«AGWATER Options for sustainable agricultural production and water use in Cyprus under global change Scientific Report 6 Deliverable D15, D16 Zomenia ...»
Digital soil map of Cyprus (1:25,000)
Options for sustainable agricultural production and water use in Cyprus under global change
Scientific Report 6
Deliverable D15, D16
Zomenia Zomeni1, Corrado Camera2, Adriana Bruggeman2,
Andreas Zissimos1, Irene Christoforou1, Jay Noller3
1 Geological Survey Department of Cyprus
2 Energy, Environment and Water Research Center, The Cyprus Institute
3 Department of Crop and Soil Science, Oregon State University Nicosia, 15 November 2014 Table of Contents
DERIVED SOIL SERIES MAPS
DERIVED SOIL PROPERTIES MAPS
SOIL SERIES MAP
SOIL PROPERTY MAPS
i ABSTRACT Considering the increasing threats soil are experiencing especially in semi-arid, Mediterranean environments like Cyprus (erosion, contamination, sealing and salinisation), producing a high resolution, reliable soil map is essential for further soil conservation studies.
This study aims to create a 1:25.000 soil map covering the area under the direct control of the Republic of Cyprus (5.760 km2).
The study consists of two major steps. The first is the creation of a raster database of predictive variables selected according to the scorpan formula. It is of particular interest the possibility of using, as soil properties, data coming from three older island-wide soil maps and the recently published geochemical atlas of Cyprus. Electric conductivity, pH, total carbon and the Mafic Index of Alteration (MIA-R) were selected to represent soil properties;
maximum and minimum temperature for climate; organic carbon for organic matter; the DEM and related relief derivatives (slope, aspect, curvature, landscape units); and bedrock and surficial geology for parent material and age.
In the second step, the digital soil map including soil series and soil properties (depth and texture) is created using the Random Forests package in R. Random Forests is a decision tree classification technique where many trees, instead of a single one, are developed and compared to increase the stability and the reliability of the prediction. The model is trained and verified on areas where a 1:25.000 published soil maps obtained from field work is available and then it is applied for predictive mapping to the other areas.
Results shown that the average error of the model, both for soil series and soil properties, is around 10%, demonstrating the robustness of the methods proposed.
INTRODUCTIONThe soils of Cyprus are unique due to the geological complexity, the Mediterranean island climate and the long presence of man on the landscape. The geology of Cyprus is dominated by the Troodos Ophiolite, which is a fragment of a fully developed oceanic crust, consisting of plutonic, intrusive and volcanic rocks and chemical sediments. Sedimentary formations cover the coastal plains in the south and the intermountain plain in the north. The soils on Cyprus vary between lithosols, leptosols, regosols, gypsisols, solonchaks, solonetz, vertisols, and cambisols based on the WRB (World Reference Base) of FAO (Food and Agriculture Organization of the United Nations) soil classification system (FAO, 1989). They are generally poor in organic matter (Koudounas, and Makin, 1981; Grivas, 1988) and closely associated to parent material and landscape position. An incomplete series of soil surveys and maps at a scale of 1:25,000 have been prepared by the Soil Section of the Department of Agricultural from 1967-1985, using traditional field survey methods The soils are mapped and classified based on their development stage, origin and parent material. These maps formed the basis of the development of a digital soil map of Cyprus at a scale of 1:250,000.
The aim of the study was to create, using digital soil mapping techniques, a soil map of Cyprus (including soil series and soil properties) at 1:25,000 scale. This involved the creation of digital soil data to be used as training data and the creation of other data of physical parameters involved in the soil forming process to be used as predictors. The analysis was run for areas under the effective control of the government of the Republic of Cyprus were data were available.
