GIDA - CiteSeerX

GIDA_______________________ Journal of Geographic Information and Decision Analysis 2002, Vol. 6, No. 1, pp. 1-16

Integrating Multi-criteria Analysis and GIS for Land Condition Assessment: Part I – Evaluation and Restoration of Military Training Areas Guillermo A. Mendoza Department of Natural Resources and Environmental Sciences University of Illinois 1102 S. Goodwin Avenue, Urbana, Illinois 61801, USA [email protected]

Alan B. Anderson U.S. Army Engineering Research and Development Center Construction Engineering Research Laboratory Champaign, IL 61821, USA [email protected]

George Z. Gertner Department of Natural Resources and Environmental Sciences University of Illinois Urbana, Illinois, 61801, USA [email protected]

ABSTRACT This paper describes multiple criteria

models that can be used to assess land condition in general, and military training areas in particular. Three measures of land condition are used, namely: 1) erosion status, which is estimated based on the Revised Universal Soil Loss Equation (RUSLE), 2) percent vegetative cover, and 3) range condition. In addition, the paper also describes site-specific thresholds that can be used to classify land condition after training. Thresholds can be set based on individual factors, or as a composite measure based on the cumulative impacts of all the

© GIDA 2002

factors. The multi-criteria methods are integrated or linked with Geographic Information Systems (GIS) to make land condition assessment geographicallyspecific. Finally, a GIS-based land repair allocation model is developed that can be used to identify and prioritize critical areas in need of restoration. The methodologies are demonstrated in a case study involving the Fort Hood military training area located in Texas. KEYWORDS: Multi-criteria analysis, GIS, land condition, critical thresholds, land allocation

ISSN 1480-8943

Integrating Multi-criteria Analysis and GIS for Land Condition Assessment: Part I

1. Background The Department of Defense (DoD) is responsible for administering more than 25 million acres of federally owned land in the United States (Public Land Law Review Commission 1970), making it the fifth largest federal land managing agency. The Integrated Training Area Management (ITAM) Program is the Army’s program for managing training lands. A major objective of the ITAM program has been to develop a method for estimating training land carrying capacity. Training land carrying capacity is defined as the amount of training that a given parcel of land can accommodate in a sustainable manner, based on a balance of use, condition, and maintenance practices. The Army Training and Testing Area Carrying Capacity (ATTACC) methodology (Anderson, 1999; U.S Army Environmental Center, 1999) is an initiative to estimate training land carrying capacity. The ATTACC methodology is also used to determine land rehabilitation and maintenance costs associated with land-based training. The ATTACC methodology consists of three main components: training load characterization, environmental characterization, and cost analysis. The training load component characterizes training load in terms of Maneuver Impact Miles (MIM). One MIM is the equivalent impact of an M1A2 tank traveling 1 mile while participating in an armor battalion field training exercise. The environmental component characterizes land condition in terms of measures of land condition that include erosion, vegetative cover, and species composition. The cost analysis component characterizes installation land maintenance and repair practices in terms of the type of practice, costs, area affected, and effectiveness. Land repair cost factors are estimated and expressed in dollars per mile for each vehicle type. Maneuver Impact Miles are based on the number and type of vehicles, number of miles traveled, and a series of Training Impact Factors (Equation 1). Training Impact Factors include the Event Severity Factors (ESF), Vehicle Severity Factors (VSF), Vehicle Off-Road Factors (VOF), Local Condition Factors (LCF), and Vehicle Conversion Factors (VCF). The ESF is a multiplier that represents the relative impact of an event, as compared to the standard event (Armor Battalion FTX). The VSF is a multiplier that represents the relative impact of a vehicle, as compared to the standard vehicle (M1A2 tank). The VOF is a multiplier that represents the percentage of vehicle miles typically driven off improved roads. The VCF is a multiplier that represents the area impacted by a vehicle, as compared to the area impacted by the standard vehicle. The LCF is a multiplier that represents the relative impact of vehicle traffic due to different site conditions including soil moisture. e    v  MIM = ∑  ∑ ( NumberV * MileageV * VSFV * VOFV * VCFV ) * DurationE * ESFE * LCFE   E =1  V =1  where: MIM = normalized training load (maneuver impact miles) E = event (dimensionless) e = number of events (dimensionless) V= vehicle type (dimensionless) v = number of types of vehicles in event E (dimensionless) Mileage = daily mileage for vehicle type V for event type E (miles) Number = number of vehicles of type V (dimensionless) VSF = vehicle severity factor for vehicle type V (dimensionless) VOF = vehicle off-road factor for vehicle type V (dimensionless) VCF = vehicle conversion factor for vehicle type V (dimensionless) LCF = local condition factor for event E (dimensionless) Duration = number of days for event type V (days) ESF = event severity factor for event type V (dimensionless)

(1)

