Operationalizing digital soil mapping for nationwide updating of the 1:50,000 soil map of the Netherlands

Highlights

• For the first time pedometric mapping was used in a Dutch nationwide mapping program.

• The national soil map was updated through updating of quantitative soil properties.

• A two-step simulation approach was implemented to allow modeling of zero-inflated data.

• Legacy point data were updated and integrated with newly acquired data.

• Uncertainty of updated legacy data was quantified and accounted for by the model.

Abstract

This paper presents a pedometric approach to updating the Dutch 1:50,000 national soil map for the peatlands, and illustrates this approach for a 187,525 ha area in the northern peatlands. This is the first time that digital soil mapping replaces conventional soil mapping in a nationwide, government-funded soil survey program in the Netherlands. Soil classes were updated indirectly through mapping two quantitative diagnostic soil properties: the thickness and starting depth of the peat layer. From these, five major soil groups could be constructed. Because the point data were zero-inflated, a two-step simulation approach was implemented. First, peat presence/absence indicators were simulated from probabilities of peat occurrence that were predicted with a generalized linear model. Second, conditional peat thickness values were simulated from kriging with external drift predictions. The indicator and peat thickness simulations were combined to obtain simulations of the unconditional peat thickness. A similar approach was followed for the starting depth. From the simulated soil properties, probability distributions of soil groups were derived. These groups were refined with information on (static) soil properties derived from the 1:50,000 map to obtain soil classes according to the 1:50,000 legend. The updated raster map was then incorporated in the 1:50,000 polygon map. The prediction models were calibrated with legacy point data, that were updated for peat thickness before being used, in addition to a set of newly acquired point data. The uncertainty associated to the updated peat thickness values in the legacy dataset was quantified and accounted for by the prediction models. The peat thickness map and a map with three soil orders were validated with independent probability sample data. The overall purity of the soil order map was 66% for both subareas. For subarea 1 this was a 12% purity improvement compared to the current 1:50,000 map, for subarea 2 this was 3%. For subarea 1, the mean absolute error of the predicted peat thickness was 23.5 cm, and the R² is 0.50. For subarea 2 these accuracy measures were 30.9 cm and 0.65. We conclude that nationwide updating the 1:50,000 map with pedometric techniques is feasible. In order to increase the value and usability of the legacy point data as well as the large set of newly acquired field observations and the updated 1:50,000 map, we recommend installation of a soil monitoring network in the Dutch peatlands.

Introduction

The national soil map at scale 1:50,000 (Steur and Heijink, 1991) is the main source of soil information in the Netherlands. This map was initially created for soil suitability analysis of various land-use systems (van Lynden et al., 1985, Sonneveld et al., 2010), but is since the 1990s increasingly used for environmental and agro-economic analyses in support of policy-making. Examples include modeling of nutrient and pollutant fluxes in the soil (van der Salm et al., 1996, Hack-ten Broeke et al., 1999, Kros et al., 2011), inventories and monitoring of carbon stocks (Schulp and Veldkamp, 2008, Reijneveld et al., 2009), modeling soil subsidence (Hoogland et al., 2012), implementation of the EU Thematic Soil Strategy (Bouma and Droogers, 2007) and simulation studies on greenhouse gas emissions from peat soils (Nol et al., 2010, van Beek et al., 2011).

Organic soils cover 527,000 ha, or almost 16% of the land surface area of the Netherlands. The Dutch soil classification system (de Bakker and Schelling, 1966) distinguishes two main types of organic soils: peat soils (peat layer > 0.4 m thick and starting within 0.4 m from the surface) and peaty soils (peat layer 0.05–0.4 m thick and starting within 0.4 m from the surface). Intensive agricultural use and deep drainage of these soils have resulted in major changes in soil conditions since the completion of the 1:50,000 survey in 1995 (the first map sheets date from the 1960s). The peat oxidation and compaction rate is estimated between 5 and 10 mm year^− 1 (van den Akker et al., 2008, Hoogland et al., 2012). As a consequence, peat soils may have changed into peaty soils, and peaty soils into mineral soils. A reconnaissance survey of peat soils in the eastern part of the Netherlands showed that the acreage of peat soils was reduced by about 50% (van Kekem et al., 2005, de Vries et al., 2009). This clearly illustrated the need for an updated soil map.

