C9 Towards a Guideline for Digital Soil Mapping in Ecuador.
PI(s) for this project:
Research and experience in digital soil mapping (DSM) shall be extended and deepened in two further research areas in order to develop a guideline for DSM in Ecuador. The guideline will give an overview: a) about the DSM approach, b) the different sampling designs developed according to the area size, accessibility and terrain complexity, c) the various methods from the field of supervised machine learning to develop digital soil maps, and d) the implementation with open source software. The soil-landscapes of the three investigation areas are analysed and soil-landscape models are developed by supervised machine learning techniques, in order to spatially predict soil properties from point data based on environmental prediction parameters. By using the so developed digital soil maps as principal input, a functional soil-landscape analysis is carried out to determine landslide, erosion and anthropogenic disturbance risk zones as well as estimate the soil organic carbon stocks and soil fertility.
Since long, soils are understood as a function of their genetic factors: parent material, relief, climate, organisms and time. The complex interaction of these factors activate particular soil forming processes, which in dependence of their intensity and duration, lead to a characteristic disitribution of soil properties in space. Through these general understanding, first conceptual soil-landscape models were established that are still widely applied. In addition, soil hydrology provided a major advance in understanding soil systems by investigating how water moves through landscapes; soil development is closely linked to these water flows that provide transport mechanisms for soil particles. These early conceptual soil-landscape models have by now resulted into quantitative models, which do not only make the spatial prediction of continuous soil properties possible, but include model uncertainty as a must. Their development is described by the term “digital soil mapping” (DSM). Soil maps form the basis for soil carbon stock and erosion risk estimation as well as for land use planning. They provide a useful tool for policy makers and municipalities. Tropical mountain areas pose severe problems to traditional soil mapping approaches, due to their heterogeneity and complex lithological composition. However, DSM provides an effective means.
Research experience from the DFG research unit 816, namely the subprojects
- A3.3 "Spatiotemporal dynamics of shallow landslides and their biotic and abiotic controls" (funding 2007–2010),
- B2.2 "Soil physics and hillslope hydrology" (funding 2007–2010) and
- D5 “Functional soil landscape modelling in the Andean mountain forest zone: impact of land use and natural disturbances” (funding 2010–2013)
shall be deepened and extended to two further research areas in order to develop the prototype for this transfer project: A guideline for DSM in Ecuador. The guideline will give an overview: a) about the DSM approach, b) the different sampling designs developed according to the area size, accessibility and terrain complexity, c) the various methods from the field of supervised machine learning to develop digital soil maps, and d) the implementation with open source software.
The soil-landscapes of the three investigation areas (Figure 1) are analysed and soil-landscape models are developed by supervised machine learning techniques, in order to spatially predict soil properties from point data based on environmental prediction parameters. By using the so developed digital soil maps as principal input, a functional soil-landscape analysis is carried out to determine landslide, erosion and anthropogenic disturbance risk zones as well as estimate the soil organic carbon stocks and soil fertility.
The objectives are as follows:
- Develop digital soil maps for different soil-landscapes of Ecuador: In addition to the already well-studied Reserva Biológica San Francisco (RBSF) and the San Francisco catchment, where data-mining and functional soil-landscape analysis will be extended, two further soil-landscapes shall be investigated in order to prepare a guideline for DSM in Ecuador: Cajas National Park and Laipuna Natural Reserve. It is expected that the two areas constitute completely different soil-landscapes because of their different climate and vegetation. Digital soil maps of multiple soil parameters will be developed for the three different mountain landscapes. Pedodiversity as well as the adapted models will be compared between the three areas.
- Perform a functional soil-landscape analysis: The digital soil maps displaying soil type distribution, soil horizontation and various soil properties will be used for a functional landscape analysis in order a) to analyse the relation between pedodiversity and biodiversity, b) to perform a statistical risk estimation, c) to analyse the potential land use/ anthropogenic disturbance impact and d) to estimate the soil organic carbon stocks and soil fertility.
- Develop a guideline for DSM in Ecuador: The guideline will give an overview about the different sampling designs which will be developed according to the area size, accessibility and terrain complexity. Furthermore, it will provide a detailed analysis concerning the necessary amount and distribution of soil data. An overview shall be given regarding the various available methods in the field of DSM and their implementation with open source software.
