Fuzzy logics, digital soil mapping and the hydropedological behaviour of the Cathedral Peak research catchments
By Rowena Harrison, PhD Student, SAEON Grasslands-Forests-Wetlands Node
By Rowena Harrison, PhD Student, SAEON Grasslands-Forests-Wetlands Node
Anyone who has ever been on a soil mapping exercise, especially in a montane environment, will know that it is a strenuous, laborious task and a long day. This is due to access difficulties, the topography of the environment and the time required to conduct the field investigation (Ismail and Yacoub, 2012; Martín-López et al., 2019).
Recently, digital soil mapping (DSM) has become a key tool, reducing the time and financial aspects needed for conventional soil-based mapping. DSM is the interpretation of spatial and temporal soil property variations using mathematical models based on quantitative relationships between environmental information and soil measurements (Martín-López et al., 2019).
DSM has been developed since the late 1960s, when there was shift to soil mapping based on more geographic, topographic and spatial approaches to interpreting the position of soils in relation to the landscape (McBratney et al., 2003). This more recent interest is driven by an increasing recognition of the ecological, economic and societal benefits of understanding soil properties, their spatial distribution and the value of this knowledge for use in the management objectives of a variety of industries and land uses (Kimsey et al., 2020).
The geographic and topographic nature of DSM makes it relevant to hydropedology (Ma et al., 2019, Lin et al., 2006), a discipline of soil science that combines the expertise of soil physicists and pedologists and plays a key role in realising sustainability goals focused on food, water, climate and ecology. The realisation of these goals requires interdisciplinary research (Bouma, 2016).
Understanding the basic flow paths and hydropedological behaviour of a catchment or area can therefore be the first step in identifying the systems which create the ecosystems on which we all rely. Using a combination of remote sensing techniques and DSM, the hydropedology soil group classification of three montane research catchments was mapped.
The study aimed to improve the knowledge of the relationships between topography and the location of certain soil forms within the catchments to better understand the various flow dynamics of these catchments. The information obtained from the mapping of the hydropedological soil groups and the interactions between these groups will further enhance land use and water management planning in these areas.
The study took place in the Cathedral Peak experimental research catchment site, which forms part of SAEON and is situated on the Little Berg plateau in the northern part of the uKhahlamba-Drakensberg escarpment, KwaZulu-Natal. Three similar catchments were selected for this study and are named CP-III, CP-VI and CP-IX (Figure 1). These catchments are similar in shape and size but present different histories and land management practices.
Figure 1: Locality of the catchments selected for the study
PhD student and author Rowena Harrison engaged in soil mapping
The first aim of the DSM study was to broadly identify and map the watercourses and wetlands located within each of the three catchment sites. To achieve this, the use of satellite imagery and the analysis of this imagery was undertaken.
After pre-processing the imagery, a normalized difference vegetation index (NDVI) analysis was conducted. Based on prior knowledge of the catchments, areas which are known to be wetlands and watercourses are associated with denser vegetation compared with terrestrial areas. Therefore, initial classification of wetlands and watercourses within the catchments was taken as areas with NDVI values higher than 0.4 (Figure 2).
Figure 2: NDVI analysis results for the three catchments CP-III (left), CP-VI (centre) and CP-IX
ArcSIE (Soil Inference Engine) was used for the creation of the digital soil maps for the three catchment areas. It is a knowledge-based raster soil mapping tool, which uses rule-based reasoning as well as case-based reasoning to facilitate the creation of soil-landscape models.
In addition, ArcSIE is a toolbox that functions as an Extension of ArcMap and generates soil maps based on a simple equation which states that soil is a function of the environment. This tool was designed for creating soil maps using fuzzy logic.
A fuzzy logic model is based on the concept of fuzzy sets (Zadeh, 1965) and considers the complex nature of soils and landscapes. This complex nature creates an uncertainty in the allocation of boundaries between one soil group and another and should therefore not be represented by the abrupt lines depicted in polygon-based maps.
Fuzzy logic modelling attempts to represent the uncertainty between different classes created by polygon maps by predicting the soil groups and therefore allowing a continuous graduation from one class to another (Hellwig et al., 2016, Shi et al., 2004). It helps DSM model existing relationships between soil properties and landforms (de Menezes et al., 2014). This creates a more gradual and continuous change in soils and their properties within the map.
A rule-based approach was first undertaken within the ArcSIE tool. This involved understanding the relationships between the soil and landscape and was based on knowledge of the catchments and previous soil surveys. Furthermore, the aim of this study was to group the identified soil types into hydropedological groups based on the behaviour of the soils in relation to the flow dynamics of the catchment.
The hydropedological classes were therefore classified as (i) recharge shallow soils; (ii) recharge deep soils; (iii) interflow soils; and (iv) responsive saturated soils (van Tol and Le Roux, 2019). The relationship between these hydropedology soil groups and the slope, elevation, topographical curvature and inverse wetness index was identified.
A number of rules were applied to the inference engine within the ArcSIE tool. These rules were based on the outcome of various digital terrain models (DTMs) as well as knowledge of the catchments, with various parameters overlapping with each other due to the fuzzy logic nature of the rules applied.
