4th International Conference on Integrating GIS and Environmental Modeling (GIS/EM4):
Problems, Prospects and Research Needs. Banff, Alberta, Canada, September 2 - 8, 2000.

Effect of Soil Landscape Parameterization for Watershed Modeling

GIS/EM4 No. 98

A-Xing Zhu
D. Scott Mackay

Abstract

This paper examines the effects of different approaches to soil landscape parameterization on hydro-ecological modeling at watersheds of the meso-scale size. Two different approaches were used to characterize soil landscape. The first approach was the conventional soil mapping under which soil landscape is portrayed as distinct and discrete soil bodies. The second approach is a soil similarity approach based on fuzzy logic. Under this approach, the soil is represented as a continuum both in its geographic space and attribute domain. The spatial variation of soil hydraulic properties within a Montana watershed was characterized using each of these two approaches. Each characterization was then used to drive a hydro-ecological simulation model (RHESSys). The examination was done in two modeling approaches: the lumped parameter approach and the distributed approach. It was found that the lumped parameter approach is highly sensitive to the way in which the soil landscape is parameterized while the distributed approach is less sensitive and the sensitivity became obvious only during the time of moisture stress.

Keywords

GIS, Hydro-ecological modeling, Soils, Soil Mapping, Fuzzy Logic, Soil similarity Model, Soil landscape model, SoLIM, Watershed modeling, Environmental modeling

Introduction

Soil maps produced from conventional soil surveys are currently the major source of soil spatial information for hydro-ecological modeling of mesoscale watersheds. However, standard soil surveys were not designed to provide the detailed (high-resolution) soil information required by this kind of environmental modeling (Band and Moore, 1995; Zhu, 1999a) due to the model and process used in producing conventional soil maps. Soil spatial variation portrayed in conventional soil maps is often highly generalized (Zhu, 1997) and are incompatible with other landscape data derived from detailed digital terrain analyses and remote sensing techniques (Band and Moore, 1995; Zhu, 1997; Zhu, 1999a). This incompatibility has important ramifications for hydro-ecological modeling at the watershed scale (Zhu, 1999a).

Zhu and Band (1994), Zhu et al. (1996), Zhu et al. (1997), Zhu (1997), Zhu (1999b), and Zhu (2000) have developed a soil-land inference model (SoLIM) to overcome the limitations in conventional soil surveys (Figure 1). This approach combines the knowledge of local soil scientists with geographic information systems (GIS) techniques under fuzzy logic to map soils at a finer spatial detail and higher attribute accuracy. This detailed soil spatial information is more compatible with the other landscape data derived from detailed digital terrain analyses and remote sensing techniques (Zhu, 2000).

Figure 1: Overview of SoLIM

Study area and methods

Study area

The study area is part of the Lubrecht Experimental Forest, which was established in 1937 to foster research on natural resources (Nimlos, 1986). The area is about 50 km northeast of Missoula and is in the mountainous terrain of western Montana with a moderate to strong relief. High elevations are in the east and southwest. Low elevations run from southeast through northwest. The climate is considered to be semi-arid to semi-humid. There is a strong contrast in terms of moisture conditions between north and south facing slopes and between low and high elevations. Slopes facing south at low elevations have poor moisture conditions while the moisture conditions on north-facing slopes at high elevations are better.

There are three major bedrock types in the area: Belt rocks, granite, and limestone. Each bedrock type is contiguous with Belt rocks in the north, granite materials in the south and limestone in the middle. Belt rocks are the oldest rock in the area and were deposited during the Precambrian period about one billion years ago. The sediments from which they were formed were deposited in a shallow sea, subsequently buried and then metamorphosed into quartzites, argillites and siltites (Nimlos, 1986). Soils on these materials are formed from a mantle of colluvium. Soils formed from the Belt rock materials have a finer texture than soils from the other two materials. Within each bedrock area, soils on slopes with poor moisture conditions have a shallower soil profile than soils on slopes with better moisture conditions.

The RHESSys Model

RHESSys (Regional Hydro-Ecological Simulation System) is a spatial data processing and simulation modeling environment designed to scale up the spatial extent of water and carbon processes from stand through watersheds to regional extent (Running et al., 1989; Band et al., 1991, 1993; Nemani et al., 1993; Mackay and Band, 1997). The system is specifically designed to represent the surface soil, topographic and vegetation patterns along with certain hydro-ecological processes at the landscape level so that the necessary parameters can be realistically estimated to reproduce the dominant patterns of hydro-ecological dynamics (such as runoff generation, evapotranspiration (ET), and canopy photosynthesis) over the landscape (Band et al., 1993).

