An evaluation of supervised classifiers for indirectly detecting salt-affected areas at irrigation scheme level

https://doi.org/10.1016/j.jag.2016.02.005Get rights and content

Highlights

  • Supervised classifiers were tested for detecting salt accumulation at irrigation scheme level.

  • Spectral and spatial features derived from pan-fused SPOT5 imagery were used.

  • The methods were assessed in two distinct South Africa irrigation schemes.

  • Weak statistical relationships between features and salinity levels were observed.

  • Good classifications (>90%) were achieved using machine learning algorithms.

Abstract

Soil salinity often leads to reduced crop yield and quality and can render soils barren. Irrigated areas are particularly at risk due to intensive cultivation and secondary salinization caused by waterlogging. Regular monitoring of salt accumulation in irrigation schemes is needed to keep its negative effects under control. The dynamic spatial and temporal characteristics of remote sensing can provide a cost-effective solution for monitoring salt accumulation at irrigation scheme level. This study evaluated a range of pan-fused SPOT-5 derived features (spectral bands, vegetation indices, image textures and image transformations) for classifying salt-affected areas in two distinctly different irrigation schemes in South Africa, namely Vaalharts and Breede River. The relationship between the input features and electro conductivity measurements were investigated using regression modelling (stepwise linear regression, partial least squares regression, curve fit regression modelling) and supervised classification (maximum likelihood, nearest neighbour, decision tree analysis, support vector machine and random forests). Classification and regression trees and random forest were used to select the most important features for differentiating salt-affected and unaffected areas. The results showed that the regression analyses produced weak models (<0.4 R squared). Better results were achieved using the supervised classifiers, but the algorithms tend to over-estimate salt-affected areas. A key finding was that none of the feature sets or classification algorithms stood out as being superior for monitoring salt accumulation at irrigation scheme level. This was attributed to the large variations in the spectral responses of different crops types at different growing stages, coupled with their individual tolerances to saline conditions.

Introduction

The term salinity is used to describe the processes and impacts of salt and water, while also being a measure of the amount of salt in soil or water (McGhie and Ryan, 2005). Soil salinity is defined as the accumulation of soluble salts in the soil (Al-Khaier, 2003). Salt accumulation can occur naturally (primary salinity) or as a result of human interference (secondary salinity). Human activities such as vegetation clearing, irrigation and landscape reshaping through earth works can, for example, increase the volume of water and salt in the soil and change how they move and where they accumulate (Ghassemi et al., 1995, McGhie and Ryan, 2005, Metternicht and Zinck, 2009). Determining salinity risk goes hand in hand with the understanding of groundwater movement, since this controls the mobility and transfer of salt deposits, among other landscape factors (Spies and Woodgate, 2005).

According to Hillel (2000) soil salinity is a severe environmental hazard that affects the growth of various crops. Although salinity might not be as dramatic or damaging as earthquakes or landslides, it is an environmental hazard that greatly reduces crop yields and agricultural production (Ghassemi et al., 1995, Metternicht and Zinck, 2009, Umali, 1993, Zinck, 2000). Salt accumulation also has additional secondary negative impacts, such as the devaluation of farm properties, land degradation, eutrophication of rivers, damage to infrastructure, increased soil erosion and engineering difficulties (Metternicht and Zinck, 2003).

It is estimated that nearly one billion hectares (ha), which is equivalent to about 7% of the earth’s continental extent, is associated with salt-affected areas (Ghassemi et al., 1995). According to the Food and Agriculture Organization of the United Nations (FAO), of the 230 million ha land available for irrigation, 45 million ha (19.5%) are salt affected (FAO, 2016). Most of these affected areas are the consequence of human activities, in particular, irrigation (Metternicht and Zinck, 2003). This is an alarming statistic and Abbas et al. (2013) estimates that, at global scale, soil salinization is increasing at a rate of up to 2 million ha per annum.

Proactive monitoring of salt accumulation is needed to keep its negative effects under control. Salinity monitoring involves the identification of areas where salts concentrate and the detection of temporal and spatial changes in this occurrence (Zinck, 2000). Remotely sensed data can contribute a great deal to monitoring these processes because of its ability to capture information in both spatial and temporal scales (Abbas et al., 2013). Bastiaanssen et al. (2000) states that remote sensing has the potential to predict soil salinity, perform diagnosis and asses its impact. Compared to solely relying on regular field surveys for monitoring salt accumulation, the synthesis of remote sensing with field surveys can potentially save labour, time and effort (Eldiery et al., 2005, Metternicht, 1996).

