- Email: [email protected]
Fine-resolution remote-sensing and modelling of Himalayan catchment sustainability
Remote Sensing of Environment 107 (2007) 430 – 439 www.elsevier.com/locate/rse
Fine-resolution remote-sensing and modelling of Himalayan catchment sustainability D.J. Quincey a,⁎, A. Luckman a , R. Hessel b , R. Davies a , P.L. Sankhayan c , M.K. Balla d a
School of the Environment and Society, Swansea University, Singleton Park, Swansea, SA2 8PP, UK b Soil Science Centre, Alterra, WUR, P.O. Box 47, 6700 AA Wageningen, The Netherlands c Department of Ecology and Natural Resource Management, Norwegian University of Life Sciences, P.O. Box 5003, N-1432 Aas, Norway d Tribuvan University, Kirtipur, Kathmandu, Nepal
Received 21 February 2006; received in revised form 20 September 2006; accepted 23 September 2006
Abstract A number of studies have reported on environmental degradation in the Nepal Himalaya as a result of large-scale deforestation and the associated agricultural extension. In contrast to many previous regional scale studies, we consider land cover and its environmental impact on an individual catchment-scale, using fine-resolution Quickbird data and a soil erosion model. First, using a detailed land cover map generated from Quickbird imagery, we establish basic relationships between land cover, dwelling density and topographic variability, which exist in a typical midelevation Nepalese catchment, the Pokhare Khola. These data suggest a strongly subsistence type of household economy based predominantly on terraced arable farming. We demonstrate using multitemporal vegetation indices that this farmland has existed in the region since the late 1980s, and that widespread deforestation has not taken place since then, possibly as a result of specific forest conservation policies of the government coupled with efforts by local communities. Further, through the use of soil erosion modelling we demonstrate the role that the terraced farming practices can play in reducing runoff and hence soil nutrient loss, thereby enabling restoration of vegetation in the previously deforested land terrains. Finally, by combining this information with regional land cover data, we show that the findings of this research can be scaled up to draw conclusions about environmental degradation across the Nepal Himalayan region. © 2006 Elsevier Inc. All rights reserved. Keywords: Remote-sensing; Quickbird; Deforestation; Environmental degradation; Soil erosion modelling; Nepal Middle Hills
1. Introduction Environmental degradation poses a serious threat to the longterm sustainability of ecologically fragile mountainous Himalayan watersheds characterised by subsistence farming (Tiwari, 2000). Socio-economic and environmental changes have impacted heavily on land use in the region, mainly manifested by the extension of cultivation into marginal land and forested areas. Such changes have depleted and eroded the natural resources around populated areas, thereby reducing groundwater recharge and increasing surface runoff. The problem is well pronounced in ⁎ Corresponding author. 35 Spon Green, Buckley, Flintshire, CH7 3BH, UK. Tel.: +44 1244 540942. E-mail address: [email protected] (D.J. Quincey). 0034-4257/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2006.09.021
mid-elevation (i.e. 1000–3000 m) areas of Nepal, where steep slopes and narrow valleys are conducive to high levels of erosion, particularly following a period of large-scale deforestation from approximately 1950–80. During this time, forested areas were placed under government ownership, which loosened controls on natural resource management and thus led to large-scale devastation of forest cover (Bajracharya, 1983). A key consequence of this land cover change was an increase in the rate of soil erosion in the region, which is now known to be very high in comparison to other parts of the world (Sen et al., 1997). This in turn has led to a gradual decrease in soil productivity, agricultural production and the incomes of subsistence farmers of the region. Several studies have documented the extent of land use change in the area (Balla et al., 2003; Gautam et al., 2003; Rao & Pant, 2001), reporting on trends of deforestation and, in more
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
recent decades, a reversal of this trend following forest conservation efforts by local communities and external agencies. Others have focussed on the controlling factors of the observed environmental change using bio-economic modelling (e.g. Sankhayan et al., 2003; Sitaula et al., 2005). Some studies have reported on the effects of degradation in the region, mostly focussing on river and reservoir siltation (Sthapit & Balla, 1999) and increases in the incidence and severity of flooding (Tiwari, 2000) although others have also considered the long-term effects on economic prosperity (Semwal et al., 2004). With the exception of a few studies (e.g. Saxena et al., 2005; Thapa, 1996), little attention has been paid to the management and conservation of agricultural lands, which account for a substantial proportion of all land resources in the region. Until recently, detailed assessments of Himalayan catchment dynamics have depended mostly on the feasibility of extensive field investigations and surveys. Now, however, readily available and affordable fine-resolution satellite sensor imagery offers an alternative method for extracting such information, for analysis by itself or as input to models for simulating the impact of altering the environmental conditions within a catchment. Repeated satellite images can be used both for the visual assessment of land resources as well as the quantitative evaluation of land cover changes over time (Tekle & Hedlund, 2000). Furthermore, hydrological models can now be employed to evaluate the impact of land cover change on a catchment, such as the expansion of agriculture into previously forested areas. This approach allows us to better quantify the trends observed across the wider Himalayan region. The impact of terrace farming practices on ecologically sensitive land is currently poorly understood in the Himalayan region. Thus, the aim of this paper is to provide an up-to-date analysis of the key issues surrounding environmental degradation in the Himalaya. We aim to demonstrate, for the first time, how the integration of fine-resolution remote-sensing data with GIS-based soil erosion modelling can be used to establish trends in land cover and land use, and evaluate how changes in both impact on land sustainability. This requires the fulfilment of several key objectives: 1. To use Quickbird remote-sensing data to map land cover in the Pokhare Khola catchment, Nepal, and analyse the extent to which it has changed over recent decades, by comparison with historical imagery and map sheet information. 2. To compare these catchment-scale land cover trends with regional trends, derived from various sources including previous studies, satellite data at a more coarse resolution (but with greater coverage) and classified historical map sheet data. 3. To couple the fine-resolution land cover maps with elevation data to demonstrate new potential for modelling the impact of deforestation and terrace agriculture on soil erosion in Himalayan catchments. 2. Study area The Pokhare Khola catchment is situated in the Dhading district of the central development region of Nepal, and covers
431
an area of approximately 5.28 km2 (Fig. 1). Geographically, it lies between 27°46′28″ N and 27°48′06″ N latitude and 84°53′32″ E and 84°55′11” E longitude. The climate is classified as sub-tropical, and temperatures reach a maximum of 31 °C in summer months and 8 °C during winter. The average annual rainfall is 1370 mm, most of which falls during the monsoon period between May and September. The catchment is steepest in the south, reaching an elevation of 1079 m and drains to the north through its lowest point at approximately 380 m. Here, the water drains into the Trishuli River and travels westwards taking drainage into the lowlands. The catchment land cover is a mixture of forest, scrub and subsistence terrace farming. There are two predominant crop seasons, namely, rainy and winter seasons. While paddy and maize are the dominant rainy season crops, wheat and vegetables are grown in the winter season. Most families also keep livestock, mainly cows, goats and oxen, as complementary and supplementary activities to agriculture for their livelihood. 3. Data sources Satellite image interpretation is based primarily on Quickbird data acquired on 21 February 2003, at 0.6 m spatial resolution in the panchromatic band and 2.8 m spatial resolution in four spectral bands. Neither scene was hampered by the presence of cloud cover and both were geo-referenced to UTM Zone 45 (WGS 84). In addition to the Quickbird data, which covered little more than the Pokhare Khola catchment, historical data from the 21st December 2001 and 22nd October 1986 were also acquired from the SPOT 4 HRVIR sensor (10 m spatial resolution multispectral) and SPOT 1 HRV sensor (20 m spatial resolution multispectral). Both images were supplied as L1G data, which are geo-referenced and radiometrically corrected. The scenes were cloud free and largely unaffected by shadow. SRTM DEM data were acquired (post-spacing of 90 m) to facilitate a regional assessment of topographic variability, but the coarse resolution of the data was insufficient for a detailed catchment-scale analysis. Thus, an ASTER DEM was also generated (post spacing of 15 m) using ERDAS software, which was used to establish topographic trends on a much finer scale in the Pokhare Khola. Comparisons between the two DEM datasets showed an average absolute height disparity of 56 m, but visual inspection of the two datasets revealed a high relative correlation, suggesting an accurate replication of the catchment topography by the ASTER stereo model. This finding contrasts with previous reports of large ASTER DEM errors in areas of extreme topography (Kääb et al., 2003). In addition to satellite sensor imagery, eight map sheets of the catchment and its surrounding areas were acquired and scanned at a resolution of 300 dpi, equivalent to approximately 2 m pixel size. These map sheets were presented at a scale of 1:25 000 with primary elevation contours of 20 m, in a Modified Universal Transverse Mercator projection. They were compiled by the Survey Department of the Government of Nepal in 1994, in co-operation with the Government of Finland, using 1:50 000 scale aerial photography acquired in 1992.
