--- Cell 1 ---
SIMMAX.ipynb

--- Cell 2 ---
# Paleotemperatures with SIMMAX 

--- Cell 3 ---
from pangaeapy import PanDataSet
import math
import sys
import pandas as pd
analogDS = PanDataSet('10.1594/PANGAEA.77352')
sampleDS=PanDataSet('10.1594/PANGAEA.55156')
#The annual mean
#levitus=pd.read_csv('woa13_decav_t00mn01v2.csv',sep=';')
#summer
levitus=pd.read_csv('woa13_decav_t15mn01.csv',sep=';')

#Compare with:
#http://discovery.ucl.ac.uk/101363/1/2002PA000774.pdf

--- Cell 4 ---
def getLevitusTemp(lat, lon, lev):
    depth=30
    t=0
    #select the values for the upper water column
    temp=lev[['0','5','10','15','20','25','30','35','40','45','50']][(lev['LATITUDE']== (math.floor(lat)+0.5)) & (lev['LONGITUDE']== (math.floor(lon)+0.5))]
    #delete empty cells
    temp=temp.dropna(axis=1).to_dict(orient='list')
    #cast dict keys from string to int
    temp={int(key): value for key, value in temp.items()} 
    temp_keys=list(temp.keys())
    #find the closest available water depth value
    closest=min(temp_keys, key=lambda x:abs(x-depth))
    if len(temp[closest])>0:
        t=temp[closest][0]
        if t==None:
            print(str(lat)+' x '+str(lon))
            t=0
    else:
        print(str(lat)+' + '+str(lon))
        t=0
    return t
analogDS.data['Temperature']=analogDS.data.apply(lambda x: getLevitusTemp(x['Latitude'],x['Longitude'], levitus), axis=1)

--- Cell 5 ---
sampleDS.data[(sampleDS.data['Event']=='GIK16458-1')].head()

--- Cell 6 ---
#Sometimes the dataset also contains data from different depths of one event, we simplify the procedure here a bit and remove event duplicates e.g. deeper samples
sampleDS.data.drop_duplicates(subset='Event', keep='first', inplace=True)
sampleDS.data[(sampleDS.data['Event']=='GIK16458-1')].head()

--- Cell 7 ---
def getDistance(row):
    
    R = 6372800  # Earth radius in meters
    lat1, lon1 = row['Latitude1'], row['Longitude1']
    lat2, lon2 = row['Latitude2'], row['Longitude2']
    
    phi1, phi2 = math.radians(lat1), math.radians(lat2) 
    dphi       = math.radians(lat2 - lat1)
    dlambda    = math.radians(lon2 - lon1)
    
    a = math.sin(dphi/2)**2 + \
        math.cos(phi1)*math.cos(phi2)*math.sin(dlambda/2)**2
    
    return 2*R*math.atan2(math.sqrt(a), math.sqrt(1 - a))/1000
#Remove duplicates

#Merge to receive a sampl x analog matrix
Distances = pd.merge(sampleDS.data[['Event','Latitude','Longitude', 'Depth']].assign(k=1), analogDS.data[['Event','Latitude','Longitude','Temperature']].assign(k=1), on='k', suffixes=('1', '2')).drop('k', axis=1)
Distances['Event1']=Distances['Event1']+'_'+Distances['Depth'].map(str)
Distances['Distance']=Distances.apply(getDistance, axis=1)
#But we need to avoid to run into a division by zero trap, preperae by the simmean algo which will snap if distance is zero: (sj/dj)
#Therefore we need to replace zero distances by a very, very low distance e.g. 0.000000001 m or so
Distances.loc[Distances['Distance']==0,'Distance']=0.00000000001
Distances.set_index(['Event1', 'Event2'], inplace=True)  

--- Cell 8 ---
Distances.head()  

--- Cell 9 ---
#Forams used by Pflaumann et. al in SIMMAX28
#G. mentum = Globorotalia cultrata and tumida
foramCols=[]
foramParams=['G. bulloides','G. calida','G. falconensis','G. quinqueloba','G. rubescens','G. digitata','G. aequilateralis',
             'G. conglobatus','G. ruber p','G. ruber w', 'G. tenellus','G. trilobus tril','G. trilobus sac','O. universa',
             'S. dehiscens','G. crassaformis','G. mentum', 'G. hirsuta', 'G. inflata', 'G. scitula','G. truncatulinoides',
             'N. dutertrei','N. pachyderma s','P/D int','G. glutinata','P. obliquiloculata']

for fP in foramParams:
    if fP in sampleDS.data.columns:
        foramCols.append(fP)

