A hierarchical method for soil erosion assessment and spatial risk modelling: a case study of Kiambu district in Kenya.

Though a lot has been done and achieved in erosion research and control in Kenya, most of the erosion research methods have in the past put emphasis more on quantifying soil loss or measuring soil erosion, rather than pinpointing to areas that are likely to suffer soil erosion. In most cases the erosion processes have been assumed to occur in a uniform manner at all levels of the landscape hierarchy, and hence the results of one level observation can be factored to cover other levels for which data was not collected. This has resulted in many people extrapolating site-specific point data to cover wider geographic regions, assuming uniformity of the erosion process over the region. Another interesting aspect of soil erosion is that though most attention is normally put on the negative effects of soil erosion, soil erosion has also some beneficial effects. For example, the deposition of eroded soil material to lower areas has sometimes improved the quality of the soil receiving the sediment, thereby improving agricultural productivity of the depositional areas. There have however been suggestions that the problem of soil erosion has been exaggerated and not proven to actually diminish crop yields against the background of improved crop productivity improvement techniques. All theses schisms makes it necessary to engage in soil erosion research, either to disprove the sceptics or to provide other means of assessing and viewing the problem of soil erosion.

The general objective of the thesis was to develop and present a method, which can be used to assess the risk of water erosion for different levels of the landscape system hierarchy using spatial methods. The broad aim was to define relevant levels that form the basis for predicting and managing soil erosion and controlling its risk. The specific objectives were:

To conceptualize and define from the landscape continuum hierarchically ordered landscape elements whose internal characteristics and parts influence the occurrence of soil erosion and whose spatial extent and geometry enable their capture and modelling by remote sensing and GIS.

To prove that there are spatial features (erosion proxies) which are part-of, and internally contained in the hierarchically defined landscape elements that can aid in soil erosion risk assessment and modelling.

To demonstrate that the selected erosion proxies can be related to actual occurrences of soil erosion by statistical methods and similarly be differentiated as either drivers or disrupters of erosion.

To demonstrate that prediction models can be derived from field data collected on the erosion proxies and the developed models used for modelling of soil erosion risk spatially in a GIS for each of the defined levels.

To test and validate the method in Kiambu.

To address these objectives, concepts associated with hierarchy theory, landscape system construction, geographic information systems theories, and soil erosion theories were knitted together to develop a conceptual framework and a practical methodological approach to effect and realise the objectives.

Kiambu district was selected for testing the developed methodology due to its intensive utilisation for agriculture and its location in a rugged terrain in the upper footridges and footslopes of the Aberdare Mountains where below-canopy soil erosion is obscured by vegetation vigour and intensive cropping. Soil loss studies through river sediment yields in the district, indicate that there are high amounts of soil lost annually by water erosion. These range from 20 t km -2yr -1in undisturbed forests, to 3000 t km -2yr -1in cultivated to grazing lands. Soil loss studies from runoff plots in Kiambu indicate that cultivated land loses between 20 and 30 t ha -1season -1and bare soil loses more that 70 t ha -1season -1. Other justifications were prompted by the fact that soil conservation in Kenya has been focussed to the 'Catchment Approach' without necessarily defining what the catchment means. Perception of soil erosion by farmers was also biased to visible features of erosion such as gullies and tended to ignore the finer features of erosion like rills, interill erosion and other visible forms or erosion.

Recent developments in geographic information systems (GIS) technology have made it possible to model and represent geographical real world phenomena in computerised spatial databases through which they can be stored, analysed, and displayed. GIS can enable stepwise and ordered analysis of the landscape system components as deemed by the landscape researcher. Soil erosion is a product of the interaction of many geographical factors such as: soil surface cover, the erodibility of the soil mantle, the steepness and length of the eroding slope, the erosive energy of falling rain-drops and the specific management aspect of the eroding site. It can therefore be assessed and modelled in a GIS environment as is demonstrated in this thesis.

Concepts

The method integrates concepts, knowledge and implementation procedures. First there is a priori knowledge that is collated in order to facilitate the study of soil erosion in the context of a landscape system. Knowledge and practice are invoked to organise the landscape system into hierarchical levels through which equally ordered erosion processes can be studied, assessed and measured. The conceptual model revolves around the construction of hierarchical levels in a landscape system and how the landscape holons and their parts relate to soil erosion in terms of assessment, prediction and management.

Hierarchical construction of the landscape system allows an ordered organisation of the landscape components into superordinate or subordinate parts which correspond to process time-scales and their corresponding spatial extents. Such an arrangement of the landscape system allows relevant observations and measurements to be made of the ordered processes. The concept deviates from viewing the landscape as an agglomeration of parts in which processes occur presumably in a uniform manner, and where only size changes allowing for smoothing of measured results in a linear generalisation transformation. The method identifies 'individual' landscape elements in each level of the landscape hierarchy in which processes occur within comparable time scales and for which data interchange and modelling can take place. They are termed 'holons' . The holons are used to represent the levels in the hierarchy.