The system consists of two tightly integrated components: Geographic data processing and simulation modeling (Figure 2). The geographical data processing component organizes landscape parameters into a multi-tiered hierarchy based on key driving environmental variables (Band et al., 1993; Mackay and Band, 1997). For example, in mountainous areas, the landscape can be first partitioned into hillslope units that reflect differences in incident short-wave radiation (the hillslope mean parameters file in Figure 2). Each hillslope is then partitioned into a number of elevation zones capturing within-hillslope variation in adiabatic temperature lapse rate (Lammers et al., 1997). For each of the elevation zones in a hillslope partition, wetness intervals, computed from the TOPMODEL topography-soils index (TSI) (Beven and Kirkby, 1979; Beven, 1986; Sivapalan et al., 1987), are used to separate areas showing different conditions of local hydrology. These within-hillslope variations in environmental conditions are represented in the distribution files.

The simulation model combines forest ecosystem process components adapted from FOREST-BGC (Running and Coughlan, 1988; Running and Gower, 1991), a catchment hydrological model using TOPMODEL (Beven and Kirkby, 1979), and a simple, mountain climate simulator, MT-CLIM (Running et al, 1987). Simulation over large, heterogeneous watersheds is facilitated by dividing the landscape into a series of facets or hillslope partitions based on geomorphometric principles. Hillslope partitions capture most of the variance in incident short-wave radiation (Band et al., 1991), which means the partitions can be simulated in parallel. Catenary variations of topography, soils and vegetation within each hillslope are then incorporated as distributional information within hillslope partitions. The distributional information accounts for the variability of lateral subsurface flow, ET, soil drainage, snowmelt, and canopy photosynthesis (Band et al., 1993).

Figure 2: Overview of RHESSys

This paper is concerned with the impacts of different approaches to soil landscape characterization on the simulation of two hydro-ecological processes: canopy net photosynthesis (PSN) and stream flow. The simulation of these two processes in RHESSys is detailed in many papers (Band et al., 1993; Running and Coughlan, 1988; Running and Gower, 1991; Mackay and Band, 1997) and is beyond the scope of this paper.

Two key soil properties are needed in RHESSys: Rooting zone depth and soil saturated hydraulic conductivity (K_o). In this illustration, we use depth to the bottom of B-horizon (solum depth) as a surrogate to rooting zone depth. Although rooting zone may very well extend beyond the B-horizon into C, fully describing rooting zone depth across landscape can be extremely difficult, if not impossible. Soil saturated hydraulic conductivity (K_o) is approximated by soil texture based on Clapp and Hornberger (1978).

Soil landscape parameterizations

The soil landscape over the study area was characterized using two different schemes for this study. The first scheme was to derive the spatial distribution of the needed soil properties from a conventional soil map. Each soil polygon was assigned the typical soil property value of its respective assigned soil class.

The second scheme was the SoLIM approach. Zhu et al. (1997) described an approach to derive spatially continuous soil property maps using the SoLIM model. In general, Zhu et al. (1997) used a linear and additive weighting function to estimate A-horizon depths.

The spatial distributions of solum depth derived from these two schemes are shown in Figure 3. Figure 3a shows the spatial variation of solum depth derived from the SoLIM approach and Figure 3b depicts the distribution of solum depth based on the soil map. Both images show a general pattern, which is that solum is deep at high elevation (areas labeled as A) and on north facing slopes (areas labeled as B), and is shallow at low elevation (areas labeled as C) and on south facing slopes (areas labeled as D). However, the differences between the two images are quite strong. Figure 3a shows a gradual variation of solum depth while Figure 3b depicts sharp changes at the boundaries of soil polygons. Although these sharp changes are possible at locations where major geological material changes (such as areas labeled as E), often these sharp changes are not realistic and are the artifact from the polygon-based approach in soil mapping.

(a)	(b)
Figure 3: Images of solum depth using two different approaches (light tone means deep solum. (a) Using the SoLIM approach; (b) Based on the conventional soil map.

