def data_collection():
rawDataPath = '../../Data/'
derivedDataPath = '../../DataOutput/'
forecastVar = 'conc'
iceType = 'extent'
hemStr = 'N'
siiVersion = 'v3.0'
startYr = 1990
forecastYr = 2010
startMonth = 2
stopMonth = 11
regionMask = np.load(derivedDataPath + 'Regions/regionMaskA100km')
numFeat = 18
numSamples = (forecastYr - startYr) * (stopMonth + 1 - startMonth)
#data collection
#start off with just the summer months and without ice thickness
feat = []
groundTruth = []
for year in range(startYr, forecastYr):
for index, month in enumerate(range(startMonth, stopMonth + 1)):
_, extent = ff.get_ice_extentN(rawDataPath, month, year, year,
iceType, siiVersion, hemStr)
groundTruth.append(extent[0])
#set month to array
sample = np.zeros(numFeat)
sample[0] = (month - 1) #gridded months are actually indexed at 0
#divide by 100 to put in the same magnitude as the other values
sample[1] = np.ma.mean(
ff.get_pmas_month(rawDataPath, year, month - 1))
#set region features to array
iceConc = ff.get_gridvar(derivedDataPath, forecastVar, month,
array(year), hemStr)
for regInd in range(16):
regionData = iceConc.data
regionData[iceConc.mask == True] = 0 #get rid of nan
desiredRegion = regionMask == regInd
sample[regInd + 2] = np.ma.mean(
np.multiply(regionData, desiredRegion))
#sample[regInd+1] = np.ma.mean(np.multiply(regionData, desiredRegion))
feat.append(sample)
feat = np.reshape(feat, (numSamples, numFeat))
groundTruth = np.reshape(groundTruth, (numSamples, 1))
return feat, groundTruth
region = 0
hemStr = 'N'
anomObs = 1
rawDataPath = '../../Data/'
derivedDataPath = '../../DataOutput/'
yrForecast = 2015
randSeedNum = 50
#thick, forecastThickMean = ff.get_ice_thickness(rawDataPath, startYr, yrForecast, forecastMonth)
yrsTrain, extentTrain = ff.get_ice_extentN(rawDataPath,
predMonth,
startYr,
yrForecast - 1,
icetype=iceType,
version=siiVersion,
hemStr=hemStr)
extentDetrendTrain, lineTrain = ff.get_varDT(yrsTrain, extentTrain)
varTrain = ff.get_gridvar(derivedDataPath, forecastVar, forecastMonth,
yrsTrain, hemStr)
varForecast = ff.get_gridvar(derivedDataPath, forecastVar, forecastMonth,
array(yrForecast), hemStr)
years, extentYr = ff.get_ice_extentN(rawDataPath,
predMonth,
yrForecast,
yrForecast,
icetype=iceType,
def main(endYear, fmonth, pmonth):
rawDataPath = '../../Data/'
derivedDataPath = '../../DataOutput/'
saveDataPath = derivedDataPath + '/Arctic/'
figPath = '../../Figures/' + '/Arctic/Predictions/'
fvars = ['conc']
iceType = 'extent'
hemStr = 'N'
siiVersion = 'v3.0'
startYear = 1980
startYearP = 1990
#endYear=2017
weight = 1
region = 0
varStrsOut = ''.join(fvars)
forecastVarsM = np.array([]).reshape(0, 7)
for year in range(startYearP, endYear + 1):
outStr = 'forecastDump' + iceType + varStrsOut + 'fm' + str(
fmonth) + 'pm' + str(pmonth) + 'R' + str(region) + str(
startYear) + str(year) + 'W' + str(weight) + 'SII' + siiVersion
forecastVars = load(saveDataPath + outStr)
forecastVarsM = np.vstack([forecastVarsM, forecastVars])
years, extent = ff.get_ice_extentN(rawDataPath,
pmonth,
startYear,
year,
icetype=iceType,
version=siiVersion,
hemStr=hemStr)
yearsP = np.arange(startYearP, endYear + 1)
"""Plot forecast data """
