Final Report: Application of Remotely-sensed Data To Regional Analysis and Assessment of Stream Temperature in the Pacific NorthwestEPA Grant Number: R827675
Title: Application of Remotely-sensed Data To Regional Analysis and Assessment of Stream Temperature in the Pacific Northwest
Investigators: Burges, Stephen J. , Booth, Derek B. , Gillespie, Alan R.
Institution: University of Washington - Seattle
EPA Project Officer: Packard, Benjamin H
Project Period: April 1, 2000 through March 31, 2003 (Extended to March 31, 2004)
Project Amount: $998,395
RFA: Regional Scale Analysis and Assessment (1999) RFA Text | Recipients Lists
Research Category: Ecosystems , Ecological Indicators/Assessment/Restoration
The objectives of this research project were to: (1) investigate the usefulness of thermal infrared (TIR) remote sensing, particularly the newly available Advanced Spaceborne Thermal Emission and Reflection Radiometer (ASTER) images from the latest generation earth observing satellite, Terra, launched in December 1999, as a tool for assessing regional stream temperatures; and (2) demonstrate how these methods can be applied to assess the effects of land use on stream temperature. To achieve these objectives, this project made use of ground-, aircraft-, and satellite-based observation methods. Ground-based networks of instream temperature data loggers, and volunteers taking surface temperature measurements, were used both to study the temperature regime of four Pacific Northwestern River systems, and to provide ground-truth data for the TIR images that were collected in support of this project.
This report presents findings first from the instream observations, as this serves as the baseline for all further measurements. It then describes the TIR data that were collected and the work done to produce the most accurate maps of stream temperatures possible in a way that can be replicated by other researchers. The report then discusses our ability to resolve regional streams given the resolution of the sensors used in this study. Finally, preliminary assessments of how land use effects can be identified from the TIR images are provided.
Instream observation methods, such as the network of instream temperature data loggers (see Figure 1) installed for this project, provide fine-scale temporal observations of point temperatures, but are limited in their ability to monitor spatial patterns of temperature. As these instream methods generally are installed in areas with good access to the stream, important controls on watershed temperatures may be missed. We hypothesized that space-based TIR sensors had the potential for monitoring the spatial patterns missed by the instream methods, while capturing more of the temporal patterns not observed by the growing array of aircraft-based high-resolution TIR sensors being used for short-time windows to image many streams in the Pacific Northwest. Our principal findings are based on extensive analysis of instream water temperature measurements for the Yakima River in eastern Washington, and for the Green River and tributaries in western Washington (see Figure 1).
Figure 1. Locations of In Situ Water Temperature Data Loggers in (a) the Green and Cedar River Watersheds (Inset is Enlarged View of Soos Creek Watershed); and (b) the Yakima River Watershed (Inset Shows Location of All Three Watersheds in Washington State). (From Cherkauer, et al., 2003.)
The ASTER imager on the National Aeronautics and Space Administration's Terra platform is the most highly advanced, publicly available TIR sensor currently in orbit. Its orbit provides for the possibility of obtaining repeat looks every 16 days. Unlike the Landsat ETM+ sensor, which images the same swath of the surface each time it passes overhead, the ASTER sensor must be programmed and pointed at a target on each orbit to take an image. This means that repeat images are not guaranteed even with clear skies. Additionally, problems scheduling image captures meant that an ASTER image was collected in conjunction with our intensive ground-truth surveys only once. Therefore, Landsat ETM+ images were included in this study to increase the number of comparisons available for testing satellite-based sensors. The aircraft-based Moderate Resolution Imaging Spectroradiometer (MASTER) sensor was used to collect higher resolution (5 m- and 15-m pixels) to study the effects of scaling.
Atmospheric and emissivity corrections have to be applied to the raw TIR images. These corrections generally improved agreement between in situ surface water temperature measurements. Before corrections were applied, only 33 percent of the TIR-derived river water temperatures fell within ± 1°C of the instream observations. After corrections had been applied, 100 percent of the TIR-derived temperatures were within ± 1°C of instream observations. The percentage of TIR-derived temperatures within ± 1°C of the instream temperature data loggers decreased from 89 percent to 0 percent after correction, implying that these relatively shallow (on the order of 1-m deep) streams are not well mixed vertically as was originally assumed. In general, these tests indicate that TIR-derived water temperatures can achieve the desired accuracy versus surface water temperatures; however, the number of useable comparisons was small, meaning these results are not definitive.
Figure 2. (a) Average Stream Centerline Temperature From Corrected TIR Images in a 1-km2 Area Around an Instream Observation Minus That Observation Plotted Versus the Number of Pixels Needed To Represent Stream Width, and (b) the Uncertainty in TIR-Derived Temperatures for the Same 1-km2 Area. (From Handcock, et al., submitted, 2004.)
