Raster GIS Tutorial (Earth Shelter)

MassGIS land cover layer, or the USGS National Land Cover Dataset, the USGS Hydrography Layers, or the Massachusetts Department of Environmental Protection wetlands layers, in terms of their fitness for identifying potentially scenic water features.

---

Examine a Continuous Surface Raster and some Surface Functions

The Digital Elevation Model, DEM is an example of a Continuous Value raster. Each cell value is an observation of the height of that cell. Obviously, no matter how precise you want to get spatially, you could measure a different height. So the precision of this raster is an critical question to investigate. Without knowing anything about the DEM, we can guess that fluctuations of elevation smaller than the cell size are not well represented. The DEM provides, at best, a general idea of the elevation surface in our study area. Because of the regular arrangement of cells, certain things like Slope and Aspect can be calculated by looking at the other cells in the 3x3 cell Neighborhood of the cell in question. Naturally, the precision of the new slope and aspect layers is a function of the precision of the DEM. We may be able to get a sense of the utility of the slope and aspect layers by looking at the results on top of our familiar USGS Topographic map.

This demonstration will also introduce ad-hoc use of tools in the ArcGIS toolbox.

References

• Calculating Slope
• Calculating Aspect

---

Making a Model to Combine and Save and Strings of Data and Operations

With the Apsect operation we carried out in the previous demonstration, we have seen how a geoprocessing tool can be used in an ad-hoc fashion. Tools take input data, have settings that we can set to control how things are processed, and they produce output datasets. You could imagine how the output of one tool may be the input of another, and how the settings of individual tools might be saved so that we can re-run a chain of procedures, altering some settings, without having to set up several tools again and again. To do this we will learn how to save our Workflows as Geoprocessing Models Like Map Documents, models can be set up with relative pathnames so that they can travel around with our data and be explored, re-run and ajusted by our collaborators.

References

• Geoprocessing Cheat Sheet

---

Exploring a Ready-Made Model and Reclass and Map Algebra Functions

The fact that models can be made in advance and re-used is a great advantage in teaching, since I can prepare a model and demonstrate it in class without having to spend a lot of class time fiddling with a lot of settings. For the next demonstration, we will examine a model that has already been prepared and packaged with the sample dataset. The next few demonstrations will discuss parts of the Earth_Shelter model in the pbc_tool toolbox. The first phase of this model adds Reclass functions to convert the different scales of the slope and aspect maps into a common scale. Then we will use a Map Algebra statement to calculate a new layer whose cells values are calculated as a simple function of the cell values of two input maps. This combination of operations will produce a new raster that ranks each cell in the study area on a scale of -2 through +2 according to the compatibility of the terrain with our values for steep south facing slopes.

In this demonstration we reveal a special cell value, named NoData no data is a cell value that is used as blank space. In some functions, like Map Algebra The blank areas of NoData, prevent analysis from happening.

References

• Understand Reclass
• How the Raster Calculator Works
• NoData and How it Affects Analysis

Evaluation

It is important to take a break here and understand the new information we have created. It is very easy to make a little mistake in the settings of your functions, that can have very important consequenses in the model. If you don't check each step, you can easily fold a bad mistake into several other steps, wherepon it will be very difficult to figure out where the problem is (if you ever discover it al all.) Aside from this, even if the functions operated as expected, it is important to see whether the result makes logical sense vis-a-vis other data.

---

Converting Vector Features to Raster

Now that we have found places that have landform capability to support an earth sheltered house we need to narrow our focus. We are interested especially in land that is forested or that is already residential. Later on, we may add more criteria, but this is a start. This will lead us to explore two more important activities of cartographic modeling: conversion of vector features to rasters and a new way of using the map algebra function. As with many of the datasets we have, the MassGIS Land Use data set is a vector feature class, not a raster. Since many cartographic modeling functions require their inputs to be rasters, we must convert them through a process called Sampling. When doing this, we have to keep an eye on the cell size and what we are losing when we trade polygons with vertices to cells.

References

• Converting Features to Raster

Evaluation

Once the Sample Model has been run all the way through, we should take a look at the result to see if it makes sense, and reflects our values. I will call your attention to an interesting thing that happens with raster elevation models. Use View->Bookmarks to go to Great Hill, in acton. Notice first that the model seems to make sense with regard to the aspect and the distance from the stream. But notice the horizontal syrations that are blanking out the slopes along the hill. If we examine the compoent layers, we can see that this apparent error of omission is a result of the results of the slope function. Apparently the elebation model shows a horizontal step on the hillside, that is not reflected in the USGS contour map. The fact that these striations are mostly east-west on our map arouses some suspicion about whether these may be artifacts of the process of creating elevation models. This sort of error isn't something that should cause us to doubt the entire model, but it his is something we should keep in mind when thinking about whether our data are as good as they ought to be or whether we ought to look for a better elevation model.

