Investigating River Channel Changes as a consequence of Continuous Flood Hazard in Terai Region of Napal using Remotely Sensed Data

Investigating River Channel Changes as a consequence of Continuous Flood Hazard in Terai Region of Napal using Remotely Sensed Data

Lal Samarakoon*, Akichika Ishibashi*, Yasushi Mabuchi*, Kioshi Honda**, Shigechika Miyajima***
*Nippon Koei Co., Ltd., Research & Development Center, 2304 Takasaki, Kukizaki, Japan
**STAR/SERD, Asian Institute of Technology, Bangkok, Thailand
***Water Induced Diaster Prevention Technical Center, Katmandu, Nepal

Abstract
Wide spread damage due to floods is one of the very common water induced natural disasters in Nepal that account for heavy losses during the monsoon period every year. Perennial rivers that flow through Siwalik area transport heavy sediments loads than rivers located in other areas during rainy season due to very fragile nature of the soils in Siwalik Hills. As a consequence, high sediment depositions are observed in the border of the Siwalik and Terai regions where the rugged terrain of Siwalik Hils change to relatively flat floodplain. This causes heavy losses during flood events in this areas due to overflow of river banks and planform changes of rivers. In this study, satellite data covering twenty year period was used in identifying the changes of river channel in the floodplain of Ratu river that originates from Siwalik hills in Central Nepal. Reason for changes were examined with field observation and the present trend in the planform change was established in predicting flood prone areas in future flood events.

Introduction
In July 1993 high intensity rainfall occurred in the Central Nepal in two occasions, July and August causing severe flood damage throughout Nepal. It is reported that July flood was more severe than the August flooding causing extensive damage as a consequence of more than 50mm of rainfall within 24 hours.This rainfall was concentrated in the central hills of Nepal extending from mahabarth mountain region, Siwalik hills and extending to Terai region. This natural phenomena caused heavy toll of damage. The death toll rose to 1460, and about 39495 houses were damaged. The transportation between to and from Katmandu was cutoff for few days hindering timely rescue work of affected people.

These types of storm and natural calamities have been occurred even in the past in Nepal. The causes for some of these unprecedented floods are storms, glacier lake burst and landslide. The physiographic nature of these Himalayan region, where climate is varying form tropical to alpine, and topography from alluvium plains to the mountain peak of the worl are some of the natural features that cause the flood hazards, and the ever increasing burden on the land could turn these hazards into disasters.

Proper countermeasures for flood prevention or mitigation are hardly carried out even though the disasters occur in time to time. This is partly due to the lack of understanding of flood hazards, hazard prone areas, causes of events, and consequences, as hardly any studies are carried out focusing on these factors.

This research project was aimed at studying the sedimentation in the riverbed of the watershed and in the floodplain using remotely sensed data. The analysis was focused on the process of accumulation, river channel aggradation, degradation, and formation of new channels as a consequence of heavy sediment unloading in the monsoon period of the Siwalik originate rivers. Flood prevention and controlling work that has been carried out in the past were investigated for identify their effectiveness.

Study Area and Data

Outline of the study area
Present study was carried out in the Siwalik area where the sediment transport is very high depositing enormous amount in the Terai Plain. Sub-streams comparatively short in length with higher gradient, and very flat riverbed is typical to the rivers originate in the Siwalik hills. With the increase of the gradient, the watershed become narrow. Further, the dendritic drainage pattern develops towards the upper stream of the River with the increase of riverbed gradient.

Ratu watershed selected for the study belongs to the Siwalik area located in the South of Sun Koshi and west of Kamala river. This river originates at an altitude of 708 meters, south of Sindhulimadi, and flow inn the direction of north-south from Siwalik Hills to the Terai region. Ratu river could be consider as wadji, as water flow in the upper reach is visible only during the during the monsoon season. The East-West Highway (hereinafter EWH) could be considered as the turning point of this river system dividing the watershed and floodplain, Figure 1.The length of the river in the upper stream up to EWH is about 35 km, and average gradient of the riverbed is about 1/100. Width of the river changes dramatically below the EW, where to the north of highway the width is about 600 meters, and below the highway the width has been extended to 2 to 3 km.

Figure 1 Spectral response patterns of selected sample for the Landsat MSS 1973 and TM 1993
Acquisition of Remote Sensing and Reference Information
Landsat MSS, TM and IRS-LISS-II data were considered for the present study. Long term investigation of the river planform changes could carry out with MSS and TM data acquired in the early and newest Landsat missions. IRS data was obtained for comparison with TM for their suitability in using land cover information extraction and the usability in lieu of Landsat TM data. Considering the objectives of the study, the seasonal variations that could present in the satellite data it was decided to selected the multi-sensor temporal data listed in Table 1.