432
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
4. Methods 4.1. Remote-sensing The scanned map sheets were registered to UTM zone 45 (WGS 84) and stitched together to act as a base map for the rest of the image data in the Pokhare Khola catchment. The SPOT imagery from 1986 and 2001 were orthorectified to these maps, and as a result were co-registered to one another within ± 20 m across the Pokhare Khola catchment and to within ± 40 m in the wider area. The Quickbird imagery was delivered already georegistered to the UTM co-ordinate system, and could thus be overlaid on the map sheets and SPOT imagery with maximum errors of ± 5 m. Both SRTM and ASTER-derived DEM data were accurately co-registered with the map sheet mosaic within the catchment as well as in the surrounding region. Vector GIS coverages delineating forest, shrub, terraces, rivers, dwellings and roads within the Pokhare Khola catchment were generated by visual interpretation of the panchromatic Quickbird imagery, with some help from the NIR band of the multispectral imagery for discriminating vegetation. Slope images were also prepared using the ASTER-derived DEM for the catchment, although the raw elevation data were first smoothed to eliminate any spurious values. The SRTM DEM data were used to generate slope images for the wider region. In each case a 3 × 3 pixel computation kernel was employed, which considers all surrounding eight pixels when calculating slope angles.
NDVI (Normalised Difference Vegetation Index) were calculated according to the standard algorithm NIR-Red/NIR+ Red for the multitemporal SPOT imagery. The data were not atmospherically corrected prior to calculating NDVI as it was assumed that similar atmospheric conditions prevailed at each acquisition. Failure of this assumption would lead to a small amount of atmospherically-introduced noise of insufficient magnitude to affect the coarse trends presented in this research. Once processed, a common area around the Pokhare Khola catchment was extracted from the NDVI data so that areas of deforestation or new cultivation could be rapidly identified. To enable comparison with the manually digitised land cover map, the scanned map sheets were classified using a simple maximum likelihood algorithm. Training data were collected from the six major land cover classes, with a minimum sample of 200 pixels per class. The results were subsequently smoothed using a median filter to eradicate contouring through the classification, and then resampled to the Quickbird-derived land cover map to facilitate a fair comparison. 4.2. Soil erosion modelling The land cover map derived from Quickbird data was coupled with the smoothed ASTER DEM data for use in the soil erosion model. We chose the Limburg Soil Erosion Model (LISEM; Jetten & de Roo, 2001) for our analyses, which is a physically based soil erosion model that simulates runoff and erosion for single rainfall events. It was designed for small
Fig. 1. Location of the Pokhare Khola catchment and the geographic extents of the remote-sensing and map sheet data sources.
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
catchments, in the order of a few km2 in size. Process descriptions are according to current knowledge on hydrology and erosion, and include interception, infiltration, detachment, deposition and routing of overland flow and sediment. LISEM is a distributed model, which has been integrated with a GIS. It reads raster maps as inputs, and also generates raster maps as outputs, which depict the spatial distribution of runoff and erosion. The model was first developed to test small scale soil conservation measures on soil loss in the Netherlands, but has since been applied to catchments in Europe, Asia, Africa, America and on the Pacific Islands. These catchments have included agricultural catchments, but also catchments mostly covered by scrub (Nearing et al., 2005) and forest (Van Dijck et al., submitted for publication). While there is currently no literature detailing the application of LISEM in the Himalaya, recent results from an Indian Himalayan catchment with similar topography and climatic characteristics as the Pokhare Khola have suggested the model is also applicable in such areas (R. Hessel, 2006, personal communication). A number of detailed map layers are required to run LISEM. These include catchment, vegetation and soil surface maps, as well as a detailed land cover map as its primary input. In the current study, all such layers were derived either from the satellite sensor imagery or the DEM data. Rainfall data were taken from a real event that occurred on 11th August 2003, measured by a rain gauge installed in the Pokhare Khola catchment. Other selected input parameters (predominantly plant and soil characteristics) were adapted from the Indian Himalayan catchment used in previous LISEM work. Such parameters affected the magnitude of the output erosion simulations, but not the trends exhibited between similarly controlled experiments. Three alternate environmental scenarios were modelled using LISEM that differed from one another in the following ways: Simulation 1. Current land cover without terraces: this simulation was used as the control. The land cover map derived from Quickbird imagery was used as the primary input, along with ASTER DEM data which were coarse enough to represent overall slope, rather than smaller scale terracing patterns. Simulation 2. Current land cover with terraces: the same land cover and topographic maps were used as input data as Simulation 1, but saturated conductivity of the upper soil level was increased to represent higher infiltration on terraced rather than sloping plains. The initial moisture contents of the upper and second soil layers were also increased accordingly, assuming that terraced terrain would hold greater moisture through increased infiltration. Finally, slope angles were decreased to model the effect of terracing on local scale topography. Simulation 3. Totally forested catchment: infiltration and initial soil moisture content were not altered from Simulation 1 for the totally forested catchment.
433
Instead the land cover map was modified so that all terraced and scrub terrain was replaced with tree cover. Output data from the simulations included soil erosion maps, deposition maps, hydrographs and sedigraphs. 5. Results and discussion 5.1. Local and regional land cover and land cover dynamics The land cover map derived from the panchromatic Quickbird imagery allowed environmental trends to be analysed at an unprecedented level of detail (Fig. 2). Combining all land cover data for the Pokhare Khola showed the predominant land covers to be forest (35.6%), shrub (16.3%) and agricultural land (47.7%). These statistics, derived independently of any map sheet reference data, correspond well with the classified maps of 1992, which quantify land cover within the catchment to be 49.2% forest and shrub combined and 45.3% cultivated land. Comparing the two datasets by means of a confusion matrix, it is also clear that a high level of correlation exists between the absolute locations of each land cover class (Table 1), suggesting that there has not been any significant modification of land cover during the period 1992–2003. The major disparity evident in the confusion matrix is associated with the mapping of scrub using the Quickbird sensor imagery. The majority (N80%) of the area classified as scrub from the Quickbird image is classified in the maps as forest cover. This may be accounted for by a decrease in tree density in such areas, the appearance of which is not as striking as widespread deforestation, but is significant enough to yield a separate class from forest. The general trends displayed in these results concur with several previous studies that have also used remote-sensing to study land use changes in the Himalayan region (e.g. Balla et al., 2003; Rao & Pant, 2001; Wakeel et al., 2005), but differ in the level of detail afforded by the Quickbird imagery, which takes analyses to an individual dwelling scale for the first time. It is now possible to associate habitations with parcels of land as well as accurately distinguish between forest, scrub, crops and bare soil. This level of detailed analysis is commensurate with that possible using aerial photography, opening up the potential for historical analyses of land cover change where such archive datasets exist. The mosaiced map sheets facilitated a comparison of land use trends in the Pokhare Khola with those found in the wider region. In total, the forest and shrub classes digitised from Quickbird imagery of the catchment accounted for 52.0% of the land cover. In the area covered by the map sheet data compiled in 1992, 52.2% of all land cover was classified as forest and bush (Table 2). Across the region as a whole, cultivated land accounted for 42.1% of the total area (compared to 45.3% at a catchment-scale), and the remaining 5.7% was classified as rivers and roads. These figures are in general agreement with previous work conducted in the combined Galaundu and Pokhare catchments, which calculated the percentage of cultivated land to be 43.6% (Balla, 2005). They also suggest,
434
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
Fig. 2. Land cover map digitised from Quickbird Pan and XS imagery of the Pokhare Khola catchment, Nepal.
given the correlation of local to regional trends, that land cover in the Pokhare Khola is representative of the regional pattern and that discussion relating to land cover in subsequent sections may thus be scaled up to apply to the wider Nepal Middle Hills region. Given the fine-resolution of the Quickbird panchromatic imagery, individual dwellings could be easily delineated, as could their associated terraced farmland. At the time of image acquisition (February, 2003) there were 350 separate properties in the Pokhare Khola catchment, located without exception on terraced land (Fig. 3) and on slopes with an average inclination of 16.6°, according to the ASTER-derived DEM. The catchment size was calculated as 5.28 km2, which equates to a density of 67 dwellings per km2 across the catchment as a whole (Table 2).