2

Mendoza, Anderson and Gertner

Based on the model in [1], the impacts of different levels or intensities of training were estimated using sampling plots established around the training sites. Details of the estimation of MIM’s are described in U.S Army Environmental Center (1999). In early versions of the ATTACC model, land condition was expressed only in terms of erosion status (ES). ES was estimated using a modification of the Revised Universal Soil Loss Equation (RUSLE) (Renard et al 1997, Concepts Analysis Agency 1996). Specifically, erosion rates were estimated using the following equation: A = R*K*LS*C*P (2) where: A = soil loss per unit area (tons ac-1 yr-1) R = rainfall and runoff factor ([hundreds of ft-tons] inch ac-1 hr-1yr-1) K = soil erodibility factor (tons hr [hundreds of ft-tons] -1 in-1) LS = slope length and steepness factor (dimensionless) C = cover and management factor (dimensionless) P = support practice factor (dimensionless). The R factor is the rainfall and runoff factor for a specific location. R factor incorporates the amount, intensity, and duration of precipitation patterns. The soil erodibility factor (K) is the rate of soil loss per rainfall erosion index unit under standard conditions. The slope length factor (LS) provides a quantitative representation of both slope length and steepness. The conservation (P) factor is a quantitative expression of the mitigating effect that conservation practices have on the erosion process. The C factor reflects the degree of erosion protection provided by vegetation. It describes the density and structure of the vegetative cover and the kind and amount of cover in contact with the soil. Based on the expected soil loss estimated by RUSLE above, erosion status is calculated as follows: ES = A/T (3) where: ES = Erosion Status (dimensionless) A = soil loss per unit area (tons ac-1 yr-1) T = soil loss tolerance factor (tons ac-1 yr-1). The soil tolerance (T) factor indicates the maximum level of soil erosion that will permit a high level of soil productivity to be sustained economically and is based on soil depth, rooting depth, soil organic matter reduction, and plant nutrient losses. Military training removes vegetation and exposes the soil surface to raindrop impact and surface runoff, resulting in an overall degradation of land condition (Ayers 1994, Ayers et al 1990, Braunack 1986, Burger et al 1985, Prose 1985, Shaw and Diersing 1989, Wilson 1988). The more the land is impacted, the higher the vegetative cover loss and the greater the degradation of land condition. Training impacts and natural recovery are incorporated in the land condition analyses through the C factor as shown in equation 4 (Concepts Analysis Agency 1996). C Factorp = C Factorc - ∆C Factor t + ∆C factorr (4) Where C Factorp = predicted C Facto C Factorc = current C Factor ∆C Factort = change in C Factor due to training ∆C factorr = change C Factor due to natural recovery. The ∆C Factort quantifies the change in C Factor associated with the impact of a single pass of the standard vehicle. The ∆C factor r is estimated based on the amount of time required for a sufficient amount of cover to return the soil erosion rates on a disturbed site to predisturbance rates. Land condition evaluated using equation 4 for a specified training load is compared to a management objective referred to as a threshold value. Thresholds define the minimum allowable land condition that meets management objectives. A commonly used threshold value for ES is one. A value of one implies that sufficient vegetative cover exists to ma intain soil resources.

3


Conceptually equation 4 can be used for any land condition measure if sufficient data is available to quantify current condition, impacts, and recovery rates. In fact, subsequent versions of ATTACC have included wind erosion and vehicular dust emissions. These additional land condition measures are in the same measurement units as the original land condition measure (ES). However there are other meaningful measures of land condition that have measurement units that are not easily related to ES. Currently there is a need to incorporate into the ATTACC methodology land condition measures that are inherently distinct from erosion status.

2. Multicriteria land condition analysis The methodology described above assesses land condition based only on one factor, namely, the erosion status. While this may be adequate for some military installations, it is more meaningful to expand the definition of land condition beyond erosion status and include other factors. In this study, the US Army Environmental Center management group wanted to include two other factors as indicators of land condition, namely, percent vegetative cover (PC) and range condition (RC). With these additional factors, evaluating land condition requires a multicriteria analysis procedure. Since land condition is now determined by a number of factors, it is best to describe it as an index expressed in terms of the three factors as follows: LCI = Σ w j LCj (5) where LCI = Land Condition Index LCj = Land Condition based on factor j w j = relative weight of the factor j such that the sum of all j equals 1 The linear additive model in [5] reflects a composite land condition involving multiple factors. Individual factors are weighted to reflect their relative importance with respect to the overall land condition. Hence, factors that are deemed more significant indicators of land condition for a given location can be assigned higher weights thereby giving them greater importance in the estimation of the Land Condition Index. The additive model also allows estimation of the impact of each factor separately. At the same time, the model offers a simple way to estimate a composite land condition index that can be viewed as the collective impacts of all the different factors. The relative weights, wj, are central to the multicriteria method described in [5]. If the factors are sufficiently known and supported with adequate data, and the interactions among the factors are also sufficiently understood, then the weights may be estimated using sophisticated methodologies that take into account the dynamic processes that link erosion status with vegetative cover and range condition analysis. For most military installations, this level of information is not readily available. However, there are procedures that allow estimation of these weights using qualitative and quantitative measures that do not rely heavily on ‘hard’ data, but can use mixed data sources, including expert opinions (Mendoza, 2000). Over the last few years, the integration of GIS and multi-criteria analysis has attracted significant interest as evidence by the increasing number of publications reported in the literature (Carver 1991; Jankowski 1995; Malczewski 1999; Eastman et al 1995; Janssen and Rietveld 1990). This integration has also led to the development of spatial decision support systems that take advantage of the analytical capabilities of multi-criteria analysis on one hand, and the information processing and display capabilities of GIS (Densham, 1991, Jankowski, et al 1997).

