Greenland Ice Glacier Analysis - Data Visualization Project

Group project done for the Data science course

Introduction

The single ice sheet or glacier that covers nearly 80% of the island of Greenland is known as the Greenland Ice Sheet, also known as Inland Ice. It is the largest ice mass in the Northern Hemisphere and is only second in size globally to the ice sheet that covers Antarctica. The Greenland ice sheet is adversely affected by climate change. It is more vulnerable to climate change. The ice sheet glacier has been melting and discharging and also it has been accumulating ice for centuries. But we have the data of its discharge and accumulation since 1840 to till date. We have analyzed different features of this glacier like discharge, Mass Balance, Surface mass Balance (SMB), Basal Mass Balance (BMB) and represented using various graphs. The goal of this project is to determine the loss or gain of discharge and mass balance of the Greenland glacier ice sheet for the next decade i.e., 2023-2032.

Importing packages and loading data

• This project uses numpy, pandas to wrangle the data and matplotlib to visualize the data
• I am loading the data from my google drive by using the mount method in colab

import pandas as pd
import numpy as np
from matplotlib import pyplot as plt
from google.colab import drive
drive.mount('/content/drive', force_remount = True)

Mounted at /content/drive

df = pd.read_csv(f"/content/drive/MyDrive/DataScience/Data/MB_SMB_D_BMB.csv")
df.head()

	time	MB	MB_err	SMB	SMB_err	D	D_err	BMB	BMB_err
0	1840-01-01	0.267185	0.373477	1.407972	0.302918	1.079574	0.215092	0.061212	0.038220
1	1841-01-01	0.300225	0.357615	1.438253	0.284199	1.079574	0.213832	0.058454	0.037356
2	1842-01-01	-0.284109	0.357375	0.857085	0.281043	1.079574	0.217393	0.061620	0.038363
3	1843-01-01	-0.320739	0.372270	0.822635	0.301422	1.079574	0.214927	0.063799	0.039190
4	1844-01-01	-0.473110	0.369937	0.670860	0.298957	1.079574	0.214297	0.064395	0.039434

df.info()

<class 'pandas.core.frame.DataFrame'>
RangeIndex: 13629 entries, 0 to 13628
Data columns (total 9 columns):
 #   Column   Non-Null Count  Dtype  
---  ------   --------------  -----  
 0   time     13629 non-null  object 
 1   MB       13629 non-null  float64
 2   MB_err   13629 non-null  float64
 3   SMB      13629 non-null  float64
 4   SMB_err  13629 non-null  float64
 5   D        13629 non-null  float64
 6   D_err    13629 non-null  float64
 7   BMB      13629 non-null  float64
 8   BMB_err  13629 non-null  float64
dtypes: float64(8), object(1)
memory usage: 958.4+ KB

Data Wrangling

• As we can see, the data doesn't have any missing values.
• The attribute time is of the type object, we should convert it to the datetime data type.
• And we should drop some columns as we don't need all of them for furthur classification, this is called future selection.
• For the visualization we don't need to consider the error in calculation, so we can drop all the err columns.
• And if we observe the time column the data till 1985 is annual and from then it's daily so, I choose to group data to make the whole data similar.

df["time"] = pd.to_datetime(df["time"]) #changing data type 
dfs = df[["MB","SMB","BMB", "D"]].copy() #feature selection
dfs["year"] = df["time"].dt.year #adding a year column by extracting year from time attribute
dfm = dfs.groupby(dfs["year"]).mean() #grouping by year

After cleaning the data, we endup like this

• As we grouped data on year, year is the index now

dfm.info()

<class 'pandas.core.frame.DataFrame'>
Int64Index: 183 entries, 1840 to 2022
Data columns (total 4 columns):
 #   Column  Non-Null Count  Dtype  
---  ------  --------------  -----  
 0   MB      183 non-null    float64
 1   SMB     183 non-null    float64
 2   BMB     183 non-null    float64
 3   D       183 non-null    float64
dtypes: float64(4)
memory usage: 7.1 KB

Outlier Analysis

We plot the boxplots for all the attributes to find the outliers.

fig, ax = plt.subplots(2,2, figsize=(12,12))
fig.suptitle("Outlier Analysis",fontsize='large', fontweight='bold')
fig.tight_layout(pad=5.0)

dic = ax[0][0].boxplot(dfm["D"])
ax[0][0].set_title("Discharge",fontsize='large', fontweight='bold')
ax[0][0].set_ylabel("D",fontsize='large', fontweight='bold')

ax[0][1].boxplot(dfm["MB"])
ax[0][1].set_title("Mass Balance",fontsize='large', fontweight='bold')
ax[0][1].set_ylabel("MB",fontsize='large', fontweight='bold')

ax[1][0].boxplot(dfm["SMB"])
ax[1][0].set_title("Surface Mass Balance",fontsize='large', fontweight='bold')
ax[1][0].set_ylabel("SMB",fontsize='large', fontweight='bold')

ax[1][1].boxplot(dfm["BMB"])
ax[1][1].set_title("Basal Mass Balance",fontsize='large', fontweight='bold')
ax[1][1].set_ylabel("BMB",fontsize='large', fontweight='bold')

plt.show()