The spatial distributions of hydraulic conductivity parameterized using the two schemes are shown in Figure 4. Figure 4a is the image of hydraulic conductivity based on the SoLIM approach and Figure 4b shows that based on the soil map. In general, the soil hydraulic conductivity is high on the granite materials (southern part) and low on the Belt (northern) and limestone materials (middle area). The conductivity also shows contrasts between north and south facing slopes, and between high and low elevation due to the level of soil profile development. It is high on south facing slopes and at low elevations where soil profile development is weak and the accumulation of fine particles in the sub soil horizon is weak. On the other hand, the accumulation of fine particles in the sub soil horizon is much stronger on north-facing slopes and at high elevations. As a result, the soil hydraulic conductivity in soils on north-facing slopes and at high elevations is often low. Both Figure 4a and 4b show this general pattern. However, Figure 4a portrays much greater detailed spatial variation of hydraulic conductivity than Figure 4b. This detailed variation depicted in Figure 4a is a realistic approximation of variation of hydraulic conductivity over the landscape in this area since this variation agrees with the variation in environmental conditions that dictate soil formation in the area.

Experiment design

The impacts of the aforementioned differences in soil landscape parameterizations on the simulation of overall hydro-ecological responses are examined in the following way. The overall responses are simulated using two different approaches: The lumped parameter approach and the distributed parameter approach (Maidment, 1993). For both approaches the watershed was first partitioned into a number of hillslopes using the hillslope partition algorithm described in Band (1989). The difference between the two approaches is how information is organized for each hillslope. For the lumped parameter approach, only the mean conditions of model parameters for each hillslope are represented, but no information on their variations within the hillslope is retained. This is accomplished under RHESSys by allowing only one big elevation band and one wetness interval within each hillslope.

(a)	(b)
Figure 4: Images of hydraulic conductivity over the study area based on the different approaches (light tone means high conductivity). (a) based on the SoLIM approach; (b) derived from the conventional soil map.

For the distributed parameter approach, each hillslope was divided into elevation zones with each zone extending 100 meters of elevation. Each elevation zone (band) was then divided into wetness interval with interval size of 2. This way of parameterization allows for the representation of the co-variation of model parameters within each hillslope.

For each modeling approach, two sets of model parameter files were derived. The two sets only differ from each other in how information on solum depth and hydraulic conductivity were obtained. For one set, the soil information was derived from a soil map of the area. For the other, the soil information was derived from the SoLIM approach using the methodology described in Section 2.3.

Results and Discussions

Lumped parameter approach

The simulated stream flows from the watershed using the lumped parameter approach are shown in Figure 5. Under the lumped parameter approach both simulated stream flows fluctuate dramatically with response to raining and snow melt events. This fluctuation can be explained by the fact that the lumped hillslope contributes runoff only when it is fully saturated. However, the stream flow based on the soil information from SoLIM fluctuates much less than that based on the soil map. This can be explained by the improved spatial co-variation of soil properties and other landscape variables. The soils in both the spatial and attribute domains are highly generalized under the conventional soil mapping approach, but are less generalized with SoLIM. As a result, the simulated hydrograph is not as “spiky” with the soil information derived using the SoLIM approach.

Figure 5: Comparison of stream flows simulated with RHESSys for the whole watershed using the lumped parameter approach.

Annual variations of net photosynthesis (PSN) simulated using the lumped parameter model are shown Figure 6. PSN is not influenced by soil parameter variability during the first five and half months of the year. However, it starts to differ towards later part of June and the difference continues throughout the summer into early September, then the simulated PSN productions become the same again. The three episodes coincide with the three periods of climatic conditions for the area. The year starts for the area with low temperature and low precipitation (mostly in the form of snow). As the year progresses, the temperature and precipitation increases. With higher temperature snow starts to melt. As a result, soil water contents are high. There is no moisture stress for the area. When the area enters early summer (later June), the temperature still increases but the amount of precipitation decreases. By that time the snow has melted, and subsequent soil water recharge is limited to infrequent rainfall events. However, during summer months there is a soil water draw down via evapotranspiration. Eventually, the area experiences moisture stress and photosynthetic activities reduces. By the end of summer temperature decreases and the amount of precipitation increases. Soil water contents increase and moisture stress dissipates. The simulated PSN differs between the two soil parameterization schemes during the period of moisture stress when stomatal closure occurs. It is understandable that during the time when recharge to soil water equals or surpasses soil water depletion the amount of available water stored in soil profile is not important. However, the ability of storing water in a soil profile (a function of soil water holding capacity and solum depth) becomes very critical when soil water depletion is faster than recharge since plants draw water from soil storage. As time goes on soil water over areas with lower soil water storage will be depleted faster and moisture stress shows up earlier and more severe. As a result, PSN production will be reduced more. Figure 6 shows that the PSN simulated based on the soil information from the soil map is much lower over the period of moisture stress than that simulated using the information from SoLIM. This can be explained by the fact that the soil map typically underestimates solum depth and overestimates hydraulic conductivity on south facing slopes. Since south facing slopes dry faster than north facing slopes due to higher radiation interception, the soil water storage in the soils over these south facing slopes are lower based on the soil map and the moisture stress shows up earlier and more severe than that based on SoLIM (Figure 6).

