No abstract

No abstract

Abstract:An overview is given on the predictive quality of spatially distributed runoff and erosion models. A summary is given of the results of model comparison workshops organized by the Global Change and Terrestrial Ecosystems Focus 3 programme, as well as other results obtained by individual researchers. The results concur with the generally held viewpoint in the literature that the predictive quality of distributed models is moderately good for total discharge at the outlet, and not very good for net soil loss. This is only true if extensive calibration is done: uncalibrated results are generally bad. The more simple lumped models seem to perform equally well as the more complex distributed models, although the latter produce more detailed spatially distributed results that can aid the researcher. All these results are outlet based: models are tested on lumped discharge and soil loss or on hydrographs and sedigraphs. Surprisingly few tests have been done on the comparison of simulated and modelled erosion patterns, although this may arguably be just as important in the sense of designing anti-erosion measures and determining source and sink areas. Two studies are shown in which the spatial performance of the erosion model LISEM (Limburg soil erosion model) is analysed. It seems that: (i) the model is very sensitive to the resolution (grid cell size); (ii) the spatial pattern prediction is not very good; (iii) the performance becomes better when the results are resampled to a lower resolution and (iv) the results are improved when certain processes in the model (in this case gully incision) are restricted to so called 'critical areas', selected from the digital elevation model with simple rules.The difficulties associated with calibrating and validating spatially distributed soil erosion models are, to a large extent, due to the large spatial and temporal variability of soil erosion phenomena and the uncertainty associated with the input parameter values used in models to predict these processes. They will, therefore, not be solved by constructing even more complete, and therefore more complex, models. However, the situation may be improved by using more spatial information for model calibration and validation rather than output data only and by using 'optimal' models, describing only the dominant processes operating in a given landscape.

Runoff may be reduced by temporal water storage in depressions at the soil surface. Depressional storage is difficult to measure and is usually estimated from some roughness index. The objective of this study was to compare the ability of selected roughness indices to describe maximum depressional storage (MDS). Height measurements were taken on 221 tilled soil surfaces across a range of roughnesses. Maximum depressional storage was determined from digital elevation models (DEMs). The MDS values ranged from 0 to 13 mm. Five roughness indices were calculated from transects across these DEMs: random roughness (RR), tortuosity (T), limiting elevation difference (LD) and slope (LS), and mean upslope depression (MUD). Regression analysis of MDS on each of five roughness indices showed that RR best described depressional storage Prediction of MDS in the field based on RR has an uncertainty of ± 3 mm (95% confidence interval). Variation was due to RR and its nonspatial nature. To improve predictions of MDS, the spatial configuration of the surface has to be taken into account.

To detect landslides by object-based image analysis using criteria based on shape, color, texture, and, in particular, contextual information and process knowledge, candidate segments must be delineated properly. This has proved challenging in the past, since segments are mainly created using spectral and size criteria that are not consistent for landslides. This paper presents an approach to select objectively parameters for a regiongrowing segmentation technique to outline landslides as individual segments and also addresses the scale dependence of landslides and false positives occurring in a natural landscape. Multiplescale parameters were determined using a plateau objective function derived from the spatial autocorrelation and intrasegment variance analysis, allowing for differently sized features to be identified. While a high-resolution Resourcesat-1 Linear Imaging and Self Scanning Sensor IV (5.8 m) multispectral image was used to create segments for landslide recognition, terrain curvature derived from a digital terrain model based on Cartosat-1 (2.5 m) data was used to create segments for subsequent landslide classification. Here, optimal segments were used in a knowledge-based classification approach with the thresholds of diagnostic parameters derived from K-means cluster analysis, to detect landslides of five different types, with an overall recognition accuracy of 76.9%. The approach, when tested in a geomorphologically dissimilar area, recognized landslides with an overall accuracy of 77.7%, without modification to the methodology. The multiscale classification-based segment optimization procedure was also able to reduce the error of commission significantly in comparison to a single-optimal-scale approach.Index Terms-Disaster support, feature extraction, India, K-means cluster analysis, object-oriented analysis (OOA), segmentation.