OR/15/030 Results and their interpretation

From MediaWiki
Jump to navigation Jump to search
Nicole Archer, Barry Rawlins, Stephen Grebby, Ben Marchant and Bridget Emmett. 2015. Identify the opportunities provided by developments in earth observation and remote sensing for national scale monitoring of soil quality. British Geological Survey Internal Report, OR/15/030.

We have presented our findings by first summarising the counts of the tagged keywords in the reference database through combining the information from the literature search, the internet search and the expert survey. Secondly, we have assessed each of the indicators separately through addressing the list of 5 key questions which were presented in the introduction. We summarise the information in a table describing the most promising remote sensing approaches which might be used for each indicator and the practicality of implementation. We chose to combine some of the indicators into two groups: i) nutrients (extractable K, extractable Mg, Olsen extractable P, total N) and ii) metals (Cd, Zn, Cu, Ni).

Tagged database summary

Figure 2 The number of soil monitoring and remote sensing studies which refer to or utilise the specified a) soil indicators, and b) sensors/platforms (based on a total of 185 references selected from peer-reviewed published literature).

Figure 2a shows that overwhelmingly the most common soil indicator which has been measured (or referred to) by soil remote sensing approaches is SOC (n=118 keyword citations). This dominance is likely for two main reasons: i) the importance of soil organic carbon as a mediator of global climate change and its general use as a soil quality indicator, ii) organic carbon signatures can be detected using hyperspectral data because the frequency of the associated bonds occur within the visible and near infra-red spectral range (0.35–2.5 µm).

Soil organic carbon (SOC) and total nitrogen (N)

SOC: There are many published studies which show that remotely sensed reflectance data can be used, via the construction of statistical models between the spectra and direct measurements of SOC, to make predictions of the concentration of SOC (status) at sites where soils samples have not been collected (Ladoni et al., 2010). The spectral reflectance (or spectral absorption) data encompasses the ultra violet (UV), visible (VIS), near infrared (NIR) and mid infrared (MIR) regions. However, there are currently few satellite-based observations that encompass the MIR range which typically have stronger correlations with SOC than the NIR range (Ladoni et al., 2010).

One of the problems of using spectral reflectance data is that mineral absorption features typically overlap with organic compounds so statistical models must be developed for particular regions and cannot necessarily be applied with the same confidence to other regions. Soil spectral reflectance and SOC relationships are not universal, and therefore require new statistical models to be constructed for each study area and field samples need to be collected and analysed (Ladoni et al., 2010). Weak statistical correlations may be observed between SOC and soil reflectance when soil samples are taken from large geographic areas with different parent materials or different landscapes where reflectance response is dominated by soil factors other than organic matter content (Henderson et al., 1992[1]).

Gomez et al. (2008)[2] used a combination of proximal spectrometer data and Hyperion hyperspectral information to construct a statistical model and map SOC. In their study, Gomez et al. (2008)[2] concluded that predictions of SOC using ground-based spectrometer data were more accurate than the Hyperion spectra, although the SOC map predicted using Hyperion data showed similarity with field observations. Selige et al. (2006)[3] showed that airborne hyperspectral remotely sensed data could be used to accurately map SOC concentrations at field scales,. In their study encompassing Europe, Stevens et al. found that the errors in predicting SOC concentrations from visible and near infra-red (VNIR) spectra were 5 times greater than for traditional soil sampling and laboratory analysis. The best spectral calibrations achieved a root mean square error ranging from 4 to 15 g C kg-1 for mineral soils and 50 g C kg-1 for organic soils.

A direct relationship between SOC and soil reflectance is only measurable when SOC is greater than 2%, otherwise the SOC signal is concealed by the presence of other biochemical components such as iron and manganese (Al-Abbas et al., 1972[4]). Such low SOC concentrations can cause difficulty in deriving absorption features through visual inspection alone, due to a combination of overlapping absorption features. Consequently, more complex spectroscopic modelling approaches are needed to derive mathematical relationships between soil spectra and soil constituents (Croft et al., 2012[5]).

Two reflectance bands (near infra-red and red) have been used from both airborne and remotely sensed imaging based on the ‘soil line’ concept. For any un-vegetated soil surface, there is an approximate linear relationship between the NIR and red reflectance bands referred to as the ‘soil line’ with which other soil properties are related. This shows bright soils are more reflective in both the IR and red, whilst dark soils are the least reflective in both reflectance bands (Richardson and Wiegand, 1977[6]). As concluded by Baret et al. (1993)[7] and Nanni and Dematte (2006)[8], experimental results show that a 'global soil line' does not apply and that the main factor of variation of ‘soil line’ parameters appears to be soil type. In their review, Ladoni et al. (2010) indicate that more recent research has shown that ‘SOC is correlated to a pixel’s location along the soil line’ using the soil Euclidean distance (SLED) technique (Fox and Sabbagh, 2002[9]). Soil sampling and measurement of these samples for SOC provide a descriptive curve relating SLED and field SOC. This method has been shown to reduce field sampling and was found to provide good correlations with SOC (R2 = 0.72 to 0.76) (Fox and Sabbagh, 2002[9]). Both NIR and red reflectance bands are available from airborne infra-red cameras and satellite platforms that encompass the UK, but these data have not, to date, been combined with ground based analyses for application to national scales in England and Wales.

