This document is part of a project evaluating the importance of hill farming in Scottish communities. There is no existing classification of hill farming in Scotland, therefore an index of hill farming needed to be created to evalute its contribution. We chose to do this at an agricultural parish level using datasets describing landscape.
Results will be aggregated to agricultural parishes, as this administrative level is widely understood and is appropriate for anonymising data. The 2016 agricultural parish data were downloaded from: https://data.gov.uk/dataset/939fdd5e-7322-4ab7-9dc9-bbfc538c4477/agricultural-parishes.
## OGR data source with driver: GPKG
## Source: "/home/mspencer/Cloud/Michael/SRUC/hill_farms/data/spatial/ag_parishes_2016.gpkg", layer: "simplified_parishes"
## with 932 features
## It has 4 fields
## OGR data source with driver: GPKG
## Source: "/home/mspencer/Cloud/Michael/SRUC/hill_farms/data/spatial/ag_parishes_2016.gpkg", layer: "Scotland"
## with 71 features
## It has 1 fields
2016 agricultural parishes
The above map shows the 886 agricultural parishes in Scotland. These vary in size between 4.4 and 1121 km².
A simple view of hill farming is of an activity which takes place in hills, but how are hills defined? Measuring hilliness by elevation would diminish the challenges of farming smaller hill areas like the Pentlands and Campsies, because they are a third of the height of the biggest mountains in Scotland. Incorporating slope into a measure of hill farming would help overcome using only elevation as a measure of hilliness because slopes can be steep without hills needing to be high.
There are many datasets available which we can use to further form an idea of hill farming areas. We have collected data from a variety of sources to help define hill farming:
## Warning: Column `PARName` joining factor and character vector, coercing
## into character vector
## Warning in `[<-.factor`(`*tmp*`, list, value = 0): invalid factor level, NA
## generated
## Warning in `[<-.factor`(`*tmp*`, list, value = 0): invalid factor level, NA
## generated
Data described in the previous section were available in a range of formats, including raster (grid), vector (polygons) and plain text. In order to assess all these data at an agricultural parish level, these data were aggregated and converted to a common format. Following this process data were converted to a proportion of hilliness in each parish, e.g. for rural urban classification the proportion of each parish classified as rural was calculated.
Exact specification for this process is detailed in:
grass_read.R
which reads files into GRASS GISgrass_analysis.R
clips polygon data to agricultural parishes and measures subsequent polygon sizes. Terrain values (elevation and slope) are also extracted from raster data.read_hill_class.R
combines data from the GRASS analysis and aggregates to parish levels. Crucially, at a parish level proportion of land is reported as being less viable, i.e. higher proportions mean a parish is more likely to be hill farming.The Ordnance Survey Terrain 50 digital elevation model was used to get terrain data for each parish. This gridded model has a 50 m cell resolution and each cell is attributed an average height of land contained within it. This elevation model was used to calculate slope, also on a 50 m grid. For the slope and elevation datasets summary values were extracted for each agricultural parish; these values were minimum, mean, median and maximum.
The four measures of elevation were compared in order to select a single indicator. These comparisons are shown below, with the numeric values being Pearson correlation coefficients.
Mean (average), median and maximum elevation are closely correlated, but there is variability between each of these and minimum elevation. This is due to within-parish elevation variation, e.g. some parishes will have a range from sea level to mountain top, but other parishes a occupy more level terrain. The elevation range within a parish could have been used, but this would largely duplicate slope, i.e. a big elevation range within a parish should produce a steeper slope.
Mean parish elevation has been chosen as a single measure of parish elevation. This because it is a central measure which captures the variation of the minimum, median and maximum best (correlations of 0.77, 0.99 and 0.94 respectively).
Map of mean parish elevation
As for elevation, slope measures were compared to each other (below). This plot matrix shows scatter plots between the measures in the lower left and Pearson’s correlations in the upper right.
As with elevation, the average (mean and median) slope measures correlate well. There is greater variability when compared to maximum slopes, but the largest variation is when comparing to minimum slopes.
