A geospatial approach of estimating vegetation roughness for flood modelling

2D hydrodynamic modelling has become a powerful tool to simulate the interaction between flow and floodplains to balance the environmental requirements and flood risks. However, vegetation roughness remains a major uncertainty. Although roughness is known to vary with depth, it is seldom implemented due to its intricacies. This research developed a practical method to estimate depth-varying vegetation roughness using GIS and remote sensing. Since high point density LiDAR is not widely accessible due to its cost, the low point density LiDAR data was combined with QuickBird satellite image using supervised and rule-based Object-based Image Analysis (OBIA) techniques to map the 14 km2 tropical vegetated floodplain in Malacca, Malaysia. The rule-based results showed an 8% improvement in the overall accuracy to 88.14% compared to the supervised classification. The McNemar results further demonstrated that the rule-based classification accuracy was highly significant compared to the supervised classification with 617 matches compared to 556 for supervised. It was shown that even with low point density, the nDSM derived from LiDAR still retains a good quality in order to improve the classification of paved surface as well as grass and cropland. Thereafter, a regression analysis was conducted for the trees and shrubs in combination with field measurements to estimate the vegetation widths with high correlations. Vegetation width is the main variable in calculating the vegetation density and consequently, the roughness coefficient. The derived canopy covers for the shrubs were found to be representative of the field measurements. The linear relationship of shrubs was found to be very strong at 0.98 and 0.95 for the Pearson correlation coefficient and R2, respectively. This implied that the shrub widths can be estimated based on the canopy covers as the widths are generally uniform throughout their heights and can be discriminated spatially. Therefore, it is assumed that the shrubs with 100% canopy cover to have the width equivalent to the plot width. On the other hand, the tree widths cannot be discriminated spatially due to the obstruction by the canopy. Accordingly, the relationship between the tree widths and NDVI was decided to be the best indicator. As a result, the tree widths can be calculated using a regression equation with the accuracy of 0.76 for Pearson correlation and 0.58 for R2. Consequently, ArcGIS routines were developed to automate the methodology to conveniently transfer to other ArcGIS interfaces for other study areas and datasets. ArcGIS routines can generate roughness maps at any preferred spatial resolution for 2D hydrodynamic modelling input. The calculated depth-varying vegetation roughness coefficients were subsequently compared against the literature. These values were plotted against the calculated Manning’s roughness coefficients which were very close to the range of experimental values of 0.055 to 0.180. A minimum value of 0.03 was found for vegetation with the lowest density of 0.01 m-1 at 0.2 m depth and a maximum value of 0.20 for vegetation with the highest density of 0.20 m-1 at 2 m flow depth. The mean absolute error (MAE) was 0.04 and 30% lower possibly due to the higher drag coefficients used by previous researchers. The software TUFLOW was used to assess the depth-varying vegetation roughness based on ecotope map and rule-based classification by comparing the modelled flood depths with those recorded during the January 2011 flood event at the reference points A, B, C and D. The simulation results showed improvements whereby the errors were reduced using the depth-varying roughness approach regardless of land cover maps. However, the details in the rule-based classification map contributed to better estimates. The P values of t-test revealed that the overall differences of flood depths and velocities on vegetated floodplains between the constant and depth-varying roughness were statistically significant, wherein the maximum differences in flood depths and velocities were 0.40 m and 0.25 m/s, respectively. The flood depth difference was significant as it was bigger than the accuracy of LiDAR data (+/- 0.15m). This underscores the importance of spatially explicit and depth-varying vegetation roughness. This research bridges between theoretical and practical applications for evaluating vegetation restoration and thinning practice to optimise vegetated floodplains as natural flood storage systems. It is useful in providing less fieldwork and offers greater certainty over vegetated floodplains.

Preview Text FK 2017 84 - IR.pdf Download (2MB) | Preview

Actions (login required)

View Item