Final+report

=GEO9016 Final report=

Introduction
The University of New South Wales is looking to expand its accommodation facilities at the Cowan Field Station. This reports proposes locations for the expansion, based on the principles that the site should be far enough from endangered species, close to roads and power lines, and of low pollution and fire danger. Additional requirements include minimised construction costs, high solar radiation and pleasant views. Location: The following map identifies the Cowan Field Station, in the north of Sydney, New South Wales, Australia (latitude: -33.592563, longitude: 151.156133). The field station environment is open woodland in a temperate climate.

Map 1: Cowan field station

Available Data: The datasets projected to MGA 56 and incorporated in the modelling using ArcGIS 9.3 software are: vegetation contours creeks rivers infrastructure DEM spot heights soils geology fire trails bark fuel load surface fuel load ground fuel load flow accumulation soil erodibility mangroves threatened flora threatened fauna power lines

Overview of the Report: The remainder of the report includes mappings of the four most important criteria for site location: building (proximity to infrastructure), erosion, fire and conservation. The developed maps convey the suitability of every 10 square metres for the new site, based on one individual criterium at a time. The maps are then combined to show the suitability of every 10 square metres for the new site, based on all the criteria. Optimum sites are shown for the cases of low conservation standards and high conservation standards.

Use of fuzzy data sets in the report
We used fuzzy sets in the derivation of the models as it allowed us to move beyond simple boolean logic. It allowed us to represent geographical phenomena where it is often difficult to partition the real world into data sets synonymous with boolean logic (Burrough and McDonnell, 1998). For example using fuzzy sets allowed us to assign diminishing value to land the steeper it got, after which we could not build. If we did not use fuzzy sets all values up to the steep value would have the same value. No "recognition" would have been given to land that was flatter than other land that was steeper (but still within the cut-off value). They would have all had the same value (boolean logic either in the range or not). This does not reflect how slope should be modelled, as land that was flatter than other land (but still within the cut-off value) would be more attractive and the model should reflect this. In this report we used two different fuzzy membership equations. The monotonically increasing function and the trapeziodal function. The first was a monotonical increasing function, where values between two values (a and b) where given values of (0 and 1) respectively. Any value less than a was given a value of 0, any value more than b was given a value of 1. Any value between a and b were scaled accordingly. **See figure x**


 * Figure x: Montomical increasing function**

We also used a trapeziodal function that had four values (a, b, c, d). any values between a and b were scaled from 0 to 1, values between b and c represented “ideal” values and were given the value of 1. Any values between c and d where scaled from 1 to 0. **See figure x.**


 * Figure x: Trapeziodal function used in the modelling**

The building model
The building model was developed to determine the most suitable location to build the field station, taking into account the slope of the land, distance from existing infrastructure including roads and electricity. In addition we needed to ensure that no building occurred on existing infrastructure.

Assumptions
For this model we assumed the extension to the field station would be built:
 * on a slope between 0 and 10 degrees
 * ideally not closer than 27 metres and not further than 77 metres, but no building after 104 metres from the fire trail
 * close to existing powerlines (within 1 km).

Derivation of the model
For the building model four rasters where created ( Slope, Distance from Roads, Distance from walking track, and Distance from electricity) and then combined.

The slope raster (Slope) was calculated from the DEM that was developed using vector data from the contours, creeks, rivers and spot heights data sets. The DEM was created using the ANUDEM algorithm developed by (Hutchinson, 1989). This DEM has an error of + or – 5 metres on the horizontal plane and + or – 12 metres on the vertical plane. The error from the DEM would have been translated to Slope.
 * The Slope raster**

The use of fuzzy data sets in Slope allowed us to “soak” up this error. Although there was an error in Slope from the DEM, the transition zone in the fuzzy data was sufficient to take this into account, “softening” the error.

Once the slope was derived from the DEM we implemented a monotonically increasing function between the minimum slope 0 (degrees) and the maximum slope 10 (degrees). We inverted this calculation so that a value of 1 was obtained for slopes of 0 degrees and a value of 0 was obtained for slopes of 10 degrees. This is because we wanted a higher value attributed to flatter pieces of land. The use of fuzzy sets allowed us to assign higher values (ie closer to 1) to land that was flatter than other land. See figure x for the Slope raster.


 * The Roads raster**

We used the firetrails data set to create the Roads raster. First distance from the firetrails dataset was calculated for all points. Then we applied the trapeziodal function to create a fuzzy membership dataset, rescaling the data from 0 to 1.

The firetrails data set had its accuracy assessed through field studies in March this year. It was found that the firetrails data set had an root mean squared error (RMSE) of 27 metres. We took this into account in creating the fuzzy membership dataset.

The width of the transition zone for roads cannot be less than the precision in the measuring of the roads (Burrough and McDonnell, 1998). As the RMSE for the fire trails data set was approximately 27 metres we needed to ensure that the range that we used in the building model was no less than 27 metres. Hence we used a transition zone of 27 metres. This “softened” the effect of the error on this data set. See figure x for the Roads raster.