Based on a combination of ground-based sampling and airborne radiometric survey, Rawlins et al. (2009)[10] showed that SOC concentration could be mapped with smaller prediction uncertainties by including measurements of radiometric K data as a secondary covariate, along with elevation data. This approach can only be applied from low-flying airborne or ground-based systems, so it may have limited applicability to national-scale approaches. All the published studies have demonstrated that remotely sensed data can contribute to improved mapping of SOC concentration [status], but none have shown that remote methods can monitor change, either with or without ground-based measurements. In their review, Ladoni et al. (2010) conclude that the main advantage of remotely sensed data (suggested by most researchers) is that 'they can be used to design a sampling scheme for mapping SOC [status] with the smallest number of samples, and with greater accuracy'. They also propose that:

  1. Remotely sensed data should first be a guide for field soil sampling, rather than directly predicting soil properties.
  2. Satellite data should be used as secondary information as geostatistical analysis has shown the potential to improve predictions of SOC.

Total N: There is a relatively restricted range of element ratios between total C (carbon) and total N (nitrogen) in soil organic matter and it is possible to use pedo-transfer functions (PTFs) to estimate total N on the basis of other measured soil properties, including soil carbon and bulk density, with only modest prediction errors (Glendining et al., 2011[11]). However, errors for predicting total N based on the application of PTFs, using EO measurements of these soil properties as predictors, would likely be too large for practical use. It has been shown that total soil nitrogen can be predicted with reasonable accuracy based on statistical models of visible and near infra-red (VNIR) laboratory spectra (Viscarra Rossel et al., 2006[12]) and also using airborne, hyperspectral remote sensing data in combination with ground-based measurements (Selige and Schmidhalter, 2005). There were two studies in which remotely sensed data had been used to enhance predictions of ground-based measurements of total nitrogen. First, a study by Wu et al. (2009)[13] applied this approach in Hengshan County, the northern Shanxi Province of China, using Hyperion satellite data. Second, a study by Stamatiadis et al. (2013)[14] applied the technique to three, 10 hectare fields in Greece. They showed that the greatest negative correlation coefficient (coefficient of determination R2 = 0.77) was obtained between satellite NIR reflectance (0.5 metre multi-spectral World View 2 image) and soil N content. In both cases, it was necessary to develop statistical models for the prediction of total soil nitrogen, using primary field sampling in combination with EO (hyperspectral) measurements. The prediction errors for total soil N using the VNIR spectral range are typically somewhat larger than for total C (Viscarra Rossel et al., 2006[12]). Based on the published literature, airborne or spaceborne hyperspectral soil reflectance measurement is the only current EO technique which is likely to provide reasonable predictions of total soil N. However, because the remote sensor must be able to detect reflectance directly from the soil surface, this approach requires non-vegetated soil which are most effectively calibrated by ground-based soil sampling and direct laboratory measurements of soil properties. Although we might expect statistically significant relationships between EO-derived vegetation canopy productivity indices and total soil nitrogen, there were no published studies which demonstrated that such relationships were sufficiently strong to accurately predict soil total N under a range of vegetation types.

Our findings relating to the most effective EO technique for predicting state or change of total soil N were very similar to those for SOC, the most promising technology is hyperspectral remote sensing.

Opportunities

The strong relationships between reflectance spectra and both SOC and total N, using multivariate statistical techniques, particularly laboratory and proximal (ground-based) approaches, illustrates their potential for improved mapping of their concentrations [state] (Croft et al., 2012[5]). The larger uncertainties in statistical models developed to predict SOC and total N, using remotely sensed spectral information by comparison to ground-based (proximal or laboratory) methods, reflects: i) the site-specific nature of relationships between remotely sensed measured variables and target properties, and ii) atmospheric interference effects. However, the uncertainties in the prediction of soil properties such as SOC can be substantially reduced by combining digital mapping using combinations of spectral information and other data sources, such as soil and vegetation maps. Such an improvement in SOC prediction is shown in the use of digital mapping of SOC, which has quickly moved from a research stage to being operational, with maps of carbon concentration and carbon stock from field to regional scale (Minasny et al., 2013[15]).

The review of Ladoni et al. (2010) suggests that given the uncertainty of direct measurements of SOC by remote sensing, there is a far greater possibility to use remotely sensed data to guide sampling schemes, rather than for direct measurement. Obtaining representative soil samples is useful for the application of geostatistical models, and remotely sensed data has been found to be advantageous in designing a sampling scheme for mapping SOC, with the smallest number of samples and with greater accuracy, at lower costs (Ladoni et al., 2010).

