Geomorphological Appraisal of the Madogo-Mororo Floods using Neuro-Fuzzy Inference and the Multinomial Logistic Regression Models

Abstract

The Tana River County has had serious issues relating to management of its riparian corridors and gazetted catchments. Floods have been an annual episode in Madogo-Mororo areas, with disastrous consequences every time they occur. Lives of both humans and livestock have been lost in every such episode, alongside the destruction of infrastructures and damages to properties, valued in the range of millions of Dollars. The causes do not directly relate to the rainfall volume registered whenever it rains, but rather to myriads of anthropogenic, geomorphological and hydrological variables. A combined statistical - and Artificial Intelligence-aided mapping of the study area was deemed a necessity as a DESK-TOP planning tool for flood risk assessment in the Study Area, as well as for detailed flood-risk mapping. The objective of this study was to identify all these three components, which involved mapping the longitudes, latitudes, elevations, soil types prevalent, vegetation density, grainsize of rocks or soil particles deemed prevalent in the surficial geology, culvert types constructed in the areas visited, hosing types, nearness of settlement to rivers, erosional intensity status of an area, as the input variables, to predict the output variable, flooding probability of an area. Visits were made to the Madogo -Mororo areas and the coded weightings of the aforementioned variables assigned, as deemed appropriate by researcher. For instance, soil erosion intensity was coded as 1, 2, and 3, with the numbers respectively coding for low intensity, moderate intensity and high intensity. The R software modeling libraries were used in this exercise. The neuro-fuzzy algorithms were then used to develop prediction models for floods, based on these coded parameters to assess then accuracy levels of the algorithm test datasets were used, generating over 98.4 percent levels of accuracies. The logistic regression models were then used, for comparison, also recording accuracy levels of 98.4 percent.  This affirmed the supremacy of fuzzy logic and logistic regression as useful assessment tools for floods probability mapping in the study area. A code value of 1, represented a high probability of an area being flooded (flood risk areas), and a code of 2 was used to represent high-probability of moderate risk to flooding, whereas the value of 3 signaled a risk-free high-probability in an area. These probability codes were then plotted in a GIS map to help display the geospatial models of flood risk in Madogo and Mororo areas.  The study notes the role of human activities and geomorphic factors in exacerbating the propensity to flooding episodes in the study area. Summarily, this study validates the use of neuro-fuzzy inference and logistic regression as dependable mapping tools for flood episodes in Madogo Mororo areas, used alongside the GIS techniques.