3. Scaling and standardizing factors Often the factors affecting land condition are of different scales and their values are of different magnitudes. For example, vegetative cover and range condition are expressed in terms of percent; hence, their values range from 0 to 100. Erosion status, on the other hand is a ratio, and hence dimensionless

4


with values greater than 0. To make these factors comparable, and hence additive for the index described in [5], they have to be scaled. Voogd (1983) describes a variety of techniques for scaling which typically use the maximum and minimum values as scaling points or limits as follows: Xi = (Ri – Rmin)/(Rmax – Rmin) (6) where Xi = scaled value of the factor i Ri = raw value of factor i Rmin = minimum raw value of factor i Rmax = maximum raw value of factor i The advantage of [6] is that the scaled values of the factors regardless of their original magnitude will be precisely between 0 and 1; the minimum (worst) is equal to 0 and the maximum (best) is always equal to 1. The disadvantage of the simple scaling model in [6] is that the scaling limits, Rmax and Rmin, are ‘raw values’ which have little meaning as far as ‘ideal’ or ‘targets’ are concerned. In so doing, some valuable information may be lost in the scaling process. For example, consider a factor with a maximum raw value of 150. However, this value may just ha ppen to be an ‘outlier’; e.g., 99 percent of the values are between 0 to 130. Further, assume that the maximum target value (or ‘ideal’ value) of the factor is 130. If the factor is scaled using [6], then the scaling limit is automatically set at 150. The scaled values between 0 and 1 have no real meaning with respect to how far they are to the ideal, and the maximum value itself has no useful meaning. Hence, the scaling procedure should also be a tool for standardizing the factors so that: 1) the values are comparable (say between 0 to 1), 2) the scaling limits are also comparable, and 3) the scaled values have real meanings with respect to how far they are to the ‘target’ or ideal values which also must serve as the scaling limit. In view of the above, the proposed scaling procedure should be as follows: Xi = (Ri – Rmin )*(Xmax – Xmin) /(R max – Rmin ) + Xmin (7) Scaling has minimal effect on factors like vegetative cover and range condition factors because 1) their values are already between 0 to 1 (0 to 100 percent), 2) their maximum and minimum values serve as the scaling limits (i.e. minimum is 0 percent for the worst condition, and the maximum is 100 per cent for the best or ‘ideal’ value), and 3) the raw values themselves are already relative values with respect to the scaling limits of 0 and 100. Scaling is essential for factors like erosion status factor because its values are not directly comparable to factors like vegetative cover and range condition. In scaling the ES factor, two things need to be established: the scaling limits to define the range of scaled values, and the minimum (worse) and maximum (better) values for the factor. The scaling limits must also be comparable with the other land condition factors, namely the vegetative cover and range condition factors. Both factors have 0 and 100 as their scaling limits; hence, the ES factor should also adopt these as scaling limits. The ideal value is obvious; that is 0 for no erosion at all. However, the worst is not easy to determine because it is the maximum amount of erosion and is dependent on the amount and intensity of training. Examining the data, a reasonable range of ES values could be determined and the worst-case scenario can be esta blished. It is clear from [3] that an ES value of 1 implies that the amount of erosion (i.e. A in the RUSLE equation shown in [2]) is equal to the tolerance limit T. It should also be pointed out that for the two other factors, land condition is negative ly correlated with the amount of training load. That is, percent vegetative cover and range condition decreases as the amount of training load (i.e. the magnitude of MIM’s) increases. Moreover, it is clear that higher values of the percent cover and range condition factors are preferable. The opposite is true for erosion status; ES increases as training load increases. Hence, the scaling process must also transform the scaled values so that higher values are preferable because they imply less negative impact. Finally, in scaling the ES factor, the original values must be reclassified so that all ES values greater than 2 will be classified as having values of 2. Then, the second stage scaling process will transform the re-classed data so that 2 is standardized to the scaling limit of 100, and finally, the other

5


values are transformed using the model in [7] where Xmax , Xmin, Rmax, and Rmin are set at 100,0, 2, and 0 respectively.

4. Land condition thresholds Land-based ecosystems are generally flexible and capable of absorbing stresses caused by various forms of disturbance including damages from military training exercises. Most terrestrial ecosystems are able to maintain their functions despite heavy loads of disturbance as long as the intensity of disturbance is within an acceptable range denoted by the threshold values. Besides defining acceptable levels of disturbance, threshold values may also serve as guides in identifying areas in need of restoration. Hence, thresholds are essentially reference values that define acceptable levels of disturbance both from the standpoint of restoration as well as the expected costs of land maintenance and improvement. In the context of land condition analysis examined in this study, thresholds have meaning at the individual factor level, and at the composite land condition index level. At the individual factor level, say erosion status, thresholds are useful guides in determining the maximum amount of training load an area can support without compromising its ability to recover from training damages to the soil. Composite thresholds may also be specified for the purpose of establishing limits where all the factors are considered critical. These limits may be useful in determining and prioritizing critical areas that are in need of restoration. Moreover, these thresholds may also serve as the bases for estimating land maintenance and improvement costs. It is conceivable that in some areas, land condition is already at the threshold based on one factor; however, this may not necessarily be the case for other factors. In such a case, the composite threshold helps identify these areas. Resources available for restoration can then be allocated to those areas whose land condition index is closest to the composite threshold value.

5. GIS and spatial analysis Geographic Information Systems (GIS) have become an essential tool for spatial analysis. Historically, GIS functionalities have been limited to a set of tools for the input, storage and retrieval, manipulation, and output of spatial data. Lately, these tools have been expanded beyond map-making functions into the domains of problem solving and especially modeling and decision-making (Eastman, 1995). These enhanced capabilities enable users to take advantage of the display capabilities of GIS and the analytical power of models. Advanced GIS capabilities including map or image computations, map algebra and processing of logical expressions, and image processing make it possible to couple analytical models with a GIS. This combination or coupling of functionalities offers a powerful tool for analysis at the conceptual level and at the site or spatially explicit level. This study combines GIS with multicriteria modeling techniques to extend the ATTACC methodology to include multiple measures of land condition and demonstrates the approach using installation data. GIS is used to generate factor maps that are considered important for assessing land condition. Multicriteria analysis is used in generating the weights of the factors and in the development of the composite land condition as described in [5]. Derivative maps generated from the combined use of GIS and multicriteria analysis can also be done to conduct more indepth analysis of the impacts of training on land condition.

