The impact of climate change human activity on water sediment artificial neutral network modelling in the longchuanjiang catchment, upper yangtze river

Acknowledgements

This thesis is the result of four years of work whereby I have been accompanied and
supported by many people. It is a pleasant aspect that I have now the opportunity to express
my gratitude for all of them.
I’d like to sincerely thank my supervisor A/P Xixi Lu for providing me with the opportunity
to conduct PhD studies with him. Without his involvement and advice in the research, this
thesis would have never been ready in the present form. My gratitude goes to A/P Shie-Yui
Liong from Department of Civil Engineering, NUS, for giving me the confidence and support
to work with neural network. I would also like to express my sincere appreciation to A/P
David Higgitt from Department of Geography, NUS, for his stimulating suggestions and
encouragement. I am grateful for Miss Pauline Lee in Department of Geography for her
administrative assistance.
I am grateful for Prof. Yue Zhou from Kunming University of Science and Technology for his
support during the research. Thanks Youan Guo, Hongbo Li and Jingping Wang for their kind
help in the field work.
Friendship makes my life in Singapore more enjoyable. I thank all graduate students in the
Department of Geography. Thanks go to Luqiang, Jiangnan, Jianfeng and Hanbing, for
sharing the joys and trials with me; Joy, for being my first teacher in remote sensing; Shurong,
for her help in statistics; Chih Yuan and Songuang, for being my walking Chinese-English
dictionary and for fixing up the printer on the right time. A special thank goes to Gu Ming, for
making the thesis period less stressful (I miss the time of “BaGua” with her).
I feel a deep sense of gratitude for my parents. I’d like to thank my sister for talking to me on
phones for hours. I am very grateful for my husband Liang, for his love and support in
everything.


I
Table of Contents

Acknowledgements I

Table of Contents II
Summary VII
List of Tables IX
List of Figures XII
List of Plates XVII
1 Introduction 1
1.1 Background 1
1.1.1 Impact of climate and land use change on water discharge and sediment
flux 1
1.1.2 Mathematical modelling of the impact and limitation of current research 3
1.1.3 Longchuanjiang catchment for impact study 6
1.2 Aims and objectives of the study 8
1.3 Framework of methodology 9
1.4 Arrangement and structure of the dissertation 11
2 Study area 13
2.1 Physical characteristics of the catchment 14
2.2 Climate in the catchment 17
2.3 Social and economic environment 19
2.4 Problem statement of the catchment 20
3 Hydroclimatic change in the catchment 24
3.1 Data and method 25
3.1.1 Data 25
3.1.2 Mann-Kendall nonparametric trend test 26
3.1.3 Sen’s slope test 29
3.1.4 Pettitt change-point test 30
3.2 Climate change in the catchment 31
3.2.1 Spatial variety of climate change in the catchment 31
3.2.2 Catchment average rainfall change 38
3.2.3 Catchment average temperature change 42
3.3 Change of water discharge at Huangguayuan 43

3.3.1 Annual water discharge 43

II
3.3.2 Seasonal water discharge 47
3.3.3 Maximum monthly and daily water discharge 50
3.3.4 Minimum monthly and daily water discharge 51
3.3.5 Flow duration curve at Huangguayuan 52
3.4 Change of sediment flux at Huangguayuan 56
3.4.1 Annual average sediment flux 56
3.4.2 Seasonal sediment flux 60
3.4.3 Maximum monthly and daily sediment flux 62
3.5 Water discharge and sediment flux in the upper and lower reaches 62
3.6 Possible relations between changes in rainfall, water and sediment 65
4 Land cover/use change in the catchment 70
4.1 Introduction 70
4.2 Materials and method 70
4.3 Satellite image processing 74
4.3.1 Main characteristics of satellite images used 76
4.3.2 Procedure of land cover/use classification 76
4.3.3 Pre-classification work 79
4.3.4 Supervised classification 81
4.3.5 Post classification 89
4.3.6 Image classification results 95
4.4 Land cover/use change at catchment scale 101
4.5 Land cover/use change along the Longchuanjiang River 106
4.6 Conclusion 112
5 Modelling water discharge with Artificial Neural Network 114
5.1 Mathematical models for hydrological modelling 114
5.2 Basics of Artificial Neural Network 119
5.2.1 Structure of MLP 120

5.2.2 Information processing in MLP 122
5.2.3 Evaluation of MLP performance 124
5.3 Review on the application of ANN in hydrological modelling 126
5.3.1 Ability of ANN in rainfall-runoff modelling 131
5.3.2 Architecture of ANNs 132
5.3.3 Conclusion related to the selection of input variables 133
5.3.4 Data set partitioning for calibration and validation 134
5.3.5 Improvement to conventional ANN 136
5.4 Modelling water discharge of the Longchuanjiang River with ANN 139

III
5.4.1 Introduction 139
5.4.2 Materials and method 141
5.4.3 Result and discussion 149
5.5 Conclusion 154
6 Modelling suspended sediment flux with Artificial Neural Network 156
6.1 Introduction 156
6.2 Materials and data 163
6.3 Methodology 166
6.3.1 Data processing 166
6.3.2 Artificial Neural Networks 167
6.3.3 Multiple linear regression (MLR) and power relation (PR) models 169
6.4 Application and results 170
6.4.1 Results from ANNs 170
6.4.2 Results from the MLR and PR models 174
6.5 Discussion 174
6.6 Conclusion 181
7 Anthropogenic impact on sediment—qualitative analysis 183
7.1 Introduction 183
7.2 Statistical evidence of the anthropogenic impact 190