Table 1 Information on the satellite data obtained for the study

Sensor	Date	Sun Angle	Sun Azimuth	Processing Level
MSS	1973.03.14	Not Known	Not Known	System Corrected
MSS	1977.03.20	Not Known	Not Known	System Corrected
TM	1993.03.16	47°	127°	System Corrected
TM	1995.03.22	122°	46°	System Corrected
LISS-II	1995.03.15	47°	127°	System Corrected

Referring to the selected data and major flood events, aerial photographs acquired in 1979 and 1992 were obtain for comparison and establishing ground reference information for calibration and validation satellite data interpretation. Also, field investigation were carried out to collect information on extents on floodplain, change, and distribution of the sediments. A detailed discussion of the method of the field survey, obtained information, and the detail discussion of the soil erosion phenomenon, and its transportation is found in JICA, 1996.

Database Creation for the Analysis
In order to make use of all the information, satellite data, maps, aerial photographs and field visit information it was decided to create a geographic database referred to as GIS for easy reference of source of information.

The acquired satellite data had been received by sensors having different field of views resulting different ground resolution. Further, their spatial orientation also dissimilar as the ground processing are carried out independently. Therefore, the procured satellite data were brought into a common map projection (UTM) by constructing mapping function through identifying control points on 1:25,000 topographical maps. Also, the conventional topographical and vegetation map were digitized and incorporated into the GIS database. Further, the aerial photographs and conventional photographs obtained in the field were scanned, rectified and registered into UTM projection. Thus the completed GIS database for the present study included multi-sensor temporal satellite data, aerial photographs, elevation and land cover information.

Results and Discussion
Spectral Characteristics of Riverbed Material

Dry season riverbed appears very bright on aerial photographs, and on the conventional photographs indicating very high reflectance in the visible spectral region of the electromagnetic spectrum. Further, some form of different gradation was observed in the old deposits and newly deposited areas. It was attempted to observe these characteristics properties on the radiance values observed by the sensors to establish the best suitable spectral band or the combination of bands for delineating of deposited material, and assessing the change in the planform of river channel.

With reference to the aerial and other conventional photographs, reference area were established and corresponding digital counts for dominated over classes, forest, crop lands and river sediments covering most of the study area were extracted. Statistical parameters of representative sample were calculated and analyzed for differentiation of these cover classes, and the establish the spectral characteristics of deposited materials. The

Spectral response pattern of the selected sample are depicted graphically in Figure 1 for 1973-MSS, 1993-TM datasets. The spectral response patterns of the other datasets were similar to one shown here. Comparatively very high reflectance was observed for riverbed, followed by crop lands and forest. It is said that the bareland are highly reflective as there is not energy absorption capacity unless the surface is wet. Further, when the materials are very fine and sandy where the water retention power is very low, the reflectance tend to be very high compared to other earth feathers. This is clearly visible in this area as field survey reported that the very fine sandy particles are found in the lower part of the river, and there is no visible water flow during the dry period. The riverbed material in the upper stream showed less reflectance than the materials in the floodplain particle size or the composition of the materials, presence of moisture or the traceof the river flow. Further, the old deposits identified in the aerial photographs showed relative lower spectral response when compared to new deposits. The may be due to change of color of the deposited material with time, or sparsely grown grass ver the deposited materials. Spectral pattern of crop lands resembled the patterns of old-riverbeds. Most of the crop lands are partially cultivated or abandoned during the dry season accounts for this similarity. It could be said that any of the TM spectral bands, specially, band 1, band 1, band 3, band 3, band 5 and band 7 are very much suitable for classification of river bed degradation and aggradation or planform changes, and there is no necessary to establish any other derived index. For MSS, the best band would be band 4, band 5 or band 6, band 7 may not be very much suitable due to its narrow dynamic range.

Channel changes and sedimentation in the Ratu Watershed
Using the statistical parameters of the samples selected over interested cover classes explained in the previous section, satellite datasets were classified into sediment deposited and non-deposited areas. In the present study, radiometric correction or rectification was not considered. The spectral response patterns of sediments and riverbeds are significantly different from those of surrounding permitting easy classification. Further, the objectives of the work was to identify the spread of the deposition, and the riverbed changes, hence the compassion of digital counts of multi-temporal data was not required. The comparison was carried out with the classified images to identify changes. The riverbed extents obtained from these classification are shown in Table 1.