6. Site description Figure 1. Case study location.

The study site consists of Fort Hood property that occupies approximately 87,890-hectare (ha) of land (U.S. Department of

6


the Army 1987) in central Texas in Bell and Coryell Counties (Figure 1). Fort Hood’s climate is characterized by long, hot summers and short, mild winters. Average temperatures range from a low of about 8 °C in January to a high of 29 °C in July. Average annual precipitation is 81 cm. Elevation at Fort Hood ranges from 180 to 375 m above sea level with 90 percent below 260 m. Most slopes are in the 2 to 5 percent range with slopes in excess of 45 percent occurring as bluffs along the flood plain and as the sides of slopes on the hills. Soil cover is generally shallow to moderately deep and clayey, under-lain by limestone bedrock. Fort Hood lies in the Cross Timbers and Prairies vegetation area (Gould 1975). The area is normally composed of oak woodlands with grass undergrowth. The primary mission of Fort Hood is the training, housing, and support of the III Corps and its two divisions (1st Calvary Division and 2nd Armored Division). Support is also provided to other assigned and tenant organizations, as well as the U.S. Army Reserve, the National Guard, the Reserve Officer Training Corps, and the reservists from other services. Of the 22,700 ha live-fire and impact areas, 8,700 ha are multi-purpose maneuver live-fire areas. The range areas serve as familiarization and qualification firing ranges for all individual weapons, crew-served weapons, and the major weapons systems of active units assigned or attached to the III Corps and Fort Hood. Maneuver areas comprise 52,400 ha (not including the multi-purpose live-fire area). Maneuver areas are used for armored and mechanized infantry forces in the conduct of task force and battalion-level operations, and for company and platoon-level dismounted training, along with engineer, amphibious, combat support, and combat services support training.

7. Data set descriptions A modification of the RUSLE was implemented following the work of Renard et al (1997) with modifications as described in Concepts Analysis Agency (1996). R factor values were obtained from published isoerodent maps (Renard et al, 1997). K and T factor values for each soil series that occurred at study site were obtained from published soil surveys of Coryell and Bell counties (McCaleb 1985, Huckabee et al 1977). Using these values, a digitized soil series map of the study site was reclassified by assigning the K and T values to the respective soil types. Slope length factors (LS) were estimated using field inventory data (Diersing et al 1992) and methods described in Renard et al (1997). Slope and slope length measurements were measured in the field for 320 points. LS factor values were estimated for each point using equations from Renard et al (1997). The soil series data layer was reclassified using the average LS factor value for each soil series. The conservation (P) factor was set to a default value of one (Warren et al 1989). The C factor was estimated using field inventory data and nomographs described in Wischmeier and smith (1978). C factor values were estimated for each field plot. An unsupervised classification of a LANDSAT multispectral scanner (MSS) image of the study site was used to identify distinct land-cover categories. Each land-cover category was assigned a C Factor value defined as the average C Factor value of all plots occurring in the land-cover category (Warren and Bagley 1992). Impact and recovery rates (∆C Factort and ∆C factorr) were estimated using data from Thurow et al (1993). Vegetative Cover was estimated using field inventory data (Diersing et al 1992). An unsupervised classification of a LANDSAT multispectral scanner (MSS) image of the study site was used to identify distinct land-cover categories. Each land-cover category was assigned a vegetative cover factor value defined as the average percent vegetative cover of all plots occurring in the land-cover category (Warren and Bagley 1992). Impact and recovery rates were estimated using data from Thurow et al (1993). Range Condition was estimated using field inventory data (Diersing et al, 1992). Range condition values were estimated for each inventory plot using methods described in Stoddart et al (1975). A range condition GIS data layer was reclassified using the average range condition of all plots occurring in a range site. Impact and recovery rates were estimated using data from Thurow et al (1993).