Table 1 Confusion matrix comparing map sheet and manually digitised land cover classifications Map class
Digitised class Farming
Forest
Scrub
Total
Farming 79768 (81.05) 11190 (12.18) 9145 (18.69) 100103 (41.84) Forest 13832 (14.05) 76647 (83.43) 39233 (80.17) 129712 (54.22) Bush, 4824 (4.90) 4028 (4.38) 558 (1.14) 9410 (3.93) bamboo, grass Total 98424 91865 48936 239225 (100.00) (100.00) (100.00) (100.00)
In the area covered by the map sheet data a total of 30,821 dwellings were classified. This equates to a dwelling density of 23 properties per km2, considerably lower than that shown in the Pokhare Khola catchment itself (Table 2). This can be accounted for largely by the presence of rivers and floodplains in the wider region, which are generally uninhabitable. The Pokhare Khola catchment does not have many such areas. Another striking difference between regional and local trends relates to dwelling locations. The Himalayan foothills, located in the northern part of the studied area, appear to steer habitation towards relatively steep slopes. On average, properties in the wider region are located on slopes of 20.3°, based on SRTM DEM data, much steeper than in the Pokhare Khola catchment. Table 2 Comparison of catchment-scale and regional scale land cover parameters Catchment-scale (derived from Quickbird) Catchment size Forested area (digitised) Scrub area (digitised) Total forest + scrub (digitised) Percentage (forest and scrub) of total catchment No. of dwellings Dwellings per km2
Regional scale (derived from map sheets)
5,275,812 m2 Total map coverage 1,357,702,850 m2 1,880,478 m2 Forested area 510,075,200 m2 862,047 m2 2,742,525 m2 51.98
350 67
Bush, bamboo, grass Total bush + forest Percentage (forest and bush) of total catchment No. of dwellings Dwellings per km2
198,593,775 m2 708,668,975 m2 52.20
30,821 23
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
435
Fig. 3. Farming terraces in the Pokhare Khola catchment. Note the location of dwellings (highlighted in red), which are without exception found on terraced terrain. The extents of the image are shown on Fig. 2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
The land use does not change however; all dwellings are still found on land classified as cultivated. NDVI calculations applied to SPOT imagery from 1986 and 2001 show that there has been little modification of land cover in the Pokhare Khola catchment since the late 1980s (Fig. 4). The greatest differences in NDVI values between the two datasets can be accounted for by illumination variations and shadow (in places so dense that NDVI could not correct the data), suggesting that in this catchment at least, large-scale deforestation is not currently problematic, nor has been since 1986. These trends are in contrast to some reports that suggest
the restrictions on tree-felling in lower elevation areas have exacerbated problems in middle- and high-elevation terrain (MOPE, 2000). They do, however, support other studies that show forest covers in similar catchments in the area have stabilised in recent decades, and in some cases have even increased (Gautam et al., 2003). This relative land cover stability provides evidence of ecological sustainability, possibly as a result of land use management and rehabilitation programmes at a local scale (Neupane, 2004), but may also be at the cost of soil nutrient replenishment as local populations turn to cattle dung as a source of fuel instead of firewood, as has
Fig. 4. NDVI of the Pokhare Khola catchment (delineated in white) from a) SPOT 1 imagery of 22nd October 1986 and b) SPOT 4 imagery of 21st December 2001.
436
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
Fig. 5. Discharge at the catchment outlet for a simulated rainfall event in the Pokhare Khola, for three different land cover scenarios.
been the trend at lower elevations (MOPE, 2000). NDVI for the region as a whole show a similar trend, i.e., very little change in vegetation cover between image acquisitions, with small variations attributable to illumination variations and small areas of dark shadow. This again suggests that the area has reached an environmental equilibrium, with natural re-growth of forested areas being cancelled out by small scale tree-felling for personal use. However, these results could also reflect a consistency in the total forested area, but an overall reduction in tree density. Indeed, other studies have suggested that forest degradation did continue in the late 1980s and 1990s, but in the
form of thinning forests rather than widespread tree-felling (Rao & Pant, 2001; Wakeel et al., 2005). Without historical imagery of similar resolution to Quickbird data, it is not possible to confirm either of the hypotheses; it is clear, however, that the widespread deforestation reported during the period 1950–80 is no longer problematic in the study area. Combined with vegetal cover, slope inclination is a major factor influencing soil erosion processes in the Himalayan region (Kimoti & Juyal, 1996). The Pokhare Khola catchment is dominated by steep slopes with 70% of all terrain steeper than 15.0°. This factor may partly explain the lack of agricultural
Fig. 6. Differences in catchment erosion between LISEM Simulations 1 and 2.
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439 Table 3 LISEM outputs from Simulation 1 (no terracing), Simulation 2 (with terracing) and Simulation 3 (forested) scenarios Unit Catchment area Total rainfall Total discharge Total interception Total infiltration Mass balance error (water) Total discharge Peak discharge Peak time Discharge/rainfall Splash detachment Flow detachment (land) Deposition (land) Erosion channel Deposition channel Total soil loss Average soil loss
Simulation 1 Simulation 2 Simulation 3
ha 585.6 mm 58.9 mm 1.1 mm 1.7 mm 56.1 % 0.0 m3 6640.3 l/s 3946.7 min 71.0 % 1.9 ton 105.2 ton 8572.9 ton − 5548.8 ton 2381.5 ton − 418.4 ton 5082.1 kg/ha 8678.7
558.3 61.8 0.9 1.7 59.2 0.0 4825.3 2007.2 82.0 1.4 115.9 2039.2 − 1484.9 5544.7 − 3242.2 2965.1 5310.9
585.6 58.9 0.2 1.4 57.3 0.0 1045.1 183.2 88.0 0.3 89.9 0.0 −89.9 972.3 −405.4 556.7 950.7
437
selection of less-steep land for agricultural purposes by local farmers. These trends are also reflected in the wider region. The slope map derived from SRTM DEM data shows that outside the Pokhare Khola catchment, forest is found on terrain with an average slope of 27.0° and cultivation is practiced on terrain with an average slope of 22.6°. These figures are distinctly higher than the corresponding catchment-scale results, because of the more extreme topography of the surrounding region, particularly in the high mountain areas to the north and south of the catchment. Nevertheless, they support the suggestion that flatter land is cultivated in preference to steeper terrain. The implication of this for environmental degradation is that those slopes most susceptible to soil erosion from surface water runoff remain protected; on the other hand, flatter slopes are more susceptible to splash erosion, particularly in the summer months when the terrain has been baked dry. 5.2. Soil erosion modelling
expansion in recent decades both within the catchment and the region as a whole, particularly if all land deemed suitable for farming has been exhausted. Previous studies have shown that in catchments of extreme topography, lower slope inclinations are preferentially cultivated where expansion of cultivated land area has been observed (Rao & Pant, 2001). Within the Pokhare Khola, topography certainly appears to influence the distribution of land cover type, with forest found on terrain with an average slope of 20.2° and terraced farmland located on terrain with an average slope of 18.2°. This appears to reflect the
By running three simulations in LISEM with identical rainfall events, it was possible to compare the effect of the current terraced farmland situation with two extreme scenarios: a completely unmodified catchment (i.e. totally forested) and a catchment partially deforested, but without terraced slopes. The effect of land cover on discharge is immediately apparent by comparing the simulated catchment outlet hydrographs (Fig. 5). Discharge is decreased significantly by terraced slopes, when compared to bare, ungraded slopes, but the greatest decrease is produced by a totally forested catchment,
Fig. 7. Differences in catchment erosion between LISEM Simulations 1 and 3.