 * **Threshold** || **Threshold value (Distance (m))** || **Value in fuzzy data set** ||
 * a || 0 || 0 ||
 * b || 27 || 1 ||
 * c || 77 || 1 ||
 * z || 104 || 0 ||

Figure x: threshold values used in the trapeziodal function for the Roads raster in the Building model

We inverted this calculation so that a value of 1 was obtained for distances of 0 metres (ie very close to power lines) and a value of 0 was obtained for distances 1000 metres or more. This is because we wanted a higher value attributed to being closer to the powerlines. See figure x for the Powerlines raster. **
 * The Powerlines raster**
 * The Powerlines raster was created by extracting the powerlines data from the infrastructure dataset. Then distance from the powerlines for all points were calculated. Next we used the monotonically increasing function between the minimum value 0 (metres) and the maximum value 1000 (metres) to create a fuzzy membership dataset.

Finally we needed to ensure that the building did not take place to close to existing infrastructure. We determined that the area where the Great North Walk passed through might also be a good potential area for building. We had to exclude the Great North Walk from our calculations in the Building model. First we selected the section of the Great North Walk near the fire trail in the infrastructure dataset. Next we applied a tool to calculate the distance for all points from this track. Next we applied a ** monotonically increasing function between the minimum value 0 (metres) and the maximum value 10 (metres) and inverted it creating a fuzzy membership dataset. This was so that building could occur where values were 1 but not where values where 0 (ie on the track). We chose 10 metres as this was the minimum cell size used in the model. See figure x for the Great North Walk raster. **
 * The Great North Walk raster**

The next step was to combine all four rasters to create the the Building model raster. As all the rasters had been developed as fuzzy membership data sets, the value of 1 represented the best place to build whilst 0 represented an area that was not to be built on. We combined all the rasters (Slope, Roads, Powerlines and Great North Walk) and took the minimum value of each cell. This ensured that any cell that had a value of 0 in one of the four rasters, also had a value in the Building model raster. See figure x for the Build model raster.
 * Derivation of the Building model**

The fire model
Assumptions

** The pollution model **
The pollution model was developed using Universal Soil Loss Equation (USLE). The USLE is designed to predict the longtime average annual soil loss in runoff from specified land units (Rosewell, 1988). Soil loss is determined by multiplying six factors values together. The six factors are erosivity (R), soil erodibility (K), slope length (L), slope steepness (S), support practice (P) and crop management (C).
 * Assumptions **
 * The R, K, P factors are derived from Rosewell (1993)
 * erosion will not increase once the slope length exceeds 150 m
 * roads at cowan filed station have no erosion.

This model has four separate processes that were combined: slope steepness, slope length, crop practice and management. Slope steepness ** Digital Elevation Model was used and calculated in degree and converted to radians, then raised the power to 1.35 (Selby 1993, Moore and Burch 1986a,b). S= (sin(slope/57.296)/0.0896)1.35
 * Derivation of the model **

The result from this equation showed that slope steepness at study area is 0 to 25.512% Slope length ** Slope length was measured by using formula from Moore and Burch (1986a, b) and flow accumulation data which generated based on Tarboten (1997) L = (flow accumulation * (Cellsize/22.13))0.4

The result from this equation showed that slope steepness at study area is 1.828 to 25.512 meters Crop management and Practice factor ** This model using combination on both factors. Vegetation data was used classified by using and tables D-3 and D-4 from Rosewell (1993) Results: ** After combine the input together based on USLE. Erosion = R*K*S*L*CP. and we used constant value of Rain factor = 3000 (based on Rosewell 1993) then the erosion at Cowan Field Station approximately from 0 to 512.436 tn/ha/year.

The conservation model
This model was used to determined the area that was to be off limits for building due to conservation concerns in the area. The model derived a prohibited area within which it was not possible to build. We used four data sets to create this model. They were streams, mangroves, endangered flora and endangered fauna. To provide a range of options we developed two different conservation scenarios depending on the importance of conservation to the stakeholders of the project. The two scenarios were: Scenario A: High conservation value Scenario B: Low conservation value


 * Assumptions**

Scenario A: High conservation value
 * Dataset || Threshold value (m) ||
 * Streams || 100 ||
 * Mangroves || 250 ||
 * Flora || 250 ||
 * Fauna || 250 ||

Scenario B: Low conservation value
 * Dataset || Threshold value(m) ||
 * Streams || 50 ||
 * Mangroves || 120 ||
 * Flora || 120 ||
 * Fauna || 120 ||