Can Earth Observation replace field sampling?

Based on current sensor technology, EO cannot replace field sampling but can complement it by: i) helping to improve sampling schemes, and ii) providing an exhaustive covariate (available across large parts of the landscape) that if used in a statistical model can substantially reduce the uncertainty in predictions of total concentrations. For mapping the state of these properties it would be possible to devise efficient sampling schemes to achieve a specified accuracy based on the analysis of a specified number of ground-based samples with associated analyses, thus reducing survey costs. In addition to spectral reflectance data, there are other environmental covariates which can reduce the prediction uncertainties of SOC and total N, such as terrain indices.

Can change be measured?

There have been no published studies to date which have demonstrated that a change in SOC or total N can be measured [at national scale], using only remotely sensed methods. We noted in the introduction that the challenge of monitoring change is quite different to mapping the state of soil properties. The errors associated with predicting SOC and total N concentrations by remotely sensed methods are substantially larger than for direct soil sampling and analysis, and it is therefore not practical to apply this approach to monitoring change for those areas where vegetation is sufficiently sparse for periods of the year for reflectance measurements to be made.

Integration with National Surveys?

The most promising technique for applying remote sensing of both SOC and total N is combining the use of hyperspectral satellite data with ground-based sampling and analysis of topsoils from a national survey (such as the National Soil Inventory or Countryside Survey) to improve predictions of SOC concentrations [state] at un-sampled locations. However, reflectance data will only be available for a subset of those sites which are cultivated annually, limited largely to eastern and southern parts of England. Without first testing this approach, it is not possible to state the magnitude of the improvement in prediction accuracy through using remotely sensed data. A number of other factors would likely influence this; the quality of the remotely sensed data (e.g. the extent of atmospheric interferences) and soil-dependent interferences between mineral and organic carbon and nitrogen absorption bands.

Bulk density and compaction

The bulk density (of topsoil) is the mass of soil material per unit volume and is strongly related to the soil organic carbon concentration (and thus land cover type), land management and to a lesser extent soil texture (Hollis et al., 2012[16]). Soils which have been subject to compaction have larger bulk densities than those which have not and so we consider compaction in this section.

There were no published papers in the scientific literature (nor was there evidence elsewhere) which showed that soil bulk density or compaction could be predicted on a defined scale with a specified accuracy using remotely sensed signals. In their study on prediction of soil physical properties using spectral approaches, Minasny et al. (2008)[17] state that 'soil physical properties that are based on pore-space relationships such as bulk density, water retention and hydraulic conductivity cannot be predicted well using [MIR] spectroscopy'. The soil property which accounts for the greatest (typically around 65%) proportion of variation in bulk density is SOC concentration (Hollis et al., 2012[16]). Hence, it may be possible to use pedo-transfer functions based on SOC concentration to predict bulk density with errors somewhat larger 0.17 gm cm-3 (Hollis et al., 2012[16]) However, such large errors would mean that monitoring change by remote sensing would be impractical, and therefore it is advisable to use field-based sampling and analysis to monitor this property.

Given the current absence of effective techniques for measuring bulk density by remote methods, and no immediate prospect of this being addressed in the short-term by new technologies, we have not addressed the set of five questions relating to this indicator, but summarise them in Table 1. Soil compaction leads to small reductions in soil surface elevation so it may also be possible to detect compaction-induced changes through remote monitoring of elevation using the InSAR methods described in the study by Rawlins et al., (2014)[18], however such an approach would need considerable research and testing before it could be deployed.

Soil depth and soil erosion

Soil/peat depth: In a recent study funded by the Welsh Government, interferometric synthetic aperture radar (InSAR) remote sensing data was used to monitor changes in the surface elevation of vegetated peat over an area of 10 km2 (and other land cover types across a wider area) over a period of seven years (1993–2000; (Rawlins et al., 2014[18]). This study involved the application of a new InSAR processing technique which considers pixels within the input radar stack that are only coherent for subsets of the total period of processing. The authors found that variations in the change in peat surface elevation across the study area were similar over shorter (24 hours) and longer periods (>100 days); the longer periods included periods of prolonged wet and dry weather. The standard deviation of the peat surface elevation change was around 5.5 cm, in both cases highlighting considerable movement which the authors consider may be caused by gas generation and loss in peat soil. The standard deviation in elevation change for other soil types with differing vegetation (grassland, forest, heather) was substantially smaller (1.7–3.2 cm). The mean elevation of peatland (and its vegetation canopy) did not appear to change markedly over the seven year period, but there was clearly a considerable amount of short-term variation. These data suggest that it would be possible to monitor peat canopy elevation over many years to detect change, as a proxy for measuring changes in peat thickness, assuming there were limited changes in vegetation canopy height. A substantial period of monitoring (>20 years) would likely be required to detect a statistically significant change in elevation of blanket peats, given the long timescales over which these soils respond to changes in local conditions.