7


8. Result and discussion Table 1 contains a summary of the average values of the different factors affecting land condition. Note that the values under RUSLE reflect the expected amount of erosion (tons/acre/year) while ES is the erosion status which is dimensionless. It should also be noted that these values take into account the intensity of training (expressed in terms of MIM’s), the recovery process, and the amount of time necessary to restore the area after training. As shown in [4], the equation incorporates changes due to training impacts as well as the changes due to the recovery process. Hence, the temporal dimensions as well as the biophysical aspects of the recovery process are imbedded into the model described in [4]. The values contained in Table 1 are average values obtained for the entire Fort Hood Installation. The expected amount of erosion as calculated by RUSLE range from a low of 0.519 tons per acre per year for 100,000 MIM’s and progressively increases as the MIM’s increases to 3.643 tons per acre per year for a training load of 1,200,000 MIM’s. Percent cover, on the other hand, has a high of 92.94 % for a training load of 100,000 MIM’s and progressively got lower up to only 49.39 % for a training load of 1,200,000 MIM’s. These two examples confirm what we expect: 1) amount of erosion increases as MIM’s increases, and vegetative cover decreases as training load increases. At the landscape or Table 1. Average impact of training load on land condition factors. installation level, one could use the average values of the Training Load Erosion Vegetative Range 50 x 50 grid cells as shown in Status Cover Condition Table 1 as a basis for ('000 MIMs) (%) developing a predictive 100 0.386 92.94 31.6 equation of land condition as 200 0.506 87.54 22.91 determined by the individual 300 0.658 81.59 17.04 factors. Table 2 shows the 400 0.841 75.24 13.86 regression-based models of 500 1.052 68.95 11.45 these predictive models. 600 1.284 64.03 9.42 The predictive 700 1.531 60.53 7.98 models described in Table 2 800 1.791 57.53 6.83 can be used to predict the 900 2.054 55.02 6.03 expected impact of training 1000 2.316 52.79 5.32 load at the landscape level. 1100 2.582 50.9 4.71 For example, consider again 1200 2.852 49.39 4.2 the model described in [4], that is ES = A/T. Hence, if Table 1. Regression analysis of land condition predictive models. ES is equal to 1, the amount Dependent Variable Intercept Coefficient R-square of erosion is exactly equal to RUSLE 0.0178 0.002921 99.1 the amount tolerable for that ES -0.011 0.002306 99 particular area. One may Percent Cover 92.49 -0.0402 95 view this as a meaningful Range Condition 25.45 -0.0214 81 threshold; that is, find the maximum training load that can be supported by, or applied to, an area such that the expected erosion status does not exceed 1. Using the ES predictive equation in Table 2, an ES value of 1 yields a corresponding training load of 438,000 MIM’s. Hence, this training load can serve as a guide to determine the maximum acceptable amount of training that should be applied to the area. Likewise, if the military sets aside a threshold value of 60% for percent cover, then the maximum acceptable training using the predictive equation shown in Table 2 is about 808,000 MIMs.

8


9. Critical thresholds: Individual factor and composite thresholds The previous section illustrated a simple case of a threshold applied at the landscape level (i.e., the entire installation). At the macro level, this type of analysis may be sufficient to guide training in terms of estimating the maximum training load the entire installation can support. Note that the analysis above is based on average values of grid cells whose actual values may exhibit wide range of variability. A more focused analysis of training load and the extent of the impacts can provide more detailed insights for planning and scheduling training activities within the installation. Table 3 shows a summary of results based on some subjectively set thresholds for each of the factors. Table 3: Impacts of Training Loads at different Threshold Values Training Loads (‘000 MIMs)

Erosion Status 40

50

Percent Cover 60

100 Area 1 72,172 70,497 70,023 Percent2 0.95 0.90 0.92 600 Area 60,510 58,519 56,889 Percent 0.80 0.77 0.75 1,200 Area 50,070 46,638 44,942 Percent 0.66 0.61 0.59 1 Hectares of land in acceptable condition. 2 Percent of area in acceptable condition.

Range Condition

40

50

60

30

40

50

74,650 0.98 54,383 0.72 42,704 0.56

74,638 0.98 50,717 0.67 38,638 0.51

74,585 0.98 45,893 0.60 32,947 0.43

41,857 0.55 7,544 0.10 1,937 0.03

8,799 0.12 1,823 0.02 713 0.01

6,640 0.09 1,195 0.02 324 0.00

In Table 3, a summary of sample impact analysis is shown for three levels of training intensity, namely, low with 100,000MIMs, medium with 600,000MIMs and high with 1,200,000 MIMs. Note that the ES value is scaled (so that the values are between the scaling limits 0 to 2 tons per acre per year), then standardized (so that its value is between 0 and 100 just like percent cover and range condition), Finally, it is transformed (so that ‘more’ is desirable; i.e. make it compatible and additive with the other two factors). Table 3 contains the estimated impacts of different training loads expressed in terms of the total area classified as ‘acceptable condition’ for different threshold values. For example, consider the percent cover factor and 600,000 MIMs. Suppose the threshold value is set at 50% vegetative cover. This means that an area is considered to have an ‘acceptable land condition’ if its percent cover is at least 50%. If a training load of 600,000 MIMs is applied to the entire installation (has about 76,000 actual training area), approximately 50,717 hectares or 67% of total area is predicted to be in acceptable land condition. Similarly, consider a threshold value of 50 for the transformed ES factor. This corresponds to an actual ES value of 1 which, as pointed out earlier, implies that the expected amount of erosion (A in [2] or the actual RUSLE value) is equal to the tolerable limit specified for the area. If a training load of 1,200,000 MIMs is applied to the entire installation, approximately only 46,638 hectares or 61% of total is predicted to have ‘acceptable land condition’, or approximately 30,000 hectares will be degraded. Examining the contents of Table 3 more closely, two general observations can be noted. First, land condition gets worse as training load increases as shown by the diminishing areas classified under ‘acceptable condition’ as MIMs increases. Second, the total area classified as having ‘acceptable condition’ also decreases as the threshold value increases; that is, the requirement for acceptable condition increases. Third, in terms of the ES and percent cover factors, a training load of 100,000 MIMs resulted in marginal negative impact as shown by the negligible differences in acceptable or nonacceptable land condition for different threshold values. Range condition, on the other hand, shows noticeable negative impact even at a low training load such as 100,000 MIMs. In fact, setting a threshold