7.3 Impact of deforestation and reforestation 195
7.4 Impact of agriculture intensification 202
7.5 Impact of engineering projects 207
7.6 Impact of dams and reservoirs 211
7.7 Conclusion 218
8 Anthropogenic impact on sediment—ANN modelling 222
8.1 Introduction 222
8.2 Differentiating influences from climate change and human activity 228
8.2.1 Data 228
8.2.2 Double mass curve method 229
8.2.3 Linear regression method 232
8.2.4 ANN method 233
8.2.5 Discussion 238
8.3 Differentiating influences from individual human activity 240
8.3.1 Method 240
8.3.2 Model application and result discussion 250
8.4 Conclusion 254

IV
9 Sensitivy of water discharge and sediment flux to climate change 256
9.1 Introduction 256
9.2 Methodology 260
9.2.1 Climate scenarios 260
9.2.2 Data and method 262
9.3 Performance of ANNs 264
9.4 Sensitivity of water discharge to climate change 265
9.4.1 Overall changes in water discharge 265
9.4.2 Seasonal changes in water discharge 267
9.5 Sensitivity of sediment flux to climate change 269
9.5.1 Results of ANN prediction 269

9.5.2 Sensitivity of sediment flux to rainfall and temperature 273
9.5.3 Sediment flux under possible future climate in the Longchuanjiang
catchment 277
9.6 Sensitivity of water and sediment to climate change under changing human activity
278

9.6.1 Introduction 278
9.6.2 Sensitivity of water discharge under Level I and Level II human activity
280
9.6.3 Sensitivity of sediment flux under Level I and Level II human activity
282
9.7 Conclusion 285
10 Conclusions 289
10.1 A brief overview of the study 289
10.2 Main findings of the study and their implication 290
10.2.1 Land use, climate and hydrological change in the catchment 290
10.2.2 Anthropogenic impact on sediment—qualitative analysis 292
10.2.3 Anthropogenic impact on sediment—ANN modelling 294
10.2.4 Sensitivity of water and sediment to climate change 295
10.3 Application of ANN in hydrological modelling 297
10.4 Limitation of the study 299
10.4.1 Causal variables considered 299
10.4.2 Land use/cover data 300
10.4.3 Applicability of the study results to other catchments 301
10.5 Prospects and future work 301
10.5.1 Examination of hydrological change on shorter temporal scale 301
10.5.2 Influence of reservoir and road construction 302

V
10.5.3 Modification to ANN 302

Appendix I Data collection and quality control at gauging station in the
Longchuanjiang catchment 303
Bibliography 306


VI

Summary

Climate change coupled with intensified human activity could significantly affect the
hydrological processes and have posed a serious threat to the sustainable management of the
river system. This research aimed to investigate the impact of climate change and human
activities on river water discharge and, particularly, suspended sediment flux with a case
study in the Longchuanjiang catchment, the Upper Yangtze River, China. Non-updating
artificial neural network (ANN) was used as a modelling tool to assess the influence from
human activities and to project the response of water discharge and sediment flux under
hypothetical climate scenarios.
The study area had experienced a sharp increase in suspended sediment flux in the post-1990
period. The research indicated that compared with the background condition (1960-1990), the
intensification of human activity had lead to an increase of 2.76 million t yr
-1
in years from
1991 to 2001. Of the total change in sediment flux this period, the contribution of the
intensified human activity exceeded that of the increased rainfall, with the former accounting
for 66~75% and the latter for 25~34%. Among the various human activities, road construction
was the dominant variable for the increase of the sediment. During the period from 1991 to
2001, road construction was estimated to have result in an increase of 30.01 million t in the
total sediment flux. But meanwhile, conversion of barren land to range land in areas along the
channel resulted in a reduction of 9.88 million t in sediment. Reservoir was another factor that
contributed to reduce the sediment in the river. The trapping efficiency of the reservoirs in

was estimated to be approximately 90%. The change of forest in the catchment was failed to
be related to sediment in the river due to various reason like the immaturity of the trees, the
lack of the undergrowth and the location of the reforested area.
Climate change will affect water and sediment in a river. The sensitivities of water discharge
and sediment flux to 25 hypothetical climate changes were predicted by ANNs. Under the
possible future climate change in the catchment till 2050, the change of sediment flux was
estimated to be between -0.7%~13.7%. In addition, sediment under intensified human activity
was found to be more sensitive to the climate change.
ANN provides a competitive alternative to the physical and conventional empirical models in
hydrological modelling, especially in sediment modelling. The current study indicated that

VII
ANN is capable of modelling the monthly water discharge and sediment flux with fairly good
accuracy when proper input variables representing drivers and their lag effect are included.
One significant feature of the ANN in the current study is that it relates sediment directly to
the drivers that have physical influence on it. Such ANN can be used to investigate the
physical relationship between the drivers and the water/sediment and it permits the
assessment of hydrological responses to climate change and human activity.
The current research demonstrated a method for use in studying the impact of climate change
and human activity on water discharge and sediment flux. The conclusion drawn may provide
information for understanding the complicated hydrological system and its response to the
changing climate and human activity. Further research on the influences from variables such
as gully erosion, sediment re-transportation, location of road and reservoir retention may help
to elucidates hydrological change in the catchment.


