Multi-temporal Soil Erosion Risk Assessment in N. Chalkidiki Using a Modified Usle Raster Model

The aim of this work was to test a modified version of the Universal Soil Loss Equation (USLE) for assessing the risks of erosion in N. Chalkidiki, Greece. USLE estimates the severity of erosion, thus assisting the decision process in selecting erosion control measures. Although USLE has several limitations, it was selected because it is the simplest approach while remaining robust, and it partially solves the problem of data availability. Modifications referred to here concerned the estimation of factors C and k (representing land management and soil vulnerability, respectively). More specifically, the C-factor was estimated using multi-temporal NDVI layers derived from LANDSAT images, while the k-factor was estimated based on the geological map. All USLE factors were calculated as grid layers after processing the original data, then they were multiplied together (according to the USLE) in order to derive the final risk map for three different seasons. A scale of 1:100 000 was selected and the mapping unit was set at 1 ha. The results, assessed for their accuracy by experts, showed that the use of multi-temporal NDVI gave a better insight than a single date approach in understanding the erosion procedures in the study area and facilitated the comparison between seasons and areas. INTRODUCTION About soil erosion Erosion is a natural geological phenomenon resulting from the removal of soil particles by water or wind, transporting them elsewhere, while some human activities can significantly increase erosion rates. Erosion is triggered by a combination of factors such as steep slopes, climate (e.g. long dry periods followed by heavy rainfall), inappropriate land use, land cover patterns (e.g. sparse vegetation) and ecological disasters (e.g. forest fires) (1). Moreover, some intrinsic features of a soil can make it more prone to erosion (e.g. a thin layer of topsoil, silty texture or low organic matter content) (2). The Mediterranean region is particularly prone to water erosion due to its physical factors: climate, topography and soil characteristics (3,4). Serious erosion is generally irreversible. Different indicators of soil erosion have been identified and it is a common opinion that the area actually affected by erosion is in fact the best indicator for soil erosion. Equally interesting to the actual erosion rate is the risk of future erosion in a specific area. The area at risk can be estimated using an appropriate model of soil erosion. Effective modelling can provide information about current erosion, its trends and allow scenario analysis. The integration of existing soil erosion models, field data and data provided by remote sensing technologies, through the use of geographic information systems (GIS), appears to be an asset to further exploit (5). From this perspective, it is not surprising that the European Commission adopted an action plan on GMES Global Monitoring for the Environment and Security in 2004. The plan outlines firm steps towards the establishment of a system that will harness, co-ordinate, and enhance existing Earth Observation (EO) data and monitoring information from satellites and Earth-based sensors, in order to support decision making for the environment and security. EARSeL eProceedings 8, 1/2009 41 The awareness of the fact that EO data have not been used to their maximum capability for policy support in the EU has produced a considerable effort to correct this situation (6). This is true for many areas of application including land use, land cover, terrestrial ecology, geology and soils. Considering EO data (i.e. data acquired with Remote Sensing), scale and spatial resolution are two of the most important factors, especially in the land resource area, because of the spatial heterogeneity of some complex mosaic of patches and transitional forms, typical of the EuroMediterranean region (7). In many cases, problems with the availability of multi-temporal EO data also create obstacles. Multi-temporal data are needed both to aid classification (e.g. of land cover) and to detect changes. However, gaps in time series data often occur because of cloud cover and sensor failure. These add to the cost and difficulty of data acquisition and may limit data use. Soil erosion risk models and USLE A large variety of models can be found in the literature that could be used in soil erosion risk assessment. These models can be classified as follows: • Empirical and mechanistic models: The empirical models describe a process based on empiricism. In contrast, mechanistic models attempt to represent the physical causes of responses to conditions. • Static and dynamic models: the difference between static and dynamic models is that dynamic models take into account time as an extra variable. • Deterministic and stochastic models: Deterministic models make definite predictions for quantities without any associated probability distribution. Stochastic models, on the other hand, contain some random elements or probability distributions. Except for the predicted value, stochastic models can also predict the variance. • Spatial dimensions in models: Any model can be distinguished between one-dimensional (1D), two-dimensional (2D) and three-dimensional (3D) models. • Qualitative and quantitative models: Qualitative models predict values on quality levels such as not risky, risky or highly risky. The input data for a qualitative model can be both qualitative and quantitative. On the other hand, a quantitative model produces a numerical output. • Long-term or event-based models (temporal scale). • Single point or spatially distributed models (spatial scale). Because of the complexity of real world processes, the models which try to simulate these processes often contain combinations of the aforementioned model types; as an example, the model “ANSWERS” is a dynamic two-dimensional (2D) model. The main criteria in order to choose one of the above models are: the purpose of use, the available data, the available time, and the cost. Most erosion models have