Mapping Wetlands for Railroad Environmental Engineering and Operations

Transcription

1 Mapping Wetlands for Railroad Environmental Engineering and Operations Brian J. Huberty U.S. Fish & Wildlife Service 5600 American Blvd West; Suite 990 Bloomington, MN (612) Steven M. Kloiber Minnesota Department of Natural Resources 500 Lafayette Road North St. Paul, MN (651) Joseph F. Knight Department of Forest Resources University of Minnesota 1530 Cleveland Ave N Saint Paul, MN (612) jknight@umn.edu Number of Words: 3088 ABSTRACT Railroads need to visualize watersheds and landscapes with wetland and surface water mapping tools for environmental engineering and operations. Water dynamics in wetlands, rivers, and lakes can all have significant economic and environmental impacts to any railroad operation. Climate changes are causing severe swings in our weather where wetlands, lakes and rivers have grown rapidly over the last twenty years. In response, BNSF is investing millions of dollars to raise track beds through the Devils Lake region of North Dakota. Here we describe a recent large-area application with a state-of-the-art, semi-automated process used to create the new Minnesota National Wetland Inventory (MNWI) for east-central Minnesota. The new MNWI incorporated high resolution, multi-spectral aerial imagery from multiple seasons; high resolution elevation data derived from lidar; and other image and geospatial datasets. Image object segmentation and random forest classification techniques were used along with a final digital image interpretation review to create these new wetland maps. Validation data points (more than 1000) were acquired using both independent image interpretation and field reconnaissance. Overall accuracy for wetland identification was 90% compared to field data and 93% compared to image interpretation data. Railroads can use this data as well as this approach to improve the currency of their geospatial wetland and surface water infrastructure. INTRODUCTION Recent climate, disaster, and infrastructure episodes continue to bring the spotlight on the underlying environmental impact by our railroad systems. Impacts can range from track flooding, tank car spills, to railway expansion. For example, the report by BARR Engineering (1) illustrates the need to raise a track bed to avoid flooding by Devils Lake Chain in North Dakota. AREMA

2 Figure 1. Churchs Ferry, ND 1990 aerial image overlaid with National Wetland Inventory polygons derived in The polygons basically match the image of water bodies and wetlands. Figure 2. Churchs Ferry, ND 1997 aerial image overlaid with National Wetland Inventory polygons derived in The expansion of the Devils Lake chain can be seen through the larger waters and wetland expansion. Note the BNSF main line designated with the black line running lower right to upper left. AREMA

3 Figure 3. Churchs Ferry, ND 2012 aerial image overlaid with National Wetland Inventory polygons derived in The expansion of the Devils Lake chain has consumed half of the image. The three decades old NWI map is essentially obsolete. Figures 1, 2, & 3 illustrate the expansion of lakes and wetlands over the Devils Lake watershed in North Dakota. Earlier this year, an oil train derailment spilled oil into nearby wetlands in Heimdal, ND (2). A Canadian Pacificproposed rail yard expansion into a wetland in St. Paul, MN is now being built (3). Finally, recent passenger railroad corridor studies (4,5,6) refer to outdated U.S. Fish & Wildlife Service, National Wetland Inventory (NWI) maps being used to design high-speed railway corridors in the Midwest. Basing multi-billion dollar projects on outdated wetland and surface water maps can be done much better. Current knowledge about the location and extent of wetlands and surface waters requires the use of updated and current wetland maps. Updating and maintaining wetland and surface water maps like NWI have become more expensive and technically challenging as we have moved into the digital age. More importantly, the 2-D NWI map represents just a snapshot in time. NWI does not map water level elevation or changes in type or area over time which is needed to represent the true 4-D nature of wetlands. Annual funding levels for the U.S. Fish & Wildlife Service s National Wetland Inventory (7) program have essentially remained unchanged at about $5 million dollars nationally since Typically it takes about $15 million dollars (including lidar and remote sensing imagery) to map a state such as Minnesota. This publication seeks to primarily highlight the methods and approach (8) used for the Minnesota National Wetland Inventory update using image object mapping methodologies (9) and to transfer this knowledge for better and faster railway infrastructure mapping. AREMA