9


value of at least 40% would almost result in the installation classified as having unacceptable or bad range condition. From the discussions above, a number of issues are noteworthy from the standpoint of planning and scheduling training activities. First, the threshold values are pivotal in terms of the projected areas classified as having acceptable land condition. Second, different factors reflect the impact of training load in varying degrees; hence, the impact of training load may be minimal for some factors at certain levels on one hand while the same load would impact other factors adversely. This creates a challenging problem in scheduling or allocating areas for training. The problem is multi-faceted for two reasons. First, the maximum training load (i.e., thresholds) that can be supported by the installation varies for different factors. Second, it is likely that not all factors are of equal importance so that a certain degree of land degradation with respect to a less significant factor may be tolerable if it meant improvement in other more important factors (i.e. trade-off among factors is possible). Third, after satisfying the threshold of the most important factor, it may still be possible to increase training load by balancing the negative effects of the other remaining factors. The problems described above point to the need for a composite land condition index which can serve as a tool to balance the impacts of the different factors, and to harmonize the different threshold values of the different factors. The land condition index described in [6] was used in generating the data contained in Table 4. Note that the LCI values, including the LCIthreshold were estimated using the relative weights of each factor as shown in Table 4. For example, the LCI threshold value of 48 for MIMs 100,000 for the factor thresholds, .50, .50, and .30 for ES, percent cover and range condition, respectively, was calculated using [6] as follows: .48 = .70*.50 + .20*.50 + .3 * .10. Table 4: Impacts of Training Loads at different Threshold Values Training Loads (‘000 MIMs)

Erosion Status

Percent Cover

Range Condition

Land Condition Index1

100

40

40

20

38

0.96

72,920

50

50

30

48

0.95

72,014

60

60

40

58

0.92

69,772

40

40

20

38

0.79

60,171

50

50

30

48

0.76

58,104

60

60

40

58

0.70

53,235

40

40

20

38

0.63

47,898

50

50

30

48

0.60

45,339

60

60

40

58

0.55

42,002

600

1,200

Acceptable Land condition Percent Area (Has)

1

LCI calculated with the weights of 0.70, 0.20, and 0.10 for erosion status, percent cover, and range condition, respectively.

The contents of Table 4, much like Table 3, also reflect the extent of damage associated with training loads and threshold values for the different factors. The extent of damage is characterized by the total acreage of land considered acceptable or in good condition as a result of the training load and the 10


limits defined by the threshold. Clearly, the total acreage is represented by those areas or grid cells whose LCI values are greater than the LCIthreshold. As can be expected, the higher the threshold (i.e. the higher the standard), the lower the total acreage for areas with acceptable or good land condition. Conversely, the lower the standard, the higher the total acreage of land considered to be of good condition becomes. This is consistent for all three levels of training loads. Note again that at low training loads, say 100,000 MIMs, most of the training areas (about 90%) are expected to be in good condition. On the other hand, at 1,200,000 MIMs, approximately 60% of the training area are expected to be in good or acceptable land condition. The importance of weighting can be illustrated by examining the results based on the erosion status and range condition factors. Note from Table 4 that the weights of .70 and .10 are assigned to the erosion status factor and range condition factor, respectively. Consider the results contained in Table 3, particularly under range condition and 1,200,000 MIMs. It is clear from these results that at this level of training load, the entire installation is expected to exhibit unacceptable or bad land condition at thresholds ranging from 30% to 50%. However, the results shown in Table 4 indicate that based on the composite land condition index, even at 1,200,000 MIMs, only about 60% of the installation will exhibit bad land condition. The reason for this is quite transparent; the more optimistic result is because of erosion status. Note that the erosion status factor is not as restrictive as range condition as shown in Table 3 because even at 1,200,000 MIMs, a significant area is expected to have acceptable land condition (i.e. 50,070, 46,638 and 44,932 hectares for threshold values of 40%, 50% and 60%, respectively). Therefore, because the erosion status has a higher weight, the composite land condition index yielded a more optimistic land condition scenario consistent with the less restrictive erosion factor.

10. Land repair allocation The discussions above offer some useful insights with respect to the impacts of training loads in terms of total acreage and their projected land condition. These impacts are projected at different levels of training loads and three threshold values of the three factors. Table 3 summarizes the impacts given the threshold values of each individual factor while Table 4 gives the impacts at the composite land condition index level. These results lead to two critical questions that must be addressed in planning and scheduling military training activities. The first can be stated as: how much training load can be applied to a training section in particular, or an installation in general? Tables 3 and 4 offer some meaningful insights that can be used to address this first question. The second but closely related question is: given a planned amount of training load, how can land repair be allocated and scheduled so that the most critical areas (i.e. those that have the lowest land condition after training) are restored first? Answering the second question requires addressing a number of issues such as environmental costs, costs of restoration, maintenance, or land improvement. One approach is to use the land condition index as a basis for identifying and prioritizing areas where land repair practices can be applied. Conceptually, the land repair allocation problem can be stated as follows: Minimize LCI = Σ wj LCj (8) The LCI value in [8] can be estimated for all grid cells within the installation so that each grid cell will have an associated LCI value. Hence, model [8] can be implemented by ‘ranking’ the cells based on their LCI value. Land repair activities can then be made on those areas with lowest LCI values progressively; that is, start with grid cells with low LCI values and progressively consider higher ranked cells (i.e. with lower LCI values) based on the need for more land repair. For example, suppose a training load of 1,200,000 MIMs must be applied on the installation. Based on the information in Table 4, approximately 45,000 hectares can be expected to have good land condition based on the three threshold values of the three factors (i.e. 47,898, 45,339, and 42,002 hectares in Table 4). Hence, approximately 31,000 hectares of the 76,000-hectare installation is expected to be in poor or unacceptable land condition

11


after scheduling 1,200,000 MIMs. The land repair allocation problem then is to identify the Figure 2: Training Regions first 31,000 hectares from the entire installation that have the lowest LCI values. This can be done using the ‘ranked’ grid cell values. For instance, Eastman et al (1995) describes a GIS functionality that can easily do this cell ranking. From a tactical and administrative point of view, it may be more convenient to schedule training within contiguous units; for example, within the three training regions, namely: east range, west range or live fire training regions (See Figure 2). Hence, the 31,000 hectares of unacceptable lands should be identified within these three regions. The 31,000 hectares is roughly equivalent to 124,000 grid cells since the map resolution is 50 x 50 meters (1 grid cell is about 0.25 hectares). Hence, implementing model [8] simply involves ranking the grid cells within the three regions according to their LCI values, then identifying the first 124,000 cells with the lowest LCI values. Figure 3 shows the location of the training areas that have the lowest LCI values given the restriction that the 31,000 hectares must be within the three training regions.

