PHOTOGRAMMETRIC ENGINEERING AND REMOTE SENSING
Vol. 56, No. 3, March 1990, pp. 365-370.
© 1990 American Society for Photogrammetry and Remote Sensing
Reproduced with permission.

 Using Remote Sensing and GIS Technology to Help Adjudicate Idaho Water Rights

Anthony Morse, Thomas J. Zarriello, and William J. Kramber
Idaho Department of Water Resources, Boise, ID 83720

ABSTRACT: The Idaho Department of Water Resources (IDWR) is using remote sensing and GIS in the nation's largest water rights adjudication. Landsat MSS images are geometrically controlled, transformed to principal components, clustered, and classified to identify six information classes: irrigated agriculture, dryland agriculture, non-agricultural land, riparian vegetation, water, and clouds and/or cloud shadows. The Public Land Survey System (PLSS) is being converted to a digital format subdivided to the quarter-quarter section (QQ) level. The classified Landsat image is overlaid with the PLSS digital file, and the acreage of each land-cover class is computed for each QQ. The output products are Mylar film positives registered to 1:24,000-scale quads showing the land-cover classification, and tabular files listing the land cover acreage per QQ. These products are designed to assist water users in filing accurate water-right claims, and to help IDWR personnel in processing those claims. Classification accuracy is measured using regression analysis of Landsat-based irrigated acreage against acreage measured from USDA Agricultural Stabilization and Conservation Service aerial slides, for a random sample of irrigated sections. Results from the first five counties processed show an average r² of 0.90.

INTRODUCTION

The Idaho Department of Water Resources (IDWR) is in the process of adjudicating all the water rights in the Snake River Basin of Idaho. This is the largest water-right adjudication attempted in the United States and includes an estimated 140,000 water rights within about 72,000 square miles, as illustrated by Figure 1. The purpose of this paper is to explain the background of this project and discuss the methodology developed at IDWR to help facilitate this huge undertaking using remote sensing, image processing, and geographic information system technology.

The Snake River Plain, which supports most of Idaho's irrigated agriculture, covers approximately 18,000 square miles of and land. Over the last 130 years, irrigation projects have enabled farmers to bring progressively more acreage into production so that, by 1989, water is being pumped or diverted onto about 4 million acres. While Idaho farmers depend on Snake River water to supply them with three quarters of their irrigation supply, the Idaho Power Company depends on the same Snake River water for about 57 percent of its electrical generating capacity. These competing needs inevitably led to conflict, and, in 1982, the Idaho Supreme Court decided "Idaho Power Co. vs. The State of Idaho, et al." The issues were complex, as is typical of western water-right controversies.

Costello and Kole (1985) describe the case in detail, but it can be summarized as follows. The Court upheld the Idaho Power Company's contention that its early 1900s water right for 8,400 CFS at the Swan Falls Dam, which is at the western (downstream) end of the Snake River Plain, was not necessarily subordinate to the water rights of upstream irrigators. With the stroke of a pen, water on the Snake River Plain went from partially appropriated to over-appropriated.

In responding to the decision, the Idaho Power Co. did not seek to confiscate any water being put to beneficial use. Instead, it filed suit against about 7500 holders of upstream permits and water-right applications for which beneficial use had not yet been proven. After two unsuccessful attempts by the State Legislature to resolve the conflict, and facing millions of dollars of costs for years of litigation, the Governor, the Attorney General, and the Idaho Power Co. negotiated a settlement that the Legislature ratified in 1985. The affected parties agreed to several points, one of which being that all water rights in the Snake River drainage must be adjudicated. The Idaho Legislature required that IDWR provide the presiding Court with all the technical information necessary for the Court to make a decision about each water right.

In order to comply with the Court's mandate, IDWR is using remote sensing and GIS technology to estimate the irrigated acreage associated with each water right (Morse et al., 1988). That estimate can be compared to the existing water rights file and to the claims submitted as part of the adjudication to help assess the accuracy of the claims.

PROCEDURES

The staff of the Idaho Image Analysis Facility (IIAF), a remote sensing/GIS section within IDWR, devised a three-step methodology for using image processing and GIS to help complete the adjudication in the planned ten years. For the first step, Landsat multispectral scanner (MSS) digital data are classified to produce land-cover maps with six classes, including irrigated land. Second, section corners of the Public Land Survey System (PLSS) are digitized from 1:100,000-scale, stable-base, Mylar maps, and the sections are subdivided into quarter-quarter sections, government lots, Bureau of Reclamation tracts, patented mining claims, and homestead entry claims, which are referred to here collectively as "QQs." QQs are used because the location of Idaho water rights is keyed to quarter-quarter sections. Third, the Landsat data are digitally overlaid with the corresponding subdivided PLSS data, and the irrigated acreage is computed for each QQ.