Polygon masks were furthermore added to the ArcSIE tool to delineate the wetlands and watercourses identified from the NDVI analysis. These rules and polygon masks were used to produce the first maps showing the fuzzy membership for the four hydropedological classes (Figure 3).
Figure 3: Fuzzy membership maps for each hydropedological soil group as well as the draft combined map for CP-III
The ArcSIE interface allows for the input of case-based features with the aim of adjusting the predictive models created within the rules-based phase, to better reflect the knowledge about a particular landscape. Existing soil maps and soil information gathered at specific points within the catchments were added as individual cases to the ArcSIE interface and this information was utilised to refine the mapping rules.
A total of 49 validation points were utilised for CP-III, 46 for CP-VI and 27 for CP-IX. Soil sampling points were chosen to cover the range in altitude, planform curvature, slope and topographical wetness of the individual catchment areas. Figure 4 shows the final maps produced.
Figure 4: Refined Hydropedological Soil Group Maps for the three catchments: CP-III (left); CP-VI (centre) and CP-IX (right)
Statistical analysis of the performance of the ArcSIE interface to create the combined hydropedological maps for each of the catchments was undertaken. The Kappa coefficient of agreement was used. The overall Kappa coefficient for CP-III is 0.57, for CP-VI it is 0.59 and for CP-IX it is 0.74. These values range from a weak (CP-III and CP-VI) to a moderate (CP-IX) agreement between the predicted maps and the ground-truthed points.
Results of the DSM study therefore indicate that while ArcSIE is a powerful tool to be used to gain a general idea of the relationship between the soils and the landscape from a hydropedological stance, the knowledge of the soil-landscape relationship, as with the traditional soil mapping exercise, is still necessary to improve the accuracy of the tool.
The quality of the digital soil map produced is furthermore dependent on the method employed (rule-based versus case-based) as well as the environmental covariates used for the model predictions. This is true for all inputs to the prediction model. Therefore, the more knowledge one has of the landscape processes under study, the more accurate the outcome of the digital soil mapping exercise.
However, the hydropedological soil group maps achieved a more correct representation of the complex nature of the soil-landscape relationship, with changes between one soil group and the next being gradual and continuous. These soil group correlations are not well represented in polygon-based maps and therefore require the fuzzy membership logic utilised by the ArcSIE interface to depict these relationships.
Of importance is that the accuracies and inaccuracies within the fuzzy-membership maps can be quantified, allowing for a confidence rating in the use of these maps. These maps can therefore be used in further applications in water and land management for the area.
Bouma, J. 2016. Hydropedology and the Societal Challenge of Realizing the 2015 United Nations Sustainable Development Goals. Vadose Zone Journal 15, p. 12. doi:10.2136/vzj2016.09.0080.
de Menezes, M.D., Silva, S.H.G., Owens, P.R. & Curi, N. 2014. Digital Soil Mapping Approach Based on Fuzzy Logic and Field Expert Knowledge. Ciência e Agrotecnologia Vol. 37. No. 4. Lavras.
Hellwig, N., Graefe, U., Tatti, D., Sartori, G., Anschlag, K., Beylich, A., Gobat, J. & Broll, G. 2016. Upscaling the spatial distribution of enchytraeids and humus forms in a high mountain environment on the basis of GIS and fuzzy logic. European Journal of Soil Biology 79, pp. 1–13.
Ismail, M. & Yacoub, R. K. 2012. Digital soil map using the capability of new technology in Sugar Beet area, Nubariya, Egypt. The Egyptian Journal of Remote Sensing and Space Science 15(2), pp. 113–124. doi:10.1016/j.ejrs.2012.08.001.
Kimsey M.J. 2020. Soil Mapping, Monitoring, and Assessment. In: Pouyat R., Page-Dumroese D., Patel-Weynand T., Geiser L. (eds) Forest and Rangeland Soils of the United States Under Changing Conditions. Springer, Cham. https://doi.org/10.1007/978-3-030-45216-2_9.
Lin, Y. & Liquan, Z. 2006. Identification of the spectral characteristics of submerged plant Vallisneria spiralis. Acta Ecologica Sinica 26, pp. 1005–1011.
Ma, Y., Minasny, B., Malone, B.P. & McBratney, A.B. 2019. Pedology and digital soil mapping (DSM). European Journal of Soil Science 70, pp. 216–235. doi: 10.1111/ejss.12790.
Martín-López, J.M., Da Silva, M., Valencia, J., Quintero, M., Keough, A. & Casares, F. 2019. A comparative Digital Soil Mapping (DSM) study using a non-supervised clustering analysis and an expert knowledge-based model – A case study from Ahuachapán, El Salvador. Presented at: Joint Workshop for Digital Soil Mapping and Global Soil Map March 12–16 2019.
Shi, X., Zhu, A., Burt, J.E., Qi, F. and Simonson, D. 2004. A Case-based Reasoning Approach to Fuzzy Soil Mapping. Soil Science Society of America Journal 68, pp. 885–894.
van Tol, J.J. & Le Roux, P.A.L. 2019. Hydropedological grouping of South African soil forms. South African Journal of Plant and Soil 36(3), pp. 233–235. DOI: 10.1080/02571862.2018.1537012.
Zadeh, L.A. 1965. Fuzzy sets. Information and Control 8, pp. 338–353.