METHODS The soil series and soil property maps are calculated using Random Forest. Random Forest is a multiple tree classification and regression method developed by Leo Breiman (2001). A clear overview of the method’s functioning is presented by Boulesteix et al. (2012) and summarized in Figure 1. Each tree is a standard classification tree. At each node the code randomly samples N (mtry) predictors and it picks the predictor that ensures the best split, evaluated by the decrease of Gini impurity (DGI). A bootstrap sample from the original data set is used to build a tree. Each target point is then classified aggregating the trees and picking the class that received the major number of votes. A very relevant feature of Random Forest is the out-of-bag (OOB) error. As stated, trees are calculated using a bootstrap sample from the original data set, this means that some values are not actually used to construct the trees. Therefore these data can be used for validation purpose. The OOB error is the average error, calculated for each target class, coming from the comparison of the observations that have been left out and the model output. An additional feature of Random Forest is the capacity to rank the relative importance of the variables in the prediction.
3 Figure 1: flow chart explaining the functioning of the Random Forest algorithm (from Boulesteix et al., 2012) DATA Training Data The sample data for deriving the train and out-of-bag data were based on published soil maps.
These detailed soil maps have a 1:25,000 scale with only ten out of the forty-some possible sheets having been published to date (Figure 2). By far the most detailed soil reference on the island are these ten 1:25.000 scale soil sheets (Soteriades and Georgiades, 1967, Soteriades and Grivas, 1968, Soteriades, Koudounas and Markides 1968, Soteriades.and Markides, 1969, Grivas and Georgiades, 1972, Markides, 1975, Koumis, 1980a, Koumis, 1980b, Koumis, 1980c, Markides, 1985) which are always accompanied by a land suitability for agriculture map. Two of them, the Pafos sheet (Soteriades and Koudounas, 1968) and the Polemi sheet (Markides, 1973) are also accompanied by extensive soil memoirs. These ten soil sheets form the basis for the most detailed and thorough digital soil information on the island.
4 Figure 2: Soil map of Cyprus and availability of soil maps at the 1:25,000 scale (from the Soil Map of Cyprus (1999), by the Soil and Water Use Section, Cyprus Department of Agriculture).
The maps were scanned, georeferenced and digitized in a GIS environment. The resulting dataset is a merge of the 10 soil maps and consists of with 11.000 polygons classified in 52 soil series with further 4-8 subseries classification for each series. The dataset was converted to raster format with cell size of 25 x 25 m2 and used as training data for the training areas in building the multiple tree classification.
Predictors Predictors have been selected according to the scorpan formula (McBratney et al., 2003) and include physical variables like relief, climate, geology, geomorphology, and geochemistry (Table 1). Some graphical examples of these data are shown in Figure 3.
Derived Soil Series maps In detail, 52 soil series were recognized across the island and all of them are present in the training areas. To train the model, a random selection from 50% of the area covered by the existing 1:25,000 soil maps (approximately one million points) and a total of 350 trees have been created. The number of selected training points and trees is the maximum that the available computing facilities allowed. The model has been run at the High Performance Computing (HPC) facility of the Cyprus Institute (Cy-Tera) in Lefkosia.
Derived Soil Properties maps From the existing ten 1:25,000 sheets (Soteriades and Georgiades, 1967, Soteriades and Grivas, 1968, Soteriades, Koudounas and Markides 1968, Soteriades.and Markides, 1969, Grivas and Georgiades, 1972, Markides, 1975, Koumis, 1980a, Koumis, 1980b, Koumis, 1980c, Markides, 1985) 8 soil depth classes have been identified from the legend descriptions of all the 52 soil series and many subseries on the maps. Each class is characterized by a depth interval. To end up with a single value of soil depth for use in the modeling applications, the average of the interval has been calculated (Table 2).
In the same fashion, 17 soil texture classes have been recognized. However, among the 17 classes different nomenclature and classification systems were used. Therefore, data was harmonized and reclassified in 9 consistent classes (Table 3). Available water capacities (AWC) were assigned for all standard textures according to Saxton and Rawls (2006), except for clay, which was taken from Allen et al. (1998).