As the data is not so large, I have manually identified the outliers with this code and droped those rows

dfm["rank"] = dfm["MB"].rank(ascending = True)
dfm = dfm[(dfm["rank"] != 1.0)]
dfm = dfm[(dfm["rank"] != 2.0)]
dfm = dfm[(dfm["rank"] != 3.0)]
dfm["rank"] = dfm["D"].rank(ascending = False)
dfm = dfm[(dfm["rank"] != 1.0)]
dfm["rank"] = dfm["SMB"].rank(ascending = False)
dfm = dfm[(dfm["rank"] != 1.0)]
dfm["rank"] = dfm["BMB"].rank(ascending = True)
dfm = dfm[(dfm["rank"] != 1.0)]
dfm.shape

(177, 5)

After deleting the outliers the boxplots looks clean, so this data is good to fit the models.

fig, ax = plt.subplots(2,2, figsize=(12,12))
fig.suptitle("Distribution of features")
fig.tight_layout(pad=5.0)

ax[0][0].boxplot(dfm["D"])
ax[0][0].set_title("Discharge")
ax[0][0].set_ylabel("D")

ax[0][1].boxplot(dfm["MB"])
ax[0][1].set_title("Mass Balance")
ax[0][1].set_ylabel("MB")

ax[1][0].boxplot(dfm["SMB"])
ax[1][0].set_title("Surface Mass Balance")
ax[1][0].set_ylabel("SMB")

ax[1][1].boxplot(dfm["BMB"])
ax[1][1].set_title("Basal Mass Balance")
ax[1][1].set_ylabel("BMB")

plt.show()

dfm.shape

(177, 5)

Data visualization

How the data is distributed

• By using the histograms we can see how the data is centralized towards the mean.

fig, ax = plt.subplots(2,2, figsize=(12,12))
fig.suptitle("Distribution of features",fontsize='large', fontweight='bold')
fig.tight_layout(pad=5.0)

ax[0][0].hist(dfm["D"])
ax[0][0].set_title("Discharge Distributuion",fontsize='large', fontweight='bold')
ax[0][0].set_xlabel("D",fontsize='large', fontweight='bold')
ax[0][0].set_ylabel("Density",fontsize='large', fontweight='bold')

ax[0][1].hist(dfm["MB"])
ax[0][1].set_title("Mass Balance Distributuion",fontsize='large', fontweight='bold')
ax[0][1].set_xlabel("MB",fontsize='large', fontweight='bold')
ax[0][1].set_ylabel("Density",fontsize='large', fontweight='bold')

ax[1][0].hist(dfm["SMB"])
ax[1][0].set_title("Surface Mass Balance Distributuion",fontsize='large', fontweight='bold')
ax[1][0].set_xlabel("SMB",fontsize='large', fontweight='bold')
ax[1][0].set_ylabel("Density",fontsize='large', fontweight='bold')

ax[1][1].hist(dfm["BMB"])
ax[1][1].set_title("Basal Mass Balance Distributuion",fontsize='large', fontweight='bold')
ax[1][1].set_xlabel("BMB",fontsize='large', fontweight='bold')
ax[1][1].set_ylabel("Density",fontsize='large', fontweight='bold')
plt.show()

• We can see that Discharge has the long right tail, which means the discharge is irregular and it is high sometimes.
• As discharge D is right tailed, mass balance MB is left tailed which means the MB is low for most of the times as a result of high discharge
• other attributes looks good and properly distributed.

How was the Mass balance across the time

• I have used the bar chart to see the mass balance vs year distribution.
• I also colored the bars as green if mass balance is increased compared to previous year and red if not.

mask = dfm["MB"]<0
plt.figure(figsize = (15,10))
plt.bar(dfm.index[mask],dfm["MB"][mask], color = 'red', label = "Ice Loss")
plt.bar(dfm.index[~mask],dfm["MB"][~mask],color = 'green', label = "Ice Gain")
plt.axhline(y = 0, linestyle = '-', color  = 'black')
plt.xlabel('Year',fontsize='large', fontweight='bold')
plt.ylabel('Mass Balance',fontsize='large', fontweight='bold')
plt.title("Yearly Mass Balance of Ice",fontsize='large', fontweight='bold')
plt.xticks(np.arange(1842,2023,10), rotation = 90)
plt.legend()
plt.show()