The Dutch national government recognizes the importance of good quality, up-to-date soil information and has acknowledged the need for a map update. The government commissioned an extensive updating program that aims to have an updated map for 365,000 ha of peatlands ready by the end of 2014. The deep peat soils (peat layer starting within 0.4 m below the surface and extending deeper than 1.2 m below surface) of the western and northern fen peat areas (162,000 ha) were not part of the updating program. The peat layer in these areas can be up to 6 m deep. Here, oxidation and compaction of peat do not directly result in a change of soil class. The national soil map was, therefore, considered to be up-to-date for these areas. Updating by means of conventional soil mapping started in 2009, but this soon turned out to be too expensive and too slow. In 2011 it was decided to continue the map update program by means of digital soil mapping (DSM) (McBratney et al., 2003). Kempen et al. (2012c) have shown that DSM can be an efficient alternative to conventional soil mapping for updating the national soil map in terms of accuracy and costs, but the use of DSM for map updating in the Netherlands has so far been experimental and applied to small case study areas only (Kempen et al., 2009, Kempen et al., 2012b). This was the first time that DSM was going to be made operational in a government-funded, nationwide soil mapping program.

Making DSM operational for updating the national soil map for the peatlands brings two challenges. The first relates to the use of point data from different sources and different quality. The Dutch soil information system BIS stores spatially referenced soil profile descriptions from over 325,000 locations, which were collected during surveys and research projects since the 1950s. These data are an important resource for DSM (Carré et al., 2007, Sulaeman et al., 2013), but, given their age, may not properly represent actual field conditions for dynamic soil properties such as the thickness of the peat layer. Though this might limit their utility for updating, these data may still provide relevant information. Since we aimed to make most out of existing data, we decided to update the legacy soil point data and use these data, together with newly acquired field data, in our modeling framework. Hereby taking into account the uncertainty associated to the updated point data.

The second challenge concerns the mapping methodology. Kempen et al., 2009, Kempen et al., 2012b have shown that pedometric mapping of soil classes can be challenging, especially when one wants to model the spatial correlation structure. Calibrating a generalized linear geostatistical model (GLGM) is complex and computationally demanding (Diggle et al., 1998, Christensen, 2004). Alternatively, multinomial logistic regression, which is much easier to implement than the GLGM, has as disadvantage that certain structures in the data can cause numerical problems when fitting the model in the presence of categorical covariates (Hosmer and Lemeshow, 2000). Furthermore, the spatial correlation structure is not accounted for. The same holds for (boosted) classification trees and random forests that become increasingly popular for mapping categorical soil attributes (e.g. Heung et al., 2014, Odgers et al., 2014, Pahlavan Rad et al., 2014, Subburayalu et al., 2014). We, therefore, decided to take a different approach to updating soil class maps. Instead of mapping soil classes, we mapped (continuous) key diagnostic soil properties as proposed by Kempen (2011). The soil classes of the 1:50,000 legend are defined by a set of measurable soil properties. We focus on those properties required to distinguish peat soils, peaty soils and mineral soils, which are the thickness and starting depth of the peat layer. The type of peat, peaty or mineral soil class according to the 1:50,000 legend can then be reconstructed by refining the predictions with information on (static) properties obtained from the current soil map. This mapping approach also better suits the information demand by several soil data users, who have expressed interest in quantitative information about the thickness of the peat layer instead of qualitative information in the form of a peat soil class.

The aim of this paper is to describe, illustrate and validate the geostatistical mapping approach for updating the Dutch national soil map at scale 1:50,000 for the northern peatlands.