8 RESULTS Soil Series map The calculated 1:25,000 soil series map of Cyprus is presented in Figure 4. The mean OOB error in classifying the 52 soil series over the training area is 0.1, meaning that 10% of the soils are wrongly classified. In detail, only four soil series show errors higher than 0.2 (Figure 5), demonstrating the value of the applied method. The four soil series that are worst classified are: Troodos (class error 0.28), Quarries (error 0.23), Argaki (error 0.22), and Rivers (error 0.20). However, while the errors are low for the training areas, this does not mean that all the soils of the mapped areas outside the training areas are correctly predicted. The methodology implicitly assumes that the 10 soil maps (training areas) are representative for the full mapped area. An obvious problem is the lack of soil maps for the Troodos massif. Thus, the soils for the Troodos may be not perfectly predicted.
The importance of the different predictors is shown in Figure 6. It is interesting to notice how the first three predictors, in terms of relative importance, are all geochemistry variables (pH, OrgC, EC). Therefore, these variables are crucial in the classification process. In Figure 7 the correlations between the geochemistry variables are shown. These figures demonstrate that the four predictors are independent of each other.
Figure 4: 1:25,000 soil map of Cyprus obtain with digital soil mapping techniques.
Soil Property maps The predicted soil depth and soil texture maps are presented in Figure 8 and Figure 9, respectively.
Similar to the soil series map, the results are very good with the mean error equal to 0.10 (range 0.05 – 0.14) and 0.11 (range 0.05-0.25) for soil depth and texture, respectively. For both maps the previously predicted, the soil series data, pH and Aspect are the three predictors with the highest importance in the classification process. It is also worth pointing out that the results in the Troodos area could not be as good as in the training areas, for the reasons explained in the previous section (Soil map Series).
In Table 4 a summary of the areal percentages for each class of soil depth and texture is presented.
This gives a quick overview of the amount of land suitable for agriculture (soil depth 30 cm and no stoniness).
13 REFERENCES Allen, R.G., Pereira, L.S., Raes, D., and Smith, M.: Crop Evapotranpiration: Guildlines for computing crop water requirements, FAO Irrigation and Drainage Paper No 56. Food and Agriculture Organisation, Land and Water. Rome, Italy, 1998 Boulesteix, A.L., Janitza, S., Kruppa, J., and König, I.R.: Overview of random forest methodology and practical guidance with emphasis on computational biology and bioinformatics. WIREs Data Mining Knowl. Discov. 2, 493–507, 2012 Breiman, L.: Random forests. Mach. Learn. 45, 5–32, 2001 Büttner, G., Kosztra, B.: CLC2006 Technical guidelines. Technical Report No. 17/2007. EEA,
2007. Available from http://www.eea.europa.eu/publications/technical_report_2007_17 Camera, C., Bruggeman, A., Hadjinicolaou, P., Pashiardis, S., and Lange, M.A.: High resolution gridded datasets for meteorological variables: Cyprus, 1980-2010 and 2020-2050, AGWATER Scientific Report 5, 70 pp., 2013 Cohen, D.R., Rutherford, N.F., Morisseau, E., and Zissimos, A.M.: Geochemical Atlas of Cyprus.
Sydney: UNSW Press, 2011.
Department of Agriculture, 1999, Section of Soil and Water use, Soil Map of Cyprus, scale 1:250,000 FAO-UNESCO, 1989, Carte mondiale de sols, Legende revisee, Rapport sur les resources en sols du monde, No. 60, FAO, Rome Grivas, G.C., and Georgiades, M., 1972, 1:25.000 Sheet 30 Lakatamia, Soil Section, Department of Agriculture, Ministry of Agriculture and Natural Resources, Cyprus Grivas, G., 1988, Development of land resources in Cyprus, in Proceedings - Workshop on conservation and development of natural resources in Cyprus - case studies - soils - groundwater mineral resources, Zomenis, S.L., Luken, H., Grivas, G., (editors), published by: Ministry of Agriculture, Cyprus & Federal Institute for Geosciences and Natural Resources, W. Germany, pp.
7-16 Koudounas, C.; Makin, J., 1981, A study of representative soil profiles from the Limassol - Paphos districts, Ministry of Agriculture and Natural Resources, Nicosia, 61 p.