Targeting critical source areas for phosphorus losses: Evaluation with soil testing, farmers’ assessment and modelling

The farms selected for the study were 16 demonstration farms included in the project Farming in Balance (FiB, http://​www.​odlingibalans.​com/​). These farms are located in arable areas of southern and central Sweden stretching from Skåne in the south to Dalarna in the north (Fig. 1). The characteristics of each farm are summarised in Table 1. They cover a wide range of climate, pedological, hydrological and production conditions. These particular farms within the FiB project are intended to serve as a bridge between research and practical farming, making them very suitable as study areas for the present analysis. Data on all fields and parcels belonging to each farm were downloaded as GIS vector layers from the Swedish Board of Agriculture database in the form of agricultural blocks. These blocks were then imported to Google Earth to create high-resolution images of each farm. These images were then sent to the 16 farmers for identification of CSAs based on their own experience and observations.

In order to quantify potential P mobilisation from the most vulnerable parts of the fields, soil sampling was carried out in the vulnerable areas identified through distributed modelling with high-resolution DEM performed for this study (see Section “Modelling and farmer evaluation of overland flow and erosion” below for a detailed description of the modelling process). A total of 163 soil samples were collected in the immediate vicinity of the modelled erosion pathways, identified as the top 2% of all 2m × 2 m cells with the highest erosion values on each of the 16 farms. The selection of sampling points was also based on soil maps in order to cover different soil textural classes. Each sample (10 cm deep) consisted of 15 soil cores collected from an area of 1 m². These soil samples were air-dried, gradually broken down by hand and sieved (<5 mm) before analysis with DESPRAL test, and the content of plant-available P was determined by extraction with ammonium lactate/acetic acid (P-AL) at pH 3.75 (Egnér et al. 1960). The risk of sediment, dissolved and particle-bound P mobilisation was estimated with the DESPRAL test, performed as described by Withers et al. (2007), who showed that the results of the DESPRAL test correlated well (r ² = 0.7–0.8) with the amounts of SS, total P and dissolved P in overland flow generated by indoor simulated rainfall. DESPRAL is an environmental soil test developed to estimate the intrinsic risk of sediment and P mobilisation from agricultural soils, where both dispersed particles and P are simultaneously quantified (Villa et al. 2014). Suspended solids (SS), total P (TP) and dissolved P (DP) were determined in DESPRAL aliquots in accordance with the methods issued by the European Committee for Standardization (European Committee for Standardization 1996). Suspended solids were determined by filtration through 0.2-μm pore membrane filters dried at 105 °C while turbidity, which is highly correlated to SS (Villa et al. 2014), was measured on post-dispersion aliquots using a Hach 2100AN instrument (Hach Company, CO) and expressed as nephelometric turbidity units (NTU). Total phosphorus was determined colorimetrically after digestion of unfiltered samples in acid persulphate solution, DP was determined on filtered samples (0.2-μm pore membrane filters) using Gallery Plus Photometric Analyzer Thermo Fisher. Unreactive P (UP) was calculated as the difference between TP and DP.

Plant-available soil P, potassium (K), calcium (Ca) and magnesium (Mg) concentrations were determined by extraction with ammonium lactate/acetic acid (P-AL) at pH 3.75 (Egnér et al. 1960), which is the standard agronomic soil P test used in Sweden. The same extraction was also used to analyse iron (Fe) and aluminium (Al) as indicators of soil P sorption capacity (Ulén 2006). Simple and multiple regression analyses were performed to study relationships between the results from AL analyses and DESPRAL tests. One-way analysis of variance (ANOVA) with a Fisher comparison test was used to test differences between P-AL classes and soil textural classes regarding DP and UP release in DESPRAL tests. All statistical analyses were performed using Minitab 16.1.1.

