Keywords

Introduction

There is increasing recognition of the importance of safeguarding land health—the capacity of land to sustain delivery of essential ecosystem services—for global development and environment (MA 2005). Despite much attention paid to land degradation indicators and assessments (e.g. Vogt et al. 2011; Winslow et al. 2011), there is still a lack of scientifically sound approaches for defining, measuring, and monitoring land health at different scales. Basic problems include lack of proper sampling frames that permit inference to the population level from well-defined sample units (Cochran 1977) and lack of consistent measurement methods that can be applied across diverse landscapes so that results can be compared in wider spatial and temporal contexts. There is little hope for making progress on assessing land degradation until proper baselines and monitoring systems are put in place. Land health surveillance, developed by the World Agroforestry Centre, is an approach designed to overcome these limitations and is modelled on scientific approaches that have been used for decades in public health monitoring (Box 1).

Land health surveillance aims to provide high-quality spatial data on land health problems and risks to (1) help target sustainable land management interventions and (2) provide a baseline and method for monitoring changes in land health over time to be able to assess trends and impacts of interventions. The approach combines (1) consistent, georeferenced field measurements of vegetation and soil conditions and soil sampling, (2) use of new light-based (spectral) techniques for analyzing large numbers of soil samples at low cost, (3) use of remote sensing data to provide land health metrics at national to regional scales, and (4) scientific workflows that combine field, laboratory, and remote sensing data in statistical models to provide digital maps and quantitative information designed to support specific decision problems.

The Land Degradation Surveillance Framework (LDSF) is the basic field sampling methodology being applied in the Africa Soil Information Service which is sampling 60 randomized sentinel sites (10  ×  10 km sampling sites) throughout sub-Saharan Africa (Fig. 1). This is the first ever attempt at taking a ground-based, unbiased, population-level sample of African land condition. For example, these data enable quantitative analysis of the relationships between woody cover and distribution and soil health (Fig. 1).

Fig. 1
figure 1

Land health surveillance sampling frame deployed in the Africa Soil Information Service (www.africasoils.net). Sixty 10  ×  10 km sentinel sites are randomly located, stratified on major climate zones. Each sentinel site contains 160 spatially stratified randomized 1,000-m2 plots in which ground measurements are made. Within each plot, observations and soil samples are taken from 100-m2 subplots. Soil is characterized using infrared spectroscopy

Full size image

In this chapter, we illustrate some results of applying the LDSF methods in a rangeland management project in Laikipia, northern Kenya. The field measurements were conducted by a small private conservation company, Wajibu MS, in partnership with local communities, as part of a holistic rangeland management project in North Laikipia. A key feature of holistic management is the use of livestock to improve degraded and denuded lands, for example, by planned grazing to allow recovery periods for grass growth, rotation of bomas for manure management, and removal of unwanted invasive plant species. The land health surveillance methods were deployed to provide a baseline and monitoring framework for measuring impacts on land health at a landscape scale. The focus of this chapter is mapping soil carbon, as this is a key soil quality indicator.

Study Area

Field sampling was conducted in two LDSF sites encompassing 10,000 ha each within a 40,000 ha area, centred on the Sanctuary at Ol Lentille. One site was in the south (Ol Lentille) and one in the north (Kipsing) of the project area (Fig. 2). The project site is located 75 km northwest of Nanyuki in north central Kenya, transcending the borders of North Laikipia and Isiolo Districts, in the Mukogodo Division of Laikipia District, and Oldonyiro Division of Isiolo District. The study area falls under semiarid to arid agroecological zones and is a part of the Upper Ewaso Ng’iro North River Basin. The southern part is characterized by vertiginous highlands and rolling plateaus with an altitude of 1,600–1,800 m, while the northern section sits on an alluvial flood plain with altitudes of 1,100–1,250 m. Rainfall distribution is bimodal with peaks of “long rains” mid-March to mid-June and “short rains” mid-October to mid-December. Six broad vegetation types can be characterized: dense woodland, open woodland, sparsely shrubbed woodland, open shrubbed grassland, open grassland, and closed grassland. Woody species in the study area predominantly belong to the Mimosoideae family, while Eragrostea, Chloridea, and Stipeae families are the most dominant Gramineae. Historically, the area was used for dry-season grazing by nomadic pastoralists, but in recent years, burgeoning population and government development policies have led to increasing permanent settlements in the area.