rcParams['xtick.major.size'] = 2
rcParams['ytick.major.size'] = 2
rcParams['axes.linewidth'] = .5
rcParams['lines.linewidth'] = .5
rcParams['patch.linewidth'] = .5
rcParams['axes.labelsize'] = 8
rcParams['xtick.labelsize'] = 8
rcParams['ytick.labelsize'] = 8
rcParams['legend.fontsize'] = 8
rcParams['font.size'] = 7
rc('font', **{'family': 'sans-serif', 'sans-serif': ['Arial']})
fig = figure(figsize=(3.5, 2.2))
ax1 = subplot(1, 1, 1)
im1 = plot(years, extent, 'k')
#im2 = plot(Years[start_year_pred-start_year:], lineT[start_year_pred-start_year:]+ExtentG, 'r')
#im3 = plot(years[-1], extent[-1], marker='o', markersize=2, color='k')
im3 = plot(yearsP,
forecastVarsM[:, 2],
marker='x',
markersize=1,
alpha=0.5,
color='k')
im3 = plot(yearsP,
forecastVarsM[:, 3],
marker='o',
markersize=1,
alpha=0.5,
color='r')
ax1.errorbar(yearsP,
forecastVarsM[:, 3],
yerr=forecastVarsM[:, 6],
color='r',
fmt='',
linestyle='',
alpha=0.5,
lw=0.6,
capsize=0.5,
zorder=2)
im3 = plot(yearsP[-1],
forecastVarsM[-1, 2],
marker='x',
markersize=2,
color='k')
im3 = plot(yearsP[-1],
forecastVarsM[-1, 3],
marker='o',
markersize=2,
color='r')
ax1.errorbar(yearsP[-1],
forecastVarsM[-1, 3],
yerr=forecastVarsM[-1, 6],
color='r',
fmt='',
linestyle='',
lw=0.6,
capsize=0.5,
zorder=2)
#errorbar(YearsP, array(lineTP)+array(ExtentG) , yerr=prederror, color='r',fmt='',linestyle='',lw=0.4,capsize=0.5, zorder = 2)
#if (np.isfinite(forecastVars[4])):
#ax1.errorbar(yearsP, extentPredAbs , yerr=[1.96*x for x in perr], color='r',fmt='',linestyle='',lw=0.3,capsize=0.5, zorder = 2)
forecastStr = '%.2f' % (forecastVarsM[-1, 3])
linearStr = '%.2f' % (forecastVarsM[-1, 2])
observedStr = '%.2f' % (extent[-1])
ax1.annotate('Year: ' + str(year) + '\nObserved: ' + observedStr +
r' M km$^2$' + '\nTrend: ' + linearStr + r' M km$^2$',
xy=(0.7, 1.02),
xycoords='axes fraction',
horizontalalignment='left',
verticalalignment='top')
ax1.annotate('\nForecast: ' + forecastStr + r' M km$^2$',
xy=(0.7, 0.85),
xycoords='axes fraction',
color='r',
horizontalalignment='left',
verticalalignment='top')
ax1.annotate('June forecasts of September sea ice',
fontsize=5,
xy=(0.02, 0.02),
xycoords='axes fraction',
horizontalalignment='left',
verticalalignment='bottom')
ax1.set_ylabel(iceType + r' (Million km$^2$)')
#ax1.set_xlabel('Years')
ax1.set_xlim(1980, 2020)
ax1.set_xticks(np.arange(1980, 2021, 10))
ax1.set_xticks(np.arange(1980, 2021, 5), minor=True)
#ax1.set_xticklabels([])
ax1.set_ylim(3, 8)
ax1.spines['right'].set_visible(False)
ax1.spines['top'].set_visible(False)
plt.tight_layout()
#subplots_adjust(left=0.15, right=0.90, bottom=0.17, top=0.96, hspace=0)
savefig(figPath + '/forecast' + outStr + 'multi.png', dpi=300)
close(fig)
def main(year,
fmonth,
pmonth,
fvars=['conc'],
iceType='extent',
hemStr='N',
siiVersion='v3.0',
startYear=1979,
weight=1,
region=0,
numYearsReq=5,
plotForecast=1,
plotSkill=1,
outSkill=1,
outLine=1,
outWeights=1):
"""
Main sea ice forecast script. Can be run here (and looped )
Args:
startYear=1980 # Start of forecast training
endYear=2018 # End of forecast
fmonth=11 #6=June, 9=Sep # Forecast month
pmonth=2 #9=SEP # Predicted month
fvars: forecast variables (e.g. 'conc'). Need to be in brackets.