4 Project Areas Figure 4. Project area map with a comparison of the original NWI to the updated Minnesota NWI on a shaded relief map derived from lidar. The Canadian National Railroad runs from the bottom left to the upper right of the image map with a couple road crossings. As one can see with the black dashed lines, the original NWI inaccurately mapped wetlands as crossing over railroads. The new wetland polygon location is much more accurately defined as conforming to the landforms as illustrated by the solid gray lines. This paper will primarily focus on the east-central portion of Minnesota as illustrated in Figure 4. This is the metropolitan core for Minnesota representing Minneapolis, St. Paul, and the surrounding suburbs. From a railway perspective, this is the first major urban area intersected by Bakken oil trains from both BNSF and CP as they transition down the Mississippi River. Having updated wetland maps are very useful for expansion, accident cleanup and restoration efforts. Methods There are many more intensive approaches to mapping and surveying railroad right-of-ways such as terrestrial lidar, photogrammetric, and intensive GPS ground surveys of which we are not going to address. However, the approaches described here can be used for these types of intensive surveys to improve the accuracy and speed of the surveys. Image object mapping can be thought of as a digital approach to analog, stereo, aerial photo interpretation. Image object mapping software such as ecognition is used to develop rulesets to semi-automatically extract informational elements or objects based on their size, shape, shadow, color, height and context in the imagery. Segments are automatically generated to create the line work. This is a tremendous time savings and less expensive approach. Head s-up digital approaches (10) are still expensive and time-consuming. Original NWI mapping techniques are obsolete due to the lack of aerial analog film cameras and interpreters. The primary base aerial imagery used for the MNWI update was spring, leaf-off, digital aerial imagery with four spectral bands (red, green, blue, and near infrared) covering the metropolitan area. The aerial imagery was acquired using a Z/I DMC camera in early April of 2010 and late April to early May of Imagery for 60% of the project area was acquired at a spatial resolution of 30 cm, while imagery for the other 40% was acquired at 50 cm resolution. For the image segmentation process, the 30cm images were resampled to 50cm resolution using a bilinear interpolation algorithm. Satellite radar imaging systems have also been used for wetland mapping (11, 12, 13). These referenced works may be of interest since radar systems provide a more rapid approach for mapping water and oil features due to its ability to see through clouds. For this project, PALSAR L-band satellite radar images were acquired to cover AREMA

5 the project area to aid in the identification of forested wetlands. The scenes available were a combination of single and dual polarization during a leaf-off seasonal window. The Alaska Satellite Facility MapReady Remote Sensing Tool Kit was used for terrain correction and geo-referencing. Additional geo-referencing was performed in ArcGIS using control points selected from the aerial imagery. Radar imagery was classified using a 10-class maximumlikelihood ISODATA clustering routine implemented in ERDAS Imagine software. The classes associated with wet forest training sites were identified and the classification was applied to all clusters within the radar image. Lidar data were used to derive digital elevation models (DEMs) for about 60% of project area, while DEMs from 10-meter resolution DEMs were obtained from the USGS National Elevation Dataset for the rest of the project area. The typical lidar point spacing was about 1 point per square meter. The Minnesota Department of Natural Resources (DNR) processed the bare earth points into a digital elevation model using 3D Analyst for ArcGIS by importing the points into a terrain data set and then interpolating a 1-meter DEM that was subsequently resampled to a 3-meter DEM. ArcGIS Spatial Analyst was used to calculate slope, curvature, plan curvature, profile curvature, topographic position index - TPI and compound topographic index CTI (See Figure 5). Figure 5. MNWI lidar processing workflow to derive 7 GIS layers at 3m resolution for input into the rule set. TPI was calculated by subtracting the mean elevation for a given pixel from the mean elevation of its neighborhood (14). We used an annulus neighborhood with radii of 15 and 20 meters. The CTI (15) was calculated using a skinless version of the DEM. A slope grid and upstream catchment area grid were calculated using the D-Infinity flow directions tool from TauDEM (16). CTI was then computed from slope and contributing drainage area using a custom python script. SSURGO digital soil maps were also incorporated by extracting variables from the soil regime class and the percentage of hydric soils. Drainage class, flood and pond frequencies for April, and pond frequency for August were derived from these variables. AREMA

6 All of these layers were formatted into rule sets within ecognition image object software. All layers were clipped to the boundary as shown in Figure 4. Figure 6. Minnesota NWI Update Data Analysis Workflow showing the process to create wetland maps. Manual image interpretation is still required to edit objects for final map delineations as illustrated in the far right column. As seen in Figures 5 & 6, the workflow is a complex geospatial process. Further details about the entire process can be obtained by referring to reference (8). AREMA