A mean (average) slope has been used as a single parish slope indicator because it has the best correlation with median or maximum slope (0.98 and 0.71 respectively).
Map of mean parish slope
The common grazing data were supplied by RESAS at a parish level in a csv format, so needed no format conversion or aggregation preparation. Common grazing data were sent with units of Hectares per parish, these were converted to m² before deriving proportion of common grazing within each parish. Common grazing is most dominant in the Outter Hebridies and North West Scotland.
Map of parish common grazing proportion
Rough grazing areas were taken from the June Agricultural Census (2011).
Map of parish common grazing proportion
The SNH landscape character assessment has three levels. Level 1 has the most different landscape classes and level 3 has the least. As this work is concerned with aggregating data to define parishes, level 3 data were used as a hill farming indicator. Level 3 data were split as follows to define hill farming areas:
Flat or Rolling, Smooth or Sweeping, Extensive, High Moorlands of the Highlands and Islands Inland Loch
Highland Straths
Moorland Transitional Landscapes of the Highlands and Islands
High, Massive, Rolling, Rounded Mountains of the Highlands and Islands
High Massive Mountain Plateau of the Cairngorms
Smooth Upland Moorland Hills
Highland Foothills
Upland Igneous and Volcanic Hills The Ochil, Sidlaw, Cleish and Lomond Hills
High, Massive, Rugged, Steep-Sided Mountains of the Highlands and Islands
Sea Lochs of the Highlands and Islands
Highland and Island Rocky Coastal Landscapes
Peatland Landscapes of the Highlands and Islands
Rocky Moorlands of the Highlands and Islands
Rugged, Craggy Upland Hills and Moorlands of the Highlands, including the Trossachs
Low Coastal Hills of the Highlands and Islands
Coastal Hills Headlands Plateaux and Moorlands
Lowland Hills
Foothills and Pronounced Hills
Upland Hills, The Southern Uplands and Cheviots
High Plateau Moorlands
Rugged Granite Uplands
Rocky Volcanic Islands
Rugged Moorland Hills
Upland Fringe Moorland
Upland Hills, The Lammemuir, Pentland and Moorfoot Hills
Rocky Coasts Cliffs and Braes of the Lowlands
Knock or Rock and Lochan of the Islands
Highland Cnocan
Coastal Island
Highland and Island Forested Landscape
Highland and Island Crofting Landscapes
Farmlands and Estates of the Highlands and Islands
Highland and Island Glens
Lowland Loch Basins
Lowland Cities, Towns and Settlements
Lowland River Valleys
Low, Flat, and / or Sandy Coastal Landscapes of the Highlands and Islands Highland and Island Cities, Towns and Settlements
Agricultural Lowlands of the North East
Lowland Coastal Landscapes of the North East
Lowland Valley Fringes
Lowland Rolling or Undulating Farmlands, Hills and Valleys
Flatter Wider Valleys and Floodplains of the Lowlands
Coastal Raised Beaches and Terraces
Lowland Hill Margins and Fringes
Lowland Coastal Flats Sands and Dunes
Lowland Loch and Shore
Narrow Valleys in the Lowlands
Coastal Margins
Lowland Urbanised Landscapes
Upland Glens, Valleys and Dales
Upland Fringe Valleys and Farmlands
Low Coastal Farmlands
Lowland Plateaux and Plains
Drumlin Lowlands
Upland Basin
As can be seen, some of these descriptions are a little ambiguous. For example “Upland Basin” describes both a hill and glen feature.
Following this separation of landscape classes, the proportions in each parish were calculated, these are mapped below. The upland areas of the Cairngorms, Western Highlands and North West Scotland dominate this map.
Map of parish SNH upland landscape proportion
SNH produce a map of carbon rich soils and peatland, these are polygon data with each polygon taking a categorical variable. These classifications were split into those with peat soils and those without. These classifications are listed below and are taken from the following report.