This model was derived from stream lines including rivers and creeks, mangrove, threatened flora and fauna datasets. The model delimited an area around the feature (ie threatened flora) within which we were unable to build. There were limitations in the use of these datasets however. For example in the endangered flora and endangered fauna datasets they represented single values, or in the case of mangroves a definitive boundary, however this was not an accurate representation of the biological community on the ground. The data indicated that there was endangered flora or fauna in a 10 metre by 10 metre cell or the mangroves where present in one cell, but in the adjacent cell there were no mangroves. However this is not how biological communities would be distributed. Instead they may centre on the location represented by the cell (centre cell), but would also likely be distributed in the neighbouring cells slowly tapering off until a boundary was met. There is uncertainty as to exactly where the range of the endangered flora or fauna community extended to. In addition there may be inaccuracies in the reported location of the threatened community. However the use of fuzzy datasets allowed us to model this uncertainty and to reduce the effect of errors in the data. This is because Fuzzy data sets are less sensitive to small errors in data, particularly when the values are close to boundaries (Burrough and McDonnell, 1998).
 * Derivation of the model**

First, we calculated the distance from each dataset (endangered flora, endangered fauna, rivers and mangroves) for all points in the area. Then we created fuzzy membership datasets using the threshold values from the two scenarios for each dataset. This was done by using a monotonically increasing function. As the conservation model was an exclusion model we inverted the rasters that were created. The inversion of the dataset was necessary so that when we subtracted it from the building model we could determine which locations gave greater importance to building compared to conservation (see combination of the models). We created Streams, Flora, Fauna and Mangroves rasters see figures x, x, x and x.
 * Process**
 * Finally all these rasters were combined together and took the maximum value of each cell. This ensured that any cell that had a value of 1 in one of the four rasters, also had a value in the Conservation model raster. See figure x for the Conservation model raster. **

**Fire Model**

The fire intensity raster was developed by quantifying weather conditions and fuel load, the most important factors controlling the fire intensity (Bessie & Johnson 1995). Since direction of wind changes frequently and influences the effect ridge tops have on fire danger, the wind direction and ridge tops were not accounted for in this model (Pyne et al. 1996). Thus there is a significant error range for the raster developed.

Fire intensity was calculated as the following product: (energy with which the forest fire will burn)*(fuel load)*(rate of spread * e^(0.069*slope))*(0.02777)

The energy with which the forest fire will burn was estimated as a constant, 18600 kJ/kg as suggested in the fire model lab (Gill 1998).

The rates of spread were determined using the method from Noble et al. (1980). The assumed environmental conditions were:

Temperature: 37.2 degrees C Relative Humidity: 15% Average Wind Speed at 10 m height: 40 km/hr McArthur Drought Index: 10

These estimations may underestimate the worst case scenario if global warming persists and increases the climatic extremes (decreasing humidity and increasing temperature and drought index).

For grassland, the degree of curing was assumed to be 80% and the rate of spread calculated as 0.811 km/hr. Roads (including the freeway, highway and smaller roads) were estimated as 10 m wide and assumed to have zero fire danger. This would have underestimated the fire danger around the smaller roads. See the map with buffer below.

Map 1: Road buffer represented at Cowan field station

For eucalypt and other plants that burn well (assumed to include Sheltered Hawkesbury Forest, Banksia Heath Scrub and LW/Low Hawkesbury LW Slope Plat), the rate of spread was calculated as 0.0802 km/hr.

For poorly burning vegetation (assumed to include Low Woodland Shrubland, Exposed Hawkesbury Woodland and Plateau Low Open Forest), the rate of spread was estimated to be 0.02 km/hr (Jenkins et al. 2008). The rates (in raster format) were multiplied by e^(0.069*slope). The rate_slopespd = rate of spread * e^(0.069*slope) is shown in the following map:

Map 2: Fire spreading rates represented over Cowan field station

Multiple students estimated the fuel load across the Cowan field station using the Overall Fuel Hazard Guide (McCarthy et al. 1999). Each group covered two 500 square metre blocks; there was some error due to people judging the bark, surface and elevated fuel from unique perspectives. The observed fuel load is shown below.

Map 3: Fuel load represented over Cowan field station

The product of all these parameters is shown is the fire intensity map below.

Map 4: Fire intensity represented over Cowan field station

Some vegetation, such as mangroves, had not been represented in the vegetation dataset. Mangroves have a low fire spreading rate, and hence present a low fire danger; they do not need to be incorporated in this model because the areas in which they reside are already regions of high fire danger due to surrounding vegetation and fuel loads (so they would not reduce the fire danger in those areas if the model is assumed accurate).

Since the fire is assumed to be easily controllable below 3000 kW/m and uncontrollable above 4000 kW/m, the fire intensity raster was clipped to these limits. Then it was rescaled to a 0-1 range and inverted so that 1 is best for building on and 0 worst. This fire model (shown below) was later combined with the erosion, conservation and building models for site determination.

Map 5: Fuzzy fire intensity representation for Cowan field station

The fuzzy representation does not show much grey; perhaps the precision was sufficient to provide a range of fire intensities between 3000 and 4000 kW/m. Further discriminating the rates of spread of the different plant types may increase the precision.


 * Combining the models**

The ranking process

 * Recommendations**

The high conservation standards should be upheld since the candidate sites for both low and high conversation scenarios are of similar location, size and shape. There appears to be no additional cost for choosing high conservation.