The approach described by Rezatec (see expert survey) to develop a peat spotter service will be based on a combination of satellite-based interferometry and LiDAR techniques to quantify tropical peat depths and volumes, in combination with ground-based measurements using field sites in Indonesia. Based on the findings of this work, which are in the early stage, it is hoped that the methodologies might be transferable to temperate zones such as the UK.

Soil erosion: The current, Defra-funded pilot of a soil erosion monitoring network for England and Wales (project SP 1311), will utilise a fixed-wing UAV at a series of selected sites as a means of identifying the extent of erosion features, but there are no plans currently to process the captured images to quantify erosion losses. A fixed-wing UAV was used by d'Oleire-Oltmanns et al. (2012) to quantify gully and rill erosion in 2D and 3D for a region of Morocco, based on the creation of high-resolution Digital Terrain Models. In their review paper on monitoring, soil degradation by remote sensing, Shoshany et al. (2013)[19] identified three RS technologies which were most likely to be effective: InSAR, LiDAR and close-range photogrammetry. For national scale erosion monitoring we suggest that deployment of direct approaches to detect either gully erosion features or overall lowering of parts of the landscape (with associated accumulation downslope) is the most promising approach.

Opportunities

InSAR methods applied to satellite data may be able to detect a lowering of surface elevation by millimetres or centimetres. This is a similar approach to that used by Rawlins et al. (2014)[18] in their assessment of changes in peat elevation. After processing, the ERS satellite data used by Rawlins et al. (2014)[18] had a support (ground footprint) of a square, with side length 100 metres. Although these data are costly, more recent satellites (TerraSAR-X and RADAR-SAT) have the potential for finer resolution imagery (square of side length 1 or 10 metres) which may offer greater potential for detecting changes within particular fields or identifying gully erosion features. The SAR data from ESA’s soon-to-launch (Spring 2014) Sentinel-1 satellite will be freely available at resolutions of 10 metres making this a practical means of obtaining the relevant data to assess the InSAR technique. One of the limitations of applying InSAR methods to the assessment of soil erosion/soil depth is the need to have a consistent (or preferably absent) vegetation canopy. The method could be applied effectively to cultivated land when crops have been removed, but it may be problematic to apply it in areas of grassland where sward heights change by season and between years. Another study also used a combination of LiDAR data, InSAR and air photographs for detecting landslides and erosional features evolving at rates similar to those represented by water erosion of soil (Roering et al., 2009[20]). The approach was of sufficient resolution for the authors to compute downslope motion rates and also to estimate denudation rates.

Timing and further development

There are no current technical barriers to testing the use of a combination of InSAR with IsBAS processing for assessing changes in surface elevation as a means of estimating soil erosion in cultivated areas. To access the most appropriate data it would be necessary to make a scientific application to those responsible for satellite data capture to ensure the relevant scenes were acquired on a regular basis. It would also be necessary to ensure that areas where the technique is deployed are not subject to subsidence which could confound predictions of erosion loss or overall changes in surface elevation (Sowter et al., 2013[21]).

Can Earth Observation replace field sampling?

Based on the evidence presented, we believe that a combination of InSAR with IsBAS processing for assessing changes in surface elevation, as a means of estimating soil erosion or change in soil depth, could replace field-based measurements such as terrestrial LiDAR. However, the method would need to be rigorously tested in areas where cultivated soils were known to have eroded to demonstrate that the method was fit for purpose.

Assuming there were no technical difficulties with access to and processing of the data, it should be possible to measure changes in the elevation of cultivated land, which could be attributed to erosion processes or changes in soil depth. However, this approach/technology would need thorough testing with ground-based validation before it could be implemented.

Integration with National Surveys?

There is currently no established national-scale soil erosion monitoring scheme, although a pilot is currently being undertaken (Defra SP1311). It may be possible to integrate earth observation with such a survey, although the InSAR technique is likely to be limited to vegetation-free, cultivated areas initially.

Metals

There is currently no technical basis for estimating the four total heavy metal soil indicators directly from remotely sensed data. Quantitative assessments of soil properties from remote sensing largely rely on the correlation of particular soil properties with infra-red spectroscopic reflectance measurements (Mulder et al., 2011[22]) which relate to the chemical bonds within organic and inorganic compounds in the soil. The bonds that respond markedly to infra-red spectroscopy are predominantly between carbon, nitrogen, hydrogen and oxygen, which are present in substantial quantities in certain soil phases. The four heavy metal soil indicators (Cu, Zn, Ni and Cd) either do not exhibit diagnostic spectral features in the infra-red wavelength region, or in the case of extractable components, the relationships between the bonds and extracted phases are too complex to be resolved in a quantitative measurement. Nor do these indicators typically have simple relationships with common soil properties that can be identified by their infra-red spectral characteristics (Viscarra-Rossel et al., 2006). One study on soil heavy metals refered to the application of remotely sensed data (Chen et al., 2012[23]), however, the hyperspectral imagery was used as a means of classifying land use, not as the basis for measuring or improving the prediction of heavy metal concentrations in the soil.