weight: Spatially weighting the data (1=True)
iceType: Ice type being forecast ('extent' or 'area')
hemStr: Hemipshere (N or S)
siiVersion: version of the NSIDC sea ice index (if used as the observed ice state)
startYear: Start year of training data (default = 1980)
region: Region being forecast (0 is pan Arctic/Antartctic)
#0 implies pan-Arctic or Antarctic
#2 Weddell Sea
#3 Indian Ocean
#4 Pacific Ocean
#5 Ross Sea
#6 Amundsen/BHausen Sea
#A Alaskan
numYearsReq: (defaul 5) Number of years required in a grid cell for it to count towards the training data
plotSkill: =1 for plotting the skill
outSkill: =1 far saving the skill values
outLine: =1 for saving the forecast time series
outWeights: =1 for saving the weightings
Returns:
A dumped Python array including the folowing variables:
Observed ice extent
Observed detrended ice extent from linear trend persistence (LTP)
LTP extent
Absolute forecast of sea ice extent (added LTP)
Detrended (forecast) ice extent from LTP
Forecast anomaly from observed
Estimated forecast error (1 SD)
"""
repoPath = '/Users/aapetty/GitRepos/GitHub/SeaIcePrediction/'
rawDataPath = repoPath + '/Data/'
derivedDataPath = repoPath + '/DataOutput/'
if (hemStr == 'S'):
saveDataPath = derivedDataPath + '/Antarctic/'
figPath = repoPath + '/Figures/Antarctic/YearlyPredictions/'
elif (hemStr == 'N'):
saveDataPath = derivedDataPath + '/Arctic/'
figPath = repoPath + '/Figures/Arctic/YearlyPredictions/'
if (region == 'A'):
saveDataPath = derivedDataPath + '/Alaska/'
figPath = repoPath + '/Figures/Alaska/YearlyPredictions/'
print('Forecast year:', year)
print('Forecast data month:', fmonth, 'Predicted month:', pmonth)
print('Variables:', fvars)
print('Hemisphere:', hemStr)
print('Ice type predicted', iceType)
print('Weighted:', weight)
varStrsOut = ''.join(fvars)
outStr = 'forecastDump' + iceType + varStrsOut + 'fm' + str(
fmonth) + 'pm' + str(pmonth) + 'R' + str(region) + str(
startYear) + str(year) + 'W' + str(weight) + 'SII' + siiVersion
if (year < 2018):
anomObsT = 1
else:
anomObsT = 0
forecastVals = ff.CalcForecastMultiVar(rawDataPath,
derivedDataPath,
year,
startYear,
fvars,
fmonth,
pmonth=pmonth,
region=region,
anomObs=anomObsT,
numYearsReq=numYearsReq,
weight=weight,
outWeights=outWeights,
icetype=iceType,
hemStr=hemStr,
siiVersion=siiVersion)
print('Observed ice state:', forecastVals[0])
print('Linear trend presistence:', forecastVals[2])
print('Forecast ice state:', forecastVals[3])
print('Detrended observed ice state :', forecastVals[1])
print('Detrended observed ice state :', forecastVals[4])
print('Forecast anomaly :', forecastVals[5])
array(forecastVals).dump(saveDataPath + outStr)
if (plotForecast == 1):
if (region == 0):
years, extent = ff.get_ice_extentN(rawDataPath,
pmonth,
startYear,
year,
icetype=iceType,
version=siiVersion,
hemStr=hemStr)
ff.plotForecastOneYear(figPath,
years,
extent,
year,
forecastVals,
outStr,
iceType,
minval=2,
maxval=8)
elif (region == 'A'):
poleStr = 'A'
extent = loadtxt(derivedDataPath + '/Extent/' + 'ice_' + iceType +
'_M' + str(pmonth) + 'R' + str(region) + '_' +
str(startYear) + '2017' + poleStr)