---

Thinking about Relationships

Lets back up now and think formally about Modeling We have developed a concept of sites that are propitious for earthsheltered housing. Our conceptual model so far includes three concepts of fact:

1. Slopes steep enough for earthshelter
2. Aspect for southern exposure
3. Land Use Forested or existing Residential

Our model also includes a concept of Relationship:

1. Juxtaposition of propitious factors occurring in the Same Location

You can see that this association among facts is simulated in the Map Algebra procedure in the model. This procedure examines each location in a list of layers and produces a new layer whose cell values are each 0 a function of juxtaposed cells in the input layers. According to Tomlin's Map Algebra notation, this sort of association of cells is a Local Function. Tomlin has several other classes of associative procedures in his Cartographic Modeling language: Incremental Functions, Focal (Neighborhood) Functions and Zonal Functions YOu can check out these illustrations to see how these work.

References

• Analyzing Raster Data
• An Overview of the Distance Tools
• How Zonal Analysis Works

Incremental Functions: Distance, Movement Across the Landscape

A Local relationship describes two facts corresponding to the same location. Lets think about another relationship we may want to model. For example, in our investigation of earthshelter sites, we can see that there are many areas that our model designates as propitious for earthsheltered housing, which are actually on bluffs that are up-slope from drainage features such as streams or open water. This is a problem not only of erosion but also in terms of the long-term stability of the site, which may be involved in a landslide. We would like our model to eliminate the worst of these and to rank other sites as lower in value if they are too near water bodies. so that we don't have to waste our time and money examining these sites that are more than likely unbuildable. So we have a new concept of fact: drainage feature, and a new concept of relationship: Proximal To There are several ways we could model this relationship. The simplest is the concept of Euclidean Distance. This function takes a source layer, which defines an area of data as the area to be measured from, and an are to be measured into. The source layer may be a raster or it may be vector features. In the case of the Source Layer, the cells we intend to measure From can have any value, but the cells in the area we intend to measure Into need to have a value of No Data. The incremental functions and produces an output that measures, incrementally, some function of distance, or accumulated cost of traversing the landscape to get to the nearest source cell.

---

Weighted Overlay: Putting it all Together

Note how the last step in our simple sample model uses the Single Output Map Algebra computes the weighted average of all four factors.
( slopeval + ( 2 * aspect_val) + (0.5 * mgis_lu_val) + hy_dist_val) / 4.5
We note here that this map algebra function is a local function. We can also see that the terms in the map algebra expression are weigted, so that Aspect is weigted double with regard to slope. The land use is given half a weight. Distance from water and slope are not weighted, which is the same as giving them a weight of one. The weighted avrearge is computed by summing the terms and dividing by the sum of the weights.

References

• Buildng Expressions in the Raster Calculator

Discussion

At this point, we should take a look at the results of the model to judge whether it makes sense, based on our visual inspection of the results on top of the topographic map and aerial photograph. It is useful to keep in mind that we don;t expect this model to be perfect, but it would be nice if it did as good a job as we feel is possible given the data and the procedural tools that we have. In the next series of sections, we will explore a few more sophisticated ways that we might approach this problem.

---

Modeling Complex Incremental, Focal and Zonal Relationships

The approcah we have taken so far in this tutorial has shown us the major workflow patterns involved with cartographic modeling: Transforming data into represenataions of our own values; modeling local and incremental relationships, and also some simple terrain analysis. The next section will elaborate on our model a bit more, to showcase some fancier incremental functions for taking into account areas having variable resistance. We will also look at visibility functions, which is another sort of incremental relationship. THere are two more major types of relationships that are part of the cartographic modeling toolkit: Focal Functions, which consider the relationship of a cell with the facts represented in a data layer -- within a specified neighborhood, and Zonal functions which take measurements of phenomena within areas defined by the zones in a raster or the polygons in a feature-class.

Since our model is going to get a bit more complicated, we will break it down into separate sub-models, which will each yield a value raster which summarizes the local pluses and minuses for each cell within our study area with special regard to the development of earth-sheltered housing. These will then be given the weighted average treatment in the very last phase.

Cost-Weighted Distance

References

• What is a Mosaic?
• Cost Distance
• Conditional Tools

Model Number 2 uses a distance function to look at the proximity of commercial centers. We value good access to commercial centers. To demonstrate this sort of distance function, we will consider accessibility by car, since this is a common thing to do. We will leave it as an exercise for you to figure out how to convert this to a model for pedestrian and cycling accessibility. Our model converts the road class attribute to a value of cost for crossing a cell. Because the reclass function can only yiels an integer raster, our first transformation yields values of Minutes per Kilometer. Because people do not move only on roads, we reclass \ the value of the land between the roads from NoData, to 10. This simulates the cost of traveling overland to get from the road to your destination. The cost units, required by the cost distance function are expected to be experesed in terms of the distance units of the input layers or the geoprocessing environment. This is why we use the map algebra function to divide our cost units by 1000, converting them to minutes per meter. Now when we use this cost layer in the costdistance function, the relationship that is represented in our output for each cell, will be the cost of getting to the nearest commercial area (in minutes.)