- Capacity building: Ecuadorian academics and laymen shall be trained in soil sampling and DSM so that the guidelines serve to perform DSM in other areas of Ecuador.
Figure 1: Research areas: a) Quinuas watershed of Cajas National Park and neighbourhood, b) Cientific research station San Francisco and surrounding area, c) Laipuna Natural Reserve. Topographic data use with permission from the Ecuadorian mapping agency "Instituto Geográfico Militar" (2013, National base, scale 1:50,000).
WP 1: Data mining regarding the RBSF area
Further laboratory and data analysis as well as the development of digital soil maps is planned for the RBSF area and the San Francisco catchment, using additional methods of statistical learning theory, e.g. support vector machines (SVM), boosting and artificial neural networks (ANN). It shall to be investigated which amount of data, sampling density, spatial data distribution and DTM resolution are optimal in dependence on terrain complexity and the size of the investigation area.
WP 2: Development of a sampling design
In DSM, soil samples are used to elaborate models to relate soil attributes to environmental parameters. Because the relationships are based on the soil observations, the quality of the resulting soil map depends also on the soil observation quality. Traditionally, there are no statistical criteria for soil sampling. It is through DSM, that soil scientists became aware that this may lead to bias in the areas being sampled. Currently, the DSM community is following two approaches in order to obtain a representative dataset for the investigation of a soil-landscape with DSM approaches: 1) random sampling approaches and 2) stratified sampling approaches based on terrain analysis. However, problems for any random sampling method arise in a hardly accessible complex terrain. Therefore, in dependence of terrain complexity, area size and accessibility, various sampling methods have to be considered.
WP 3: Field work and laboratory analysis
Soil profiles shall be excavated in each of the two new research areas (Reserva Laipuna and Parque Nacional El Cajas) to get a detailed overview of the soil types present in the two soil-landscapes and their physical and chemical soil properties. Well infiltration measurements will be conducted to determine the saturated hydraulic conductivity, Ksat, of the first soil horizon, close to the excavated soil profiles.
WP 4: pedotransfer and spectrometric soil functions
A region specific pedotransfer function (PTF) regarding water retention and Ksat in dependence on soil texture, bulk density, stone content, etc. shall be developed using supervised learning algorithms as described in work package 5.
An ASD AgriSpec System will be used to perform VIS/NIR spectrometric measurements.
WP 5: Development of digital soil maps
Machine learning algorithms for supervised learning to develop digital soil maps in complex terrain comprise artificial neuronal networks, and recursive partitioning methods like CART, bagging, Random Forest and boosting. Continuous environmental variables, such as terrain parameters calculated from a DTM, information derived from remote sensing data (vegetation indices, land cover, climate) as well as geological maps can be used as prediction parameters to extrapolate soil information from point data. Methods to analyse digital soil map uncertainty include bootstrapping, Jackknifing, cross validation and Nash-Sutcliffe Efficiency.
WP 6: Functional soil-landscape and sensitivity analysis
The developed digital soil maps are used to analyse pedodiversity at various scales. Furthermore, they will be investigated regarding the impact of their uncertainty on further risk estimations and soil-landscape analysis. A functional soil-landscape analysis is carried out
- to perform spatial risk analysis (e.g. erosion) in dependence on soil properties,
- to distinguish zones of high pedodiversity
- to investigate the relation between pedodiversity and biodiversity,
- to estimate soil organic carbon stocks,
- to evaluate the landscapes’ conservation potential (protection index), and
- to evaluate the land use and anthropogenic disturbance impact on soil properties.
WP 7: Guideline for digital soil mapping
A guideline for spatial monitoring concerning DSM in Ecuador will be developed. Soil maps are the basis for any further functional monitoring and land use planning. The three investigation areas serve as training sites which allow for future transfer of the DSM methodology to further Ecuadorian soil-landscapes. The guideline will include a detailed description of the necessary approach regarding the choice of a sampling design, the necessary amount of soil data, the DSM methodology and a short introduction to open source software available to perform the analysis.
WP 8: Knowledge transfer
Ecuadorian laymen and technicians will be trained in soil sampling in the Reserva Laipuna and Parque Nacional El Cajas areas. Academic staff will be taught in DSM though the supervision of M.Sc. and PhD students. In addition, several one-week training courses will be offered at universities in Cuenca and Loja.