Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

35
“Predictive Vegetation Mapping Of Bara And Rautahat Districts Using Artificial Neural Network” Project Members Maheshwor Karki (14) David Nhemaphuki (18) Bibek Karki (13) Email id: [email protected] 1

description

Using Artificial Neural Network Model for the prediction of forest cover of Rautahat an Bara district. Analyze the changing pattern of the forest in these area.

Transcript of Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Page 1: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

“Predictive Vegetation Mapping Of Bara And

Rautahat Districts Using Artificial Neural Network”

Project Members

Maheshwor Karki (14)David Nhemaphuki (18)Bibek Karki (13)

Email id: [email protected]

Page 2: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

1.Background• Forest degradation a challenge for biodiversity conservation

• Agricultural encroachment, Forest extraction, sand and gravel extraction, Urbanisation

• Forest area 4 billion hectares in the world, 13 million hectares of forest is converting in other use each year(FRA,2010)

• In Nepal,

Wed

nesd

ay, A

pril

12, 2

023

• 20 Terai districts 0.06% during 1990/91 to 2000/2001(Nepal’s forestry outlook study,2009)

1.7% per year

1978/79 1994

2

Page 3: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

2.Introduction

Classification, Change Detection Prediction and Mapping

• Provides underlying picture of changes in Land use and land change

• Predict geographic distribution of the vegetation composition

• Managing natural resources

Wed

nesd

ay, A

pril

12, 2

023

3

Page 4: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Contd…Artificial Neural Network (ANN)• Is a feed-forward model • Back propagation learning algorithm

• Improves itself by making corrections to its internal structure

Wed

nesd

ay, A

pril

12, 2

023

4

Page 5: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

3. ObjectiveMain objective • To predict vegetation using models Artificial Neural Network

Sub objectives• To prepare the Multi Temporal Vegetation Coverage Map • To detect the changes• To find out the spatial pattern of vegetation

Wed

nesd

ay, A

pril

12, 2

023

5

Page 6: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

4. Study Area

• Geographic coordinates:Bara: 2 2’ N 8 0’E 7͘͘͘5͘7͘͘͘((( 5͘(Rautahat: 2 46’ N 8 16’E6͘5͘6͘( 5͘(

• Area:Bara: 1190 sq. kmRautahat :1126 sq. km.

• Climatic Zones:Lower TropicalUpper Tropical

Wed

nesd

ay, A

pril

12, 2

023

Map of Nepal

6

Page 7: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

5. Methods

Wed

nesd

ay, A

pril

12, 2

023

7

Page 8: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Image Classification

Wed

nesd

ay, A

pril

12, 2

023

Figure : Image Classification Algorithm

Downloaded Landsat Images

Radiometric Calibration

Dark pixel subtractions

Image Pre-processing

NDVI Image Enhancement

Tasselled Cap Brightness

Selection of ROI Image Classification (Supervised: Maximum Likelihood Algorithm)

Forest and Non Forest Map

Accuracy Assessment

Composite Image

8

Page 9: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Change Detection

Wed

nesd

ay, A

pril

12, 2

023

Classified Image 1999

Classified Image2009

Classified Image2013

Image Differencing1999-1989

 

Image Differencing1999-2009 

 

Image Differencing2009-2013 

  

Temporal Change Detection Map 

    

Figure : Change Detection Method

Classified Image 1989

9

Page 10: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Artificial Neural Network

Wed

nesd

ay, A

pril

12, 2

023

• Emulates properties of biological

nervous system and draw on the

analogies of adaptive biological

learning

Why ANNs?• More accurate than traditional statistical methods

• ANNs can learn from and generalize from experience

• ANNs are universal functional approximator

• Ability to combine data from different source

10

Page 11: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Artificial Neural Network Model Development

Wed

nesd

ay, A

pril

12, 2

023

Classified Vegetation map

DEM

Slope

Aspect

Distance from road

Distance from Settlement

Input Layer Hidden Layer Output Layer

nodes

Figure : multi-layer perceptron neural network11

Page 12: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

6. Result

Wed

nesd

ay, A

pril

12, 2

023

12

Page 13: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Accuracy Assessment

Wed

nesd

ay, A

pril

12, 2

023

2009 2013

Overall accuracy 95.83% 87.85%

Kappa coefficient 0.91 0.76

study2013_ Non_Forest Forest Ground truth

Non_Forest 60 5 65

Forest 10 63 73

Total 70 70 140

study2009 Non_Forest Forest Grount truth

Non_Forest 60 5 65

Forest 0 55 55

Total 60 60 120

Confusion Matrix: 2009 Confusion Matrix: 2013

13

Page 14: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Land Cover

Wed

nesd

ay, A

pril

12, 2

023 Year

Class1989 1999 2009 2013

Change(%) 1989-1999

Change(%) 1999-2009

Change(%) 2009-2013

Change(%) 1989-2013

Forest 48940 48003 45134 43140 -1.91 -5.98 -4.42 -11.85Non-Forest 78282 79215 82084 84078 1.19 3.62 2.43 7.40