Figure 6: Comparison of PSN simulated with RHESSys for the whole watershed using the lumped parameter approach.

Distributed parameter approach

Stream flow and PSN simulated using the distributed approach are shown in Figures 7 and 8. The stream flow hydrographs are very different from those from the lumped parameter approach (Figure 5) due to the fact the distributed approach not only considers the mean conditions, but also incorporates spatial variation within each hillslope. Under a distributed framework surface runoff generation occurs over a partial contributing area, as opposed to all or none of the hillslope area in the lumped parameter approach. The detailed comparison and discussion on the differences between the lumped versus distributed approaches can be found in Band et al. (1993) and are out of the scope of this study. However, what we want to discuss is the difference in the impacts of different soil landscape parameterization strategies on the lumped and distributed modeling approaches. The difference between the two hydrographs under the distributed approach is small compared to that under the lumped parameter approach. This is due to the fact the distributed approach considers the spatial co-variation of local topography (elevation and slope gradient) and drainage area within a given hillslope. Thus, much of local variation of soil properties (such as solum depth and hydraulic conductivity) is expressed by this detailed description of other landscape parameters. As a result the difference in simulated stream flow between the two different soil landscape parameterization schemes is small under the distributed approach.

Figure 7: Comparison of stream flows simulated with RHESSys for the whole watershed using the distributed parameter approach.

PSN simulated under the distributed approach exhibits similar temporal pattern as discussed in 3.1. However, the difference in PSN between the two different soil landscape parameterization schemes is much smaller. In particular, the difference in PSN between the two schemes starts much later in the growing season (Figure 9). This can be explained by the fact that the representation of spatial co-variation of local topography and drainage area under the distributed approach captures the major variation of local soil conditions, and the impacts of over- or under-estimation of soil properties in the conventional soil map are much reduced under the distributed approach.

Conclusion

This study reveals that the detailed soil landscape parameterization using the SoLIM scheme has significant impacts on the simulated hydro-ecological processes with the lumped parameter approach. The simulated hydrograph was less "spiky" and more realistic in this forested watershed than that simulated based on the soil information from a conventional soil map. The detailed soil information from SoLIM does not have much impact on the simulation of PSN over the period when there is a sufficient water recharge to soil profiles. However, the simulated PSN production based on the detailed soil information is higher over the moisture stress period than that based on the soil information from the soil map.

Figure 8: Comparison of PSN simulated with RHESSys for the whole watershed using the distributed parameter approach

With the distributed parameter approach the detailed soil spatial information derived from the SoLIM scheme has much less impact on the simulation of hydro-ecological processes. The simulated hydrograph based on detailed soil information is very similar to that based on the soil information from the soil map. The difference in simulated PSN production during the period of moisture stress is smaller between the two different soil landscape parameterizations and this difference also occurred much later in the moisture stress period than it was under the lumped parameter approach. The impacts of detailed soil parameterization on spatial distribution of simulated transpiration occurred mainly over south facing slopes during the period of moisture stress.

Acknowledgements

The study reported here was funded by grants from the Graduate School, University of Wisconsin-Madison. Funding to the second author from McIntire-Stennis is also gratefully acknowledged.

References used

Band, L.E., 1989. A terrain based watershed information system. Hydrological Processes, 3, 151-162.

Band, L.E., Peterson, D.L., Running, S.W., Coughlan, J., Lammers, R., Dungan, J., and Nemani, R., 1991. Forest ecosystem processes at the watershed scale: basis for distributed simulation. Ecological Modeling, 56, 171-196.

Band, L.E., 1993. Effect of land surface representation on forest water and carbon budgets. Journal of Hydrology, 150, 749-772.

Band, L.E. and Moore, I.D., 1995. Scale: Landscape attributes and geographical information systems. Hydrological Processes, 9, 401-422.