438
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
which is predominantly accounted for by factors such as increased hydraulic conductivity (infiltration). Runoff peaks are also retarded by creating a flatter and denser land cover, which is expected because of increased infiltration and interception in both forested and terraced simulations. Similar trends can be observed in the output erosion maps. By comparing Simulation 1 and Simulation 2 (without terracing and with terracing respectively) it can be seen that for large parts of the catchment, erosion is lower for terraced slopes than for bare slopes (Fig. 6). These areas correspond to the agricultural land, which was modified in Simulation 2 to give the effect of terracing. Other land uses generally do not show any change, as they remained unmodified in the input map data. The clearest indication of the effect of terracing comes from the figures related to total soil erosion across the catchment: erosion in a catchment with terraced slopes is 38.8% less than without terracing (Table 3), highlighting the beneficial effects of local farming practices to environmental sustainability, by reducing nutrient loss through increasing infiltration. However, simulating a fully forested catchment (Simulation 3, Fig. 7) and comparing it with unterraced, bare slopes (Simulation 1) shows the degree to which deforestation really damages the Himalayan environment. These data highlight the importance of the forest canopy for intercepting rainfall and protecting the terrain from splash erosion. Indeed, the model outputs show that a fully forested catchment produces less than one-fifth (17.9%) of the soil erosion of a catchment modified by terraced farming (Table 3). In comparison, a fully forested catchment produces just over one-tenth (10.9%) of the erosion of a partially deforested catchment not converted to terracing. Thus, according to the LISEM model outputs, it can be seen that while terraced farming practices in the Pokhare Khola catchment reduce the impact of deforested land on soil erosion by around 39%, they cannot alone fully compensate for the losses due to the removal of trees. This case study shows the potential for more detailed soil erosion modeling work in Himalayan catchments now that fineresolution land cover data can be so easily derived. Nevertheless, given the remote-sensing approach, there are inevitably a number of factors that cannot be considered in such work, but play an important role in the Himalayan degradation issue. For example, in addition to nutrient loss through runoff, intensive farming can contribute to soil degradation if strategic plans are not implemented for crop rotation and soil fertilisation. Different crop heights and densities may influence infiltration rates and thus soil erosion magnitudes. Further, compacting of farmland by cattle grazing may increase runoff, erosion and ultimately outlet discharge. Thus, such results should be taken as indicators of relative trends, rather than absolute figures with specific implications. 6. Conclusions This study has demonstrated that fine spatial resolution imagery facilitates land cover mapping at an unprecedented level of detail. Quickbird panchromatic imagery allowed trends of land use and farming practice to be characterised for the
Pokhare Khola catchment at an individual dwelling scale. Basic but important relationships were thus identified that give an insight into the interactions between farmers and their land, such as the controlling parameters on dwelling locations and the type of land that is preferentially cultivated. The potential availability of fine-resolution land cover data derived from remote-sensing is of particular significance for studies focussing on areas with limited accessibility, because of extreme topography or political sensitivity for example. The trends identified within the Pokhare Khola catchment were shown to be representative of the wider Nepal Middle Hills. As previous reports have suggested, deforestation and the associated extension of cultivation have not recently been problematic in the study area, as a stable land cover has been maintained over a twenty-year period. Such stability is vital if soil degradation is to be managed effectively across the region. This management will first require a full assessment of the impact of the current land cover situation on environmental sustainability. Our erosion modelling has gone some way towards this, having shown that although terracing cannot fully compensate for the effects of rainfall on steeply sloped deforested areas, it does contribute to lowering surface runoff by reducing the local hill gradient. Indeed, given that agricultural practices are essential for human survival in the region, a terraced landscape is clearly preferable and more sustainable than non-terraced, bare terrain. These initial findings suggest that with further availability of fine-resolution remote-sensing data, such as that demonstrated for the Pokhare Khola catchment, more detailed modelling will be feasible for assessing the severity of Himalayan environmental degradation in a more quantitative and rigorous manner. Acknowledgements This work was funded by the European Commission, Proposal Number ICA4-2000-10402 (Himalayan Degradation). The authors would like to thank Mr. Biswombher Man Pradhan for his help in the field and for providing expertise in the region. We are also grateful to Sanjeevi Shanmugam for help with analysing ASTER data. References Bajracharya, D. (1983). Deforestation in the food/fuel context: Historical and political perspectives from Nepal. Mountain Research and Development, 3, 227−240. Balla, M. K. (2005). Land use change in Galaundu/Pokhare Khola watershed, Dhading district, central Nepal. Research report. Pokhara, Nepal: Institute of Forestry. Balla, M. K., Awasthi, K. D., Shrestha, P. K., & Sherchan, D. P. (2003). Land use changes in two subwatersheds in Chitwan and Tanahun districts, west Nepal. Nepal Journal of Science and Technology, 5, 49−55. Gautam, A. P., Webb, E. L., Shivakoti, G. P., & Zoebisch, M. A. (2003). Land use dynamics and landscape change pattern in a mountain watershed in Nepal. Agriculture, Ecosystems and Environment, 99, 83−96. Jetten, V., & De Roo, A. P. J. (2001). Spatial analysis of erosion conservation measures with LISEM. In R. Harmon & W.W. Doe (Eds.), Landscape erosion and evolution modelling ( pp. 429445). New York: Kluwer Academic/Plenum. Kääb, A., Huggel, C., Paul, F., Wessels, R., Raup, B., Kieffer, H., et al. (2003). Glacier monitoring from ASTER imagery: Accuracy and
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439 applications. EARSeL Proceedings, LIS-SIG Workshop, Berne, March 11– 13, 2002 ( pp. 4353). Kimoti, M. M., & Juyal, N. (1996). Environmental impact assessment of a few selected watersheds of the Chamoli district (central Himalaya) using remotely sensed data. International Journal of Remote Sensing, 17, 1391−1405. MOPE (2000). Nepal's state of the Environment. Public report, His Majesty's government Kathmandu: Ministry of Population and Environment. Nearing, M. A., Jetten, V., Baffaut, C., Cerdan, O., Couturier, A., Hernandez, M., et al. (2005). Modeling response of soil erosion and runoff to changes in precipitation and cover. Catena, 61, 131−154. Neupane, K. P. (2004). Community management of natural resources in Galaudi/Pokhare Khola sub-watershed area of Dhading district of Nepal. Uganda Journal of Agricultural Sciences, 9, 414−424. Rao, K. S., & Pant, R. (2001). Land use dynamics and landscape change pattern in a typical micro watershed in the mid elevation zone of central Himalaya, India. Agriculture, Ecosystems and Environment, 86, 113−123. Sankhayan, P. L., Gurung, N., Sitaula, B. K., & Hofstad, O. (2003). Bioeconomic modelling of land use and forest degradation at watershed level in Nepal. Agriculture, Ecosystems and Environment, 94, 105−116. Saxena, K. G., Maikhuri, R. K., & Rao, K. S. (2005). Changes in agricultural biodiversity: Implications for sustainable livelihood in the Himalaya. Journal of Mountain Science, 2, 23−31. Semwal, R. L., Nautiyal, S., Sen, K. K., Rana, U., Maikhuri, R. K., Rao, K. S., et al. (2004). Patterns and ecological implications of agricultural land-use changes: A case study from central Himalaya, India. Agriculture, Ecosystems and Environment, 102, 81−92. Sen, K. K., Rao, K. S., & Saxena, K. G. (1997). Soil erosion due to settled upland farming in Himalaya: A case study in Pranmati watershed. International Journal of Sustainable Development and World Ecology, 4, 65−74.
439
Sitaula, B. K., Sankhayan, P. L., Bajracharya, B. M, & Singh, B. R. (2005). A system analysis of soil and forest degradation in a mid-hill watershed of Nepal using a bio-economic model. Land Degradation and Development, 15, 435−446. Sthapit, K. M., & Balla, M. K. (1999). Sedimentation monitoring of Phewa lake, Kaski, Nepal. In P. Mathema, I. C. Dutta, M. K. Balla, & S. N. Adhikary (Eds.), Sustainable forest management. Proceedings of an International Seminar, Pokhara, Nepal, 31 August to 2 September 1998. Pokhara, Nepal: Institute of Forestry. Tekle, K., & Hedlund, L. (2000). Land cover changes between 1958 and 1986 in Kalu District, Southern Wello, Ethiopia. Mountain Research and Development, 20, 42−51. Thapa, G. B. (1996). Land use, land management and environment in a subsistence mountain economy in Nepal. Agriculture, Ecosystems and Environment, 57, 57−71. Tiwari, P. C. (2000). Land-use changes in Himalaya and their impact on the plains ecosystem: Need for sustainable land use. Land Use Policy, 17, 101−111. Van Dijck, S., de Coninck, H., Keizer, J., Ferreira, A., & Stolte, J. (submitted for publication). The relevance of interception in event-based rainfall-runoff modeling. Hydrological Processes. Wakeel, A., Rao, K. S., Maikhuri, R. K., & Saxena, K. G. (2005). Forest management and land use/cover changes in a typical micro watershed in the mid elevation zone on Central Himalaya, India. Forest Ecology and Management, 213, 229−242.