7 Results Figure 7. Image Object mapping of wetlands as shown with the spring aerial image in the upper left, shaded relief in the upper center image, initial image segmentations in the upper right, the refined image objects in the lower left, with the final wetland inventory map on the lower right. White areas in the last image are upland areas. Figures 7 best illustrates steps of extracting wetland classes derived from the various datasets. By automating initial segmentation and delineation of wetland boundaries with initial identification of broad wetland classes, we were able to allow the image interpreters to focus more of their efforts on the most difficult components of the process, such as the assignment of detailed NWI wetland classes and modifiers (17). This project adapted automation approaches developed at the University of Minnesota for use in map production over large areas (18). There is more time invested in setting up the rulesets for processing but the time (labor) saved for large area mapping shows a net gain in efficiency, reduced costs, higher accuracies and more detailed classifications. Table 1. MNWI compared to the Original NWI Accuracy Assessment Original NWI Updated NWI Feature Accuracy Field 75% 90% Image-interpreted 76% 93% Class Accuracy Field 53% 72% AREMA

8 Image-interpreted 52% 78% Table 1. This table shows the accuracy comparison of the original NWI as compared to the updated Minnesota NWI. Both field and image-interpreted ground truth is compared to the original and newly mapped projects. Feature accuracy is best described as wetland vs. non-wetland. Class accuracy is a step more detailed distinguishing forested, shrub, or emergent wetlands for example. Further details can be found on the web in the Resources section. Table 1 shows the accuracies gained as compared to the original NWI. Gains of 15% using field checks (90% overall accuracy) and 17% for image interpreted ground checks (93% over all accuracy) from the original NWI were gained for defining wetlands vs non-wetlands which is the Feature Accuracy. Gains of 19% from 53% (NWI) to 72% (MNWI) compared to field checks and gains of 26% from 52% (NWI) to 78% (MNWI) for image interpreted ground checks for defining specific classes of such as forested, shrub, or emergent wetlands. Further class definitions can be found on the internet listed in the Resources section below. Our results showed that when compared to current field data we achieved a 15% increase in wetland-upland discrimination and a 19% increase in wetland class accuracy. With the limited funding for these types of mapping efforts, additional work is needed to continue to increase the efficiency of wetland mapping, while at the same time producing results that meet the needs of environmental engineers. CONCLUSION Our wetlands and surface water systems are four dimensional. They change in area, and elevation over time. The time component can be a matter of minutes from a flash flood washing out bridges to years for change where slowly raising water levels are negatively impacting the Devils Lake region in North Dakota. Our railroad infrastructure performance is a result of many outside factors including weather and water. Our wetlands and surface water systems can severely impact both the design and maintenance of our railroads and our environment. More importantly, railroads need to know where our wetlands and water systems are flowing in order to clean-up after derailments. The results shown here is just for one project over a large area. Figure 8 illustrates the next project area already being developed for Northeast Minnesota using the image object approach. AREMA

9 Figure 8. This is a draft MNWI map of the Canadian National rail yard and roundhouse in Proctor, Minnesota. This area is part of the Northeast Minnesota NWI update project. The forested wetlands in Northeast Minnesota are one of the most difficult areas to map wetland features in the country. The image object approach provides a more current baseline but more importantly, this new approach can be replicated in time as new imagery is acquired to continually update the wetland and surface water maps. Using the image object approach has proven to be more accurate with finer delineations at a reduced cost than previous methods. Railroads need to take action so they know the current state of wetland and water systems impacting their rightof-ways. They also need to understand all the watersheds in which they cross in case of an accident. Any spill can negatively degrade fish and wildlife resources in wetlands and surface waters. Having current geospatial knowledge is essential for any cleanup and restoration. Railroads may wish to collaborate with other wetland mapping organizations to develop better and current wetland and surface water maps. The techniques presented here can be adopted to improve the geospatial infrastructure for their organization. The image object approach may also be useful for other types of railroad infrastructure mapping. Disclaimer The findings and conclusions in this article are those of the authors and do not necessarily represent the views of the U.S. Fish and Wildlife Service, the Minnesota Department of Natural Resources, or the University of Minnesota. Mention of any trade names does not constitute an endorsement of their products. Acknowledgements This paper represents an accumulation of highly technical approaches to mapping wetlands with collaboration and funding by a host of organizations over the past decade too numerous to mention here. However we do AREMA