been designed to predict point soil loss, because they were developed on a field scale. As a result, this kind of models cannot estimate accurate soil erosion loss values when they are applied over large geographic scales. Besides, most models have been developed to predict a specific type of soil erosion (e.g. rill-, inter-rill erosion, gully erosion, etc). As a result, a model cannot perform well in an area where the dominant type of erosion is not the one for which the model was designed. The main problem in relation to the erosion risk models is the validation of their estimates, because there are not always reliable data for comparing the calculations of the models with actual soil losses. One of the most widely applied empirical models for assessing the sheet and rill erosion is the Universal Soil Loss Equation (USLE), developed by Wischmeier and Smith in 1978 (8). This model takes into consideration several determining factors, such as the soil erodibility factor, rainfall intensity factor, slope length and steepness factor, cover and management factor and support practice factor. USLE was developed mainly for soil erosion estimation in croplands or gently sloping topography. USLE estimates soil loss from a hillslope caused by raindrop impact and overland flow (commonly termed "interrill" erosion), plus rill erosion. It does not estimate gully or stream-channel erosion. Although USLE has many shortcomings and limitations, it is widely used, especially at reEARSeL eProceedings 8, 1/2009 42 gional and national level, because of its relative simplicity and robustness (9) and because it represents a standardised approach. USLE has not been designed to operate at field scale, however, it was noted that there is room for improving the accuracy of results by using more detailed digital elevation models, satellite data, with enhanced geometric characteristics, and more detailed soil information. A Revised Universal Soil Loss Equation (RUSLE) followed the same formula as USLE, but got several improvements in the determining factors and a broader application to different situations, including forests, rangelands and disturbed areas compared to USLE (10). RUSLE is a computation method that may be used for site evaluation and planning purposes and also for assisting in the decision process of selecting erosion control measures. It provides an estimate of the severity of erosion and also numerical results that can validate the benefits of planned erosion control measures in the risky areas (11). Aim and objectives In many countries including Greece, data availability plays a crucial role in selecting an appropriate erosion prediction model. Especially, the lack of reliable soil and climatic data is the main obstacle in implementing most models. Detailed soil properties maps are missing or they are not available in Greece, while the meteorological stations are not as dense as needed for a mountainous country. Additionally, as erosion is a multi-temporal procedure, seasonal land cover and use affect the accuracy of any mapping results. Therefore, land cover maps of static nature, such as CORINE, are not enough for deriving multi-temporal information. As a result of all the above, several potential data sources for obtaining seasonal land cover, soil characteristics, and rainfall estimations should be considered for any specific case. The main aim of this work was to test a modified version of the Universal Soil Loss Equation (USLE) for assessing the risk of soil erosion in N. Chalkidiki, Greece. For this purpose, the C-factor was calculated using multi-temporal vegetation indices (namely NDVI) derived from satellite images, whereas it is originally estimated by expertise. The k-factor was calculated based on the geological map, while it is originally based on soil analysis data. The NDVI was selected for estimating the C-factor because NDVI can express the condition of vegetation in different seasons (seasonal land cover), thus providing a reliable temporal dimension in soil erosion risk assessment. This is of great importance in Mediterranean regions, such as the studied one, where dry summers are followed by heavy winter rainfalls. The reason for calculating the k-factor from geological data was the lack of precise and reliable soil datasets in the study area. The testing hypothesis of the work was that the above modifications of C and k-factors were necessary and efficient for improving insight in understanding the erosion procedures in the study area. Study area The study area is located in the north part of Chalkidiki prefecture, Greece (Figure 1). Chalkidiki is a peninsula in the Aegean Sea, made up of three smaller peninsulas which from west to east are: Kassandra, Sithonia and Athos. Cholomon Mountain dominates the landscape. The region has very warm and dry summers: average temperature is between 23°C and 34°C during summertime; and between 4°C to 19°C during the winter. The area’s population is 110 000, and its capital city is Polygyros. The lowlands are covered by crop patterns of arable land, olive trees, grapes, and vegetables (mainly tomato and cucumber), while on the mountainous areas coniferous, broadleaved, and mixed forests are altered with bushes and pastures. DATASET AND METHODS Dataset The data that was available for the implementation of this work comprised the following: • A set of three (diachronic) LANDSAT-7 images acquired on 24 August 2000, 5 April 2001, and 2 November 2002 (Figure 2). • A representation of the relief in terms of Digital Terrain Model (DTM) derived from the Greek Ministry of Agriculture, on a scale of 1:50 000. EARSeL eProceedings 8, 1/2009 43 • A geological map derived from the Greek Institute of Geological Surveys (IGME), on a scale of 1:50 000. • A land cover/use map based on the CORINE nomenclature, used for assisting interpretation of the satellite images (Figure 3). • Monthly climatic average observations derived from three meteorological stations of the Greek Institute of Forest Surveys. 0 50 100 Kilometres 400 Figure 1: The study area is located in Chalkidiki, Greece and comprises forest and agricultural patterns.