Fine-resolution remote-sensing and modelling of Himalayan catchment sustainability D.J. Quincey a,⁎, A. Luckman a , R. Hessel b , R. Davies a , P.L. Sankhayan c , M.K. Balla d a
School of the Environment and Society, Swansea University, Singleton Park, Swansea, SA2 8PP, UK b Soil Science Centre, Alterra, WUR, P.O. Box 47, 6700 AA Wageningen, The Netherlands c Department of Ecology and Natural Resource Management, Norwegian University of Life Sciences, P.O. Box 5003, N-1432 Aas, Norway d Tribuvan University, Kirtipur, Kathmandu, Nepal
Received 21 February 2006; received in revised form 20 September 2006; accepted 23 September 2006
Abstract A number of studies have reported on environmental degradation in the Nepal Himalaya as a result of large-scale deforestation and the associated agricultural extension. In contrast to many previous regional scale studies, we consider land cover and its environmental impact on an individual catchment-scale, using fine-resolution Quickbird data and a soil erosion model. First, using a detailed land cover map generated from Quickbird imagery, we establish basic relationships between land cover, dwelling density and topographic variability, which exist in a typical midelevation Nepalese catchment, the Pokhare Khola. These data suggest a strongly subsistence type of household economy based predominantly on terraced arable farming. We demonstrate using multitemporal vegetation indices that this farmland has existed in the region since the late 1980s, and that widespread deforestation has not taken place since then, possibly as a result of specific forest conservation policies of the government coupled with efforts by local communities. Further, through the use of soil erosion modelling we demonstrate the role that the terraced farming practices can play in reducing runoff and hence soil nutrient loss, thereby enabling restoration of vegetation in the previously deforested land terrains. Finally, by combining this information with regional land cover data, we show that the findings of this research can be scaled up to draw conclusions about environmental degradation across the Nepal Himalayan region. © 2006 Elsevier Inc. All rights reserved. Keywords: Remote-sensing; Quickbird; Deforestation; Environmental degradation; Soil erosion modelling; Nepal Middle Hills
1. Introduction Environmental degradation poses a serious threat to the longterm sustainability of ecologically fragile mountainous Himalayan watersheds characterised by subsistence farming (Tiwari, 2000). Socio-economic and environmental changes have impacted heavily on land use in the region, mainly manifested by the extension of cultivation into marginal land and forested areas. Such changes have depleted and eroded the natural resources around populated areas, thereby reducing groundwater recharge and increasing surface runoff. The problem is well pronounced in ⁎ Corresponding author. 35 Spon Green, Buckley, Flintshire, CH7 3BH, UK. Tel.: +44 1244 540942. E-mail address: [email protected] (D.J. Quincey). 0034-4257/$ - see front matter © 2006 Elsevier Inc. All rights reserved. doi:10.1016/j.rse.2006.09.021
mid-elevation (i.e. 1000–3000 m) areas of Nepal, where steep slopes and narrow valleys are conducive to high levels of erosion, particularly following a period of large-scale deforestation from approximately 1950–80. During this time, forested areas were placed under government ownership, which loosened controls on natural resource management and thus led to large-scale devastation of forest cover (Bajracharya, 1983). A key consequence of this land cover change was an increase in the rate of soil erosion in the region, which is now known to be very high in comparison to other parts of the world (Sen et al., 1997). This in turn has led to a gradual decrease in soil productivity, agricultural production and the incomes of subsistence farmers of the region. Several studies have documented the extent of land use change in the area (Balla et al., 2003; Gautam et al., 2003; Rao & Pant, 2001), reporting on trends of deforestation and, in more
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
recent decades, a reversal of this trend following forest conservation efforts by local communities and external agencies. Others have focussed on the controlling factors of the observed environmental change using bio-economic modelling (e.g. Sankhayan et al., 2003; Sitaula et al., 2005). Some studies have reported on the effects of degradation in the region, mostly focussing on river and reservoir siltation (Sthapit & Balla, 1999) and increases in the incidence and severity of flooding (Tiwari, 2000) although others have also considered the long-term effects on economic prosperity (Semwal et al., 2004). With the exception of a few studies (e.g. Saxena et al., 2005; Thapa, 1996), little attention has been paid to the management and conservation of agricultural lands, which account for a substantial proportion of all land resources in the region. Until recently, detailed assessments of Himalayan catchment dynamics have depended mostly on the feasibility of extensive field investigations and surveys. Now, however, readily available and affordable fine-resolution satellite sensor imagery offers an alternative method for extracting such information, for analysis by itself or as input to models for simulating the impact of altering the environmental conditions within a catchment. Repeated satellite images can be used both for the visual assessment of land resources as well as the quantitative evaluation of land cover changes over time (Tekle & Hedlund, 2000). Furthermore, hydrological models can now be employed to evaluate the impact of land cover change on a catchment, such as the expansion of agriculture into previously forested areas. This approach allows us to better quantify the trends observed across the wider Himalayan region. The impact of terrace farming practices on ecologically sensitive land is currently poorly understood in the Himalayan region. Thus, the aim of this paper is to provide an up-to-date analysis of the key issues surrounding environmental degradation in the Himalaya. We aim to demonstrate, for the first time, how the integration of fine-resolution remote-sensing data with GIS-based soil erosion modelling can be used to establish trends in land cover and land use, and evaluate how changes in both impact on land sustainability. This requires the fulfilment of several key objectives: 1. To use Quickbird remote-sensing data to map land cover in the Pokhare Khola catchment, Nepal, and analyse the extent to which it has changed over recent decades, by comparison with historical imagery and map sheet information. 2. To compare these catchment-scale land cover trends with regional trends, derived from various sources including previous studies, satellite data at a more coarse resolution (but with greater coverage) and classified historical map sheet data. 3. To couple the fine-resolution land cover maps with elevation data to demonstrate new potential for modelling the impact of deforestation and terrace agriculture on soil erosion in Himalayan catchments. 2. Study area The Pokhare Khola catchment is situated in the Dhading district of the central development region of Nepal, and covers
431
an area of approximately 5.28 km2 (Fig. 1). Geographically, it lies between 27°46′28″ N and 27°48′06″ N latitude and 84°53′32″ E and 84°55′11” E longitude. The climate is classified as sub-tropical, and temperatures reach a maximum of 31 °C in summer months and 8 °C during winter. The average annual rainfall is 1370 mm, most of which falls during the monsoon period between May and September. The catchment is steepest in the south, reaching an elevation of 1079 m and drains to the north through its lowest point at approximately 380 m. Here, the water drains into the Trishuli River and travels westwards taking drainage into the lowlands. The catchment land cover is a mixture of forest, scrub and subsistence terrace farming. There are two predominant crop seasons, namely, rainy and winter seasons. While paddy and maize are the dominant rainy season crops, wheat and vegetables are grown in the winter season. Most families also keep livestock, mainly cows, goats and oxen, as complementary and supplementary activities to agriculture for their livelihood. 3. Data sources Satellite image interpretation is based primarily on Quickbird data acquired on 21 February 2003, at 0.6 m spatial resolution in the panchromatic band and 2.8 m spatial resolution in four spectral bands. Neither scene was hampered by the presence of cloud cover and both were geo-referenced to UTM Zone 45 (WGS 84). In addition to the Quickbird data, which covered little more than the Pokhare Khola catchment, historical data from the 21st December 2001 and 22nd October 1986 were also acquired from the SPOT 4 HRVIR sensor (10 m spatial resolution multispectral) and SPOT 1 HRV sensor (20 m spatial resolution multispectral). Both images were supplied as L1G data, which are geo-referenced and radiometrically corrected. The scenes were cloud free and largely unaffected by shadow. SRTM DEM data were acquired (post-spacing of 90 m) to facilitate a regional assessment of topographic variability, but the coarse resolution of the data was insufficient for a detailed catchment-scale analysis. Thus, an ASTER DEM was also generated (post spacing of 15 m) using ERDAS software, which was used to establish topographic trends on a much finer scale in the Pokhare Khola. Comparisons between the two DEM datasets showed an average absolute height disparity of 56 m, but visual inspection of the two datasets revealed a high relative correlation, suggesting an accurate replication of the catchment topography by the ASTER stereo model. This finding contrasts with previous reports of large ASTER DEM errors in areas of extreme topography (Kääb et al., 2003). In addition to satellite sensor imagery, eight map sheets of the catchment and its surrounding areas were acquired and scanned at a resolution of 300 dpi, equivalent to approximately 2 m pixel size. These map sheets were presented at a scale of 1:25 000 with primary elevation contours of 20 m, in a Modified Universal Transverse Mercator projection. They were compiled by the Survey Department of the Government of Nepal in 1994, in co-operation with the Government of Finland, using 1:50 000 scale aerial photography acquired in 1992.