All modelling simulations were performed prior to the farmers’ self-evaluations as described in Djodjic and Villa (2015). The base layer for the modelling work was a DEM in raster format. A 2-m grid based on LiDAR data was used, with a density of 0.5–1 point m⁻² and accuracy which is usually better than 0.1 m (Lantmäteriet 2014). The modified USPED model (Mitasova et al. 2001) was implemented within a frame of PCRaster software for environmental modelling (Schmitz et al. 2009). In brief, USPED is a simple model which predicts the spatial distribution of erosion and deposition patterns based on the change in overland flow depth and on the local geometry of terrain, including both profile and tangential curvatures. The slope length factor (LS) of the RUSLE equation is replaced with upslope contributing area in the modified model and the LS factor is calculated usingwhere A is the upslope contributing area and b is the slope angle. Exponent values of 1.6 and 1.3 were used here, as recommended by Mitasova et al. (2001). In the modified USPED, the erosivity factor (R), soil erodibility factor (K) and vegetation cover factor (C) were used in accordance with the following: the catchment-specific mean annual runoff (Table 1) was used as the rainfall erosivity factor (R), the values of soil erodibility factor (K) were based on the new soil map of textural classes of Swedish agricultural soils (Paulsson et al. 2015), in combination with soil maps from the Geological Survey of Sweden for non-agricultural areas, and each soil textural class was assigned a specific K value according to Stone and Hilborn (2012). Land use map and cover factor (C) values from Stone and Hilborn (2012) were combined to spatially distribute the effects of vegetation cover. Since the aim of the modelling was to compare and rank relative long-term erosion and overland flow risk, all arable soil was assigned the same cover factor (C) representative for cereal crops (C = 0.35), without consideration for the actual crop distribution. In order to separate and better visualise the sub-areas of agricultural land most prone to overland flow, erosion and water ponding, the results obtained in erosion modelling were post-processed as described in Djodjic and Villa (2015). In short, using the “Slice” tool with the “Equal Area” method and 50 output zones within ArcGIS 10.2.1 (©1999-2013 Esri Inc.), 2-m grid cells were reclassified and ranked according to modelled erosion vulnerability. This approach allows incremental identification of CSAs starting with the top 2% of total agricultural area with the highest erosion values according to modelling results and thereafter, if necessary, stepwise 2% increases.

A meeting was organised to present modelling results and allow farmers to work with self-evaluations of CSA risk based on their own knowledge of their fields. At the meeting, the farmers were first given a short introduction and examples of how to consider and report (describe and draw on the map) different types of CSAs, such as frequent overland flow pathways, erosion channels and routes, frequent occurrence of flooding and ponding water on the fields, inadequate drainage and compacted soils. All observations drawn on maps by farmers were then digitised for comparison with the modelled values.

It should be stressed that the modelling work was completed prior to farmers’ self-evaluation. After the self-evaluations were completed, the model results were then compared with the self-evaluations directly at the same meeting. No calibration of the model or its parameters was made to achieve a possible better fit with farmer observations. The abovementioned 2% top-ranked cells were then compared against the CSAs identified by farmers, using the “Selection by location” tool within ArcGIS 10.2.1 (©1999-2013 Esri Inc.), which identified all observed areas that intersected with the modelled areas. The comparison was discussed with farmers and farmers’ reactions and impressions were documented.

At the same meeting mentioned above, all farmers included in the FiB project (n = 16) were asked to perform a SWOT (strengths, weakness, opportunities and threats) analysis (Learned et al. 1969) to assess not only the internal strengths and weaknesses of their farm, but also external opportunities and threats (Grošelj and Zadnik Stirn 2015). The SWOT analysis is widely applied in strategic decision support for business management, but recently it has also been used for environmental management and assessment (Scolozzi et al. 2014). Prior to the SWOT analysis, the farmers were introduced to the concept of P index (Lemunyon and Gilbert 1993), where factors governing P losses were grouped into two categories: P sources and P transport pathways. The farmers were asked to categorise their statements in the SWOT analysis into one of these two categories or a third category (Other).