IMAGE PROCESSING

The image processing aspect of the project is based on digital analysis of Landsat MSS data using in-house software, clustering and classifying software from Spectral Software Associates, Inc., and other image processing software from ERDAS, Inc. The analysis included four steps: (1) geometric control, (2) principal component analysis, (3) unsupervised classification, and (4) post-classification sorting.

Fifteen Landsat MSS scenes are required to cover the Snake River Basin. IDWR purchased the scenes from EOSAT based on minimum cloud cover and dates that corresponded to peak agricultural crop maturity. The dates of 14 of the scenes ranged from 3 July 1986 to 23 August 1986. One scene dated 16 July 1984 was acquired because no acceptable 1986 or 1985 scene was available. EOSAT radiometrically and geometrically corrected all scenes, re-sampling pixels to 57 by 57 meters.

IIAF personnel geometrically referenced each full Landsat scene to the UTM coordinate system with an affine transformation. The process involved using the ERDAS programs GCP, COORD2, and RECTIFY. An analyst selected approximately 25 well distributed ground control points within each scene and on 1:24,000-scale USGS quads, set the error tolerance for both the x and y directions to +1.0 pixels, and, based on the results of Logan and Strahler (1979), chose bilinear interpolation to resample the images.

Each scene was transformed to its principal components as the second step in image processing. Principal component analysis (PCA) is a statistical procedure that transforms a set of data into a new system of axes based, in this case, on the variance-covariance matrix (Chatfield and Collins, 1980). PCA was used to reduce the dimensionality of the data, which in turn reduced both the volume of data to be processed and the CPU time needed to process them. The ERDAS program PRINCE transformed the geometrically corrected images based on transformation coefficients computed from a 7 by 7 sub-sample of image data. The principal components were not scaled in order to maintain their relative magnitudes of variance. PCA effectively reduced the four MSS bands to two components containing at least 98 percent of the variance for each scene processed. The remaining image processing steps used principal components one (PC1) and two (PC2).

Spectral Software Associates software clustered and classified county sub-scenes using histograms and look-up tables for extremely fast processing of ERDAS-format images. The clustering software is based on an iterative, converging algorithm (Forgy, 1965), and generates up to 255 clusters from all image vectors. The classifier uses a table look-up method similar to those described by Shlien (1975) and Wharton (1983). Selecting the number of spectral classes to generate in unsupervised classifications is often difficult and subjective depending on the algorithm used. IIAF analysts decided that, for a project of this scale, it was inappropriate to experiment with different numbers of spectral classes for each classification, and adopted the practical and consistent solution of generating the maximum number (255) possible for each classification.

The 255 spectral classes were identified and aggregated to five land-cover types: irrigated agriculture, dryland agriculture, non-agricultural land, riparian vegetation, water, and one class of clouds and/or cloud shadows. This was done, first, by generating a scatterplot of PCI and PC2 from the statistics file that stores the mean vector and variance-covariance matrix for each cluster. The cluster number from the statistics file was also written to the scatter plot. Next, PC1 and PC2, which approximate the brightness and greenness of Kauth and Thomas (1976), were read into the green and red image display planes, respectively. Then the ERDAS program CLASOVR wrote the classified image into the blue band. This allowed one or more spectral classes to be highlighted and labeled to one of the six information classes with the corresponding false-color image as background. This procedure allows efficient class identification and labeling.

Post-classification sorting is a common technique used in remote sensing to improve classification accuracy (Cibula and Nyquist, 1987; Hutchinson, 1982). For this project, an image interpreter delineated irrigated agriculture, dryland agriculture, and riparian vegetation using image interpretation techniques to interactively screen digitize these class boundaries with the ERDAS program DIGSCRN, while displaying a 512 by 512 sub-scene of the Landsat data. Areas outside of these boundaries were non-agricultural land. When the Landsat data were inconclusive, the interpreter analyzed U-2 color-infrared photographs to assist in drawing these boundaries. After all boundaries were completed, the vector file was converted to raster using the ERDAS program GRDPOL. An in-house program, TOL, overlaid the 255-class image with the file created from image interpretation using a GIS matrix operation to produce the final classification.

Using post-classification sorting reduces total classification error by preventing certain cases of commission error. For example, in many parts of Idaho, harvested fields of irrigated small-grains are spectrally similar to native rangeland. This results in spectral classes that represent both the irrigated-agriculture and non-agriculture information classes. Post-classification sorting allows each spectral class to be labeled to more than one information class and eliminates commission errors of irrigated agriculture from occurring in non-agricultural areas, thus increasing classification accuracy.