Due to a lack of appropriate methods, we do not consider it appropriate to separately address the five questions posed in our introduction. To summarise, there is no current or medium-term potential earth observation technique which could be deployed to monitor total (aqua regia extractable) soil heavy metal concentrations. It is likely that the only realistic approach is field based sampling and laboratory analysis, likely based on existing monitoring networks.

Extractable nutrients

The three extractable nutrients (and the forms they are present in extracted solutions that concern us are: phosphorus (Olsen P), magnesium (extractable), potassium (extractable).

Olsen P, extractable MG, extractable K

The extractable concentrations of these three nutrients inform us about the fertility of the soil. Extractable concentrations are quite different from total concentrations, because the former are held in ‘exchangeable’ forms associated with the surfaces of soil minerals and organic matter. The concentrations of extractable nutrients do not have such strong relationships with bulk soil properties, because they can reflect recent applications of manufactured fertilisers and organic materials or transitory soil hydrological conditions (e.g. wetting and drying cycles). The basis of predicting soil properties based on VNIR spectroscopy relies on the overall reflectance properties of the dominant soil mineral and organic phase composition, not the composition of ‘exchangeable’ elements on their surfaces. A review study by Ge et al. (2011)[24] summarised approaches to estimate a range of soil properties by use of remote sensing including total P, total K and Mg, but not extractable concentrations. The bonds that respond to infra-red spectroscopy are predominantly between carbon, nitrogen, hydrogen and oxygen which are present in substantial quantities in certain soil phases, whilst bonds for P, K and Mg are much less responsive.

Olsen P

There were no published studies which showed that Olsen P could be measured using any remote sensing technique. Nor had any study shown that Olsen P could be predicted accurately from soil based spectroscopic measurements. A study by Baojuan (2008)[25] in the USA showed that total soil P could be predicted using Hyperion hyperspectral, remote sensing imagery; the statistical model based on the spectra accounted for 67% of the variation in laboratory measured total P from field based sampling. However, although there are typically positive correlations between total P and Olsen P in soil (Owens et al., 2008[26]), the uncertainties associated with predicting the latter based on the former are far too large for this approach to be practical.

Extractable Mg

Hyperspectral VNIR data have been used to construct statistical models to predict extractable Mg in soil. In a study using 750 diverse soils from South Africa, (Shepherd and Walsh, 2002) accounted for 81% of the variance in extractable Mg using VNIR spectra, suggesting this approach had potential for wider application. However, no studies have confirmed this relationship in temperate environments such as the UK. None of the other remote sensing technologies have been shown to have sufficiently strong statistical relationships with extractable Mg for practical deployment in predicting state or change.

Extractable K

The only study to have demonstrated that extractable K could be measured based on the VNIR spectra of soil samples was published by Daniel et al. (2003)[27] based on 41 samples collected from a region of Thailand. In their study, Daniel et al. (2003)[27] found that soil spectra accounted for 80% of the variance in extractable K; there was no attempt to relate the ground-based spectra to remotely sensed sources such as hyperspectral satellite data. A total sample size of 41 is a relatively small number upon which to develop and independently validate predictions of extractable soil K. It is somewhat surprising that, if there was a general relationship between soil VNIR spectra and extractable K, it has not been reported more widely given its potential importance for soil fertility and agricultural production. We believe that the relationship reported by Daniel et al, (2003)[27] was unusual (confined to these local soils) and that if tested more widely, the relationship could not be deployed to accurately predict extractable K in soils across broad landscapes. Furthermore, the relationship with hyperspectral satellite data and extractable K would likely be even more tenuous.

By contrast, it is possible to accurately predict total K based on airborne radiometric survey data which relies on the detection of gamma radiation from 40K emitted from the ground surface. Airborne radiometric survey data have been used to determine the distribution of plant available K at farm-scale (Pracilio et al., 2006[28]); gamma-K concentrations accounted for 60% of the variation in bicarbonate-extractable K across three farms in Australia. Specific regions of England and Wales have been flown by airborne radiometric survey by the British Geological

Survey, including parts of the Midlands (Rawlins et al., 2007[29]) and Wales, the Isle of Wight, and in 2013 south-west England. However, the relationships between total and extractable potassium have not, to date, been investigated across England and Wales using these data. Such an investigation could be undertaken by combining the data from the original National Soil Inventory of England and Wales (McGrath and Loveland, 1992[30]) with the airborne survey data held by BGS. This would establish if radiometric K is an effective predictor of extractable K at the landscape scale (i.e. to predict its state for mapping), and/or to infer change in this property.