Class | Description |
---|---|
1 | All vegetation cover is priority peatland habitats. All soils are carbon-rich soils and deep peat. |
2 | The vegetation cover is dominated by priority peatland habitats. All soils are carbon-rich soil and deep peat. |
3 | Dominant vegetation cover is not priority peatland habitat but is associated with wet and acidic type. Occasional peatland habitats can be found. Most soils are carbon-rich soils, with some areas of deep peat. |
5 | Soil information takes precedence over vegetation data. No peatland habitat recorded. May also show bare soil. All soils are carbon-rich soil and deep peat |
Class | Description |
---|---|
-1 | Unknown soil type: information to be update when new data are released. |
-2 | Non-soil: loch, built up area, rock and scree. |
0 | Mineral soils: peatland habitats are not typically found on such soils. |
4 | Area unlikely to be associated with peatland habitats or wet and acidic type. Area unlikely to include carbon-rich soils. |
Those believed to indicate hill farming constraints were used to calculate the proportion of agricultural parish within them and are plotted below. North West Scotland appears as the highest proportion using these measures.
Map of parish SNH peat map proportion
The James Hutton Institute produce a land capability index for agriculture based on soil, climate and relief. We downloaded the 1:250k resolution version, as it covers the whole of Scotland. The vector dataset classifies areas of Scotland by their agricultural capability, these classes are:
Polygons classed as 6 or 7 were aggregated and the proportion of agricultural parishes lying within them calculated. These are shown in the following map, which has the highest proportions in the central and western Highlands.
Map of parish JHI agricultural capability proportion
The previous section describes a range of indices which can indicate hill farming areas. We have combined to produce a single index of hill farming areas.
The majority of the indicies have maximum values of one, because they show proportions. However, elevation and slope have maximum values above one. To create an unweighted index slope and elevation have been normalised by dividing values by their series maximums. These individual indicies were then summed for each parish to create a single index of hill farming for Scotland.
The results of this summation can be seen in the map below. The Cairngorms score highest, with areas in the central and western Highlands, North West Scotland and the Outer Hebridies also scoring highly.
Map of parish hill farming index
The index of hill farming was ranked to show relative positions of each parish. These results are below:
Map of parish hill farming index rank
The top 15 ranked parishes are:
## # A tibble: 15 x 1
## PARName
## <chr>
## 1 Crathie and Braemar
## 2 Ardgour
## 3 Alvie
## 4 Laggan
## 5 Harris
## 6 Glenelg
## 7 Kilmallie
## 8 Lochbroom
## 9 Applecross
## 10 Lochcarron
## 11 Contin
## 12 Glenshiel
## 13 Kintail
## 14 Assynt
## 15 Eddrachilles
We can use the individual indicies to group similar parishes together. First, we can inspect the relationships between indicies in the plot matrix below. Numbers in the upper right panels are Pearson correlations.
Comparison of hill farming indices by parish
The most well correlated variables are the JHI land capability indicies and the SNH Peat and Carbon map. These indicies use JHI soil data, so relationships are expected.
We can use a k-means model to group similar parishes together. K-means works by looking for common averages to cluster data points around. The following sections detail this work.
A k-means model was built using standarised elev_mean, slope_mean, grazing_prop, landscape_class_prop, land_cap_ag_prop, land_cap_forest_prop, wildland_prop, rural_prop, NNR_prop and peat_prop variables. The optimum number of cluster were determined using the elbow method, where we try to minimise the total intra-cluster variation. In this case, the below plot indicates two clusters may be appropriate, however these data are a little messy.
Elbow plot for determining optimal cluster number
These two clusters can then be mapped to our parishes as two levels: hill farming areas and not hill farming areas (below). As we expect, these are largely similar to the hill farming index, but partitioned into two groups.
Map of parishes grouped by hill farming indices
Map of parishes grouped by hill farming indices
We can see the difference in hill farming index against the two cluster groups below:
Plot comparing hill farming index to clusters of hill farming
Plot comparing hill farming index to clusters of hill farming