432
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
4. Methods 4.1. Remote-sensing The scanned map sheets were registered to UTM zone 45 (WGS 84) and stitched together to act as a base map for the rest of the image data in the Pokhare Khola catchment. The SPOT imagery from 1986 and 2001 were orthorectified to these maps, and as a result were co-registered to one another within ± 20 m across the Pokhare Khola catchment and to within ± 40 m in the wider area. The Quickbird imagery was delivered already georegistered to the UTM co-ordinate system, and could thus be overlaid on the map sheets and SPOT imagery with maximum errors of ± 5 m. Both SRTM and ASTER-derived DEM data were accurately co-registered with the map sheet mosaic within the catchment as well as in the surrounding region. Vector GIS coverages delineating forest, shrub, terraces, rivers, dwellings and roads within the Pokhare Khola catchment were generated by visual interpretation of the panchromatic Quickbird imagery, with some help from the NIR band of the multispectral imagery for discriminating vegetation. Slope images were also prepared using the ASTER-derived DEM for the catchment, although the raw elevation data were first smoothed to eliminate any spurious values. The SRTM DEM data were used to generate slope images for the wider region. In each case a 3 × 3 pixel computation kernel was employed, which considers all surrounding eight pixels when calculating slope angles.
NDVI (Normalised Difference Vegetation Index) were calculated according to the standard algorithm NIR-Red/NIR+ Red for the multitemporal SPOT imagery. The data were not atmospherically corrected prior to calculating NDVI as it was assumed that similar atmospheric conditions prevailed at each acquisition. Failure of this assumption would lead to a small amount of atmospherically-introduced noise of insufficient magnitude to affect the coarse trends presented in this research. Once processed, a common area around the Pokhare Khola catchment was extracted from the NDVI data so that areas of deforestation or new cultivation could be rapidly identified. To enable comparison with the manually digitised land cover map, the scanned map sheets were classified using a simple maximum likelihood algorithm. Training data were collected from the six major land cover classes, with a minimum sample of 200 pixels per class. The results were subsequently smoothed using a median filter to eradicate contouring through the classification, and then resampled to the Quickbird-derived land cover map to facilitate a fair comparison. 4.2. Soil erosion modelling The land cover map derived from Quickbird data was coupled with the smoothed ASTER DEM data for use in the soil erosion model. We chose the Limburg Soil Erosion Model (LISEM; Jetten & de Roo, 2001) for our analyses, which is a physically based soil erosion model that simulates runoff and erosion for single rainfall events. It was designed for small
Fig. 1. Location of the Pokhare Khola catchment and the geographic extents of the remote-sensing and map sheet data sources.
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
catchments, in the order of a few km2 in size. Process descriptions are according to current knowledge on hydrology and erosion, and include interception, infiltration, detachment, deposition and routing of overland flow and sediment. LISEM is a distributed model, which has been integrated with a GIS. It reads raster maps as inputs, and also generates raster maps as outputs, which depict the spatial distribution of runoff and erosion. The model was first developed to test small scale soil conservation measures on soil loss in the Netherlands, but has since been applied to catchments in Europe, Asia, Africa, America and on the Pacific Islands. These catchments have included agricultural catchments, but also catchments mostly covered by scrub (Nearing et al., 2005) and forest (Van Dijck et al., submitted for publication). While there is currently no literature detailing the application of LISEM in the Himalaya, recent results from an Indian Himalayan catchment with similar topography and climatic characteristics as the Pokhare Khola have suggested the model is also applicable in such areas (R. Hessel, 2006, personal communication). A number of detailed map layers are required to run LISEM. These include catchment, vegetation and soil surface maps, as well as a detailed land cover map as its primary input. In the current study, all such layers were derived either from the satellite sensor imagery or the DEM data. Rainfall data were taken from a real event that occurred on 11th August 2003, measured by a rain gauge installed in the Pokhare Khola catchment. Other selected input parameters (predominantly plant and soil characteristics) were adapted from the Indian Himalayan catchment used in previous LISEM work. Such parameters affected the magnitude of the output erosion simulations, but not the trends exhibited between similarly controlled experiments. Three alternate environmental scenarios were modelled using LISEM that differed from one another in the following ways: Simulation 1. Current land cover without terraces: this simulation was used as the control. The land cover map derived from Quickbird imagery was used as the primary input, along with ASTER DEM data which were coarse enough to represent overall slope, rather than smaller scale terracing patterns. Simulation 2. Current land cover with terraces: the same land cover and topographic maps were used as input data as Simulation 1, but saturated conductivity of the upper soil level was increased to represent higher infiltration on terraced rather than sloping plains. The initial moisture contents of the upper and second soil layers were also increased accordingly, assuming that terraced terrain would hold greater moisture through increased infiltration. Finally, slope angles were decreased to model the effect of terracing on local scale topography. Simulation 3. Totally forested catchment: infiltration and initial soil moisture content were not altered from Simulation 1 for the totally forested catchment.
433
Instead the land cover map was modified so that all terraced and scrub terrain was replaced with tree cover. Output data from the simulations included soil erosion maps, deposition maps, hydrographs and sedigraphs. 5. Results and discussion 5.1. Local and regional land cover and land cover dynamics The land cover map derived from the panchromatic Quickbird imagery allowed environmental trends to be analysed at an unprecedented level of detail (Fig. 2). Combining all land cover data for the Pokhare Khola showed the predominant land covers to be forest (35.6%), shrub (16.3%) and agricultural land (47.7%). These statistics, derived independently of any map sheet reference data, correspond well with the classified maps of 1992, which quantify land cover within the catchment to be 49.2% forest and shrub combined and 45.3% cultivated land. Comparing the two datasets by means of a confusion matrix, it is also clear that a high level of correlation exists between the absolute locations of each land cover class (Table 1), suggesting that there has not been any significant modification of land cover during the period 1992–2003. The major disparity evident in the confusion matrix is associated with the mapping of scrub using the Quickbird sensor imagery. The majority (N80%) of the area classified as scrub from the Quickbird image is classified in the maps as forest cover. This may be accounted for by a decrease in tree density in such areas, the appearance of which is not as striking as widespread deforestation, but is significant enough to yield a separate class from forest. The general trends displayed in these results concur with several previous studies that have also used remote-sensing to study land use changes in the Himalayan region (e.g. Balla et al., 2003; Rao & Pant, 2001; Wakeel et al., 2005), but differ in the level of detail afforded by the Quickbird imagery, which takes analyses to an individual dwelling scale for the first time. It is now possible to associate habitations with parcels of land as well as accurately distinguish between forest, scrub, crops and bare soil. This level of detailed analysis is commensurate with that possible using aerial photography, opening up the potential for historical analyses of land cover change where such archive datasets exist. The mosaiced map sheets facilitated a comparison of land use trends in the Pokhare Khola with those found in the wider region. In total, the forest and shrub classes digitised from Quickbird imagery of the catchment accounted for 52.0% of the land cover. In the area covered by the map sheet data compiled in 1992, 52.2% of all land cover was classified as forest and bush (Table 2). Across the region as a whole, cultivated land accounted for 42.1% of the total area (compared to 45.3% at a catchment-scale), and the remaining 5.7% was classified as rivers and roads. These figures are in general agreement with previous work conducted in the combined Galaundu and Pokhare catchments, which calculated the percentage of cultivated land to be 43.6% (Balla, 2005). They also suggest,
434
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
Fig. 2. Land cover map digitised from Quickbird Pan and XS imagery of the Pokhare Khola catchment, Nepal.
given the correlation of local to regional trends, that land cover in the Pokhare Khola is representative of the regional pattern and that discussion relating to land cover in subsequent sections may thus be scaled up to apply to the wider Nepal Middle Hills region. Given the fine-resolution of the Quickbird panchromatic imagery, individual dwellings could be easily delineated, as could their associated terraced farmland. At the time of image acquisition (February, 2003) there were 350 separate properties in the Pokhare Khola catchment, located without exception on terraced land (Fig. 3) and on slopes with an average inclination of 16.6°, according to the ASTER-derived DEM. The catchment size was calculated as 5.28 km2, which equates to a density of 67 dwellings per km2 across the catchment as a whole (Table 2).