VIII
List of Tables

Table 2.1 Annual water and sediment discharge at Xiaohekou and Huangguayuan (1957-

2001) 16

Table 2.2 Forest areas in the catchment 17
Table 2.3 Summary of some climate indicators in the Longchuanjiang catchment (1960-2001)
18

Table 2.4 Reservoirs in the Longchuanjiang catchment 20
Table 2.5 Change of road and railway length in the catchment 20
Table 2.6 Soil erosion affected area in the catchment 21
Table 3.1Estimated coefficients of linear regression equations for annual rainfall and annual
average temperature 32

Table 3.2 Estimated coefficients of linear regression equations for maximum monthly and
daily rainfall 33

Table 3.3 Estimated coefficients of linear regression equation for annual potential evaporation
and annual average humidity 36

Table 3.4 Trend in rainfall and temperature time series (Mann-Kendall test and Sen’s test) 40
Table 3.5 Change-point in rainfall and temperature time series in the Longchuanjiang
catchment (with Pettitt test) 40

Table 3.6 Trend in seasonal rainfall time series (Mann-Kendall test) 42
Table 3.7 Estimated coefficients of linear regression equations for annual, maximum
monthly/daily and minimum monthly/daily water discharge 44

Table 3.8 Trend in water discharge time series (Mann-Kendall test and Sen’s test) 46
Table 3.9 Change-point in water discharge time series at Huangguayuan (Pettitt’s test) 47
Table 3.10 Trend in seasonal water discharge time series (Mann-Kendall test) 49
Table 3.11 Statistics of flow duration curves at Huanguayuan in 1960-1974, 1975-1990 and

1991-2001 56

Table 3.12 Estimated coefficients of linear regression equations for annual, maximum
monthly/daily sediment flux 58

Table 3.13 Trend in sediment flux time series (Mann-Kendall test and Sen’s test) 59
Table 3.14 Change-point in sediment flux time series at Huangguayuan 59
Table 3.15 Trend in seasonal sediment flux time series (Mann-Kendall test and Sen’s test) 61
Table 4.1 Data sources for land use/cover time series analysis 71
Table 4.2 Main characteristics of Landsat MSS, TM and ETM+ images 77
Table 4.3 Land use/cover classification scheme 80
Table 4.4 Jeffries-Matusita and Transformed Divergence between land cover pairs 87
Table 4.5 Confusion matrix for image MSS 1974 of the Longchuanjiang catchment 92
Table 4.6 Confusion matrix for image TM 1989 of the Longchuanjiang catchment 93
Table 4.7 Confusion matrix for image ETM+ 1999 of the Longchuanjiang catchment 94

IX
Table 4.8 Land cover/use of the Longchuanjiang catchment in 1974, 1989 and 1999 96
Table 4.9 Medium sized reservoirs in the Longchuanjiang catchment 102
Table 4.10 Arable land in the Longchuanjiang catchment (from statistical year book) 103
Table 4.11 Forest land in the catchment—from document and satellite image classification105
Table 4.12 Changes between forest, range land and barren land 107
Table 4.13 Land cover/use along the Longchuanjiang River 109
Table 5.1 Details of hydrologic models reviewed 117
Table 5.2 Review of papers on the application of ANN in hydrological modelling 127
Table 5.3 Cross-correlation (r) between climate variables and water discharge (average
interval: month) 145

Table 5.4 Inputs combinations for water discharge modeling (average interval: month) 146
Table 5.5 Statistical characteristics of the calibration, testing and validation data sets (average

interval: month) 148

Table 5.6 Performance of ANN for water discharge modelling in the Longchuanjiang basin
(average interval: month) 151

Table 6.1 Characteristics of previous works on modelling sediment discharge with ANN 160
Table 6.2 Statistical parameters of hydro-climatic data for the Longchuanjiang catchment.165
Table 6.3 Correlation coefficients (r) of the hydro-climatic data for the Longchuanjiang
catchment 166

Table 6.4 Performances of ANNs, MLR and PR models for sediment flux modelling in the
Longchuanjiang basin (average interval: month) 171

Table 6.5 Estimated MLR and PR models for sediment flux modelling in the Longchuanjiang
basin (average interval: month) 174

Table 7.1 P values for linear regression of annual rainfall, water discharge and sediment flux
in 1960-1990 and 1991-2001 periods 192

Table 7.2 Input combination and performance of ANN_spatial 194
Table 7.3 Ecological projects in the Longchuanjiang catchment 197
Table 7.4 Change of soil erosion area between 1987 and 1999 197
Table 7.5 Land use/cover change in the dry-hot valley 205
Table 7.6 Sediment deposition in reservoirs 213
Table 7.7 Trapping efficiency of medium-sized reservoirs in the Longchuanjiang catchment
215

Table 7.8 Soil erosion rate estimated from plot (after Yunnan Hydraulic Bureau, 1987) 216
Table 7.9 Change of human activity in the pre- and post-1990 periods and it hydrological
effect 220

Table 8.1 Characteristics of methods reviews of sediment flux change related impact
assessment 227

Table 8.2 Statistical parameters of hydro-climatic data in pre- and post-1990 period 229
Table 8.3 Annual sediment flux under background, only climate variation and both
climate/human activity change — double mass curve method 231


X
Table 8.4 Annual sediment flux under background, only climate variation and both
climate/human activity change — linear regression method 233