Band, L.E., Patterson, P., Nemani, R. and Running, S.W., 1993. Forest ecosystem processes at the watershed scale: incorporating hillslope hydrology. Agricultural and Forest Meteorology, 63, 93-126.

Beven, K. 1986. Runoff production and flood frequency in catchments of order n: an alternative approach. In: V.K. Gupta (Editor), Scale Problems in Hydrology. D. Reidel Publishing Company, Hingham, Mass. 107-31.

Beven, K.J. and Kirkby, M.J., 1979. A physically based, variable contributing model of basin hydrology. Hydrological Sciences Bulletin, 24, 43-69.

Clapp. R.B. and Hornberger, G.M., 1978. Empirical equations for some soil hydraulic properties. Water Resources Research, 14, 601-604.

Lammers, R.B., Band, L.E., and Tague, C.L. 1997. Scaling behaviour of watershed processes. In: P. van Gardingen, G. Foody and P. Curran (eds.). Scaling-Up: From Cell to Landscape, Cambridge University Press, Cambridge, 295-317.

Maidment, D.R., 1993. GIS and hydrologic modeling. In: Michael F. Goodchild, Bradley O. Parks, and Louis T. Steyaert (Editors), Environmental Modeling with GIS, Oxford University Press, New York, pp. 147-167.

Mackay, D.S. and Band, L.E., 1997. Forest ecosystem processes at the watershed scale: dynamic coupling of distributed hydrology and canopy growth. Hydrological Processes, 70, 1197-1217.

Nemani, R., Pierce, L., Running, S., and Band, L.E., 1993. Forest ecosystem processes at the watershed scale: sensitivity to remotely sensed leaf area index estimates, International Journal of Remote Sensing, 14, 2519-2534.

Nimlos, J.T., 1986. Soils of Lubrecht Experimental Forest, Miscellaneous Publication No. 44, 36 pp., Montana Forest and Conservation Experiment Station, School of Forestry, University of Montana, Missoula, Montana.

Running, S.W. and Coughlan, J.C., 1988. A general model of forest ecosystem processes for regional applications: I. Hydrologic balance, canopy gas exchange and primary production processes. Ecological Modeling, 42, 125-154.

Running S.W. and Gower, S.T. 1991. FOREST-BGC, a general model of forest ecosystem processes for regional applications. II. dynamic carbon allocation and nitrogen budgets. Tree Physiology, 9, 147-60.

Running, S.W., Nemani, R.R., Hungerford, R.D., 1987. Extrapolation of synoptic meteorological data in mountainous terrain, and its use for simulating forest evapotranspiration and photosynthesis. Canadian Journal of Forest Research, 17, 472-483.

Running, S.W., Nemani, R.R., Peterson, D.L., Band, L.E., Potts, D.F., and Pierce, L.L., 1989. Mapping regional forest evapotranspiration and photosynthesis by coupling satellite data with ecosystems simulation. Ecology, 70, 1090-1101.

Sivapalan, M., Beven, K. and Wood, E.F. 1987. On hydrologic similarity 2: a scaled model of storm runoff production. Water Resources Research, 23(12), 2266-2278.

Zhu, A.X., 1997. A similarity model for representing soil spatial information, Geoderma, 77, 217-242.

Zhu, A.X., 1999a. Fuzzy inference of soil patterns: implications for watershed modeling. In: D.L. Corwin, K. Loague, and T.R. Ellsworth (Editors), Advanced Information Technologies for Assessing Nonpoint Source Pollution in the Vadose Zone, Geophysical Monograph 108, AGU Publication, pp. 135-149.

Zhu, A.X., 1999b. A personal construct-based knowledge acquisition process for natural resource mapping. International Journal of Geographical Information Science. 13, 119-141.

Zhu, A.X., 2000. Mapping soil landscape as spatial continua: the neural network approach. Water Resources Research, 36, 663-677.

Zhu, A.X. and Band, L.E.,1994. A knowledge-based approach to data integration for soil mapping. Canadian Journal of Remote Sensing, 20, 408-418.

Zhu, A.X., Band, L.E., Dutton, B., Nimlos, T.J., 1996. Automated soil inference under fuzzy logic. Ecological Modelling. 90, 123-145.

Zhu, A.X., Band, L.E., Vertessy, R., and Dutton, B., 1997. Derivation of soil properties using a soil land inference model (SoLIM). Soil Science Society of America Journal, 61, 523-533.

Authors