Opportunities

In general, there are limited opportunities to implement national scale RS techniques to contribute to monitoring changes in the three soil nutrients (Olsen P, extractable K, extractable Mg). In the case of Olsen P, there is currently no technical basis for successful application of remote sensing of this soil property. In the case of extractable Mg, there is some evidence that hyperspectral RS data might contribute to indirect mapping of this property (state), but there is no evidence to suggest it could be used to monitor change even in combination with a ground-based survey. In the case of extractable K, the most promising technology for mapping this property, in combination with ground-based sampling, is airborne radiometric survey. However, further investigation is required to establish the strength of statistical relationships between radiometric K and extractable K across the landscape; data are available to do this.

Timing and further development

There do not appear to be any major development opportunities or timelines for radical advances in the detection of the three extractable soil nutrients (P, Mg and K). One potential development is the investigation of whether radiometric K has a strong correlation with extractable K, which could be used to enhance the prediction of extractable K across the landscape.

Can earth observation replace field sampling?

Based on the literature survey, it is very unlikely that earth observation could replace field sampling for the three soil nutrient indicators; it is possible that with further development the hyperspectral and radiometric RS approaches might be able to complement field sampling.

Can change be measured?

Based on the available information it is very unlikely that change in these three soil properties could be measured directly using RS approaches.

Integration with national surveys?

There are technical and practical barriers which would limit the extent to which RS approaches could be integrated with national surveys to monitor soil nutrients. First, with the possible exception of extractable Mg, there is uncertainty over the strength of direct or indirect relationships between these soil properties and their (hyperspectral) remote sensing signal; this is likely to be insufficient for practical monitoring in combination with ground based measurements. The practical limitations of applying hyperspectral remote sensing are that for much of the year vegetation prevents direct reflectance from the soil surface which would severely hamper any attempt to apply these approaches to much of the landscape.

Soil PH

Soil pH is a critical soil property which has importance for a wide range of soil processes concerning fertility and soil nutrients, water quality (acid deposition and buffering capacity), soil biology. Soil pH is controlled by a combination of soil mineralogy and land management (liming) and organic matter composition. Soil pH is typically measured using wet laboratory- based methods but there are strong relationships between VNIR spectra and soil pH.

Opportunities

The most promising approach for quantifying the state of soil pH using remote sensed methods is the application of hyperspectral remote sensing in combination with ground-based measurements (Lu et al., 2013[31]). In their study, Lu et al. (2013)[31] showed that using Hyperion imagery, 60% of the variation on pH was accounted for by a statistical model fitted between the measured pH of the field samples and the remote sensing data from the VNIR region. To robustly detect a change in soil pH, the method for its measurement would need to have a significantly smaller error than that which would be associated with spectral approaches that relay in ground-based calibration of remote sensed data which could have RMSE values as large as 0.3 pH units (Lu et al., 2013[31]). So it is unlikely that hyperspectral imagery (in combination with field sampling) could be used to measure change of soil pH. Based on our literature survey, only one study has demonstrated that remote sensed data can improve the prediction of soil pH (state) and further studies are required across a range of soil types to prove that this approach is more generally applicable.

Can Earth Observation replace field sampling?

It will be necessary to use field sampling to develop the statistical model based on the remote sensing data for the prediction of soil pH.

Can change be measured?

It is unlikely that change can be measured based on current technology; there are no technologies in the pipeline that are unlikely to mean that change can be measured by remote methods at national scale.

Integration with National Surveys?

By capturing hyperspectral remote sensed imagery for periods when cultivated areas are largely free of vegetation, it would be possible to make more accurate predictions of soil pH in these areas, but only in regions where ground-based sampling has been undertaken. It would not be possible to use this approach in year-round, vegetated areas.

Summary

Table 1 below provides a summary of the most promising earth observation techniques for each of the soil indicators.

Table 1 Summary of most promising remote sensing techniques for a range of soil indicators.
Indicator Most promising RS technique (or none available) Platform Can detect state and/or change Will also require ground-based soil sampling and analysis at national scale? Further development required? Practicality or technical considerations to be addressed
Organic carbon and total N Hyperspectral Satellite *State (not change)

Yes

 Outcome: smaller errors in predicting concentrations in bare soil (partially vegetation free regions). New satellites (e.g. HySPIRI) may enhance spectral signal-to-noise ratios.
Bulk density None available

NA

NA

NA

NA

Soil depth (and erosion) InSAR (using IsBAS) Satellite Change

No

 This technique needs further testing before it could be used to monitor either peat depth or estimate the magnitude and extent of soil erosion.
Total metals (Cu, Zn, Cd, Ni) None available

