|Mueller Warrant, George|
Submitted to: International Journal of Applied Earth Observation and Geoinformation
Publication Type: Peer Reviewed Journal
Publication Acceptance Date: 1/5/2015
Publication Date: 7/31/2015
Citation: Mueller Warrant, G.W., Whittaker, G.W., Banowetz, G.M., Griffith, S.M., Barnhart, B.L. 2015. Methods for improving accuracy and extending results beyond periods covered by traditional ground-truth in remote sensing classification of a complex landscape. International Journal of Applied Earth Observation and Geoinformation. 38:115-128.
Interpretive Summary: Obstacles that have previously limited researchers' ability to generate high quality remote sensing classifications of crops and other landuses of interest have included high costs for imagery, the general challenge of collecting necessary ground-truth data, and shortage of classification techniques robust enough to produce highly accurate results even in the face of problems with clouds, scan-line gaps, and ground-truth errors. The USGS decision in 2008 allowing free access to Landsat archives greatly lessened the first problem, leaving availability of ground-truth data and accuracy/reliability of classification procedures as the major remaining problems. In order to provide a better framework for monitoring and modeling crop production inputs and outputs, we have collected detailed landuse data for the diverse landscape of western Oregon going back to 2004. Successful use of remote sensing classification data in modeling of crop production inputs requires a high level of classification accuracy. The series of steps taken to achieve success in this endeavor included methods to effectively merge results from separate classification runs within single years, aggregation techniques substantially improving accuracy by recognition of field boundaries, and multiyear tests designed to remove erroneous results when classifications in one year are in conflict with those of other nearby years. Effectiveness of collaborations with other researchers interested in a bevy of environmental issues will benefit from our success in classifying 19 annually disturbed crops, 20 established perennial crops, 13 forests and other natural landscape elements, and 5 urban development conditions. We also succeeded in using current year ground-truth data to classify prior year crops, a concept we plan to eventually extend back in time through the entire Landsat archive period for western Oregon. Our techniques should apply in any landscapes where a majority of landuses persist from one year to the next, either because they represent established perennial stands or because they are annually disturbed crops grown on the same fields for multiple years.
Technical Abstract: Successful development of approaches to quantify impacts of diverse landuse and associated agricultural management practices on ecosystem services is frequently limited by lack of historical and contemporary landuse data. We hypothesized that recent ground truth data could be used to extrapolate previous or future landuse in landscapes where cropping systems do not generally change greatly from year to year. Our first objective was to classify 57 major landuses in the Willamette Valley of western Oregon from 1995 through 2011, elaborating on previously published work and traditional sources such as Cropland Data Layers (CDL) to more fully include minor crops grown in the region. Our second objective was to test the hypothesis that ground-truth data from one year could be useful in identifying landuse in previous or subsequent years within a complex landscape where the majority of crops are established perennials or the same annual crops grown on the same fields for multiple years. Available remote sensing data included Landsat, MODIS 16-day composites, and National Aerial Imagery Program (NAIP) imagery, all of which were resampled to a common 30 m resolution. The frequent presence of clouds and the nearly ubiquitous presence of Landsat7 scan line gaps forced us to conduct of series of separate classifications in each year. We found that merging results based on accuracy of the individual run partial area classifications was often equivalent to merging them by simply maximizing the number of cloud- and gap-free bands used in the final classification of any given pixel. Procedures adopted to improve accuracy beyond that achieved by maximum likelihood pixel classification included majority-rule reclassification of pixels within 91,442 Common Land Unit (CLU) polygons from 2004, smoothing and aggregation of areas outside the CLU polygons, and majority-rule reclassification over time of forest and urban development areas. Final classifications in all seven years separated annually disturbed agriculture, established perennial crops, forest, and urban development from each other at 90 to 95% overall 4-class validation accuracy. In the most successful use of subsequent year ground-truth data to classify prior year landuse, an overall 57-class accuracy of 75% was achieved despite the omission of 10 entire classes, most of which were annually disturbed or perennial crops grown on very few fields. Synthetic ground-truth data for the 2003-2004 cropping year based on the most common landuse classes over the following 7 y classified 49 of 57 categories with overall accuracy of over 96% for a final version that included CLU polygon majority rule, default smoothing and aggregation, and forcing of urban development and forest from multiyear majority rule as methods improving classification accuracy.