extent = extent[0:year - startYear + 1]
years = np.arange(startYear, startYear + size(extent), 1)
ff.plotForecastOneYear(figPath,
years,
extent,
year,
forecastVals,
outStr,
iceType,
minval=0,
maxval=2)
def main(endYear, fmonth, pmonth, fvars=['conc'], iceType='extent', hemStr='N', siiVersion='v3.0', startYear=1979, minval=3, maxval=8, region=0):
monthStrs=['Jan', 'Feb', 'Mar', 'Apr', 'May', 'June', 'July', 'August', 'September']
textStr=monthStrs[fmonth-1]+' forecasts of '+monthStrs[pmonth-1]+' Arctic sea ice '+iceType
repoPath='/Users/aapetty/GitRepos/GitHub/SeaIcePrediction/'
rawDataPath = repoPath+'/Data/'
derivedDataPath = repoPath+'/DataOutput/'
if (hemStr=='S'):
saveDataPath=derivedDataPath+'/Antarctic/'
figPath=repoPath+'/Figures/Forecasts/'
elif (region=='A'):
saveDataPath=derivedDataPath+'/Alaska/'
figPath=repoPath+'/Figures/Forecasts/'
else:
saveDataPath=derivedDataPath+'/Arctic/'
figPath=repoPath+'/Figures/Forecasts/'
startYearP=1990
#endYear=2017
weight=1
varStrsOut=''.join(fvars)
forecastVarsM=np.array([]).reshape(0,7)
for year in range(startYearP, endYear+1):
outStr='forecastDump'+iceType+varStrsOut+'fm'+str(fmonth)+'pm'+str(pmonth)+'R'+str(region)+str(startYear)+str(year)+'W'+str(weight)+'SII'+siiVersion
forecastVars=load(saveDataPath+outStr)
forecastVarsM=np.vstack([forecastVarsM, forecastVars])
if (region==0):
years, extent = ff.get_ice_extentN(rawDataPath, pmonth, startYear, year,
icetype=iceType, version=siiVersion, hemStr=hemStr)
elif (region=='A'):
poleStr='A'
extent=loadtxt(derivedDataPath+'/Extent/'+'ice_'+iceType+'_M'+str(pmonth)+'R'+str(region)+'_'+str(startYear)+'2017'+poleStr)
#extent=extent[0:year-startYear+1]
years=np.arange(startYear, 2017+1, 1)
yearsP=np.arange(startYearP, endYear+1)
#extentyr, extentObsDT, extTrendP, extentForrAbs, extentForrDT, anom, prstd[0]
print ('Prediction int:', forecastVars[-1])
skill = '%.2f' %(1 - (ff.rms(forecastVarsM[0:-1, 5]))/(ff.rms(forecastVarsM[0:-1, 1])))
"""Plot forecast data """
rcParams['xtick.major.size'] = 2
rcParams['ytick.major.size'] = 2
rcParams['axes.linewidth'] = .5
rcParams['lines.linewidth'] = .5
rcParams['patch.linewidth'] = .5
rcParams['axes.labelsize'] = 8
rcParams['xtick.labelsize']=8
rcParams['ytick.labelsize']=8
rcParams['legend.fontsize']=8
rcParams['font.size']=7
rc('font',**{'family':'sans-serif','sans-serif':['Arial']})
fig = figure(figsize=(4.,2.2))
ax1=subplot(1, 1, 1)
im1 = plot(years, extent, 'k')
#im2 = plot(Years[start_year_pred-start_year:], lineT[start_year_pred-start_year:]+ExtentG, 'r')
#im3 = plot(years[-1], extent[-1], marker='o', markersize=2, color='k')
im3 = plot(yearsP, forecastVarsM[:, 2], marker='x', markersize=1, alpha=0.5, color='k')
im3 = plot(yearsP, forecastVarsM[:, 3], marker='o', markersize=1, alpha=0.5, color='r')
#ax1.errorbar(yearsP, forecastVarsM[:, 3] , yerr=forecastVarsM[:, 6], color='r',fmt='',linestyle='',alpha=0.5, lw=0.6,capsize=0.5, zorder = 2)
im3 = plot(yearsP[-1], forecastVarsM[-1, 2], marker='x', markersize=2, color='k')
im3 = plot(yearsP[-1], forecastVarsM[-1, 3], marker='o', markersize=2, color='r')
