Supervised Landform Classification to Enhance and Replace Photo-Interpretation in Semi-Detailed Soil Survey

of stereoscopic landscape analysis, and then determine the soil types that occur in each map unit by field inspecA method to enhance manual landform delineation using phototion of the soil at representative sites. A common inspecinterpretation to map a larger area is described. Conventional aerial tion density is one observation per one to four map photo-interpretation (API) maps using a geo-pedological legend of 21 classes were prepared for six sample areas totaling 111 km2 in the centimeters squared (Western, 1978, Table 3.3), which Baranja region, eastern Croatia. Nine terrain parameters extracted at this scale represents 25 to 100 ha. Often, the surveyors from a digital elevation model (DEM) (ground water depth, slope, make sure that there is also at least one observation plan curvature, profile curvature, viewshed, accumulation flow, wetper each polygon. The observations are used to characness index, sediment transport index, and the distance to nearest terize the composition of photo-interpretation units, watercourse) were used to extrapolate photo-interpretation over the rather than to find or adjust every boundary. entire survey area (1062 km2). The classification accuracy was assessed One approach to semi-detailed survey is to study repusing the error matrix, calculated by comparing both the whole API resentative sample areas, typically covering about 10% maps and point samples, with the results of classification. The first of the survey area, more intensively to arrive at a better results, using a maximum-likelihood classifier, were 58.2% (hill land), understanding of the soil-landscape relations and map 39.1% (plain), and 45.3% (entire area) reproducibility of the training set. Six classes in the plain were responsible for a large proportion unit composition. Field sampling is thus concentrated of the misclassifications, due to an insufficiently detailed DEM and in comparison with the densities mentioned above, to the complex nature of landforms (point bar complexes, levees, active an observation density of one per 2.5 to 10 ha in the channel banks), which cannot be explained with the terrain parameters sample area. This is at the cost of samples over the rest only. Reproducibility for a simplified legend of 15 classes over the of the area, which is then mapped purely by photostudy area was improved to 65.8% (plain), 58.2% (hill land), and interpretation, extrapolating from the detailed under63.4% (entire area) using the whole-API training set. After the simplistanding of the soil landscape built up in the sample fication of legend (15) and with the iterative (3) selection of pointareas. Because of the low inspection density, the only sample training set, classification was able to reproduce 97.6% (hill way that such maps can be reasonably accurate is if land), 86.7% (plain), and 90.2% (entire area) of the training set. The the surveyor is able to correctly understand the soilsupervised classification showed fine details not achieved by photointerpretation. The number of manual photo-interpretations that had landscape relations in the survey area, and then map to be prepared was reduced from 84 to 6. The methodology can be these by surface features visible on the aerial photo applied by soil survey teams to edit and update current maps and to (e.g., the landform as seen stereoscopically). enhance or replace API for new surveys. Several systematic approaches to soil-landscape photointerpretation have been developed. In this study, we use the “geo-pedological” method of Zinck (Zinck, 1988; A of the semi-detailed soil survey is an Zinck and Valenzuela, 1990), which explicitly relates entity-class or polygon map of soil types at a typical landform elements to predictions of soil classes and scale of 1:50 000, with minimum legible delineations of properties. In many cases, however, including standard 10 ha and optimal delineations of 40 ha (Forbes et al., mapping procedures in the USA, surveyor’s experience 1982). This corresponds to “Order 3” to “Order 4” (Soil on soil-landscape relations is used without formalization Survey Division Staff, 1993, Table 2-1), semi-detailed (Soil Survey Division Staff, 1993, p. 219–231). Jenny’s or medium intensity soil surveys (Avery, 1987, Table 2). conceptual equation (Jenny, 1980), that is, soil property These soil maps are intended for extensive land-use or class f(climate, organism, relief, parent material, planning and to give a reasonably accurate picture of time) is thus used subjectively (and sometimes subconthe distribution of soil types in an area at relatively low sciously) as a concept to guide photo-interpretation. cost. The standard method of semi-detailed survey is to Depending on the survey area, some aspects of the equadraw preliminary boundaries on aerial photos by means tion may be more important than others; for example, on a typical hillside, the catena or topo-sequence concept may be uppermost in the surveyor’s mind, whereas T. Hengl, Dep. of Earth System Analysis, International Institute for in areas of recent deposition, parent material and time Geo-Information Science & Earth Observation (ITC), P.O. Box 6, may be more important. In many areas, landform classi7500 AA Enschede, The Netherlands. D.G. Rossiter, International Institute for Geo-Information Science & Earth Observation (ITC), fication or segmentation, usually by photo-interpretaP.O. Box 6, 7500 AA Enschede, The Netherlands. Web: http://www. itc.nl/personal/rossiter. Supplementary materials, datasets, additional Abbreviations: GIS, geographic information system; DEM, digital results and color graphics available online at: http://www.itc.nl/library/ elevation model; API, aerial photo-interpretation; ILWIS, Integrated Academic_output/2003/. Received 18 May 2002. *Corresponding auLand and Water Information System; GWD, ground water depth; thor ([email protected]). SLOPE, slope gradient; PROFC, profile curvature; TANGC, plan curvature; VSHED, viewshed reflectance; FLOW, accumulation flow; Published in Soil Sci. Soc. Am. J. 67:1810–1822 (2003).  Soil Science Society of America CTI, Compound Topographic Index; STI, Sediment Transport Index; DISTW, distance to nearest watercourse; PC, principal component. 677 S. Segoe Rd., Madison, WI 53711 USA