The limiting factor for being able to use satellite-based TIR sensors for monitoring regional stream temperatures is pixel resolution. We show that the agreement between TIR-derived and instream-observed water temperatures significantly improves when the TIR sensor resolution is such that three or more pixels represent the width of the stream channel (see Figure 2). Three pixels means that stream bank effects (i.e., warmer temperatures) generally are incorporated into the two-edge pixels, while the center pixel is much more likely to represent the thermal signal from water without any thermal signal added from the river banks and, therefore, will be more accurate. Increasing the image resolution and, therefore, the number of pixels representing stream width does not appear to improve accuracy beyond what is obtained with three pixels representing the width.
This behavior holds for all of the sensors tested. It is possible that images with resolutions of less than 5 meters could lead to improved accuracy by unmixing midstream contamination (such as rocks and large woody debris), but this also would increase the processing time significantly, as these objects need to be identified. Robust automatic classification schemes for such identification remain elusive.
The three-pixel limit on accuracy yields concrete numbers for when specific sensors would be useful. The ASTER sensor, with a TIR resolution of 90 m, is limited to minimum river widths of 270 m. Landsat ETM+, with a resolution of 60 m, is limited to minimum river widths of 180 m. Based on a survey of river reach widths in the State of Washington, ASTER and Landsat ETM+ could be used to monitor only 6 percent of all current Total Maximum Daily Load (TMDL) sites. Increasing TIR resolution to 15 m increases the number of reaches that can be adequately monitored to 12 percent. Until resolutions reach 5 m or less, satellite-based imagers will be minimally useful for monitoring regional stream temperatures and TMDL requirements.
Because of the limitations of space-based pixel resolution, we focused the assessment of effects of land-use change on water temperatures using TIR images on the 5- and 15-m resolution MASTER imagery as well as the instream network of temperature data loggers. Our preliminary results indicate that some land use impacts can be identified from the TIR images. We are conducting ongoing extensive analysis (beyond the scope of this project) of these data.
In the Yakima River Basin, river water temperatures are relatively unchanged over the 10 km between the city of Ellensburg, WA, and where the river enters the Yakima River Canyon. This is a closed ground water basin that is heavily irrigated. Ground water return flows are a significant source of warm water during the winter in this basin. Ground water flow moderates summer river water temperatures, as water temperatures increase steadily once the river enters the canyon, where additional inflows are minimal.
Water temperatures in the Green River are relatively high after leaving the Howard Hanson Reservoir, but cool as the river passes through the National Forest, lightly developed private forested lands, and the Green River Gorge (see Figure 3). Water temperatures begin to increase again once the water leaves the gorge and begins to meander on the lower slopes of its middle reach. Agricultural uses do not appear to contribute significantly to water temperatures here, probably because of the well-developed riparian buffer. Once the river enters the more heavily urbanized and channelized lower reach, temperatures increase to the same levels as when they left the reservoir.
Analysis of surface temperature variability, specifically for the Green River (see Figure 3), may prove to be more useful in assessing the impact of land use on water temperatures. This became apparent as we explored supplemental data gathering in support of the main project mission. This post-project analysis has shown that areas with more limited development have higher spatial (along-stream) temperature variability than urban areas or those areas directly downstream of the reservoir.
Figure 3. Green River Zonal (a,b) Average Temperatures and (c,d) Standard Deviations Based on 5 m Resolution MASTER Images From August 25, 2001. Temperatures and standard deviations plotted over (a,c) geology and (b,d) land use classifications.
Analysis of space- and aircraft-based TIR sensors indicates that sensor technology and relatively simple methods for correcting images for surface emissivity and atmospheric effects result in TIR-derived temperatures within ± 1°C of observed instream surface temperatures. The ability to resolve the stream itself is the limiting factor for using satellite-based TIR imagery to monitor regional stream temperatures. Accuracy is best when the image represents stream widths with three or more pixels, so that bank effects can be removed from the analysis by removing all but the centerline pixels. At the resolution of the current generation of satellite-based TIR sensors, no more than 6 percent of the river and stream reaches in Washington with active temperature TMDLs can be resolved. Therefore, satellite-based TIR imagery currently is not useful for the development or monitoring of regional TMDLs.
Resolution limitations also preclude using satellite-based TIR imagery to investigate the effects of land use on regional stream temperatures. Higher resolution aircraft-based TIR imagery was used in this study and was capable of identifying more fine-scale spatial changes in water temperatures that may be indicative of land use. Further study is underway to determine if additional information on the impacts of land use can be extracted by studying changes in spatial variability.
Journal Articles on this Report : 2 Displayed | Download in RIS Format
|Other project views:||All 20 publications||2 publications in selected types||All 2 journal articles|
||Cherkauer KA, Burges SJ, Handcock RN, Kay JE, Kampf SK, Gillespie AR. Assessing satellite-based and aircraft-based thermal infrared remote sensing for monitoring Pacific Northwest river temperature. Journal of the American Water Resources Association 2005;41(5):1149-1159.||
||Handcock RN, Gillespie AR, Cherkauer KA, Kay JE, Burges SJ, Kampf SK. Accuracy and uncertainty of thermal-infrared remote sensing of stream temperatures at multiple spatial scales. Remote Sensing of Environment 2006;100(4):427-440.||