Remote Sensing and GIS Technologies for Denudation Estimation in a
Siwalik Watershed of Nepal
5 Results and Discussion
5.1 land Degradation Assessment by Vegetation Index
The actual size of the bareland within the forest cover is almost smaller than the spatial resolution of the sensors, therefore, photographs were overlaid and compared to investigate the possibility of identifying them using satellite data. Degraded mountain ridges, and soil deposited tributaries are very prominent in 1992 aerial photographs and helicopter photographs. These photographs were classified assigning a value zero for forest covered pixels, and 100 for non-forested pixels. Subsequently, original pixels were resampled into 100x100 meter to represent the average of original pixels fell in each new pixel. Thus the digital value of each newly created pixel represented the bareland or forest area percentage within the pixel concerned. Accordingly, the Landsat TM NDVI image of 1993 was resampled into same pixel size generating average NDVI values for each of the 100
Meter pixels. The relationship of NDVI and the bareland ratio was compared to identify the potential of NDVI in recognizing the change of forest density. Categorized bareland percentages into 10% intervals, and the corresponding averages of NDVI values are show in Table 2.
Table 2 Distribution of photographs derived vareland percentage and 1993 values
This table shows that the NDVI decreases with the increase of bareland percentage. The highest NDVI for this area was 125, and lowest was 99 for complete forest and 100% bare, respectively. This inverse relationship shows that NDVI can be used to represent the land degradation in the Ratu watershed.
Similarly, NDVI values of 1973, 77,95 and 95-LISS data were compared with the bareland percentages. Distribution of the NDVI values and the aerial photographs established 1992 bareland percentages are shown in Figure 1. The 1979-MSS and 95-LISS data are shown in break lines as they show some incompatibility with the other three dates data. 1977 MSS data shows higher NDVI values than others, and the deviation is irregular. It was found the there was a thin cloud cover in this dataset, and thid might have formed higher level of scattering that was not presented in the other datasets. The LISS dataset also shows different trend of NDVI. The reason for this is not clear, and required to study the internal calibration procedures of Landsat and IRS sensors. These two dates data, 1977 MSS and 1995 LISS were excluded in the proceeding analysis for soil erosion estimation.
Figure 1 Distribution of bareland percentages and NDVI values
Further, the helicopter photographs incorporated in the GIS database were compared with NDVI image of 1993 to investigate the degree of accuracy in identifying the denuded ridges and silted tributaries in the watershed. It was observed that most of the degraded redges, and relatively larger sub-streams can clearly be the densely forested areas. This could be due to the spatial resolution limitation of the sensor and mixed land cover that could present in a single pixel of TM data. As a whole, the comparison of aerial photographs and helicopter photographs with satellite data showed that the remotely sensed data can satisfactorily be used in continuous nature of NDVI facilitate the interpretation of forest density more realistically than classifying the forest cover into discrete density classes.
5.2 Temporal Changes of Average Annual Soil Yield
Estimation of soil yield in the Ratu watershed was carried out based on the equation 2. the rate of denudation required in this was acquired from published literature. Table 3 shows denudation rate in the Himalayan region and in a regulated watershed in Japan.
Rate of erosion for East Nepal, 780-3680 tons/km
2/year represents about 0.4 mm/year, and 18.5 mm/year, respectively assuming the density of the sediment is 2 g/cm3. These figures are for extreme conditions
In the Eastern Siwalik area.
Table 3 Land denudatinfor some Nepal and Japan
| Description of the site |
Rate (tons/km2/year) |
Reference |
| SiwalikL East Nepal, S-aspect, sandstone foothills, landuse |
| from forest to grazing lands |
780-3,680 (0.4-1.8mm/year) |
Chatra 1976 |
| Siwalil: Far West Nepal, S-aspect,sandstone foothills |
| Degraded Forest |
2,000 (1 mm/year) |
Laban, 1978 |
| Degraded forest, gullied land |
4,000 (2mm/year) |
Laban, 1978 |
| Severely degraded, heavily grazed |
20,00 (10mm/year) |
Sakya |
| Middle MountainsL Katmandu valley,steep slopes 800 |
(0.4mm/year) |
Laban, 1978) |
| Japan: Asio region n Tochigi Prefecture for 30o slope |
| Bare lands |
20 mm/year |
Honda, 1993 |
| Grasslands |
1 mm/year |
Honda, 1993 |
| Forest |
0.1 mm/year |
Honda, 1993 |
In order to evaluate the soil yield, the denudation rate of each pixel has to be estimated using reference information given in Table 3, or otherwise. The values obtained by comparing aerial photographs in the Table 2 represent the average NDVI for forest density percentages in the Ratu watershed. The highest value of NDVI, 125 was for 100% forest cover in this area, and the lowest value represented bareland. In establishing denudation rate in the Rate watershed, three critical soil loss rates and corresponding NDVI values were recognized. Publish documents identify the Siwalik area as severely degraded, hence the least degraded areas in the Siwalik area was assigned a denudation rate of 0.4 mm/year (800 tons/km2/year), which is similar to Middle Mountains. The average NDVI of the Middle Mountain forest coverage was 144, and this value was used in conjunction with the denudation rate of this area. Accordingly, the 100% forest coverage within the watershed, which is an average for 100x100 meter pixel unit, was assigned a denudation rate of 2.0 mm/year (4000 tons/km2/year). Finally, bareland was assigned 20mm/year which was used for a Japanese watershed. All of these denudation rates are assumed to be for land surface with 30o slope angle. In order to estimate the soil yield for the whole watershed a relationship was considered for denudation and NDVI, (Honda, 1994). He has demonstrated that NDVI is related to common logarithm of the denudation rate. The interpretation of denudation with respect to NDVI for each pixel in their original resolution was carried out as shown in Figure 2.