Methodology

The overall methodology comprised several parts involving first the definition of landscape hierarchies, followed by field observations, and then spatial data capture for erosion risk prediction. Statistical analysis and spatial modelling of erosion risk in a GIS followed the first three steps. The first step mainly concentrated on defining the landscape hierarchies that could be used for assessing soil erosion risk spatially in a GIS. Landscape elements were identified to represent each level in the landscape hierarchy. These included the field-plot, the watershed and the landscape unit. The field observation part involved the collection of soil erosion data in a stratified and randomised strategy where 164 samples were collected for the field-plot holon types, 89 for the watershed holon types, and 104 samples for the landscape holon types. Maps were prepared for each of the levels before field data collection and data points plotted on them for field reference. Collecting soil erosion data involved measuring depths of erosion features formed by either water or gravity as agents of erosion. The erosion features included soil movement, translocated surface litter, visible flow channels, depth of stem washing, depth of root exposure, rills, gullies, and mass soil movement. Only the encountered erosion features on each proxy were observed and measured. Where there was no erosion, zero values were recorded. For prediction modelling, independent spatial data sets were prepared for the three spatial landscape holon types and their internally contained erosion proxies. Aerial photographs were used to capture the field-plot and watershed holon types while satellite images were used to capture the landscape level holon types. This was done in the spatial data capture part of the methodology. Statistical analysis involved confirming that the selected spatial erosion proxies influenced the occurrence of soil erosion, first by the occurrence of erosion features in the proxies and secondly by differentiating the contribution of each proxy to the overall erosion using the analysis of variance and the mean occurrence. The spatial modelling part of the methodology involved the production of logistic and logit regression prediction models from the initial random data, which were used to manipulate the newly created spatial data sets to produce erosion risk domains based on pre-defined erosion indicators, herein defined as erosion proxies for each of the levels. Properties of the proxies such as their slope-gradient, percentage ground cover and soil erodibility were used as the predictor variables after being tested for suitability as prediction variables by correlation analysis.

Results

It was possible to use large-scale aerial photographs to capture field-plot and watershed level erosion proxies. Their real-world positions in the broader landscape were located using a global positioning system, commonly known as a GPS. The extracted holon types representing each level and their erosion proxies captured from the aerial photographs and satellite images were digitised and stored in a geographic information system. Field plots with their lands use kinds represented the lowest level of the hierarchy. Linear features in the watershed such as field-plot boundaries, footpaths, hedges, fences, tree lines, stonewalls, forest edges, wash-stops in tea and closed field boundaries represented the second level of the hierarchy. The highest level of the hierarchical construction, the landscape unit had river valleys, roads, and built-up centres representing the highest level erosion proxies. All the levels of the landscape hierarchy represented by different holon types, i.e. field-plots, watersheds, and the landscape units were stored as area objects while the erosion proxies inside them were stored as either area objects or as line objects. The field-plots were stored as area objects while the linear watershed proxies were stored as line objects. The landscape level erosion proxies that included built-up areas, river valleys, and road networks were stored as either linear objects or as area objects.

From the descriptive statistics and analysis of variance it was possible to differentiate between the proxies that acted as the drivers of erosion as earlier postulated and those that acted as disrupters of erosion. The selected proxies for each of the levels and the frequency of their occurrences in each holon type supported the hierarchy theory of new and emergent erosion indicators for the different levels of the hierarchical organisation of the landscape system. Some proxies could be selected to act as the erosion indicators for each of the levels. Soil movement, rills, washed stems, exposed roots, flow channels and translocated surface litter occurred in all the field-plot level erosion proxies. Gullies emerged as new erosion features for the proxies of the watershed level while soil movement, rills, stem wash and root exposures occurred in one third of the watershed level erosion proxy types. Mass soil movement emerged as new erosion features for the landscape level proxies while rills dominated and occurred in all the proxies. Soil movement and gullies occurred in two-thirds of the landscape level erosion proxies.

Important outputs from the prediction models were the odds ratios that provided quantitative values of how the erosion proxy properties influenced the risk of erosion. For the field-plot level holon types, a unit percent increase in slope created an increase in the risk of occurrence of soil erosion by 24% while a unit percentage increase in cover reduced the risk of occurrence by 8%. Slope added to the risk while cover reduced the risk. The risk contributed to by slope assumes constant cover and similarly the reduction in the occurrence of risk assumes constant slopes. For the watershed level holons, slope enhanced the risk of erosion by 36%. For the landscape level holons, a unit increase in percent slope enhanced the risk by 229%. A unit increase in cover percentage reduced the risk by 5% when the slope is constant. These figures obtained by the odds ratio show the erosion hazard conditioned by each of the proxy types. Slope and cover of the different multi-level erosion proxies proved to be important variables for predicting the hierarchical risk of erosion in the landscape system.