Table 1 Confusion matrix comparing map sheet and manually digitised land cover classifications Map class
Digitised class Farming
Forest
Scrub
Total
Farming 79768 (81.05) 11190 (12.18) 9145 (18.69) 100103 (41.84) Forest 13832 (14.05) 76647 (83.43) 39233 (80.17) 129712 (54.22) Bush, 4824 (4.90) 4028 (4.38) 558 (1.14) 9410 (3.93) bamboo, grass Total 98424 91865 48936 239225 (100.00) (100.00) (100.00) (100.00)
In the area covered by the map sheet data a total of 30,821 dwellings were classified. This equates to a dwelling density of 23 properties per km2, considerably lower than that shown in the Pokhare Khola catchment itself (Table 2). This can be accounted for largely by the presence of rivers and floodplains in the wider region, which are generally uninhabitable. The Pokhare Khola catchment does not have many such areas. Another striking difference between regional and local trends relates to dwelling locations. The Himalayan foothills, located in the northern part of the studied area, appear to steer habitation towards relatively steep slopes. On average, properties in the wider region are located on slopes of 20.3°, based on SRTM DEM data, much steeper than in the Pokhare Khola catchment. Table 2 Comparison of catchment-scale and regional scale land cover parameters Catchment-scale (derived from Quickbird) Catchment size Forested area (digitised) Scrub area (digitised) Total forest + scrub (digitised) Percentage (forest and scrub) of total catchment No. of dwellings Dwellings per km2
Regional scale (derived from map sheets)
5,275,812 m2 Total map coverage 1,357,702,850 m2 1,880,478 m2 Forested area 510,075,200 m2 862,047 m2 2,742,525 m2 51.98
350 67
Bush, bamboo, grass Total bush + forest Percentage (forest and bush) of total catchment No. of dwellings Dwellings per km2
198,593,775 m2 708,668,975 m2 52.20
30,821 23
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
435
Fig. 3. Farming terraces in the Pokhare Khola catchment. Note the location of dwellings (highlighted in red), which are without exception found on terraced terrain. The extents of the image are shown on Fig. 2. (For interpretation of the references to colour in this figure legend, the reader is referred to the web version of this article.)
The land use does not change however; all dwellings are still found on land classified as cultivated. NDVI calculations applied to SPOT imagery from 1986 and 2001 show that there has been little modification of land cover in the Pokhare Khola catchment since the late 1980s (Fig. 4). The greatest differences in NDVI values between the two datasets can be accounted for by illumination variations and shadow (in places so dense that NDVI could not correct the data), suggesting that in this catchment at least, large-scale deforestation is not currently problematic, nor has been since 1986. These trends are in contrast to some reports that suggest
the restrictions on tree-felling in lower elevation areas have exacerbated problems in middle- and high-elevation terrain (MOPE, 2000). They do, however, support other studies that show forest covers in similar catchments in the area have stabilised in recent decades, and in some cases have even increased (Gautam et al., 2003). This relative land cover stability provides evidence of ecological sustainability, possibly as a result of land use management and rehabilitation programmes at a local scale (Neupane, 2004), but may also be at the cost of soil nutrient replenishment as local populations turn to cattle dung as a source of fuel instead of firewood, as has
Fig. 4. NDVI of the Pokhare Khola catchment (delineated in white) from a) SPOT 1 imagery of 22nd October 1986 and b) SPOT 4 imagery of 21st December 2001.
436
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
Fig. 5. Discharge at the catchment outlet for a simulated rainfall event in the Pokhare Khola, for three different land cover scenarios.
been the trend at lower elevations (MOPE, 2000). NDVI for the region as a whole show a similar trend, i.e., very little change in vegetation cover between image acquisitions, with small variations attributable to illumination variations and small areas of dark shadow. This again suggests that the area has reached an environmental equilibrium, with natural re-growth of forested areas being cancelled out by small scale tree-felling for personal use. However, these results could also reflect a consistency in the total forested area, but an overall reduction in tree density. Indeed, other studies have suggested that forest degradation did continue in the late 1980s and 1990s, but in the
form of thinning forests rather than widespread tree-felling (Rao & Pant, 2001; Wakeel et al., 2005). Without historical imagery of similar resolution to Quickbird data, it is not possible to confirm either of the hypotheses; it is clear, however, that the widespread deforestation reported during the period 1950–80 is no longer problematic in the study area. Combined with vegetal cover, slope inclination is a major factor influencing soil erosion processes in the Himalayan region (Kimoti & Juyal, 1996). The Pokhare Khola catchment is dominated by steep slopes with 70% of all terrain steeper than 15.0°. This factor may partly explain the lack of agricultural
Fig. 6. Differences in catchment erosion between LISEM Simulations 1 and 2.
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439 Table 3 LISEM outputs from Simulation 1 (no terracing), Simulation 2 (with terracing) and Simulation 3 (forested) scenarios Unit Catchment area Total rainfall Total discharge Total interception Total infiltration Mass balance error (water) Total discharge Peak discharge Peak time Discharge/rainfall Splash detachment Flow detachment (land) Deposition (land) Erosion channel Deposition channel Total soil loss Average soil loss
Simulation 1 Simulation 2 Simulation 3
ha 585.6 mm 58.9 mm 1.1 mm 1.7 mm 56.1 % 0.0 m3 6640.3 l/s 3946.7 min 71.0 % 1.9 ton 105.2 ton 8572.9 ton − 5548.8 ton 2381.5 ton − 418.4 ton 5082.1 kg/ha 8678.7
558.3 61.8 0.9 1.7 59.2 0.0 4825.3 2007.2 82.0 1.4 115.9 2039.2 − 1484.9 5544.7 − 3242.2 2965.1 5310.9
585.6 58.9 0.2 1.4 57.3 0.0 1045.1 183.2 88.0 0.3 89.9 0.0 −89.9 972.3 −405.4 556.7 950.7
437
selection of less-steep land for agricultural purposes by local farmers. These trends are also reflected in the wider region. The slope map derived from SRTM DEM data shows that outside the Pokhare Khola catchment, forest is found on terrain with an average slope of 27.0° and cultivation is practiced on terrain with an average slope of 22.6°. These figures are distinctly higher than the corresponding catchment-scale results, because of the more extreme topography of the surrounding region, particularly in the high mountain areas to the north and south of the catchment. Nevertheless, they support the suggestion that flatter land is cultivated in preference to steeper terrain. The implication of this for environmental degradation is that those slopes most susceptible to soil erosion from surface water runoff remain protected; on the other hand, flatter slopes are more susceptible to splash erosion, particularly in the summer months when the terrain has been baked dry. 5.2. Soil erosion modelling
expansion in recent decades both within the catchment and the region as a whole, particularly if all land deemed suitable for farming has been exhausted. Previous studies have shown that in catchments of extreme topography, lower slope inclinations are preferentially cultivated where expansion of cultivated land area has been observed (Rao & Pant, 2001). Within the Pokhare Khola, topography certainly appears to influence the distribution of land cover type, with forest found on terrain with an average slope of 20.2° and terraced farmland located on terrain with an average slope of 18.2°. This appears to reflect the
By running three simulations in LISEM with identical rainfall events, it was possible to compare the effect of the current terraced farmland situation with two extreme scenarios: a completely unmodified catchment (i.e. totally forested) and a catchment partially deforested, but without terraced slopes. The effect of land cover on discharge is immediately apparent by comparing the simulated catchment outlet hydrographs (Fig. 5). Discharge is decreased significantly by terraced slopes, when compared to bare, ungraded slopes, but the greatest decrease is produced by a totally forested catchment,
Fig. 7. Differences in catchment erosion between LISEM Simulations 1 and 3.