Year

Class1989 1999 2009 2013 Change(%)

1989-199Change(%) 1999-2009

Change(%) 2009-2013

Change(%) 1989-2013

Forest 29336 27897 25308 25509 -4.91 -9.25 0.79 -13.05

Non-Forest 74285 75726 78315 78114 1.94 3.42 -0.26 5.15

Land cover of Bara district

1989 1999 2009 201323000

24000

25000

26000

27000

28000

29000

3000029336.85

27897.39

25308.36 25509.69

Forest Area Rautahat

Year

Fore

st A

rea

(hec

tare

s)

1989 1999 2009 20134000041000420004300044000450004600047000480004900050000

48940.6548003.84

45134.91

43140.69

Forest Area in Bara

year

Area

Of f

ores

t in

hect

ares

Land cover of Rautahat district

14

Page 15: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Change Maps

Figure: Change map of Bara: 1989-1999 Figure: Change map of Bara: 1999-200915

Page 16: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Figure. Change map of Bara: 1989-2013Figure. Change map of Bara: 2009-2013

Cont…

16

Page 17: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Figure. Change map of Rautahat: 1989-1999 Figure. Change map of Rautahat: 1999-2009

Cont…

17

Page 18: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Figure. Change map of Rautahat: 1989-2013Figure. Change map of Rautahat: 2009-2013

Cont…

18

Page 19: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Deforestation RateFAO 1995 formula:q=((A2-A1)^1/(t2-t1))-1Where,

q=deforestation rate(% lost areal year)A1=initial forest areaA2=final forest areat2-t1=interval in years during which change in land cover is being assessed

Puyravad Formula:(Based on compound interest and more intuitive than FAO formula)r=1/(t1-t2)lnA2/A1Where,

r=deforestation rateA1=initial forest areaA2=final forest areat2-t1=interval in years during which change in land cover is being assessed 19

Page 20: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Deforestation Rate1989-1999:r=.0019331999-2009:r=.0061632009-2013:r=.011297

1989-1999 1999-2009 2009-20130

0.002

0.004

0.006

0.008

0.01

0.012

0.001933

0.006163

0.011297

Deforestation Rate in Bara

Time Interval

Defo

rest

ation

Rat

e

1989-1999:r=.00503121999-2009:r=.00973972009-2-13:r=-.001980 1989-1999 1999-2009 2009-2013

-0.004-0.002

00.0020.0040.0060.008

0.010.012

0.0050312

0.0097397

-0.00198

Deforestation Rate in Rautahat

Time Interval

Defo

rest

ation

Rat

e

20

Page 21: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Spatial Metrics

Wed

nesd

ay, A

pril

12, 2

023

21

Page 22: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Bara spatial metrics(1989/1999/2009/2013)

class Metrics

1989 1999 2009 2013

Forest Non-Forest Forest Non-Forest Forest Non-Forest Forest Non-Forest

CA-Class Area 48940 78282 48003 79215 45134 82084 43140 84078

NP-Number of patches

823 1779 350

2179 147 1043 233 3158

ED-Edge Density

17.45 18.71 15.83

17.22 11.43 12.74 20.57 21.98

LPI- Largest patch Index

22.15 55.30 21.66 56.21 10.63 61.43 10.46 62.51

CONTAG 43.4473 44.3296 47.0926 53.9301

Class

Metrics

1989 1999 2009 2013Forest Non-Forest Forest Non-Forest Forest Non-Forest Forest Non-

Forest

CA-Class Area

29336 74285 27897 75726 25308 7831 25509 78114

NP-Number of patches

364 1017 148 1040 117 1657 376 1609

ED-Edge Density

11.80 13.28 10.64 12.37 12.61 14.31 16.73 18.39

LPI- Largest patch Index

26.42 69.33 14.21 71.270 23.22 73.98 11.11 73.46

CONTAG 50.9184 52.3493 53.4769 51.6868

Rautahat Spatial metrics(1989/1999/2009/2013)