Another strategy may be to schedule land repair based on cost or available resources which may limit the total area that can be restored or repaired. For example, consider a land repair allocation where only 20,000 hectares can be restored because of costs and other resource requirements. The problem then is to identify the high priority areas (i.e. the first 20,000 hectares that have the lowest LCI values) that can be restored. Following the procedure described in [8], Figure 4 shows the location of the 20,000 hectares within the three regions scheduled for repair. Similarly, Figure 5 shows the land repair allocation if only 10,000 hectares can be scheduled for repair.

12


The model described in [8] is one approach to land repair allocation based on the weighted LCI values. It may also be instructive to schedule land repair according to the land condition as determined by one factor instead of the composite land condition index value. One approach is to prioritize the factors according to their ‘order’ (instead of magnitude) of importance. In this approach, no trade-off is allowed between factors. In other words, thresholds set aside for the most important factor must be satisfied at the exclusion of other factors which are considered to be of lower priority. For example, assume that the erosion status is the primary factor. In this case, the land repair allocation must satisfy land condition

13


based on thresholds set for erosion status. Allocations based on other factors are considered only after the primary factor threshold is met. This approach can be described by the model below: Minimize LCI = Σ w j LCj (9) Subject to: Erosion Status > Erosion Statusthreshold (10) The model in [9]-[10] can be extended if other factors have thresholds that must also be satisfied. For example, percent cover may be considered a secondary factor that also has thresholds that must be satisfied. In this case, the model can be described as: Minimize LCI = Σ w j LCj (11) Subject to: Erosion Status > Erosion Statusthreshold (12) Percent cover > Percent coverthreshold (13) The constrained optimization model in [11] to [13] can be generalized following the work of Eastman et al (1995) as follows: Minimize LCI = Σ w j LCj * ∏c i (14) where ci are constraints such as [12] and [13] whose values are 0 or 1. A value zero is assigned if the corresponding constraint is not satisfied, and a value 1 is assigned if the constraint is satisfied. The optimization in [14] can be implemented by simply ‘ranking’ the grid values using Eastman (1995) as described earlier. The only difference between [8] and [14] is the addition of the constraints. However, by formulating the two constraints in product form, and since the constraints are either 0 or 1, only ‘active’ constraints (all constraints with values of 1) influence the optimization or ranking of grid values. That is, only cells or grids that simultaneously satisfy all the constraints will be considered in the optimization or ranking. All other cells will have zero values and will not be considered in the ranking.

11. Summary and conclusions Land condition analysis is a major concern of the US Army consistent with its strong commitment to land stewardship of its military training installations. In order to insure that these training areas are properly maintained and capable of supporting future military training exercises, the Army is adopting the ATTACC methodology. The ATTACC methodology determines land rehabilitation and maintenance costs associated with land-based training. The methodology also helps the military match projected training loads with the capacity of the lands to support these missions. The models described in this paper are designed to enhance the capabilities of ATTACC to analyze and assess land condition as impacted by the intensity of military training exercises. Land condition analysis is conducted through an integrated use of multicriteria analysis and GIS. This integrated framework takes advantage of the analytical strengths of multicriteria methods and at the same time provide a spatially-explicit planning environment to examine the input to, and output from, the analysis using GIS. The multicriteria models described in this paper are designed to evaluate land condition as affected by training load expressed in terms of MIMs. Three factors considered in the models are: eros ion status, percent vegetative cover, and range condition. The models developed offer a number of analytical capabilities. First, individual factor models are developed that can predict land condition in terms of these factors as affected by the extent or amount of training load. Second, a composite land condition index model is developed to evaluate land condition as a composite measure reflecting the cumulative impacts of the three factors. The composite land condition index is modeled using a simple linear model where the individual factors are assigned relative weights to reflect their respective importance values. The composite land index is obtained by adding the weighted individual factor impacts. Multicriteria models often involve factors that are not directly comparable because of differences in magnitude. This paper also describes a scaling method that not only transforms the factors into

14


comparable units but also standardized them so that appropriate scaling limits are specified in a manner that is more meaningful for land condition analysis. The scaling and standardization model allows the limits to serve as ideal (or target) and worst values of the factors. Hence, the scaling limits also serve as reference points. This standardization is desirable particularly in the context of the additive and linear land condition index model. In addition to the multicriteria evaluation models, the paper also described a threshold analysis procedure that allows critical limits to be specified for individual factors. Moreover, composite thresholds involving all factors can also be specified. These limits serve as reference values for categorizing land condition into acceptable or non-acceptable conditions. This classification analysis may be useful for the purpose of assigning areas for land repair practices, and to support scheduling or assigning areas for training. The models are capable of estimating the impacts of training based on individual factors, or at the aggregate level. Impacts are measured in terms of the total acreage projected to have acceptable or non-acceptable land conditions. Finally, land repair allocation models are also developed for the purpose of identifying and prioritizing critical areas for restoration based on their land condition index values. These models are integrated with GIS to determine the location of these priority areas.