NA

NA

NA

NA

Extractable K Radiometric survey (gamma radiation) Airborne *State (not change)

Yes

 This technique would need testing in a pilot study — there are sufficient data across England and Wales for this to be undertaken.
Extractable Mg Hyperspectral Satellite *Maybe state (not change)

Yes

 This approach needs to be explored further as there are currently insufficient data from England and Wales to determine whether it could be effective.
Soil pH Hyperspectral Satellite

*State (not change)

Yes

 Smaller errors in predicting pH values in bare soil (or partially vegetation free regions). New satellites (e.g. HySPIRI) may enhance spectral signal-to-noise ratios.
Olsen P None available

NA

NA

NA

NA

*the remotely sensed data would most likely be used here as a secondary covariate to reduce the uncertainty in the prediction of the primary indicator property (measured by ground-based sampling plus analysis) at locations across the landscape where the remotely sensed data was available, but the primary soil indicator had not been measured.

NA = Not applicable

References

  1. ↑ HENDERSON, T L, BAUMGARDNER, M F, FRANZMEIER, D P, STOTT, D E & COSTER, D C. 1992. High dimensional reflectance analysis of soil organic-matter. Soil Science Society of America Journal, 56, 865–872.
  2. ↑ 2.0 2.1 GOMEZ, C, ROSSEL, R A V & MCBRATNEY, A B. 2008. Soil organic carbon prediction by hyperspectral remote sensing and field vis-NIR spectroscopy: An Australian case study. Geoderma, 146, 403–411.
  3. ↑ SELIGE, T, BOHNER, J & SCHMIDHALTER, U. 2006. High resolution topsoil mapping using hyperspectral image and field data in multivariate regression modeling procedures. Geoderma, 136, 235–244.
  4. ↑ AL-ABBAS, A, SWAIN, P & BAUMGARD, M. 1972. Relating organic-matter and clay content to multispectral radiance of soils Soil Science, 114, 477–485.
  5. ↑ 5.0 5.1 CROFT, H, KUHN, N J & ANDERSON, K. 2012. On the use of remote sensing techniques for monitoring spatio-temporal soil organic carbon dynamics in agricultural systems. Catena, 94, 64–74.
  6. ↑ RICHARDSON, A J & WIEGAND, C L. 1977. Distinguishing vegetation from soil background information. Photogrammetric Engineering and Remote Sensing, 43, 1541–1552.
  7. ↑ BARET, F, JACQUEMOUD, S & HANOCQ, J F. 1993. About the soil line concept in remote-sensing. Advances in Space Research, 13, 281–284.
  8. ↑ NANNI, M R & DEMATTE, J A M. 2006. Soil line behavior obtained by laboratorial spectroradiometry for different soil classes. Revista Brasileira De Ciencia Do Solo, 30, 1031–1038.
  9. ↑ 9.0 9.1 FOX, G A & SABBAGH, G J. 2002. Estimation of Soil Organic Matter from Red and Near-Infrared Remotely Sensed Data Using a Soil Line Euclidean Distance Technique. Soil Science Society of America Journal, 66, 1922–1929.
  10. ↑ RAWLINS, B G, MARCHANT, B P, SMYTH, D, SCHEIB, C, LARK, R M & JORDAN, C. 2009. Airborne radiometric survey data and a DTM as covariates for regional scale mapping of soil organic carbon across Northern Ireland. European Journal of Soil Science, 60, 44–54.
  11. ↑ GLENDINING, M J, DAILEY, A G, POWLSON, D S, RICHTER, G M., CATT, J A & WHITMORE, A P. 2011. Pedotransfer functions for estimating total soil nitrogen up to the global scale. European Journal of Soil Science, 62, 13–22.
  12. ↑ 12.0 12.1 VISCARRA ROSSEL, R A, WALVOORT, D J J., MCBRATNEY, A. B., JANIK, L. J. & SKJEMSTAD, J. O. 2006. Visible, near infrared, mid infrared or combined diffuse reflectance spectroscopy for simultaneous assessment of various soil properties. Geoderma, 131, 59.
  13. ↑ WU, J., LIU, Y. L., CHEN, D., WANG, J. & CHAI, X. 2009. Quantitative mapping of soil nitrogen content using field spectrometer and hyperspectral remote sensing. 2009 International Conference on Environmental Science and Information Application Technology, Vol II, Proceedings, 379–382.
  14. ↑ STAMATIADIS, S, EVANGELOU, L. BLANTA, A, TSADILAS, C, TSITOURAS, A, CHRONI, C, CHRISTOPHIDES, C, TSANTILA, E, SAMARAS, V, DALEZIOS, N & DIMOGIANNIS, D. 2013. Satellite Visible-Near Infrared Reflectance Correlates to Soil Nitrogen and Carbon Content in Three Fields of the Thessaly Plain (Greece). Communications in Soil Science and Plant Analysis, 44, 28–37.