10 specifically we wish to thank the Minnesota Environment & Natural Resources Trust Fund, the author s organizations, Ducks Unlimited, the Minnesota Pollution Control Agency, the Board of Soil & Water Resources, and the Metropolitan Mosquito Control District who all provided exceptional support in this endeavor. Resources The data can be viewed here with this ArcGIS online map viewer: The MNWI data can be obtained at the Minnesota Geospatial Commons: Further background material can be found at: References (1) BARR Engineering (2011). Railroad Grade Raise Planning and Feasibility Study: BNSF Mainline Track Raise Between Devils Lake and Churchs Ferry, North Dakota URL (2) Denver Post (2015) BNSF reopens track at ND site of oil train derailment, fire. May URL (3) StarTribune (2014). Railroad steams ahead with St. Paul rail yard expansion. Sept. 19, URL (4) MNDOT - Minnesota Department of Transportation. (2013). Northern Lights Express High Speed Passenger Rail Project from Minneapolis to Duluth, Minnesota URL (5) MNDOT - Minnesota Department of Transportation (2014). Passenger Rail Corridor Investment Plan and Tier 1 EIS. 45. URL (6) Quandel Consultants, LLC (2011). Reasonable and Feasible Passenger Rail Alternatives: Milwaukee-Twin Cities High-Speed Rail Corridor Program URL (7) Department of Interior (2014). Budget Justifications and Performance Information, Fiscal Year 2014 Fish and Wildlife Service (2014). URL (8) Kloiber, S.M., Macleod, R.D., Smith, A.J., Knight, J.F. and Huberty, B.J. (2014). A Semi-Automated, Multi- Source Data Fusion Update of a Wetland Inventory for East-Central Minnesota, USA. Wetlands. 35 (2), (9) Baatz, M., and Schäpe, A. (2000). Multiresolution segmentation: An optimization approach for high quality multi-scale image segmentation. Angewandte Geographische Informationsverarbeitung XII (10) Drazkowski, B., May, M., and Herrera, D.T. (2004). Comparison of 1983 and 1997 Southern Michigan National Wetland Inventory Data. Michigan Department of Environmental Quality, Geological and Land Management Division. 15 pp. (11) Corcoran, J.M, Knight, J.F., Brisco, B., Kaya, S., Cull, A., Murhnaghan, K. (2011) The integration of optical, topographic, and radar data for wetland mapping in northern Minnesota. Canadian Journal of Remote Sensing, 27(5): (12) Brisco, B., Touzi, R., Van der Sanden, J., Charbonneau, F., Pultz, T., and D Iorio, M. (2007). Water resource applications with Radarsat-2. International Journal of Digital Earth, 1(1): AREMA

11 (13) Bourgeau-Chavez, L.L., Kowalski, K.P., Carlson Mazur, M.L., Scarbrough, K.A., Powell, R.B., Brooks, C.N., Huberty, B.J., Jenkins, L.K., Banda, E.C., Galbraith, D.M., Laubach, Z.M., Riordan, K. (2013). Mapping invasive phragmites Australis in the coastal Great Lakes with ALOS PALSAR satellite imagery for decision support. Journal of Great Lakes Research 39(1):65 77 (14) Guisan, A., Weiss, S. B., and Weiss, A. D. (1999). GLM versus CCA spatial modeling of plant species distribution. Plant Ecology, 143(1), (15) Moore, I.D., Grayson, R.B. and Ladson, A.R. (1991). Digital Terrain Modelling: A Review of Hydrological, Geomorphological, and Biological Applications. Hydrological Processes, 5:3-30. (16) Tarboton, D. G. (2003). Terrain analysis using digital elevation models in hydrology. In 23rd ESRI international users conference, San Diego, California (Vol. 14). (17) Cowardin, L.M., Carter, V., Golet, F.C. and LaRoe, E.T. (1979). Classification of Wetlands and Deepwater Habitats of the United States. U.S. Fish and Wildlife Service Report No. FWS/OBS/-79/31.Washington, D.C. (18) Rampi, L.P., Knight, J.F., and Pelletier, K.C. (2014). Wetland mapping in the Upper Midwest United States: An object-based approach integrating lidar and imagery data. Photogrammetric Engineering and Remote Sensing. 80(5), AREMA