• As we can see the most of the bars are red, which means most of the time we are loosing the ice.

dfm.head()
plt.figure(figsize = (10,10))
# plt.title("Percentage of Years with ICE LOSS and ICE GAIN during time period 1840 to 2022",fontsize='large', fontweight='bold')
colors = ['#e83845','#32CD32']
dfm["ice"] = np.where(dfm["MB"]<0, "Ice loss", "Ice gain")
patches, texts, autotexts=plt.pie(dfm["ice"].value_counts(), labels = ["Ice loss", "Ice gain"],autopct='%.1f%% of Years',pctdistance=0.8,counterclock=True,startangle=90,colors=colors)
texts[0].set_fontsize(20)
autotexts[0].set_fontsize(20)
texts[1].set_fontsize(20)
autotexts[1].set_fontsize(20)
plt.show()

From the Pie chart we can infer that ice lost most of the times

How was Mass Balance and Discharge trends across the years

• Here we plot line graphs of MB, SMB and Discharge to find out the relation in the changes

fig, ax = plt.subplots(3,1, figsize=(15,10), sharex = True)
fig.suptitle("Distribution of features")
fig.tight_layout(pad=5.0)

ax[0].plot(dfm.index,dfm["D"],color = 'red', label = "D")
ax[1].plot(dfm.index,dfm["MB"], label = "MB")
ax[1].plot(dfm.index,dfm["SMB"], label = "SMB")
ax[0].legend()
plt.xlabel('Year')
ax[0].set_ylabel('Ice Discharge')
ax[1].set_ylabel('Mass Balance')
ax[0].set_title("Ice Discharge rate")
ax[1].legend()
ax[2].plot(dfm.index,dfm["BMB"], label = "BMB")
plt.xticks(np.arange(1842,2023,10), rotation = 90)
plt.legend()
plt.show()

• From the above plot we can see that MB, SMB and BMB follows a similar pattern where as Discharge follows the exact opposite pattern as to them.
• This is due to default reason that mass balance decreses when the ice discharge is high.
• and if we observe carefully, there is slight decrease in the mass balance across the time.

Correlation between all off them

• we plot them scatter plot between the features to find out how they are correlated.

fig,axc = plt.subplots(2,2,figsize=(12,12))
fig.suptitle("Correlation between features",fontsize='large', fontweight='bold')
fig.tight_layout(pad=5.0)

m, b = np.polyfit(dfm["SMB"],dfm["MB"], 1)
axc[0][0].set_title("SMB and MB Correlation",fontsize='large', fontweight='bold')
axc[0][0].set_xlabel("Surface MB",fontsize='large', fontweight='bold')
axc[0][0].set_ylabel("Mass Balance (MB)",fontsize='large', fontweight='bold')
axc[0][0].plot(dfm["SMB"], m*dfm["SMB"]+b, color='red')
axc[0][0].scatter(dfm["SMB"],dfm["MB"])

m, b = np.polyfit(dfm["D"],dfm["MB"], 1)
axc[0][1].set_title("MB and D Correlation",fontsize='large', fontweight='bold')
axc[0][1].set_ylabel("Mass Balance (MB)",fontsize='large', fontweight='bold')
axc[0][1].set_xlabel("Discharge (D)",fontsize='large', fontweight='bold')
axc[0][1].plot(dfm["D"], m*dfm["D"]+b, color='red')
axc[0][1].scatter(dfm["D"],dfm["MB"])

m, b = np.polyfit(dfm["D"],dfm["SMB"], 1)
axc[1][0].set_title("SMB and D Correlation",fontsize='large', fontweight='bold')
axc[1][0].set_ylabel("Surface MB",fontsize='large', fontweight='bold')
axc[1][0].set_xlabel("Discharge (D)",fontsize='large', fontweight='bold')
axc[1][0].plot(dfm["D"], m*dfm["D"]+b, color='red')
axc[1][0].scatter(dfm["D"],dfm["SMB"])

m, b = np.polyfit(dfm["D"],dfm["BMB"], 1)
axc[1][1].set_title("BMB and D Correlation",fontsize='large', fontweight='bold')
axc[1][1].set_ylabel("Basel MB",fontsize='large', fontweight='bold')
axc[1][1].set_xlabel("Discharge (D)",fontsize='large', fontweight='bold')
axc[1][1].plot(dfm["D"], m*dfm["D"]+b, color='red')
axc[1][1].scatter(dfm["D"],dfm["BMB"])

plt.show()

• The Mass Balances are positiviely correlated to eachother whereas any Mass Balance with discharge is negatively correlated except for Basel MB.
• From this we can infer that Basel Mass Balance doesn't effect for the ice discharge.
• The discharge is mainly due to the Surface Mass Balance.

After the data analysis we did apply some regression models on this data to find out trends in ice discharge for next 10 years.