438
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439
which is predominantly accounted for by factors such as increased hydraulic conductivity (infiltration). Runoff peaks are also retarded by creating a flatter and denser land cover, which is expected because of increased infiltration and interception in both forested and terraced simulations. Similar trends can be observed in the output erosion maps. By comparing Simulation 1 and Simulation 2 (without terracing and with terracing respectively) it can be seen that for large parts of the catchment, erosion is lower for terraced slopes than for bare slopes (Fig. 6). These areas correspond to the agricultural land, which was modified in Simulation 2 to give the effect of terracing. Other land uses generally do not show any change, as they remained unmodified in the input map data. The clearest indication of the effect of terracing comes from the figures related to total soil erosion across the catchment: erosion in a catchment with terraced slopes is 38.8% less than without terracing (Table 3), highlighting the beneficial effects of local farming practices to environmental sustainability, by reducing nutrient loss through increasing infiltration. However, simulating a fully forested catchment (Simulation 3, Fig. 7) and comparing it with unterraced, bare slopes (Simulation 1) shows the degree to which deforestation really damages the Himalayan environment. These data highlight the importance of the forest canopy for intercepting rainfall and protecting the terrain from splash erosion. Indeed, the model outputs show that a fully forested catchment produces less than one-fifth (17.9%) of the soil erosion of a catchment modified by terraced farming (Table 3). In comparison, a fully forested catchment produces just over one-tenth (10.9%) of the erosion of a partially deforested catchment not converted to terracing. Thus, according to the LISEM model outputs, it can be seen that while terraced farming practices in the Pokhare Khola catchment reduce the impact of deforested land on soil erosion by around 39%, they cannot alone fully compensate for the losses due to the removal of trees. This case study shows the potential for more detailed soil erosion modeling work in Himalayan catchments now that fineresolution land cover data can be so easily derived. Nevertheless, given the remote-sensing approach, there are inevitably a number of factors that cannot be considered in such work, but play an important role in the Himalayan degradation issue. For example, in addition to nutrient loss through runoff, intensive farming can contribute to soil degradation if strategic plans are not implemented for crop rotation and soil fertilisation. Different crop heights and densities may influence infiltration rates and thus soil erosion magnitudes. Further, compacting of farmland by cattle grazing may increase runoff, erosion and ultimately outlet discharge. Thus, such results should be taken as indicators of relative trends, rather than absolute figures with specific implications. 6. Conclusions This study has demonstrated that fine spatial resolution imagery facilitates land cover mapping at an unprecedented level of detail. Quickbird panchromatic imagery allowed trends of land use and farming practice to be characterised for the
Pokhare Khola catchment at an individual dwelling scale. Basic but important relationships were thus identified that give an insight into the interactions between farmers and their land, such as the controlling parameters on dwelling locations and the type of land that is preferentially cultivated. The potential availability of fine-resolution land cover data derived from remote-sensing is of particular significance for studies focussing on areas with limited accessibility, because of extreme topography or political sensitivity for example. The trends identified within the Pokhare Khola catchment were shown to be representative of the wider Nepal Middle Hills. As previous reports have suggested, deforestation and the associated extension of cultivation have not recently been problematic in the study area, as a stable land cover has been maintained over a twenty-year period. Such stability is vital if soil degradation is to be managed effectively across the region. This management will first require a full assessment of the impact of the current land cover situation on environmental sustainability. Our erosion modelling has gone some way towards this, having shown that although terracing cannot fully compensate for the effects of rainfall on steeply sloped deforested areas, it does contribute to lowering surface runoff by reducing the local hill gradient. Indeed, given that agricultural practices are essential for human survival in the region, a terraced landscape is clearly preferable and more sustainable than non-terraced, bare terrain. These initial findings suggest that with further availability of fine-resolution remote-sensing data, such as that demonstrated for the Pokhare Khola catchment, more detailed modelling will be feasible for assessing the severity of Himalayan environmental degradation in a more quantitative and rigorous manner. Acknowledgements This work was funded by the European Commission, Proposal Number ICA4-2000-10402 (Himalayan Degradation). The authors would like to thank Mr. Biswombher Man Pradhan for his help in the field and for providing expertise in the region. We are also grateful to Sanjeevi Shanmugam for help with analysing ASTER data. References Bajracharya, D. (1983). Deforestation in the food/fuel context: Historical and political perspectives from Nepal. Mountain Research and Development, 3, 227−240. Balla, M. K. (2005). Land use change in Galaundu/Pokhare Khola watershed, Dhading district, central Nepal. Research report. Pokhara, Nepal: Institute of Forestry. Balla, M. K., Awasthi, K. D., Shrestha, P. K., & Sherchan, D. P. (2003). Land use changes in two subwatersheds in Chitwan and Tanahun districts, west Nepal. Nepal Journal of Science and Technology, 5, 49−55. Gautam, A. P., Webb, E. L., Shivakoti, G. P., & Zoebisch, M. A. (2003). Land use dynamics and landscape change pattern in a mountain watershed in Nepal. Agriculture, Ecosystems and Environment, 99, 83−96. Jetten, V., & De Roo, A. P. J. (2001). Spatial analysis of erosion conservation measures with LISEM. In R. Harmon & W.W. Doe (Eds.), Landscape erosion and evolution modelling ( pp. 429445). New York: Kluwer Academic/Plenum. Kääb, A., Huggel, C., Paul, F., Wessels, R., Raup, B., Kieffer, H., et al. (2003). Glacier monitoring from ASTER imagery: Accuracy and
D.J. Quincey et al. / Remote Sensing of Environment 107 (2007) 430–439 applications. EARSeL Proceedings, LIS-SIG Workshop, Berne, March 11– 13, 2002 ( pp. 4353). Kimoti, M. M., & Juyal, N. (1996). Environmental impact assessment of a few selected watersheds of the Chamoli district (central Himalaya) using remotely sensed data. International Journal of Remote Sensing, 17, 1391−1405. MOPE (2000). Nepal's state of the Environment. Public report, His Majesty's government Kathmandu: Ministry of Population and Environment. Nearing, M. A., Jetten, V., Baffaut, C., Cerdan, O., Couturier, A., Hernandez, M., et al. (2005). Modeling response of soil erosion and runoff to changes in precipitation and cover. Catena, 61, 131−154. Neupane, K. P. (2004). Community management of natural resources in Galaudi/Pokhare Khola sub-watershed area of Dhading district of Nepal. Uganda Journal of Agricultural Sciences, 9, 414−424. Rao, K. S., & Pant, R. (2001). Land use dynamics and landscape change pattern in a typical micro watershed in the mid elevation zone of central Himalaya, India. Agriculture, Ecosystems and Environment, 86, 113−123. Sankhayan, P. L., Gurung, N., Sitaula, B. K., & Hofstad, O. (2003). Bioeconomic modelling of land use and forest degradation at watershed level in Nepal. Agriculture, Ecosystems and Environment, 94, 105−116. Saxena, K. G., Maikhuri, R. K., & Rao, K. S. (2005). Changes in agricultural biodiversity: Implications for sustainable livelihood in the Himalaya. Journal of Mountain Science, 2, 23−31. Semwal, R. L., Nautiyal, S., Sen, K. K., Rana, U., Maikhuri, R. K., Rao, K. S., et al. (2004). Patterns and ecological implications of agricultural land-use changes: A case study from central Himalaya, India. Agriculture, Ecosystems and Environment, 102, 81−92. Sen, K. K., Rao, K. S., & Saxena, K. G. (1997). Soil erosion due to settled upland farming in Himalaya: A case study in Pranmati watershed. International Journal of Sustainable Development and World Ecology, 4, 65−74.
439
Sitaula, B. K., Sankhayan, P. L., Bajracharya, B. M, & Singh, B. R. (2005). A system analysis of soil and forest degradation in a mid-hill watershed of Nepal using a bio-economic model. Land Degradation and Development, 15, 435−446. Sthapit, K. M., & Balla, M. K. (1999). Sedimentation monitoring of Phewa lake, Kaski, Nepal. In P. Mathema, I. C. Dutta, M. K. Balla, & S. N. Adhikary (Eds.), Sustainable forest management. Proceedings of an International Seminar, Pokhara, Nepal, 31 August to 2 September 1998. Pokhara, Nepal: Institute of Forestry. Tekle, K., & Hedlund, L. (2000). Land cover changes between 1958 and 1986 in Kalu District, Southern Wello, Ethiopia. Mountain Research and Development, 20, 42−51. Thapa, G. B. (1996). Land use, land management and environment in a subsistence mountain economy in Nepal. Agriculture, Ecosystems and Environment, 57, 57−71. Tiwari, P. C. (2000). Land-use changes in Himalaya and their impact on the plains ecosystem: Need for sustainable land use. Land Use Policy, 17, 101−111. Van Dijck, S., de Coninck, H., Keizer, J., Ferreira, A., & Stolte, J. (submitted for publication). The relevance of interception in event-based rainfall-runoff modeling. Hydrological Processes. Wakeel, A., Rao, K. S., Maikhuri, R. K., & Saxena, K. G. (2005). Forest management and land use/cover changes in a typical micro watershed in the mid elevation zone on Central Himalaya, India. Forest Ecology and Management, 213, 229−242.