Fig. 2
figure 2

Overview of the Sanctuary at Ol Lentille and the two LDSF (Land Degradation Surveillance Framework) sites in Laikipia, Kenya, included in this study

Full size image

Measurements

Sentinel sites may be selected at random across a region or watershed or may be placed to represent areas of planned activities (interventions) or special interest (Fig. 2). Within each site, 16 tiles (2.5  ×  2.5 km in size) are created, and random centroid locations for clusters within each tile are generated. Each cluster consists of ten plots, with randomized centre-point locations falling within a 564-m radius from each cluster centroid (Fig. 2). Thus, sentinel sites have two (or in some cases three) levels of randomization, which minimizes local biases that may arise from convenience sampling. Each plot has an area of 1,000 m2 and consists of four subplots of 100 m2 each. The coordinates of the plots are loaded onto a GPS (Geographic Positioning System), which is then used to navigate to each point.

Georeferencing and infiltration measurements are completed in the centre of the plot, and soil samples are composited to the plot level from the subplots for topsoil (0–20 cm) and subsoil (20–50 cm) depths. Soil samples are recovered using a soil auger and soil mass of samples recorded to allow calculation of soil carbon stocks on a soil volume basis. Soil depth to restrictions is recorded. A range of observations and measurements are made at the plot level to provide comprehensive information on ecosystem condition. These measurements have been designed and tested to be feasible under the difficult conditions in rural Africa. Further details of the protocol are given in Vågen et al. (2010a, b). A sentinel site can be completed by a small field team (five people) within 12–16 days, depending on the terrain and accessibility.

Soil infrared spectroscopy (Shepherd and Walsh 2007) is used as the main soil analytical tool due to its low-cost and high-throughput capability. Conventional soil testing methods are performed on 10–20 % subset of samples and calibrated to the infrared spectra, typically using partial least squares regression or multilevel regression models. For high-resolution mapping of soil organic carbon (SOC), QuickBird imagery was used and models developed based on Minimum Noise Fraction (Green et al. 1988) transposed image reflectance values, which were related to measured SOC values, allowing for prediction of SOC concentrations in the study area. The result is a wall-to-wall carbon map. Analytical methods are described in Vågen et al. (2010b) and Aynekulu et al. (2011).

Results and Discussion

The calibration of soil organic carbon concentration to near-infrared diffuse reflectance spectra was robust, as evidenced by the prediction performance of a hold-out validation set (Fig. 3). Predictions of SOC concentrations in the study area based on QuickBird MNF components are also robust, relative to near-infrared (NIR)-predicted values (Fig. 4) with an adjusted R2 of 0.82. In Fig. 5, the resulting map of SOC concentrations for the Sanctuary at Ol Lentille and surrounding areas is shown. Low carbon values are observed along the sandy river courses, and high values in hilly areas with higher tree densities. On average, Ol Lentille has about 3 g SOC kg−1 higher concentrations than Kipsing.

Fig. 3
figure 3

Measured SOC, soil organic carbon (log g kg−1), using combustion against values predicted from near-infrared spectroscopy for calibration and validation data sets

Full size image
Fig. 4
figure 4

Soil NIR-predicted values of SOC versus QuickBird Minimum Noise Fraction-predicted soil organic carbon for the study area in Laikipia, northern Kenya

Full size image
Fig. 5
figure 5

Map of topsoil SOC (g kg−1) for the Sanctuary at Ol Lentille and surrounding areas, Kenya

Full size image

The soil carbon information can be combined with other information on soil physical degradation risk, soil fertility status, topographic data, and vegetation characteristics to assist in targeting agroforestry and sustainable land management interventions. For example, areas that have low soil organic carbon levels (relative to areas under good vegetation cover and similar soil texture) but low prevalence of other constraints may be prioritized for kraaling to increase manure inputs and increase grass productivity. Areas with low soil carbon saturation in areas with otherwise good soil fertility potential may be targeted for conservation agriculture practices incorporating trees. Areas with inherent soil physical constraints such as restricted soil depth and low infiltration capacity, but with currently low tree cover, can be flagged as environmentally sensitive or marginal areas to which afforestation efforts may be directed.

Conclusion

Combining systematic ground measurements and soil sampling with high-throughput soil analytical methods based on infrared spectroscopy and remote sensing imagery, linked up using multivariate multilevel statistical modelling, has potential to enable land and soil health monitoring in landscapes. Routine use of these tools at different scales can help land users and governments make better, evidence-informed decisions on land management as part of everyday policy and practice. Further work is needed to combine the various indicators of land and soil health into a framework for providing specific land management recommendations.