evi_poly',
yield_type='rainfed',
prediction_type='leave_one_year_out'):
# Load state model performance
d_r2 = get_model_performance_state(model,
yield_type=yield_type,
prediction_type=prediction_type,
state_train=True)
# d = load_prediction(model, yield_type=yield_type, prediction_type=prediction_type,state=True)
# d_r2 = prediction_result_by_state(d)
# Load global model performance
dg_r2 = get_model_performance_state(model,
yield_type=yield_type,
prediction_type=prediction_type,
state_train=False)
d2 = load_prediction(model,
yield_type=yield_type,
prediction_type=prediction_type)
# Get mean county number of each state across years
county_n = d2[d2.year >= 2007].dropna().groupby(
['year',
'State']).count().mean(level=1)['yield_rainfed'].apply(np.ceil)
county_n.index = county_n.index.map(lambda x: x.title())
# dg_r2 = prediction_result_by_state(d2)
#
if ('evi' in model) | ('lst' in model):
year_begin = 2007
else:
year_begin = 2005
d3_r2 = get_boxplot_data(
dg_r2.loc[(slice(None), slice(year_begin, 2016)), :],
d_r2.loc[(slice(None), slice(year_begin, 2016)), :], 'R2')
d3_rmse = get_boxplot_data(
dg_r2.loc[(slice(None), slice(year_begin, 2016)), :],
d_r2.loc[(slice(None), slice(year_begin, 2016)), :], 'rmse')
d3_r2.loc[:, 'State'] = d3_r2.State.apply(lambda x: x.title())
d3_rmse.loc[:, 'State'] = d3_rmse.State.apply(lambda x: x.title())
# Begin plot
fig, [ax1, ax2] = plt.subplots(2, 1, figsize=(11, 5.5))
meanlineprops = dict(linestyle='--', color='k')
sns.boxplot(x='State',
y='R2',
hue='Model',
data=d3_r2,
fliersize=0,
ax=ax1,
meanline=True,
showmeans=True,
palette='Set3',
meanprops=meanlineprops)
sns.boxplot(x='State',
y='rmse',
hue='Model',
data=d3_rmse,
fliersize=0,
ax=ax2,
meanline=True,
showmeans=True,
palette='Set3',
meanprops=meanlineprops)
ax1.set_ylabel('R2', fontsize=12)
ax2.set_ylabel('RMSE (t/ha)', fontsize=12)
ax1.text(0.01,
0.90,
'a',
fontsize=12,
transform=ax1.transAxes,
fontweight='bold')
ax2.text(0.01,
0.90,
'b',
fontsize=12,
transform=ax2.transAxes,
fontweight='bold')
# Create fake lines for legend
mean_line = mlines.Line2D([], [],
linestyle='--',
color=ax1.artists[0].get_edgecolor())
median_line = mlines.Line2D([], [],
linestyle='-',
color=ax1.artists[0].get_edgecolor())
ax1.legend_.set_frame_on(False)
# Rotate last row xticklabels
# ax1.set_xticklabels(ax1.get_xticklabels(), rotation=10)
# ax2.set_xticklabels(ax2.get_xticklabels(), rotation=10)
ax2.legend([median_line, mean_line], ['median', 'mean'], loc='upper right')
# add county number
for i, s in enumerate(county_n.values):
ax2.text(i, -0.5, '(%d)' % s, fontsize=10, ha='center')
ax2.legend_.set_frame_on(False)
ax2.set_ylim([0, 2.5]) # 40 bu *0.0628
ax1.set_xlabel('')
ax2.set_xlabel('')
plt.subplots_adjust(left=0.05, right=0.975, hspace=0.3, top=0.95)
plt.savefig(
'../figure/figure_global_state_model_comparison_%s_%s_%s_test.pdf' %
(model, yield_type, prediction_type))
# plt.savefig('../figure/test.pdf')
print('figure saved')