References Anderson, A. (1999) Sensitivity analysis of the Army Training and Testing Area Carrying Capacity (ATTACC) model to user-sepcified starting parameters. Corps of Engineering Research Laboratory (CERL). Technical Report 99/64. Ayers, P. (1994) Environmental Damage From Tracked Vehicle Operation. J. Terramechanics, 31(3):173-183. Ayers, P., R. Shaw, V. Diersing, and J. Van Riper. (1990) Soil compaction from military vehicles. ASAE Paper No. 90-1096. ASAE, St. Joseph, MI 49085. Braunack, M. (1986) Changes in physical properties of two dry soils during tracked vehicle passage. J. Terramechanics, 23(3):141151. Burger, J., J. Perumpral, R. Kreh, J. Tolbert, and S. Minaei. (1985) Impact of Tracked and Rubber-Tired Tractors on a Forest Soil. Trans. ASAE, pp 369-373. Concepts Analysis Agency. (1996) Evaluation of Land Value Study (ELVS). US Army Concepts Analysis Agency, Study Report CAA-SR-96-5, Bethesda, Maryland. Department of the Army. (1987) Department of the Army Pamphlet (DA PAM) 350-4, Facilities Engineering and Housing Annual Summary of Operations, Fiscal Year 1987: Volume III – Installations Performance. Office of the Assistant Chief of Engineers, Washington, DC. Desham, P.J. (1991) Spatial decision support systems. In: Maguire, D.J, M.F. Goodchild and D.W Rhind (eds). Geographical information systems: Principles and Applications. Vol 1. Essex, England. 403-412. Diersing, V.E., R.B. Shaw, and D.J. Tazik. (1992) US Army Land Condition-Trend Analysis (LCTA) Program. Environmental Management 16:405-414. Eastman, R.J. P.A. Kyem, J. Toledano and J. Weigin. (1995) GIS in Decision Making. Explorations in Geographic Information Systems Technology. Vol 4. UNITAR. Geneva, Switzerland. Gould, F.W. (1975) Texas Plants—A Checklist and Ecological Summary. MP-585/Rev. Texas A&M University, Texas Agricultural Experiment Station, College Station, TX. Huckabee, J., D. Thompson, J. Wyrick, and E. Pavlat. (1977) Soil Survey of Bell County, Texas. US Department of Agriculture, Soil Conservation Service. Government Printing Office, Washington DC. Jankowski, P. (1995). Integrating geographical information systems and multiple criteria decision making methods. International Journal of Geographical Information Systems. 9: 251-273.

15


Jankowski, P., T.L. Nyerges, A. Smith, T.J. Moore, and E. Horvath. (1997) Spatial group choice. a SDSS tool for collaborative spatial decision making. International Journal of Geographical Information Systems. 576 - 602. Janssen, R and P. Rietveld. (1990) Multi-criteria analysis and geographic information systems: an application to agricultural land use in the Netherlands. In: H.J. Scholten and J.C. Stillwell (eds). Geographic information systems for urban and regional planning. Dordrecht. The Netherlands. Kluwer Academic. 129-139. McCaleb, N. (1985) Soil Survey of Coryell County, Texas. US Department of Agriculture, Soil Conservation Service. Government Printing Office, Washington DC. Mendoza, G.A. (2000) Integrating multiple criteria decision making and geographic information systems for assessing sustainability of tropical forests. Proceedings of the 2nd International Conference on Multiple Obje ctive Decision Support Systems held in Brisbane, Australia on Aug. 1–6. 1999. (In Press). Prose, D. (1985) Persisting effects of armored military maneuvers on some soil properties of the Mojave desert. Environ. Geol. Water Sci., 7(3):163-170. Public Land Law Review Commission. (1970) One third of the Nation’s Land. A report to the President and to the Congress. U.S. Government Printing Office, Washington, DC. Renard, K.G., C.R. Foster, G.A. Weesies, D.K. McCool, and D.C. Yoder. (1997) Predicting Soil Erosion by Water: A Guide to Conservation Planning With the Revised Universal Soil Loss Equation (RUSLE). U.S. Department of Agriculture, Agriculture Handbook Number 703. Government Printing Office, Washington, DC. Shaw, R., and V. Diersing. (1989) Evaluation of the effects of military training on vegetation in southeastern Colorado. Headwaters Hydrology. 223-231. Stoddart, L.A., J,A. Smith, and A.T. Box. (1975) Range Management. McGraw-Hill Book Publishing Co. New York. 532 p. Thurow, T.L., S.D. Warren, and D.H. Carlson. (1993) Tracked vehicle traffic effects on the hydrologic characteristics of Central Texas rangeland. Transactions of the American Society of Agricultural Engineers 36(6):1645-1650. US Army Environmental Center. (1999) US ARMY Training and Testing Area Carrying Capacity (ATTACC). Handbook for Installations Version 1.1. CERL, Champaign, Illinois. Warren, S.D., and C.F. Bagley. (1992) SPOT Imagery and GIS in Support of Military Land Management. Geocarto International 1:35-43. Warren, S., V. Diersing, P. Thompson, and W. Goran. (1989). An Erosion-Based Land Classification System for Military Installations. Environmental Management 13(2):251-257. Wilson, S. (1988) The effects of military tank traffic on prairies: a management mo” Environmental Management 12(3):397-403. Wischmeier, W.H., and D.D. Smith. (1978) Predicting rainfall erosion losses - a guide to conservation planning. U.S. Department of Agriculture, Agriculture Handbook 537. Government Printing Office, Washington, DC. 58pp.

16