Table 8.5 Performances of ANNs 235
Table 8.6 Comparison of the performance of S60-90 and ANN_6 235
Table 8.7 Annual sediment flux under background, only climate variation and both climate
and human activity changes — ANN method (ANN used: S60-90) 238

Table 8.8 Possible variables representing human influence on sediment flux 242
Table 8.9 Input combination tested 244
Table 8.10 Performance of ANNs tested in identifying dominant human activities 245
Table 8.11 Actual sediment flux, estimated sediment without road construction and estimated
sediment without reforestation 253

Table 9.1 Predicted climate changes by GCMs and historical data 261
Table 9.2 Performances of ANNs for the Longchuanjiang catchment 265
Table 9.3 Seasonal runoff changes (%) with changes in rainfall and temperature 268
Table 9.4 Change of sediment discharge with changes in rainfall and temperature 271
Table 9.5 Sensitivity of water discharge to climate scenarios under Level I and Level II
human influence 281

Table 9.6 Sensitivity of sediment flux to climate scenarios under Level I and Level II level
human influence 284



XI
List of Figures
Figure 1.1 Framework of methodology 10
Figure 2.1 Location of the Longchuanjiang Catchment 13
Figure 2.2 Topography of the Longchuanjiang catchment (unit: meter) 14
Figure 2.3 Spatial variation of rainfall (mm) and temperature (ºC) in the catchment 18
Figure 2.4 Time series of annual average suspended sediment load and water discharge 22
Figure 3.1 Thiessen polygons of the weather stations in the Longchuanjiang catchment 26
Figure 3.2 Time series of annual rainfall (a) and temperature (b) at individual weather stations
(straight line indicates the linear regression trend for time series with significant
changes) 34

Figure 3.3 Time series of maximum monthly (a) and daily (b) rainfall for the year at
individual weather stations (strait line indicates the linear regression trend for time
series with significant changes) 35

Figure 3.4 Time series of annual potential evaporation (a) and annual average humidity (b) at
individual weather stations (straight line indicates the linear regression trend for
time series with significant changes) 37

Figure 3.5 Time series of annual catchment rainfall in the Longchuanjiang catchment 39
Figure 3.6 Change-point in annual rainfall time series of the Longchuanjiang catchment
(Dashed line indicates the change-point) 39

Figure 3.7 Time series of maximum monthly and daily rainfall in the Longchuanjiang
catchment 41

Figure 3.8 Long term monthly average rainfall 42
Figure 3.9 Time series of annual average temperature in the Longchuanjiang catchment 43
Figure 3.10 Time series of annual water discharge at Huangguayuan 44
Figure 3.11 Change-point in annual water discharge time series at Huangguayuan (Dash line
indicates the change-point) 45

Figure 3.12 Long term monthly average water discharge and standard deviation 47
Figure 3.13 Time series of maximum monthly and daily water discharge at Huangguayuan.50
Figure 3.14 Time series of minimum monthly and daily water discharge at Huangguayuan 51
Figure 3.15 Flow duration curve at Huangguayuan (1960-1974) 53
Figure 3.16 Flow duration curve at Huangguayuan (1975-1989) 54
Figure 3.17 Flow duration curve at Huangguayuan (1990-2001) 55
Figure 3.18 Time series of annual sediment flux at Huangguayuan 57
Figure 3.19 Change-point in annual sediment flux time series at Huangguayuan (Dashed line
indicates the change-point) 58

Figure 3.20 Long term monthly average sediment flux and standard deviation 60
Figure 3.21 Time series of maximum monthly and daily sediment flux at Huangguayuan 62
Figure 3.22 Sen’s slope of changes in annual average and extreme water discharge/sediment
flux at Xiaohekou and Huangguayuan (1970-2001) (Q: water discharge; Qs:
sediment flux) 63


XII
Figure 3.23 Sen’s slope of change in seasonal water discharge 64
Figure 3.24 Sen’s slope of change in seasonal sediment flux 64
Figure 3.25 Cumulative rainfall, water discharge and sediment flux from 1960 to 2001

(straight line indicates trendline based on data from 1960-1990) 65

Figure 3.26 Scatter plot of rainfall-water discharge and rainfall-sediment flux
(Huangguayuan)(Grey line indicates the linear relation between water/sediment and
rainfall in 1960-1990; dark line for 1991-2001. Notice the higher sediment-rainfall
ratio in 1991-2001) 66

Figure 3.27 Scatter plots of cumulative water discharge against water discharge (a), rainfall
against water discharge (b), rainfall against sediment flux (c)(straight line indicates
trendline based on data from 1960-1990) 67

Figure 3.28 Sen’s slope of rainfall, water discharge and sediment flux (Huangguayuan, 1970-
2001) 67

Figure 4.1 Flow chart for land cover/use data retrieving and analysis 72
Figure 4.2 Landsat MSS, TM and ETM+ false color images covering the Longchuanjiang
catchment 78

Figure 4.3 Procedure of land use/cover classification 79
Figure 4.4 Ground reference data collected in field survey in 2004 81
Figure 4.5 DN values of the training sets for each land cover type in the Longchuanjiang
catchment (image TM 1989) 83

Figure 4.6 Sample histograms for data points included in the training areas for cover type
“forest” in the Longchuanjiang catchment (TM 1989) 85

