The methodology presented demonstrates how probabilistic climate information can be applied to long-term assessment of clay-related geohazard potential. A modified version of the UKCP09 stochastic WG was used to derive projections of daily rainfall and PET. These values were processed, providing probabilistic PSMD values for the three time periods; the methodology is summarised in Fig. 1. PSMD scenarios, incorporating climatic uncertainty, were subsequently intersected spatially with soil data and reclassified using an existing geohazard model to ascertain clay-subsidence hazard (Fig. 1).
This paper’s methodology focuses principally upon modelling the hazard (extent, severity and probability) that clay-related subsidence presents. However, in order to present a risk, the built environment needs to be both exposed to the physical geohazard (i.e., clay-related subsidence), as well as being vulnerable to damage (i.e., shallow foundation). This is best represented by the following function, after (Crichton 2001):
$$ Risk={\displaystyle \int \left( exposure,\kern0.5em hazard,\kern0.5em vulnerability\right)} $$
(1)
The methodological approach is discussed in detail for the remainder of Section 2 below.
Natural perils directory
Since the early 1990s, a range of spatial soil-related geohazard models have been developed for GB (e.g., Hallett et al. 1994). The Natural Perils Directory™ (NPD) is a geohazard thematic dataset, constructed through reinterpretation of soil-survey mapping, that provides detailed information on a series of soil-related hazards for GB. Presently, NPD combines historic (1961–75) PSMD and soil shrinkage susceptibility (SSWELL) data to interpret clay-subsidence hazard. Soils are allocated six volumetric shrink/swell (SSWELL) classes, based upon laboratory data by Reeve et al. (1980), ranging from ‘Very Low’ (0) (<3 % volumetric shrinkage) to ‘Very High’ (6) (>15 % volumetric shrinkage) (Fig. 2). A ‘High*’ SSWELL class represents soils with alluvial clay or peat at 1 m depth, but which are only prone to shrinkage when effectively drained to at least 2 m depth.
To date, climatic extremes within NPD were modelled through the addition of standard deviations around the mean PSMD available to the model. Weaknesses of this approach include both the currency of the now historical time series of the data, and the fact that no effective probabilistic element is employed in the modelling allowing for the management of potential future uncertainty.
For a detailed summary of NPD and associated data, readers are referred to the supplementary material.
UKCP09 weather generator
The standard UKCP09 WG provides simulations of weather sequences on a site-by-site (i.e., 5 km cell) basis, and so lacks spatial consistency in time over neighbouring grid cells (Jones et al. 2009; Jenkins et al. 2014). Due to specific soil properties extending beyond the 5 km resolution we adopted a modified version of the UKCP09 stochastic weather generator (WG) (Burton et al. 2013). The WG used builds upon the earlier EARWIG WG (Kilsby et al. 2007) to compile spatially-coherent daily climate values over a 30 year stationary sequence at a 5 km2 resolution for GB. The 30 year sequences included the ‘baseline’, ‘2030’ and ‘2050’ time periods. Future projections were drawn from the medium emissions scenario, equivalent to the IPCC’s (Intergovernmental Panel on Climate Change) SRES A1B scenario (IPCC 2000). As with UKCP09, scenarios did not apply urban land-use corrections.
UKCP09 baseline data were produced to reveal the extent to which the WG is able to match baseline climate calculations with known empirical data (Eames et al. 2012). Figure 3 presents annual average totals of precipitation and potential evapotranspiration for the UKCP09 WG-derived baseline (Fig. 3a) and 2050 scenarios (Fig. 3b) which are compared with observed baseline data for GB. In baseline comparisons (Fig. 3a), both observed and UKCP09 WG-derived data show the same spread. However, for the 2050s, Fig. 3b suggests higher average annual potential evapotranspiration and reduced rainfall. The 30 year WG baseline series was run 100 times based on a different randomly sampled vector of change factors, providing the probabilistic analysis. The future scenarios of 2030 and 2050 represent a higher factor of uncertainty compared with the baseline. Therefore, these scenarios were run 1000 times, based on a differently randomly sampled vector of the 10,000 UKCP09 change factors available to provide the probabilistic analysis.
Unlike its predecessors (UKCIP98 and UKCIP02), UKCP09 does not provide projections of soil moisture. However, the WG does provide daily outputs of rainfall and potential evapotranspiration (PET), fundamental for calculating PSMD (Eq. 1). The following section discusses how projections of PSMD were derived from raw WG data. We then discuss how projected PSMD data were processed and incorporated within the clay-related subsidence geohazard model. The WG was also run to obtain low, medium and high emissions scenario PSMD projections for the administrative county of Worcestershire, results being provided in Figure S1 of the Supplementary Material.
Computation of soil moisture deficit
The WG produced substantial output data (≈50 Terabytes for GB). Custom software tools were required to process the raw WG files to produce the necessary summary data products required for geohazard modelling. A series of programs were prepared using the Perl scripting language to automate calculation of PSMD values (Fig. 1).
Soil moisture accumulation and loss oscillate over the course of a year. Therefore, a temporal resolution of monthly and annual PSMD were deemed more appropriate than the raw daily data format. Future scenarios represented 1000 daily realisations of climatic parameters which, over a 30 year time series, provided 30,000 realisations of daily climate. PSMD was calculated using the following equation (after Jones and Thomasson 1985):
$$ PSMD={\displaystyle \sum \left(PPT- PET\right)} $$
(2)
Where
- PSMD:
-
Potential Soil Moisture Deficit (mm)
- PPT:
-
Daily Rainfall (mm)
- PET:
-
Daily Potential Evapotranspiration (mm)
During computation, PSMD was set to 0 each January 1st and subsequently each day’s PSMD (Eq. 2) (if not a surplus) was added to the previous days’ PSMD to give an accumulated value. If a surplus of water existed (i.e., PSMD > 0) then it was subtracted from the previous days accumulated PSMD. For consistency with UKCP09 outputs, the mean, and the 10th, 50th and 90th percentiles were calculated, over the WG change factors, for the monthly and annual soil moisture data to represent data uncertainty; the 90th percentile being taken as ‘unlikely to be more than’, the 10th percentile being ‘unlikely to be less than’, and the 50th percentile representing the ‘central estimate/tendency’. Adopting standardised approaches in the representation of UKCP09 uncertainty allows users, who are likely to be familiar with the UK climate projections, to incorporate these modelled data within climate adaptation schemes.
Integration of climatic and geohazard models
The aim of this study was to supplant historical PSMD data in NPD with projections computed from WG data. To achieve this, we spatially referenced the WG-derived PSMD data to the 5 km WG grid cells (see http://ukclimateprojections-ui.metoffice.gov.uk/ui/docs/grids/wg_5km/index.php), intersecting this with the SSWELL data. Clay subsidence hazard was then calculated from the maximum accumulated PSMD and SSWELL using a Python script running within ArcGIS (v. 10.2). Clay subsidence hazard potential in NPD is portrayed with nine classes, ranging from extremely low to extremely high. This process was undertaken for each climatic scenario and for the 10, 50 and 90th percentile annual accumulated PSMD.