  15. ↑ MINASNY, B, MCBRATNEY, A B, MALONE, B P & WHEELER, I. 2013. Chapter One — Digital Mapping of Soil Carbon.In: DONALD, L S. (ed.) Advances in Agronomy. Academic Press.
  16. ↑ 16.0 16.1 16.2 HOLLIS, J M, HANNAM, J & BELLAMY, P H. 2012. Empirically-derived pedotransfer functions for predicting bulk density in European soils. European Journal of Soil Science, 63, 96–109.
  17. ↑ MINASNY, B, MCBRATNEY, A B, TRANTER, G & MURPHY, B W. 2008. Using soil knowledge for the evaluation of mid-infrared diffuse reflectance spectroscopy for predicting soil physical and mechanical properties. European Journal of Soil Science, 59, 960–971.
  18. ↑ 18.0 18.1 18.2 18.3 RAWLINS, B G., CIGNA, F, JORDAN, C J, SOWTER, A, EVANS, C, ROBINSON, D A & TEAM, G. 2014. Monitoring changes in surface elevation of blanket peat and other land cover types using a novel InSAR processing technique.Geophysical Research Abstracts, Vol. 16, EGU2014-3047, 2014.
  19. ↑ SHOSHANY, M, GOLDSHLEGER, N & CHUDNOVSKY, A. 2013. Monitoring of agricultural soil degradation by remote- sensing methods: a review. International Journal of Remote Sensing, 34, 6152–6181.
  20. ↑ ROERING, J J, STIMELY, L L, MACKEY, B H & SCHMIDT, D A. 2009. Using DInSAR, airborne LiDAR, and archival air photos to quantify landsliding and sediment transport. Geophysical Research Letters, 36.
  21. ↑ SOWTER, A, BATESON, L, STRANGE, P, AMBROSE, K & SYAFIUDIN, M F. 2013. DInSAR estimation of land motion using intermittent coherence with application to the South Derbyshire and Leicestershire coalfields. Remote Sensing Letters, 4, 979-987.
  22. ↑ MULDER, V L, DE BRUIN, S, SCHAEPMAN, M E & MAYR, T R. 2011. The use of remote sensing in soil and terrain mapping - A review. Geoderma, 162, 1–19.
  23. ↑ CHEN, Y Y, LIU, Y L, LIU, Y F, LIN, A W, KONG, X S, LIU, D F, LI, X R, ZHANG, Y, GAO, Y & WANG, D. 2012. Mapping of Cu and Pb Contaminations in Soil Using Combined Geochemistry, Topography, and Remote Sensing: A Case Study in the Le'an River Floodplain, China. International Journal of Environmental Research and Public Health, 9, 1874–1886.
  24. ↑ GE, Y, THOMASSON, J A & SUI, R. 2011. Remote sensing of soil properties in precision agriculture: A review. Frontiers of Earth Science, 5, 229–238.
  25. ↑ BAOJUAN, Z. 2008. Using satellite hyperspectral imagery to map soil organic matter, total nitrogen and total phosphorus.Master of Sience, Indiana Univeristy.
  26. ↑ OWENS, P N, DEEKS, L K, WOOD, G A, BETSON, M J, LORD, E I & DAVISON, P S. 2008. Variations in the depth distribution of phosphorus in soil profiles and implications for model-based catchment-scale predictions of phosphorus delivery to surface waters. Journal of Hydrology, 350, 317–328.
  27. ↑ 27.0 27.1 27.2 DANIEL, K W, TRIPATHI, X K & HONDA, K. 2003. Artificial neural network analysis of laboratory and in situ spectra for the estimation of macronutrients in soils of Lop Buri (Thailand). Australian Journal Of Soil Research, 41, 47–59.
  28. ↑ PRACILIO, G, ADAMS, M L, SMETTEM, K R J & HARPER, R J. 2006. Determination of spatial distribution patterns of clay and plant available potassium contents in surface soils at the farm scale using high resolution gamma ray spectrometry. Plant and Soil, 282, 67–82.
  29. ↑ RAWLINS, B G, LARK, R M & WEBSTER, R. 2007. Understanding airborne radiometric survey signals across part of eastern England. Earth Surface Processes and Landforms, 32, 1503–1515.
  30. ↑ MCGRATH, S P & LOVELAND, P J. 1992. The Soil Geochemical Atlas of England and Wales, Glasgow, Blackie Academic and Professional.
  31. ↑ 31.0 31.1 31.2 LU, P, WANG, L, NIU, Z., LI, L H & ZHANG, W H. 2013. Prediction of soil properties using laboratory VIS-NIR spectroscopy and Hyperion imagery. Journal of Geochemical Exploration, 132, 26–33.
Retrieved from "http://earthwise-staging.bgs.ac.uk/index.php?title=OR/15/030_Results_and_their_interpretation&oldid=20091"