GIS PROCESSING

The first aspect of the GIS effort is digitizing the PLSS from stable-base Mylar maps at 1:100,000 scale. Under the best of circumstances, digitizing is an involved and time-consuming task, and this project is no exception. IIAF personnel decided to use 1:100,000-scale maps primarily because of the significant reduction in time needed for set-up and edge-matching over 1:24,000-scale maps, and because the accuracy of the 1:100,000scale maps was deemed adequate for this application. In the course of the adjudication, IIAF personnel will digitize about 2000 townships, and sub-divide them into a total of about 1,100,000 polygons.

A typical township has 36 regular sections, each of one square mile or 640 acres. Sections are divided into quarter-sections of 160 acres each. Quarter sections, in turn, are divided into quarter-quarter sections of 40 acres, so that a typical township has 36 X 4 X 4 = 576 quarter-quarter sections of 40 acres. If a township has QQs that are less than or more than 40 acres they are no longer QQs, but rather government lots, and are shown as such in the PLSS data-layer. Townships are normally lotted along their north and west boundaries, but, in fact, lotting can be found along any boundary of any section.

After the 49 section corners of each township are digitized, in-house software can automatically subdivide sections into QQs. The software calculates the UTM corner coordinates of each QQ and a corresponding attribute that includes the legal description and center-point coordinates. If a section is lotted regularly, the task is complete; if the lotting is irregular, more work is needed. Sub-dividing irregular sections is complex and time-consuming, but, fortunately, only about 20 percent of Idaho townships have irregular lotting.

Editing the irregularly lotted sections is done manually. The in-house program TOWNS sub-divides regular sections, leaving the irregular sections undivided and in a form ready for editing. An analyst enlarges and prints to paper the microfiche platmap of the irregular section, superimposes the township plot on the plat map, traces and digitizes the boundaries of the irregular lots, and adds the appropriate attributes. The analyst then appends the new data to the files produced for the regular sections of the township. At this point, the lotting of irregular sections is completely represented in the digital township-file. The job is very labor-intensive, but it produces data that very closely match the legal plat. Figure 2 illustrates part of a subdivided township that has both regularly and irregularly subdivided sections.

The second aspect of the GIS effort is to overlay the subdivided PLSS grid onto a Landsat classification. This overlay requires, first, a vector-to-raster conversion of the PLSS file. One problem with this process is that using the 57 by 57 meter pixel size of the Landsat MSS results in inaccurate acreage totals for QQs. Using a 14.25 meter pixel for the raster PLSS reduced the severity of this problem. Table 1, which summarizes the results of regression analysis using various pixel sizes, shows that a 14.25metre pixel more nearly estimates the correct QQ size than does a larger pixel size. At the same time, a pixel smaller than 14.25 meters is impractical due to hardware and software limitations on the volume of data. The in-house program TOWNCIVER overlays the PLSS raster file on the Landsat classification to produce a tabular file that lists the acreage of the six classes per QQ.

TABLE 1. THE COEFFICIENT OF DETERMINATION (r²), FROM REGRESSION
ANALYSIS, OF DIGITIZED POLYGON ACREAGE VERSUS RASTER ACREAGE
OF QUARTER-QUARTERS, FOR THREE PIXEL SIZES.

OUTPUT PRODUCTS

The two output products for the adjudication are Mylar overlays, registered to 1:24,000-scale orthophoto quads, and tabular files. Neither product is used as a court document. Rather, they provide the State with the means to assist claimants by offering a tabulation of irrigated acreage by QQ with the graphical display of familiar land features showing the irrigated acreage as of the 1986 base date. This helps both claimants and IDWR personnel in assuring that the legal description and number of irrigated acres for a water right are accurately represented on a water right claim. Figures 3a and 3b show six sections of an orthophoto quad with its associated classification overlay.

Table 2 is an example of part of a tabular file. The tabular file shows the irrigated acreage per QQ, which is compared to the state water right file and to adjudication water-right claims. This has two purposes. First, it will show if irrigated lands exist where no water right claims have been filed. Property ownership is researched to allow the State to give water users every opportunity to file a water right claim. Second, it provides an estimate of irrigated acreage that is independent of both the irrigator's claim and the recorded water right. If the claimed acreage and/or the recorded acreage differs from the Landsat estimate by a threshold, yet to be determined, the irrigator and IDWR can reconcile the difference using the Mylar overlays and corresponding orthophoto quadrangle maps, high altitude photography such as NHAP or NAPP, or, if necessary, by field verification. This process will allow IDWR to process routine, non-controversial claims quickly and efficiently while identifying those water right claims that most need the expensive attention of a field investigation.

TABLE 2. A PORTION OF THE OUTPUT TABULAR FILE SHOWING THE
ACREAGE By LAND-COVER CLASS FOR EACH QUARTER-QUARTER SECTION
AND GOVERNMENT LOT OF T2NR4WS36.