Purpose and Study Area
The purpose of this assignment is to use a LiDAR DEM of the Middleton, Nova Scotia, Canada area in order to model flooding. All of the steps in this process including hydrological flattening, hydrological enforcement, watershed delineation and incremental flooding will be discussed. This is completed at various resolutions and compared to datasets with barriers.
The data used for this study was collected by the Applied Geomatics Research Group within the Nova Scotia Community College. The lab design and workflow was supplied by the Centre of Geographic Sciences in Lawrencetown, Nova Scotia. The hydrological vectors including streams, rivers and lakes were extracted from the Nova Scotia Geomatics Centre Base Data (1:10,000 scale) which were derived from 1:40,000 scale aerial photography spanning 1987 to 1997. The LiDAR DEM was collected on August 18, 2010 with an Optech ALTM 3100 system. The DEM was made from ground poitns using TIN interpolation and have been transformed into orthometric heights (CGVD28).
Hydrological flattening involves assigning lakes and streams elevation values to remove “bad” points. What this means is when LiDAR is flown, the results over the water can be variable. For the purposes of watershed modelling we need for the lakes to be a lower elevation than the streams flowing into the lakes or it can create inaccurate flow direction models. In the case of rivers we do not want to assign a single elevation but instead provide an elevation gradient derived from the dem. Model builder was used for many of these processes.
To get a representative water level the first step is extracting shoreline elevations using the downloaded waterbodies file. If such a file is not available, the polygons can be drawn by hand. Use the polygon vertices (Feature vertices to points) to extract DEM elevations. (Add Surface Information). Next the minimum elevation is determined using Summary Statistics and Get Field Value. Essentually Summary Statistics creates a table with all of the nodes and their new DEM values and by querying the minimum value. Once this value is determined, the lake can be extracted from the DEM using Extract by Mask (use lake polygon for this). The raster calculator can be used to assign the minimum value to anything in the raster that is not null (i.e where the lake was extracted). Once the lake has been extracted from the DEM and it’s value reassigned, it can be mosaicked back into the original dem using the mosaic tool.
The river flattening is completed similar to the lake flattening, but requires that an elevation trend be applied. The river shoreline elevations are derived similar to the lakes method (Feature Vertices to Points, Add Surface Information). These points are then passed to the Trend tool to interpolate a raster that covers the water body with a flat gradient surface. The surface was then clipped to the river outline and mosaicked into the DEM that has been updated with flattened lakes.
Hydrological Flattening Discussion
This will result in flattened lakes and flattened but trending rivers. This removes irregularities and erroneous elevations that may affect water flow direction. This process could be improved by increasing the accuracy of the river and lake outlines by inputting them manually or finding a larger scale water body file. These false elevation artifacts could create false contour lines (depending on the scale). If orthorectifying an image to the dem the false elevations in the river could skew the results and affect the pixel accuracy.
Watersheds were created using the flow direction tool, flow accumulation tool, the fill tool and the watershed in ArcGIS. The pour point is placed on the area of highest accumulation. Barriers such as bridges and culverts can appear to be solid in LiDAR when in reality water can flow under or through this infrastructure. Location of Culverts would significantly aid in modelling waterflow especially if size of culvert was included. Refer to the image to the right, focusing on areas with red dots where the road crosses the river.
This process could be further improved by analyzing additional accumulation zones where there are no rivers (areas with barriers such as terrain or roads) and then using aerial photography to search for evidence of culverts and adding them into the model. More realistic models may result from larger DEM’s which cover a larger catchment area.
The flattened LiDAR DEM with barriers removed supplied the most realistic results, followed by the 20m DEM and finally the LiDAR DEM without barriers removed. The results could be better if the DEM covered a larger area and allowed for additional watersheds to be recognized. Were this the case the 2m Flattened DEM with removed barriers may show trends more similar to the 20m DEM, assuming that the 20m DEM recognizes more regional trends.
20m DEM with Barriers, no flattening
2m DEM with barriers and flattening
2m DEM with barriers removed and flattening
Incremental flooding was emulated using a for loop to cycle through min and max flood levels and setting flood increments. The Flood mask was created by using the Raster Calculater to created a mask covering a DEM cell value range (i.e <10) and a binary value for covered cells (i.e cells below that value are given a value of 1, cells outside that range are not given a value). Areas such as water treatment plants have water but are not connected to the flood points. To ensure that only areas with connectivity to the pour point are flooded, connectivity must be tested. To test connectivity the Raster to Polygon tool was used to create a polygon from the flood raster. Any polygons that were connected to the pour point were selected using the spatial join tool. Select the flood polygon as the Target Feature and the pour point as the Join Feature. One-to-One and Keep all target features was turned off. intersect was the spatial operation. The output will have only polygons that intersect the pour point.
Most of the results are similar. The unflattened DEM flooded a lesser extent likely due to a false barrier created from elevation spikes in a narrow part of the river. This is visible below and is possibly even more obvious at greater flood increments. Shown below. Additionally the unflattened imagery often resulted in a strip of elevated terrain in the middle of the river. The cause of this is uncertain.
The clearest difference occurs in the hydrological flattened images between the barrier and removed barrier models. The model with barriers removed penetrates the east west highway and floods the other side before the model with barriers intact. This is demonstrated below. This is an expected results. The barrier removal process was very conservative. Removing larger portions of the boundary may cause a larger gap between the flood results of the two models. See below.
Overall the flooding mostly matches the watershed. The extents of flooding go beyond the watershed and are less affected by ridged terrain perpendicular to the river as it is parallel to flood direction.
Flooding as it crosses the barrier. Barrier present (pink) vs barrier removed (blue)
Results and Discussion
The tools used in model builder for the construction of this project and modelling of flood data were very affective. I ran all of the tools from within model builder so that I could customize a few features. For lake flattening I ran an iterator that selected the next row in a table before running the model until the row value was null. This removed manual selection of each lake before the running of the model to flatten it. For flood modelling I used integers instead of uneven numbers so that I could avoid issues with data compatibility since the flood model toolbox accepts only integers. The flood model was set up to use metres and not centimetres. If greater accuracy was required the flood model could be set to use centimetres. Additionally instead of parsing the raster name and adding the value of water level for polygon output I manually changed the base of the name and the suffix using %title% (i.e DEM_%WaterLevel%).
Issues I encountered
1) The attempt to run a <8m flood raster with errors. Solution was that there was no data below 8m and this was resulting in no polygons. The polygon to raster extraction does not work if the shapefile is empty.
2) I did not run the fill tool on the raster before running the flow direction tool. This resulted in a black and white flow direction raster ranging from 1 to 255. I’m not sure why this is but flow accumulation analysis did not work.
The only concern with the results is that the barriers may not be a representative of the width of the actual culvert or flow under a bridge. Smaller flood increments may provide more information about the behavior of the flooding around terrain features.
It should be noted that for the 20m DEM it would have been appropriate to place a second pour point directly north of the east side of the east west road feature in the study area.
Comparison to USGS Base Specifications
This data was processed to an acceptable level as set out by the USGS LiDAR base specifications. Breaklines included Rivers and Lakes. The breaklines were used to create flat single elevation lakes and flat but trending rivers based on the breaklines. Keep in mind that this method is for terrestrial accuracy and may not represent the accurate elevation. Coupling the information with gauge data could provide more accurate results for hydrologic purposes.