ax1.errorbar(yearsP[-1], forecastVarsM[-1, 3] , yerr=forecastVarsM[-1, 6], color='r',fmt='',linestyle='',lw=0.6,capsize=0.5, zorder = 2)
#errorbar(YearsP, array(lineTP)+array(ExtentG) , yerr=prederror, color='r',fmt='',linestyle='',lw=0.4,capsize=0.5, zorder = 2)
#if (np.isfinite(forecastVars[4])):
#ax1.errorbar(yearsP, extentPredAbs , yerr=[1.96*x for x in perr], color='r',fmt='',linestyle='',lw=0.3,capsize=0.5, zorder = 2)
forecastStr='%.2f' %(forecastVarsM[-1, 3])
linearStr='%.2f' %(forecastVarsM[-1, 2])
#observedStr='%.2f' %(extent[-1])
ax1.annotate('Year: '+str(year)+'\nLinear trend forecast: '+linearStr+r' M km$^2$',
xy=(0.8, 1.02), xycoords='axes fraction', horizontalalignment='left', verticalalignment='top')
ax1.annotate('NASA GSFC forecast: '+forecastStr+r' M km$^2$',
xy=(0.8, 0.9), xycoords='axes fraction', color='r', horizontalalignment='left', verticalalignment='top')
#ax1.annotate('Year: '+str(year)+'\nObserved: '+observedStr+r' M km$^2$'+'\nTrend: '+linearStr+r' M km$^2$',
# xy=(1., 1.02), xycoords='axes fraction', horizontalalignment='left', verticalalignment='top')
#ax1.annotate(,
# xy=(0.8, 0.85), xycoords='axes fraction', color='r', horizontalalignment='left', verticalalignment='top')
#ax1.annotate('NASA GSFC forecast: '+forecastStr+r' M km$^2$'+'\nSkill:'+skill,
# xy=(0.8, 0.85), xycoords='axes fraction', color='r', horizontalalignment='left', verticalalignment='top')
ax1.annotate(textStr, fontsize=5,
xy=(0.02, 0.02), xycoords='axes fraction', horizontalalignment='left', verticalalignment='bottom')
ax1.set_ylabel('Ice '+iceType+r' (Million km$^2$)')
#ax1.set_xlabel('Years')
ax1.set_xlim(1980, 2020)
ax1.set_xticks(np.arange(1980, 2021, 10))
ax1.set_xticks(np.arange(1980, 2021, 5), minor=True)
#ax1.set_xticklabels([])
ax1.set_ylim(minval, maxval-0.1)
ax1.set_yticks(np.arange(minval, maxval, 1))
ax1.spines['right'].set_visible(False)
ax1.spines['top'].set_visible(False)
ax1.yaxis.grid(which='major', linestyle='-', linewidth='0.1', color='k')
#plt.tight_layout()
#subplots_adjust(left=0.15, right=0.90, bottom=0.17, top=0.96, hspace=0)
subplots_adjust(left=0.11, right=0.74, bottom=0.1, top=0.96, hspace=0)
savefig(figPath+'/'+outStr+hemStr+'multi.png', dpi=300)
savefig(figPath+'/'+outStr+hemStr+'multi.jpg', dpi=300)
close(fig)
#set region features to array
iceConc = ff.get_gridvar(derivedDataPath, forecastVar, month, array(year),
hemStr)
for regInd in range(16):
regionData = iceConc.data
regionData[iceConc.mask == True] = 0 #get rid of nan
desiredRegion = regionMask == regInd
sample[regInd +
2] = 100 * np.ma.mean(np.multiply(regionData, desiredRegion))
sample = np.reshape(sample, (1, -1))
feat = pcaConc.transform(sample)
#feat=sample
_, extent = ff.get_ice_extentN(rawDataPath, fmonth, year, year, iceType,
siiVersion, hemStr)
gTruth = extent[0]
#Ensemble Model run
output = mlpConc.predict(feat) #THIS OUTPUT REPRESENTS SEA ICE FOR JULY
debug = output
#Testing out penalizing small features as inaccuracies
for i in range(np.size(output)):
if (output[0][i] < 0.09):
output[0][i] = 0
debug[0][i] = 0
month = month + 1
sample = np.zeros(numFeat)