--- Cell 10 ---
#Lets do the SIMMAX
import numpy as np
#sum of squared percentages 
sampleDS.data['SQRTSUM']=np.sqrt((sampleDS.data[foramCols]**2).sum(axis=1))
analogDS.data['SQRTSUM']=np.sqrt((analogDS.data[foramCols]**2).sum(axis=1))
sampleIdx=sampleDS.data[foramCols].div(sampleDS.data['SQRTSUM'], axis=0)
analogIdx=analogDS.data[foramCols].div(analogDS.data['SQRTSUM'], axis=0)
sampleIdx['Event']=sampleDS.data['Event']+'_'+sampleDS.data['Depth'].map(str)
sampleIdx.set_index('Event', inplace=True)
analogIdx['Event']=analogDS.data['Event']
analogIdx.set_index('Event', inplace=True)
SimIndex = pd.DataFrame(
    (analogIdx[foramCols].values * sampleIdx[foramCols].values[:, None]).reshape(-1, analogIdx.shape[1]),
    pd.MultiIndex.from_product([sampleIdx.index, analogIdx.index]),
   sampleIdx.columns
)
Similars=pd.DataFrame(SimIndex.sum(axis=1), columns=['Similarity'])
Similars['Distance']=Distances['Distance']
Similars['Latitude']=Distances['Latitude1']
Similars['Longitude']=Distances['Longitude1']
Similars['Temperature']=Distances['Temperature']

--- Cell 11 ---

#s=Similars.loc['PS1231-2'].sort_values(['Similarity'],ascending=False)
#s[s['Similarity']>=0.999]
s=Similars[Similars['Similarity']>=0.999]

--- Cell 12 ---
#Original SIMMAX is using those values with similarity index >0.79 only
#mostSimilars=Similars.loc[(Similars['Similarity']>=0.79)]
#The revised SIMMAX is using the 10 top most similars with sim inded >0.9
mostSimilars=Similars[Similars['Similarity']>=0.9].sort_values(by='Similarity',ascending=False).groupby(level=0).head(10).sort_index(level=0,sort_remaining=False)
mostSimilars.index.names=['sampleEvent', 'analogEvent']  
mostSimilars

--- Cell 13 ---
analogDS.data[analogDS.data['Event'].isin(['V27-23','PS1707-1','RC13-275','RC11-79'])]

--- Cell 14 ---
#Paleotemperature calculations following Pflaumann's Formula (3)
PaleoTemp=pd.DataFrame()
PaleoTemp['TEMP']=((mostSimilars['Temperature']*mostSimilars['Similarity']/ mostSimilars['Distance']).groupby(level=0).sum())/((mostSimilars['Similarity']/mostSimilars['Distance']).groupby(level=0).sum())

PaleoTemp['LAT']=mostSimilars.groupby(level=0)['Latitude'].max()
PaleoTemp['LON']=mostSimilars.groupby(level=0)['Longitude'].max()

--- Cell 15 ---
import cartopy.crs as ccrs
import cartopy.feature as feat
import matplotlib.pyplot as plt
proj=ccrs.LambertCylindrical()
plt.figure(figsize=(10,10),dpi=200)
ax = plt.axes(projection=proj)
land_50m = feat.NaturalEarthFeature('physical', 'land', '50m',edgecolor='grey',facecolor=feat.COLORS['land'])

#PaleoTemp = PaleoTemp[np.isfinite(PaleoTemp['TEMP'])]
#print(PaleoTemp)

ax.tricontourf(PaleoTemp['LON'],PaleoTemp['LAT'], PaleoTemp['TEMP'],levels=[1,1.5,2,2.5,3.5,4,5,6,8,12,16,20,22,24,26,28],cmap = "rainbow",transform=ccrs.Geodetic())
ax.add_feature(land_50m)
ax.scatter(PaleoTemp['LON'],PaleoTemp['LAT'],label=None, alpha=0.5, s=1,cmap = "rainbow",transform=ccrs.Geodetic())

for i, point in PaleoTemp.iterrows():
        ax.text(point['LON'], point['LAT'], str(point['TEMP'].round(1)),fontsize=7, transform=ccrs.Geodetic())

--- Cell 16 ---
## References:

--- Cell 17 ---
print(analogDS.citation)
print()
print(sampleDS.citation)

--- Cell 18 ---