Figure 4.7 Two-dimensional scatter plot for separability assessment (TM 1989) 86
Figure 4.8 Land cover/use of Longchuanjiang catchment in 1974 97
Figure 4.9 Land cover/use of Longchuanjiang catchment in 1989 98
Figure 4.10 Land cover/use of Longchuanjiang catchment in 1999 99

Figure 4.11 Percentage of land cover/use type in the Longchuanjiang catchment: (a) 1974, (b)
1999 and (c) 1989 100

Figure 4.12 Land cover/use within 1,000m of the Longchuanjiang River (1999) 108
Figure 4.13 Land cover along the river system: (a) 1974, (b) 1989 and (c) 1999 111
Figure 5.1 Structure of Multilayer Perceptrons (MLP) 121
Figure 5.2 Structure of individual neuron 121
Figure 5.3 Framework of network construction 142
Figure 5.4 Thiessen polygons of the weather stations in the Longchuanjiang catchment 143
Figure 5.5 Observed and predicted monthly runoff by model W4—validation period 153
Figure 5.6 Observed and predicted monthly runoff by model W8—validation period 153
Figure 5.7 Observed and predicted monthly runoff by model W9—validation period 153
Figure 6.1 Architecture of the MLP used and the schematic representation of a neuron 168
Figure 6.2 Comparison between the observed and predicted sediment fluxes based on
validation data (a) ANN_2, (b) ANN_6, (c) ANN_9 and (d) ANN_14 173


XIII
Figure 6.3 Comparison between the observed and predicted sediment fluxes based on
validation data (a) MLR_A, (b) MLR_B, (c) MLR_C, and (d) PR. 175

Figure 6.4 Scatter plots of the observed and predicted sediment fluxes by the best performing
ANN in each group and the MLR/PR models, based on validation data (a) ANN_2,
(b) ANN_6, (c) ANN_9 and (d) ANN_14, (e) MLR_A, (f) MLR_B, (g) MLR_C,
and (h) PR. 176

Figure 6.5 Comparison between observed and predicted cumulative suspended sediment load
values by the best performing ANN of each group and MLR/PR models based on
validation data (a) MLR_A and ANN_2, (b) MLR_B and ANN_6, (c) MLR_C and
ANN_9, (d) PR and ANN_14. 177

Figure 6.6 Difference between observed and predicted sediment fluxes by selected ANNs (a)
ANN_2, (b) ANN_6, (c) ANN_9 181

Figure 7.1 Time series of annual rainfall, water discharge and suspended sediment flux (a)
annual rainfall, (b) annual water discharge and (c) annual suspended sediment flux
(straight line represents average during the corresponding time period) 191

Figure 7.2 Changing relationship between water discharge and sediment flux (a) scatter plot
of annual water discharge and sediment flux, and (b) relationship between
cumulative water discharge and sediment flux 192

Figure 7.3 Relationship between cumulative rainfall-cumulative water discharge and
cumulative rainfall-cumulative sediment flux from 1960 to 2001 (straight line
indicates trendline based on data from 1960-1990) 193

Figure 7.4 Scatter plot of observed and predicted sediment flux by ANN_6 and ANN_spatial
195

Figure 7.5 Difference between observed and predicted sediment flux by ANN_6 and
ANN_spatial 195

Figure 7.6 Location of medium-sized reservoirs and change of forest between 1989 and 1999
201

Figure 7.7 Changes of total arable land and cropping index in the catchment 203
Figure 7.8 Land use/cover in the dry-hot valley in 1974 205
Figure 7.9 Land use/cover in the dry-hot valley in 1989 206
Figure 7.10 Land use/cover in the dry-hot valley in 1999 206
Figure 7.11 Land use/cover change in the dry-hot valley and the whole catchment (a) 1974, (b)

1989 and (c), 1999 207

Figure 7.12 Change of road length and road intensity in the Longchuanjiang catchment 208
Figure 7.13 Location of Xiaohekou, Huangguayuan and medium size reservoirs 212
Figure 7.14 Sediment yield at different scale. (a), sediment yield at plots, reservoirs and rivers
in the Longchuanjiang catchment (note the decreasing sediment yield to the larger
scale); (b), comparison of sediment yield in the Longchuanjiang catchment with
other catchments in the Upper Yangtze (after Lu, 2005) 216

Figure 7.15 Increase of medium-sized reservoir storage capacity in the Longchuanjiang
catchment 217

Figure 8.1 Mass curve of sediment flux in the Longchuanjiang catchment (Black dots are
observed sediment flux in the pre-1990 period; grey dots are observed sediment flux
in the post-1990 period; the straight line is the trend line of the sediment flux in the
pre-1990 period; the equation is for the trend line). Note the difference d is the
change of sediment flux due to both climate and human activity. 230


XIV
Figure 8.2 Double mass curve of sediment flux in the Longchuanjiang catchment (Black dots
are observed sediment-water relationship in the pre-1990 period; grey dots are
observed sediment-water relationship in the post-1990 period; the straight line is the
trend line of this relationship the pre-1990 period; the equation is for the trend line).
Note the difference d is the change of sediment flux due to the intensified human
activity 231

Figure 8.3 Trend in sediment flux by linear regression (a), sediment flux–water discharge
relationship (blue dots represent data in the post-1990 period; dark dots represent
data in pre-1990 period and trend line is based on data in this period.) , (b) residuals

of the sediment flux estimation 232

Figure 8.4 Observed and estimated sediment flux during 1960-1990 (a), observed and
estimated sediment flux; (b), scatter plot. 235