22

Page 23: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

1989 1999 2009 20130

5

10

15

20

25

22.1545 21.6623

10.6344 10.4668

Largest Pitch Index Bara

Year

LPI

1989 1999 2009 20130

5

10

15

20

25

17.453415.8349

11.436

20.5733

Edge Density Bara

Year

Edge

Den

sity

1989 1999 2009 20130

200

400

600

800

1000

1200

1400

1600

823

350

1478

233

Number of forest patches in Bara

Year

NO

of f

ores

t pat

ch

23

Page 24: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

1989 1999 2009 20130

2

4

6

8

10

12

14

16

18

11.806910.6469

12.6127

16.7341

Edge Density Rautahat

Year

Edge

Den

sity

1989 1999 2009 20130

5

10

15

20

25

30

26.4208

14.2181

23.223

11.1175

Largest Patch Index Rautahat

Year

Larg

est P

atch

Inde

x(LP

I)

1989 1999 2009 20130

50

100

150

200

250

300

350

400364

148117

376

Number of Forest patches Rautahat

Year

No

of F

ores

t pat

ches

24

Page 25: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Prediction

Wed

nesd

ay, A

pril

12, 2

023

25

Page 26: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Figure: Predicted Land cover map of 2013Figure: Classified Land cover map of 2013 Figure: Change map

Prediction of 2013 land cover

Wed

nesd

ay, A

pril

12, 2

023

Classified 2013 Predicted 2013

Bara(forest) Rautahat(forest) Bara(forest) Rautahat(forest)

Area(hectare) 43140 25509 41437 23143

NP 233 376 898 699

ED 20.57 16.73 18.68 14.43

LPI 10.46 11.11 11.12 10.72

CONTAG 53.93 51.68 65.37 70.94

26

Page 27: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Figure: Predicted land cover map of the Study area

Prediction of 2020 land cover

Forest area change (2013-2020): =68648 -64587=4061

Deforestation rate during 2013-2020: 0.0057

27

Page 28: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Discussion• From the year 1989-2013, 5800 and 3800 hectare of forest area has been decreased

in Bara and Rautahat

• ICIMOD - Nepal Land Cover Map 1990 and 2010 deforestation rate is -0.0002 while the deforestation rate of Bara is 0.0009. Our study shows that there is high deforestation in both districts with deforestation rate of 0.0040and 0.0074

• Forest area has been increased in Rautahat during 2009-2013 due to people’s participation and community forest programme through collaborative forest management

• The predicted rate of deforestation is 0.0057 during 2013-2020 of study area

• There is high annual rate of deforestation rate during 1999-2009 in both districts in our case.

Wed

nesd

ay, A

pril

12, 2

023

28

Page 29: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Limitations In field:• All places are not accessible due to limited time

• Data collection time is limited in the morning and evening due to high temperature

Data

• Availability of data is a major problem of this study

• There is no data for the accuracy assessment of the classified image of the year 1989 and 1999

• Social, economic, political and cultural factors have not considered here

Wed

nesd

ay, A

pril

12, 2

023

29

Page 30: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Recommendation• Other variables like solar radiation, climatic data, soil data can be used

• Exchange between forest sub classes could be computed and predicted

• Accuracy assessment of 1989 and 1999 could be done Wed

nesd

ay, A

pril

12, 2

023

30

Page 31: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Conclusion

Wed

nesd

ay, A

pril

12, 2

023

• Change detection and prediction of vegetation mapping of Bara and Rautahat was done

• Rate of deforestation for 2013-2020 has been predicted 0.0057 and also showing significant change of forest in the past

• Timber extraction, urbanisation, agricultural encroachment, sand and gravel extraction, soil erosion are found to be major cause of deforestation

• High rate of deforestation near the forest boundary and settlement area than near the highway

31

Page 32: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Thank you!

Wed

nesd

ay, A

pril

12, 2

023

32

Page 33: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

33

Page 34: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

Explanatory Variable Cramer’s V Test

Distance from roads 0.1703

Distance from settlement 0.4782

DEM 0.7927

Slope 0.2665

Aspect 0.1609

Level of association Verbal description Comments

0.000 No Relationship Independent variable does not help in predicting the dependent variable

0.00 to 0.15 Very weak Not generally acceptable

0.15 to 0.20 Weak Minimally acceptable

0.20 to 0.25 Moderate Acceptable

0.25 to 0.30 Moderately strong Desirable

0.30 to 0.35 Strong Very desirable

0.35 to 0.40 Very strong Extremely desirable

0.40 to 0.50 Worrisomely strong Either an extremely good relationship or the two variables are measuring the same concept

0.50 to 0.99 Redundant The two variables are probably measuring the same concept

1.00 Perfect relationship If we know the independent variable, we can perfectly predict the dependent variable

34

Page 35: Predictive Vegetation Mapping of Bara and Rautahat Districts Using ANN

Wed

nesd

ay, A

pril

12, 2

023

35