Table 1 Extent of the riverbed of the Ratu river in the upper watershed

MSS-1973	MSS-1977	TM-1993	TM-1995	LISS-1995
Sq.km.5.18	sq.km.5.23	sq.km.5.48	sq.km.5.45	sq.km.5.53

There was no appreciable change was found in the extents of the riverbed area of the upper stream except for some minute difference in the extents. Given the geometric resolution of the sensors. 79m, 30m, and 34m for Landsat MSS, TM and LISS-II it is questionalble to interpret these changes as actual riverbed expansions. These observations were further analyzed for any spatial variation by considering five river segments, Figure 2. These segments were defined considered confluence of major tributaries of Ratu river. The main aim of these segmentation was to find may riverbed change with the contribution of sediments from tributaries. It also could facilitate to recognize the contribution of human activities on the sedimentation as the settlements are concentrated only in some part of the watershed.

Figure 2 River segments in the Ratu watershed
Extents of the Ratu river sub-divisions shown in Table 2. As for the entire river wihin the watershed up to EWH, no trend was found for the sub-divisions also. The highest increase was observed for the L1 segment for 1993, but the riverbed area has return back to 1973 extent in the 1995 complicating the explanation. The increase could be only due to presence of some mix-pixels in the 193 observation.

The maximum difference observed for a time span was 0.2 sq. km., and this represents 220 TM pixels or 110 MSS pixels (MSS pixels were resampled into 60m during the geometric rectification). The length of the river in the unper watershed is about 35 km, and the average width is more than 60 meters. Simple arithmetic show the ratio of the riverbed change with pixels that required to represent the Ratu riverbed within the watershed is less than 10% of TM pixels, and less than 5% of MSS pixels. Therefore, it is questionable to consider these differences in temporal estimations of the riverbed areas as a actual riverbed deviations that have been occurred within the time span under consideration. These marginal differences might have occurred because of the limitations of the spatial resolution of the sensors, and the orientation of the sensor field of view at the time of satellite pass. In concluding, it could be said that there was no appreciable change in the Ratu riverbed in the upper watershed during the 1973 to 95 time span.

Table 2 Cumulative area of the riverbed of the main stream

	1973	1977	1993	1995
L1	0.1692	0.1548	0.1917	0.1557
L2	0.4824	0.5292	0.5913	0.5868
L3	1.3932	1.4976	1.6344	1.6812
L4	5.4468	4.5976	4.8744	4.8195
L5	5.1768	5.2308	5.4855	5.4410

Channel Changes and Sedimentation in the Ratu Floodplain
As for the upper reach of the river, the sediment deposition and its change was estimated for the Ratu floodplain using spectral patterns described earlier. The extents of the sediment deposition in the floodplain for the four datasets and for the LISS dataset is shown in Table 3.

Table 3 Floodplain deposition of sediment and its change

73-MSS 77-MSS	77-MSS	93-TM	95-TM	95-LISS
9.17sq.km	11.58sq.km	9.12 sq.km	9.61sq.km	10.07 sq.km

The figure given in the table represents the whole sediment deposited are below the EWH along the downstream until the satellite sensor can discern the sediment from its surrounding. Further, the given extents exclude the old riverbed deposits that was differentiable by respective sensors of the four dates data. The estimated extents for the four dates shows the sediment deposited area is increasing gradually, except for drastic increase in the time span between 1973 to 77. This increase could be a consequence of the comparatively high precipitation prevailed in the period of 1974 to 75, where the total annual precipitation was over 6000 mm, (Karmacharya, 1995). Also, the political situation in the country during 1970 to 80 period leading to noticeable deforestation in eastern and central Nepal, (Sharma, 95) might have increased soil erosion leading to noticeable change in the Ratu floodplain. The classification accuracy of LISS shows that LISS sensor data can also be used in lieu of Landsat TM data in sediment deposition estimations.

Multi-temporal image integration was carried out to further enhance the pattern of depositin and spatial variability of river channels in the floodplain. A set of three images were produced by ingrating two dataset at a time defining three different pattern of changes. Figure 3, 4 and 5 show the newly deposited, unchanged or re-deposited, and unaffected areas during the time intervals 1973-77, 1977-93, and 1993-95, respectively. Newly deposited, and unaffected areas are shown in different shades, and the unchangedor redeposit areas are shown in hard lines. Unchanged represents sediments in both dates.