Figure 8.5 Annual sediment flux under background, under only climate variation and under
both climate and human activity change from 1991 to 2001 236

Figure 8.6 Cumulative monthly sediment flux under background, only climate variation and
under both climate and human activity change from 1991 to 2001 236

Figure 8.7 Comparison of the estimated sediment flux under only climate variation by ANN,
double mass curve and linear regression methods 239

Figure 8.8 Comparison of the estimated influence of climate variation by double mass curve,
linear regression method and ANN 239

Figure 8.9 Comparison of the estimated influence of human activity by double mass curve,
linear regression method and ANN 239

Figure 8.10 Change of variables representing human activities in the Longchuanjiang
catchment 242

Figure 8.11 RMSE of calibration, testing and validation data set by ANNs 245
Figure 8.12 CE of calibration, testing, validation and the whole data set by ANNs 246
Figure 8.13 Observed and estimated sediment flux by H1 248
Figure 8.14 Observed and estimated sediment flux by H8 248
Figure 8.15 Scatter plots of H1 and H8 249
Figure 8.16 Prediction errors of ANN_6, H1 and H8 249
Figure 8.17 Actual sediment flux, estimated sediment without road construction and estimated

sediment without reforestation (by H8) 251

Figure 8.18 Actual sediment flux and estimated flux without road construction (by H1) 252
Figure 9.1 Time series of annual rainfall, suspended sediment load and water discharge (a)
double mass plot of cumulative water discharge and suspended sediment load; (b)
annual rainfall; (c) annual water discharge and (d) annual suspended sediment load
263

Figure 9.2 Observed and predicted water discharge and sediment flux 265
Figure 9.3 Predicted water discharge under scenario T0&R0, T-1&R+20 and T+3&R-20 266
Figure 9.4 Sensitivity of runoff to the rainfall and temperature 267
Figure 9.5 Sensitivity of runoff to rainfall and temperature—in February and August 269
Figure 9.6 Predicted sediment flux under scenario T0&R0, T-1&R+20 and T+3&R-20 270

XV
Figure 9.7 Sensitivities of sediment flux and water discharge to climate scenarios (a) under T-
1 scenarios; (b) under T0 scenarios; (c) under T+1 scenarios; (d) under T+2
scenarios, and (e) under T+3 scenarios. 272

Figure 9.8 Sensitivity of sediment flux to climate scenarios (a) in November and (b) in
August 272

Figure 9.9 Cumulative rainfall (R
25
and R
50
) of 25mm- and 50mm-or-more rain days under
climate scenarios 274

Figure 9.10 Change of sediment concentration under climate scenarios 277

Figure 9.11 Possible range of sediment flux change till the 2050s 278
Figure 9.12 Relationship between rainfall and (a) water discharge or (b)sediment flux 280
Figure 9.13 Sensitivity of water discharge to climate scenarios under Level I and Level II
human influence 282

Figure 9.14 Sensitivity of water discharge to climate scenarios under Level I and Level II
human influence 285



XVI

XVII
List of Plates

Plate 2.1 Soil forest in Yuanmou—result of severe soil erosion 21

Plate 7.1 Reforested area (a and b) and original forest (c) in the catchment 199
Plate 7.2 Barren slope along the main channel in the middle reach (gully developed on the
slope) 202

Plate 7.3 The highly dissected area in the lower reach — dry-hot valley 202
Plate 7.4 Construction material excavation site in the Longchuanjiang catchment 209
Plate 7.5 Chuda highway along the Longchuanjiang River (the exposed cutting slope seven
years after completion) 209

Plate 7.6 Gullies on the cutting slope of the road 209
Plate 7.7 Construction waste dumped into the main channel of the Longchuanjiang River 211
Plate 7.8 Plate canal conducting the mud into the Longchuanjiang River over the Chengkun
railway 211

1 Introduction

1.1 Background
Hydrological regimes in a catchment can be described by a series of processes, such
as precipitation, interception, evapotranspiration, depression storage, overland flow,
infiltration, interflow and channel flow. Four interrelated sets of factors, including
catchment physical attributes, climate, land use and resource management system,
determine the hydrological processes in the catchment (Arnell, 1996, pp.7-60).
Changes in these factors could significantly influence water discharge and sediment
flux in the river.
1.1.1 Impact of climate and land use change on water discharge and sediment
flux
Climate is the most important driver of the hydrological cycle. During the 20th
century, the average global surface temperature increased by 0.6℃; it was predicted
that temperature would rise by about 1.4~5.8℃ by the year 2100 (Folland et al., 2001
pp.99-182). Meanwhile, there was an increase in global precipitation over the 20th

century. Instrumental records of land-surface precipitation continue to show an
increase of 0.5 to 1% per decade in much of the northern hemisphere mid- and high
latitudes. In contrast, over much of the sub-tropical land areas rainfall had decreased
during the 20th century (by -0.3% per decade), although this trend has weakened in
recent decades (Folland et al., 2001 pp.99-182).