Model 2a makes this accessiibility model more realistic, by stipulating that water cells are not so easily crossed as land cells. This model introduces a couple of important functions that are often necessary when transforming raster datasets:

• Merge function effectively overlays a list of rasters, and allows rasters listed first to supercede rasters listed subseqently. In the merge function cells having a value of NoData are treated as transparent, according to this overlay analogy.
• Conditional Functions: These allow you to create a new raster whose cell values are defined conditionally based on an evaluation of an expression.

The merge function lets us overlay the road values over the stream values, and the conditional function SetNull allow us to turn the set the wet cells to a value of NoData which will effectively dictate that the only way to cross a hydrographic feature is where a bridge already exists.

---

Incremental Relationships where Slope is a Factor

Model 3 provides a better way of modeling the relationship of water features to potential building sites. This model uses the Pathdistance tool, which wil yield the distance across the landscale (considering that travel on a slope accumulates more distance than travel on a level surface. Path Distance also will consider that some types of land cost more to cross than others, due to vegetation or roads, for example. Finally, we can adjust the Vertical Factor Properties to restrict the measuermentr so that it will only travel up-slope -- which when runnof is concerned, is the sort of relationship we want to measure.

These distance measuring tools provide a means of representing a relationship of proximity or accessibility. For example, now that we have a means of measuring te relationship between water body and potential sites upslope, we can use this distanncd layer in our evaluation of sites.

Note that this model does a couple of new things: First, we transform land-use to runoff resistence using a relational join to a lookup table. Befoe we do this, we make a new copy of the land use layer. Adding the join to a copy of a layer is a good idea, since we may want to run this model more than once, and adding the join to the new layer guarantees that the model won't encounter the situation where the join already exists. -- which would cause the model to have an error.

ANother feature of this model is that we have pulled the cell-size environment variable out of the path distance procedure. Because this process is fairly time-consuming, enlarging the cell size from 10 to 30 reduces the number of cells that need to be processed by 9 times. Once we have this model working the way we want it, we can poke this cellsize back down to 10 and un the model when we are at home in bed.

References

• Path Distance
• Add Join Function

---

Focal Functions: What is in My Neighborhood?

Sometimes a quality of a place is a function of qualities of the areas surrounding it. For example, a good location for a farm house may be related to the amount of productive land within a circular area of 500 meters in radius around the location in question. Focal functions (also known as neighborhood analysis) let us define the geometry of a neighborhood, and then will summarize the values on a data layer (e.g. productive land) within that neighborhood, as it is centered on and evaluated for every cell on the map. The value pf each cell in the output grid is the statistical summary (sum, average, variety...) of the values within that cells neighborhood.

References

• Calculating Statistics for Overlapping and Non-Overlapping Neighborhoods

Measuring the amount of trees upwind

Our simple model favors sites that are located in the forest. Part of this idea is that the trees will slow down the winter wind that comes, predominantly from the Northwest. In our more complex model we want to be more precise about the relationship between our site and trees. It is not necessary to be IN the forest to get the value of trees as a windbreak. The critical relationship is that a site should have Trees Upwind in the wintertime when the winds are predominantly from the northwest. Our Trees Upwind model creates a wedge-shaped analysis neighborhood, which can be used to represent the Upwind relationship for each cell. This neighborhood can be moved over a forest raster to count the number of forested cells upwind.

---

Potential Views of Water

One of the more interesting incremental functions is Viewshed which uses a layer defining the source (which is known as the observer locations, though may just as easily be thought of as the observed!) An elevation raster establishes the barriers to visibility. Like the distance functions, the Viewshed function attempts to spread from the observer points, across the elevation model, until it encounters a barrier. In our model of Water Views, we begin by establishing observer points in the water. For this, the Random Points function is very handy. Then we simply calculate the viewshed from these points. Note that although the output of the viewshed tool is portrayed with only two values: Visible and Invisible, if you change the symology of this raster, and use the get-info tool, you can see that the value of each cell varies -- it is a measure of the number of observers tha can see a given cell. Note that in this model, we set the raster analysis environment of the viewshed tool so that it uses a cellsize of 50 meters. Using the default cell size fo rthis project (10) meters, makes this procedure take 25 times longer. After we get comfortable with the way this is working, we may adjust this cell size down. This will be particularly important when we add buildings to the model.