Figure 3 Change of sedimentation from 1973-77

Figure 4 Change of sedimentation from 1977-93

Figure 5 Change of sedimentation from 1993-95
A consider increase during the period 73 to 79 was clearly visible in the Figure 3, specially in the most westward channel. In this channel sediment deposition has been active even active even before 1973, but the sediment deposition during the period 1973-77 is comparatively larger than that has been occurring prior to 1973. This change could be due to continuous aggradation of the riverbed just below the Highway in the eastward channel

resulting change of flow to west. Decrease in the sediment transportation in the lower portion of the eastern channel would further support this assumption.

Figure 4 shows the difference of the 16 year period 1977 to 1993. The most westward channel, which was very active before 1977 had become inactive in sediment transportation. 1979 aerial photographs shows presence of a dike at point X across this channel that was built to control the flooding along the down stream of the channel. The width of the Ratu river just below the EWH, and eastern portion of the river above it has been decreased. This is a as a result of a dike constructed in the eastern bank of the river, just above the bridge after 1977. A critical riverbed deposition could have been occurred near point Y with the introduction of the dike above the bridge changing the watercourse westward after passing the bridge. This sedimentation might have split the channel into two very distinctive river courses. Aggradation around the point Y could have occurred due to an increase in material to be carried by the river, a loss of discharge or velocity flow, or a rise in occurred due to an increase in material to be carried by the river, a loss of discharge or velocity flow, or a rise in occurred due to an increase in material to be carried by the river, a loss of discharge or velocity flow, or a rise in base level. Extension of the floodplain just below the EWH into few kilometers could support these reason, and high sediment activities just below the EWH. Similar explanation could be made to the division of the most eastern part of the river identified in Figure 4. Comparing the observations of these two time spans, it could be said that the planform deviation of the river was much active during the 1973 to 79 period than the next 16 year. No much difference were observed between 1993-95 within the floodplain. Few newly deposited areas are visible in the lower par to the most eastern sub-stream, and a decrease was observed in the westward channels. This could be an indication of the increasing activity in the eastern side of the floodplain due to continuos deposition and aggradation in the west side of the Ratu riverbed just below the EWH. These findings shows the potential of satellite data in identifying, and monitoring sedimentation process of a watershed, and tracing river channel planform changes.

Remote Seining for Flood Protection Planning
A natural hazard turns into a disaster when an event causes heavy loss of life and property damage. This is a mostly happens when risk inherent human activities take place in hazardous areas. With the increasing population in this mountainous country, the burden on the land is in the rise resulting further land degradation and risking to settle in flood prone areas for basis human needs.

The observation carried out in this study shows that heavy sedimentation is occurring in this Ratu floodplain for a long time, and increasing gradually requiring counter measures. It is required to identify the status of floods to consider appropriate prevention or mitigation strategies. Lessen the impacts of flood could be status of floods to consider appropriate prevention or mitigation strategies. Lessen the impacts of flood could be considered through structural measures or non-structural adjustments. Structural implementation consider out to change, modify or prevent the phenomenon, whereas the non-structural implementation consider considerably high within few kilometers below the EWH, hence the people living in the vicinity of the river in this area are very much vulnerable to flood hazards. The two dikes, one above the bridge, and another in the western channel (Figure 4) showed their influence in deviating and splitting the river to flow in different directions. Impact of these structural approaches is questionable as the total loss may have not reduced. Further, the twenty year period data showed that heavy sedimentation has taken place only in a limited area below the EWH. The most active sedimentation take place in the riverbed just below the EWH. Aggradation of the riverbed in this area of the river, and its spatial distribution would decide the direction of sediment transportation in a future flood event as there is not alarming change in the volume of sediment produced in the watershed for the time period considered here. These observations may be very critical in deciding appropriate counter measures, and would have not obtained without remote sensing data. Limiting the settlements in this area leaving natural overflow during monsoon may have better chance in mitigating hazards than considering expensive structural measures. Also, riverine beyond 10km form EWH may not experience severe damage as a consequence of sedimentation except for inundation due to excessive runoff during the monsoon period.

Conclusion
Potential of multi-temporal remote sensing data acquired from different sensing can satisfactorily be used in investigating and assessing the form of riverbed sedimentation, and the planform deviation of river channels. Historical satellite information are invaluable in evaluating the efficiency of flood protection measures carried out in the past, and their consequences that could useful in future flood management strategies. Also the present trend of the channel formation, spatial distribution of sedimentation observed by satellite data may infer the current sedimentation activities and future channel developments.