1
The warming in the last 50 years in China is more rapid than the average values of the
world and the Northern Hemisphere (Folland et al., 2001 pp.99-182). Mean annual

surface air temperature across China had increased by about 1.1℃ for the last 50
years (Ren et al., 2004b). Some of the effects of increasing temperature would be the
changes in the rainfall and in the magnitude and frequency of extreme weather events.
A significant drying trend was observed in the Yellow River Basin and North China
(Ren et al., 2004a). The annual precipitation in the Middle and Lower Yangtze
showed an increasing but insignificant trend from 1951 to 2002, whereas it showed
significant decreasing trend in the upper Yangtze (Liu, 2003; Su et al., 2004).
Temperature in the middle and lower Yangtze was dominated by a decreasing trend,
while in the upper Yangtze it decreased from 1955-1994 and increased thereafter
(Zhang et al., 2005). Researches show that climatic changes will alter basic
components of hydrological cycle such as evaporation, soil moisture and groundwater
availability, and thus influence the magnitude and timing of water discharge and
sediment flux (Chiew et al., 1995; Menzel and Burger, 2002; Xu, 2005; Zhang and
Nearing, 2005).
Land surface disturbance related human activity is another important factor which
directly or indirectly influences hydrological cycle. Human activities closely related
to the change of water discharge and sediment flux include water consumption and
diversion, deforestation/reforestation, urbanization, road building, dam construction,
mining, water conservation and erosion control measures, etc. It was estimated that
humans may be directly or indirectly responsible for 80–90% of the fluvial delivery to
the coastal ocean (Milliman et al., 1987). Syvitski et al. (2005) estimated that
compared with the pre-human condition, humans have increased the river transport of
sediment through soil erosion activities by 2.3 ± 0.6 billion tons per year at global

2
scale, but reduced the sediment flux to the global coastal ocean by 1.43 ± 0.3 billion
tons because of retention within reservoirs.
Many researches on the impact of climate and human activity on water discharge and
sediment flux have been conducted and some conclusions have been drawn. However,
an explicit understanding of the hydrological system is still not available because the

complexity of the hydrological system has made the analysis of the hydrological
response and its driving forces problematic. Hydrological system is a complex system
with a large number of interrelated factors. A change in one factor might induce
changes in other factors and feedbacks are also expected. The magnitude as well as
the spatial and temporal distribution of the effective variables will also influence the
hydrological processes in the watershed (Vanacker et al., 2003). It is even more
difficult in differentiating the influences of different types of human activities from
climate change, considering the overlapping of the influence and the lag effect of the
driving forces.
1.1.2 Mathematical modelling of the impact and limitation of current research
There is a growing concern about the impact of climate change and land use change
on hydrological processes (Bultot et al., 1990; Boorman and Sefton, 1997; Coulthard
et al., 2000; Fohrer et al., 2001; Brath et al., 2002; Bobrovitskaya et al., 2003;
Nearing et al., 2005). Quantifying hydrological change arising from modifications to
the climate, land use and management of environmental systems has been declared by
the British Hydrological Society as one of the three major themes to be addressed by
hydrologists. Mathematical models are among the best tools available for analysing
this kind of impact and improving our understanding of the system. These models are
put into categories like empirical, conceptual and physically-based models according

3
to the degree to which the hydrological processes are represented (Viessman and
Lewis, 1996).
Physically-based models attempt to represent the spatial heterogeneity of variables by
dividing the catchment into grids, and describe the processes of the water and
sediment transport from grid to grid with simplified partial differential equations
(Lorup et al., 1998; Andersen et al., 2001; Muzik, 2002; Vanacker et al., 2003;
Eckhardt and Ulbrich, 2003). For sediment flux, the widely used conceptual or
physically-based models include SWAT (Hanratty and Stefan, 1998; Boorman, 2003),
WEPP and its modified version WEEP-CO

2
(Nearing et al., 2005; O'Neal et al., 2005),
CREAMS and ICECREAMS (Boorman, 2003), and MEFIDIS (Jetten et al., 2003;
Nunes et al., 2005). Their distributed structure allows the evaluation of the influence
of land management measures on soil erosion. However, inadequate scientific basis
and intensive data requirment may be the major constraints on its application to larger
scale catchment (Refsgaard and Abbott, 1996; Kisi, 2004). Their application is limited
to small and heavily instrumented catchments (usually less than 100 km
2
) (Beven,
1993). For example, WEPP was applied to catchment up to approximately 4 km
2
in
size (Flanagan and Nearing, 1995; Nearing et al., 2005) and MEFIDIS was tested
from 0.05 to 120 km
2
(Nearing et al., 2005).
Compared with physically-based model, empirical models are more widely used on
sediment flux prediction in large-scale catchments due to their relatively simple
structure and mathematical methods involved, and their ability to work with limited
input data (Flaxman, 1972; Walling, 1983; Prosser et al., 2001; Verstraeten and
Poesen, 2001; Zhou et al., 2002). Empirical models relate the hydrological response
to the driving forces, such as rainfall, temperature and land use, by statistical