Our second version of this model adds buildings to the terrain model and adds offset values to the observer points to simulate that the cells on land (where we expect our earthshelter residents to be looking from) sould be considered to be offset 2 meters, to simulate the height of a person.

References

• Performing Visibility Analysis
• Create Random Points

---

Parsing Ridges, Hilltops, Valleys, and Hillsides

Take a look at the feng shui model in my latest sample earthshelter dataset. It uses two focal mean operations to find ridges, valleys and mid-slope areas. You need to tune it to find the sorts of areas you are looking for. It is tuned by playing with the radiuses of the larger and smaller focal mean operations. (Actually they should be 50 and 100 to start - there is an error in model in the on-line zip file. This model basically makes two different smoothed elevation models and subtracts them to create a difference raster where the wrinkles are (+) and creases are (-). You also tune it in the final reclass of the difference raster. IN our case, you would want to give valley areas and hilltops a lower score and valley walls something else, etc,. Actually, I have left this factor out opf our model, since I thnk that our slope criteria serve well elugh, but I have left the sample in the toolbox, just because it is a useful demonstration of Focalstatistics.

References

• Neighborhood Statistics and Focal Functions

---

Summarizing Earthshelter Potential

The Site Score model simply takes a weighted average of the scores for each of the concepts in our model. As you can see, there are a couple of fairly large potential sites. The spread of values is not as great as it could be. In order to get a broader differentiation, we could start to fine-tune our model to make sure that there are more cells that are awarded 2 or -2 in the component models. We could also try different weights on the various components in our weighted average function. But for now, we will just proceed.

---

Zonal Functions

So far, we have looked at lots of different sorts of relationships -- those based on the cell location, others focused on cell neighborhood (focal functions), relationships based on proximity or accessibility (incremental functions). The last sort of function we will look at in this model is based on the coverage and geometry of zones. If you look at the attribute table for our score layer, you will see that all the cells are divided into just three zones. We can look at the Count field to get a sense of how many cells in our study area have been assigned to each category. This is interesting, and a good reminder of what zones are in the raster context. Now one thing that we need to consider is that we require a fairly large site for oue pilgrim pueblo developments. We are looking to site at least 50 houses. And so it would be helpful to be able to evaluate each clump of contiguous cells as a potential site.

The Large Sites model uses the Regiongroup function to create a new zone for each clump of contiguous cells in our site-score layer. The model then reclasses this layer based on the cell count, to make a mask that will weed out potential sites that do not have at least 50 cells.

Now, lets pretend we have some parcels that have come been reported to be on the market, and we want to evaluate them based on how many high-quality cells of earth-sheltered housing. If you look at the Parcel-Score model, you will see that this uses the parcels as zones, and calculates summary statistics for each parcel, based on the values of the cells on our large_sites raster. The output of this function is a table, which can be joined back to the parcel table, to evaluate every parcel several statistics that will help us to evaluate whether the particular parcel merits further study.

References

• Regiongroup
• ZOnal Statistics
• Zonal Statistics As Table

---

Model Evaluation

Is our model a good one or a bad one? By now, you know that this is the wrong question. We should ask instead:

• Is our model useful for our purposes? We will get to this...
• What are the weakest inputs and procedures in the model?
• In what way would we expect the model to be biased -- toward errors of Omission or errors of Comission?
• Are there important considerations that are ignored by our model?
• How could the model be improved?

This model seems to do a good a job of evaluating sites for building earthsheltered houses. At least as well as I would be able to do by studying the USGS quad map -- and much faster -- paricularly since our plan is to scan and evaluate all of the real estate offerings in the united states. IN terms of our pilot study area, I'm sure that the method has identified many areas as potentially good which may not actually be good. The model has also probably missed some spots that would be great for an earthsheltered house. The weak point here is the coarseness of the terrain model (10 meters), a problem that is compounded in the slope and aspect maps which can only be seen as averages over a 3x3 cell neighborhood (90 meters) at best. This fact would undoubtedly miss a lot of slopes that might make a fine backdrop for a house. Better terrain data would no doubt make this model better in finding smaller nooks for a house (and perhaps smaller is better?) But no matter how fine the terrain model, it would be difficult to model all of the site conditions, especially the ones that can potentially be created. One can always regrade a site to an extent. <\p>

Of course I would actually spend a lot of time walking around in the study area before I actually decided on a house site. I think that this analysis is a useful aid in narrowing down areas that I would visit. By finding many of the more obvious spots in a fairly easy way, it gaves me more time to study the finer effects that may make a place more or less fit -- and this activity would probably give me more insight to fine-tune the model or add new data (soil type, access to roads) that would be useful. This knowledge may also help to learn a general method or rating criteria for establishing the thermal propensity of a site.