Semi-automated GIS techniques for detecting floodplain earthworks

Levees, channels and water storages built on the world's floodplain wetlands control flows for irrigation, flood mitigation and erosion management. Assessing their distribution and hydrological impacts through time and across broad extents is limited by significant costs and technical challenges. We tested the effectiveness of three new semi-automated geographic information systems and traditional visual interpretation techniques for detecting earthworks. We used commercially or freely available two-dimensional and three-dimensional spatial imagery within 19 quadrats in an agricultural floodplain of the MurrayDarling Basin, southeastern Australia. Semi-automated digital elevation model (DEM) analysis performed best for spatial accuracy (78% of earthworks correctly predicted within 25?m), overall classification accuracy (97.7%) and kappa (0.64), compared with traditional visual interpretation techniques using Landsat TM (52%, 96.3%, 0.39), SPOT (53%, 95.8%, 0.27) and aerial photography (72%, 97.2%, 0.31). DEM analysis also outperformed semi-automated image segmentation (16%, 93%, 0.29) and integrated analysis (75%, 96.0%, 0.43) that used spectral information. Semi-automated techniques were slow (DEM analysis: 27?418?s/km2; integrated analysis: 27?737?s/km2; and image segmentation: 1439?s/km2) compared with visual interpretation (Landsat TM: 109?s/km2; SPOT: 166?s/km2; and aerial photography: 276?s/km2); however, processing speed of semi-automated techniques can be further increased without compromising accuracy. Semi-automated techniques also offered operational autonomy following model calibration. High quality, cost-effective earthwork mapping techniques, particularly the semi-automated techniques in this study, are critical for understanding and managing ecosystem health, flood risk and water security in developed floodplains worldwide and should be implemented by governing institutions. Copyright (c) 2012 John Wiley & Sons, Ltd.