Graduate Project #1: Fire Vegetation Mapping (G.R.U.N.G.E.)
Summary:
High-resolution vegetation mapping for county fire operations support was accomplished using photointerpreted vegetation manuscripts derived from black-and-white, and color infrared imagery. The main focus of the effort was to produce a high-accuracy Image-to-Vector conversion package that could be run by non-expert volunteers. A production oriented Arc Macro Language (AML) library was developed and made available via the internet. The base imagery furnished by the county was getting old, so a routine to update the vegetation coverage with remotely sensed data (LANDSAT) was developed. This project made use of 22 unpaid students who helped test AML code, cleaned up digitally scanned manuscripts, and produced high accuracy Arc/Info coverages with the Image-to-Vector program. Students received class credit if they met reporting and productivity requirements.Accomplishments:
At the end of the project, about twenty-six standard 7.5' quadrangles had been processed. Mistakes in the manuscripts were identified. A geospatial library of semi-automated, and well-documented, AML routines had been developed. The routines had excellent error trapping features, especially for attributing. Minimal attributing effort was needed for these coverages to become complete. Maps were identified that needed further interpretation or error resolution to be useful. The semi-automated process had been improved to the point where complete maps could be converted from manuscript to digital coverages in a few hours. Most of this time was spent in the tedious correction of small anomalies found in polygon corners. (Later, ESRI improved on this technique) Landcover updates from LANDSAT imagery was accomplished, and tested on three quadrangles. Some small sample areas were ground-truthed as part of a graduate class exercise. A fuzzy-logic routine for updating coverages was found to be subjective, but very useful. The updating routines clearly showed where there were errors on the original vegetation manuscripts, usually due to recent urban development, land conversion, creation of roads, and fire. Some manuscript attributing mistakes were found. A number of students received class credit for their dilligent efforts.Image-to-Vector programs developed by vendors with large resources eclipsed our programming effort by a substantial degree during this time. However, the project was an excellent way to get a real-world feel for GIS programming and production issues not covered by classes at our university. For my professional development, it was an essential step in learning to work with non-technical natural resource employees, and learning something about the human aspects of implementing production GIS techniques.
Issues:
We had a number of issues that caused difficulty. Few volunteers were able to complete the requirements for class-credit, yet consumed training time. The manuscripts had defects that could not be resolved without further consultation with the authors. The manuscript authors and metadata on the original images were not available. It was clear that much of the imagery used for the manuscripts was older than fifteen years, and had not been updated. Spot checking of the manuscripts revealed serious differences in the accuracy of interpretation. (This is not a criticism of the work performed, just an observation of the inherent limitations in aerial interpretation) Field verification and sampling was prohibitively expensive without funding. 56 volunteer person-days of wilderness field data collection was lost by an agency computer failure.This project was ended after two years, when it was clear that financial resources would not be acquired by the project hosts. Some landuse managers apparently believed that Federally-funded vegetation coverages would be sufficient for County fire uses. In addition, a vegetation map had been created by the Department of Wildlife which covered part of this area, and promised to be of high quality.
(Update: A project using the manuscript data was later funded and partially completed by an outside vendor at substantial expense. We do not know how the missing data elements were acquired; however, the resulting product was described as 'low quality', and did not have updates from remote sensing techniques.)