Figure 2 NDVI and Log E30 relationship
This relationship gives the rate of denudation for any pixel in the watershed with respect to 1993 land cover condition. The other factor that is required for soil yield estimation is the slope gradient, and it was calculated from the digital elevation data incorporated in the GIS database. Considering no appreciable change of the topography during the 22 year period from 1973 to 95, proposed equation for soil erodibility was used for estimation of the denudation rate and the volume of the soil production based on NDVI derived forest cover densities. The estimated values are tabulated in Table X.
Table 4 soil yield and denudation estimated for the Ratu watershed
| |
1973 |
1993 |
1995 |
| Soil Yield m3 |
271.110 |
321,156 |
292,096 |
| Rate of denudation mm/year |
3.29 |
3.89 |
3.55 |
Reliable date for comparison with satellite data estimation was not available. The estimated denudation rates are very much comparable with published materials presented earlier justifying the satisfactory application of present model in soil yield prediction in Ratu watershed.
5.3 Soil Yield in a Storm Event
The foregoing section evaluated the annual average soil yield on the basis of land use and the surface topography. This model is not suitable for estimation of soil production during a cloud bust incidence that could bring about 500 mm of rainfall within few hours. The excessive rainfall could accelerate the surface erosion, but it was identified that the landslide along the banks of the streams, undercutting, or earth topple are some of the main factors highly contribute to soil erosion during 1993 flood event. Also, it was found the areas around the sub-streams are very much vulnerable to soil erosion during a flood event due to these factors. Attempt was made to modify the equation used in the previous section to use in storm event.
In establishing a storm factor for the present mode, it was considered the field observation are true and correct. Therefore, the soil yield estimated for the 1993 storm event for each sub-watershed was considered as the actual amount of sediment produced, (UICA, 1996). Hence, equating these with the estimations of the model that is adjusted for a storm event prodeuced a storm event factor (a) of the Raut watershed for a cloud burst even similar to that of 1993. The calculation was carried out as shown below:
Here (v)
i is the estimated production of watershed i by field investigation
ai Storm event factor for watershed i
åialpixels Estimation of soil production of watershed i using annual average model
The average of ai for the sub-watersheds was 11.2 with a standard deviation of 6, Very high deviation were found in sub-watersheds with larger extents. This only relates field observation with the model estimation therefore, further analysis was carried out to study the relationship of ai with topographical characteristics of subwatersheds.
Soil production carried out by field observation was based on stream length, size of landslide scars, depth of deposited materials, vertical drop of valleys, and stream gradient, (UICA, 1996). Therefore , the stream length could be considered as a suitable factor to represent the soil production during a storm event. A strem length factor (Si) of each watershed was calculated as shown in equation 4, and compared with respective ai to identify the possibility of replacing the ai with the stream length factor.
Si = (Li x Li) / Ai (4)
Whee L
i is stream length and A
i is the area of a sub-watershed i
The average value of the above calculation was 9.2. This figure is comparable with the factor estimated for the 1993 storm event (11.2), hence a stream length factor as shown in equation 4 could replace the ai in the equation 3 in estimating the soil yield for a storm event for the Ratu watershed. In the present calculation the main stream were excluded as it was observed that the explained soil lumps are mainly distributed in the tributaries with relatively steep riverbanks.
6. Conclusion
In concluding it can be said that satellite data can satisfactorily be used in land degradation estimations. Multi-temporal data from the same satellite can use through graphical correction for spectral bandwidth and atmospheric dissimilarities. Further analysis are warranted for comparison of multi-sensor spectral data. The model proposed to average annual soil yield produced land denudation rate that is very much similar to the values published by intensive field work. This validates the potential of this model in estimating soil yield in a watershed where the accurate field measurements are scarce to find. Modification of the model for a storm event was suggested incorporating stream densities in the watershed. It is important to say that the present model was used in a watershed in the Siwalik area, and more case studies are required to identify it as a general model.
References
-
Honada, K., Murai, S., and Shibasaki, R.: Prediction of vegetation restoration by erosion control works in Asio copper mine Japan, 13th IGRSS, 1993
- lves, J.D. and Messerli, B.L The Himalayan Dilemman, Routledge London and New York, 1989
- JICA; Research Report on the Investigation of Landslides and Soil Erosion in Nepal using Remote sensing Technology, March 1996
- Summerfield, M.A.: Global Geomorphlogy, Longman Scientific & Technical, 1991