The spatial data on predicted erosion risk provided useful information for erosion risk management. The spatial extent attribute made it possible to determine the amount of time and energy required for tackling the risk associated with any proxy. The object position helped in precisely targeting intervention measures, while the distribution showed the relative intensity of the risk in any given area and the proximity of the risk to known locations of the geographic feature space. The independently produced logistic and logit prediction models produced spatial patterns that corresponded very well with the predictor variables of slope and cover of the proxies. This obtained relationship gave credence to the predictive power of the models. The logistic regression models showed the probabilities of erosion risk in the proxies, while the logit models refined the risk zones due to the transformation of the probabilities to a longer linear stretch. The logit models therefore provided a broader spectrum of the predictions making it possible to distinguish areas with low risk from areas with very high risks of erosion that required immediate and prioritised attention.

Strengths of the methodology and conceptual model

The methodology seeks to establish a means by which soil erosion can be assessed and modelled at multiple levels in a landscape. It deviates from other single-level erosion assessment and risk prediction methods that are of common practice in Kenya and in many other parts of the world. It views the landscape as a medium in which water erosion processes are taking place in an intricate manner at different spatial hierarchies and which are also the entities for land use and erosion management. These spatial attributes offer the opportunity to predict and view the distribution of the risk of erosion on the broader landscape and at different attention levels. These create management opportunities are powerful tools for preventing or controlling soil erosion. The beneficiaries of the methodology outputs are seen as farmers, government departments, and other non-governmental organisations that are involved in soil and water management.

Weaknesses of the conceptual model

The following were the observed weaknesses during implementation:

In intensive annual farming systems, the features of erosion are usually obscured by frequent tillage practices making their assessment only possible immediately after tillage and after the event of an eroding rainfall or during minimum tillage periods;

The above reason means that temporal considerations must be embedded in the methodology where the observations in the field are timed to coincide with rainfall and field preparations; and

Any organisations opting to use the method must be fully equipped with GIS facilities and technical capacity to manipulate the spatial databases and modelling their attributes. Technical capacity is also required in the fields of soil erosion assessment and statistical analysis. Remote sensing knowledge and data offer the best opportunities for capturing landscape objects for integration into a GIS. Acquisition of the data and capacity to manipulate the images both digitally and manually are unavoidable requirements of the methodology.

Overarching discussions and conclusions

After collecting data on soil erosion and analysing it statistically, it has been shown that soil erosion in the field plot holon, watershed holon and landscape holon can be distinguished according to different erosion proxies for each of the holons. This finding can be used to refine the erosion prediction models such that the erosion processes are linked to the landscape holons with which they are closely associated. Scientific observations and measurements can also be ordered according to the holon type such that any extrapolations are directed to equal levels within the structure of the hierarchy. Different holons represent different levels of the landscape hierarchy.

Due to the techniques of assessing soil erosion in the field as applied in this work, it was possible to show that soil erosion goes on even in tea plantations, coffee plantations and in forests. If erosion assessment is only based on farm fields, then the erosion in the in higher level holons such as watershed or the landscape unit will be masked somewhere during erosion impact evaluations. This thesis has also enhanced the recognition of soil erosion by using visible features to capture and measure incidences of past erosion. Soil movement, flow channels, translocated surface litter, are features of erosion, which are not usually discussed nor used for either assessing or quantifying soil erosion. Their use can now be integrated into further works. With the presented methodology of assessing soil erosion, more data will be generated and demonstrated to the farmers and other interested people on soil erosion and its hazard. The presented assessment and definitions in this work have also removed the problem of terminology especially in the definition of the landscape features for studying and managing soil erosion. The watershed has presented itself as the best management unit for many members of a community and can be adopted by any organisation concerned with collective management of soil erosion. The field plot remains the preserve of a single farmer while the landscape unit requires more of state interventions and deliberate management policy.

Prediction modelling made it possible to assign a quantitative value to the risk of erosion. The landscape unit had the highest risk (229%), followed by the watershed (36%) and the field plot (24%). Resources and efforts must therefore be availed and directed towards addressing soil erosion risk occasioned by the landscape level erosion processes and proxies. Managing the landscape level alone will not have a net benefit if the two other levels are not also attended to since they equally suffer soil erosion risk. A holistic management approach will provide a more effective management approach to the soil erosion risk. Government priorities can be directed at the landscape level. Intermediate agencies such as Community Based Groups and organisations can tackle the watershed level risk while farmers and farming communities can tackle the field-plot and watershed levels.

Researchable areas include in depth social studies that link biophysical spatial properties of the landscape to the socio-economic and policy instruments for each of the levels of the landscape hierarchy, i.e., field-plots, watershed units and the landscape units. Others are studies that link these landscape system holons with soil erosion proxies and soil erosion processes especially on their relationships with sediment yields, etc. to confirm further the notion of landscape hierarchies and erosion process resolutions. Finally, calibration and further testing of the prediction models in other parts of Kenya are recommended.