4
relationship between them, without considering the actual physical processes involved.
Although they are unable to represent the spatial variability of hydrologic processes
and catchment parameters, it can provide simulations as good as those from complex
physically-based models when the interest in on e response on the entire water system
(Beven, 2000). In fact, the soil erosion modules in some of the so-called physically-

based models remain empirically-based. For example, the erosion models for
CREAMS and WEPP are based on USLE and MUSLE, respectively. However,
conventional linear or nonlinear regression models can only simulate the highly
nonlinear suspended sediment flux with limited accuracy, due to their simple model
structure and mathematical methods employed.
Artificial Neural Network (ANN) is a type of empirical model, which is based on
concepts derived from the research on the nature of human brains (Müller et al., 1995).
With its ability to approximate highly non-linear system without any priori
assumption of processes involved and the ability to give a good solution even when
input data are incomplete or ambiguous, ANN provides a promising alternative to the
conventional empirical and physical models in water discharge and sediment flux
modelling (Clair and Ehrman, 1998; Liong et al., 2000; ASCE, 2000a; ASCE, 2000b;
Rajurkar et al., 2004; Cigizoglu and Alp, 2006). There are, however, not many reports
on the application of ANN in sediment studies. The research conducted by e.g.
Abrahart and White (2001), Jain (2001), Tayfur (2002), Kisi (2004) and Agarwal et al.
(2005) may be deemed as pathfinder experiments in this area. These studies
demonstrated that the modelling of sediment, including its concentration in a river or
flux from a slope or a watershed, is possible through the use of ANN. Most
commonly, they predict sediment flux by relating it to water level/discharge and
sediment flux at previous time steps. Inclusion of antecedent previous sediment flux

5
as input may increase the accuracy of the simulation. However, ANNs established by
this method, also called updating ANNs, are unable to explain the contribution from
climatic variables. Also, they are insufficient to predict sediment flux if water and
sediment data for previous time periods are not available. The current attempt to
establish a non-updating ANN by relate the suspended sediment flux to original
driving forces, i.e., climatic variables such as rainfall, temperature, and rainfall
intensity instead of water and suspended sediment flux at previous time steps as
inputs. Such ANN could be used to explore the relationships between the climate

inputs and sediment responses. In addition, it would have a potential of filling missing
data in a suspended sediment flux time series and predicting the influence of climatic
change on suspended sediment flux.
1.1.3 Longchuanjiang catchment for impact study
The Longchuanjiang catchment, drains an area of 5560 km
2
, is located in the Upper
Yangtze River. The lower reach of the catchment is a typical dry-hot valley. Dry-hot
valley is a special environmental type in Southwest China. They are widely found
along the main streams and their tributaries in this region, notably along the upper
Yangtze (Jinsha), Dadu, Yalong, Min, Lancang (Mekong), Nu (Salween), and Yuan
(Red) and their tributaries. They usually refer to the valleys under the elevation of
1300m (northern slope of the mountain) ~ 1600m (southern slope) and are
characterized by a hotter and dryer climate, compared with their neighboring areas.
For example, the annual average temperature in the dry-hot valleys along Jinsha River
(upper part of Yangtze River) is 20~27ºC, the annual total precipitation is only
600~800mm, and the annual evaporation is 3~6 times of the precipitation.
Furthermore, the precipitation in dry season ( December to May of next year) only

6
accounts for 10% to 22.2% of the total annual precipitation, which results in a arid
index as high as 10~20 in the dry season. Despite the fragile ecosystem in the dry hot
valleys, they are among highly populated areas because of the relatively even
landscape and the abundant solar radiation and heat for agriculture industry. Due to
the harsh natural environment and the increasing pressure from human activities, most
of the dry hot valleys in Southwest China have the problem of degradation. The most
common one is soil erosion.
The Longchuanjiang catchment is in nature vulnerable to soil erosion due to several
physical factors such as intensive rainfall in rain season, fragmented topography and
surface soil which is susceptible to water erosion. In 1989, soil erosion affected area

accounted for more than 50% of its total area (Yunnan Bureau of Water Resources &
Hydropower, 1999). Soil erosion has caused many problems in this area. For example,
more than 85% of the farmland in the catchment is of medium to low productivity due
to the loss of surface soil. It also results in sedimentation downstream which reduces
the capacity of rivers and drainage ditches, blocks irrigation canals and shortens the
design life of reservoirs. According to a survey of 48 reservoirs built in the late 1950s
in Chuxiong, the total storage capacity of these reservoirs reduced by 9.88% in 1982,
due to deposition.
As population increases from 0.62 million in 1949 to 1.37 million in 2001, the
Longchuanjiang catchment had experienced changes in a wide range of human
activities, including deforestation/reforestation, intensification of agricultural activity,
engineering construction and reservoir building, as in most catchments in China.
Meanwhile, it had also experienced a climate variation characterized by higher
rainfall in the 1990s. Affected by changes in human activity and the variation in

7
climate, sediment flux showed a sharp rise since the 1990’s. Given all these processes
in the catchment, it is well suited to be a laboratory for studying the impact of
simultaneous landuse and climate change on water discharge and suspended sediment
flux.
1.2 Aims and objectives of the study
To improve understanding of the relationship of climate/human and water
discharge/sediment flux at catchment scale is crucial for both academic study and
decision making on economic and technical development. The aim of this research is
to investigate the impact of climate and human activity on water discharge and
sediment flux in a meso-scale catchment, Longchuanjiang catchment, China, with a
focus on sediment. This can be broken down into more specific objectives:
• To investigate the relationship between climate and water discharge/sediment
flux with Artificial Neural Network (ANN);
• To identify the specific causal variables, including climate and human activity,

and estimate their contributions to the change of water discharge and sediment
flux in the catchment;
• To investigate the sensitivity of water discharge and sediment flux to possible
climate changes, and
• To investigate the application of ANN, in terms of its advantage and
disadvantage, in the impact assessment study.

8