Soil salinity can be detected from remotely sensed data either directly or indirectly (Bastiaanssen, 1998, Mougenot et al., 1993). When studying the spectral properties of bare soil directly, indicators of salt accumulation includes white salt crusts, puffy soil, dark greasy surfaces, dehydrated cracks and coarser topsoil (Goodall et al., 2000, McGhie and Ryan, 2005, Metternicht and Zinck, 2003). The main limitation of the direct approach is that farming practices such as tillage and irrigation compromises the spectral properties of soils (Metternicht and Zinck, 2003, Zhang et al., 2011), particularly in highly dynamic irrigation schemes.

Techniques other than the direct approach are considered indirect methods for detecting salt accumulation. The aim of the indirect approach is to map the effect that salt accumulation has on land cover. Most indirect indicators are related to vegetation types and growth. An excess of salt in a plants’ root environment leads to a reduced ability to take up water. The salts reduce the osmotic potential of water and hinder the movement of water from the soil into the root. If saline conditions persist, the small amount of salt that enters a plant along with water, accumulates over time and becomes toxic (Hillel, 2000, McGhie and Ryan, 2005). These effects lead to spotty/uneven growth, plant wilting, blue-green tinge and moisture stress.

Several authors have successfully applied the indirect approach to identify plant stress caused by salt accumulation (Abood et al., 2011, Fernández-Buces et al., 2006, Koshal, 2010, Lenney et al., 1996a, Lobell et al., 2010, Peñuelas et al., 1997, Wiegand et al., 1994, Zhang et al., 2011). In all of these cases vegetation indices (VIs) (e.g. NDVI; EVI and SAVI) were employed. The use of VIs is, however, not without complications. Plant species with a high salt tolerance can lead to ambiguity in the response of vegetation indices (Aldakheel, 2011); poor farming practices and soil preparation can lead to poor vegetation responses which can be misinterpreted as being caused by salinity conditions (Furby et al., 1995); and using VIs in study areas with a high bare ground backscatter/noise can influence some of the vegetation indices negatively (Dehni and Lounis, 2012, Douaoui et al., 2006). Other indirect indicators of salt accumulation include soil types (García Rodríguez et al., 2007) elevation and terrain data (Caccetta et al., 2000, Farifteha et al., 2006, Furby et al., 1995, Jenkin, 1981, McFarlane et al., 2004) and image texture (Metternicht and Zinck, 2003).

Most applications of remote sensing for monitoring salt accumulation have been carried out in areas were salt accumulation occurs on a grand scale. As a result, the majority of the satellite imagery used for the identification of salt accumulation has medium to low resolutions. Landsat (30 m) (Abdelfattah et al., 2009, Aldakheel, 2011, Al-Khaier, 2003, Caccetta et al., 2000, Dehni and Lounis, 2012, Elnaggar and Noller, 2010, Fernández-Buces et al., 2006, Furby et al., 1995, Gao and Liu, 2008, García Rodríguez et al., 2007, Howari, 2003, Lenney et al., 1996a, Mohamed et al., 2011); IRS (20 m) (Abbas and Khan, 2007, Abbas et al., 2013, Dwivedi and Sreenivas, 1998, Dwivedi et al., 2001, Eldiery et al., 2005, Khan et al., 2001, Koshal, 2010) and MODIS (250mn) (Lobell et al., 2010) are most commonly used. However, agricultural activities in irrigation schemes are normally diverse in nature as many different crops are usually planted, often on a rotational basis. This poses a unique challenge for detecting salt accumulation, especially if it occurs in small patches within relatively small fields. Such areas will be difficult to detect using medium to low resolution (e.g. 20–250 m) satellite images. Few studies on the use of very high resolution (VHR) imagery for indirect salt accumulation monitoring exist. Exceptions include Douaoui and El Ghadiri (2015), and the unpublished work of Abood et al. (2011) who used WorldView-2 (WV2) (2 m) imagery; as well as Howari and Goodell (2009), Eldiery et al. (2005), Dwivedi et al. (2009) and Allbed et al. (2014) who used Ikonos (4 m) imagery. Other high resolution imagery used for detecting soil salinity include airborne multispectral (Hick and Russell, 1990, Howari, 2003, Wiegand et al., 1994) and hyperspectral imagery (Dehaan and Taylor, 2003, Dehaan and Taylor, 2002, Dutkiewicz et al., 2009, Farifteh, 2009, Naumann et al., 2009, Schmid et al., 2009). Hyperspectral imagery in particular hold key advantages over standard multispectral imagery due to its rich spectral information (Metternicht and Zinck, 2009). Due to limited coverage and low availability of these images, they were not considered for this study.

Supervised classification is a proven tool for successfully identifying salt accumulation. The maximum likelihood classification is the most commonly applied supervised method for the identification of salt accumulation. Its application stretches across various satellite platforms, namely, Landsat (Abbas et al., 2013, Castaneda and Herrero, 2009, García Rodríguez et al., 2007, Howari, 2003, Iqbal, 2011, Wu et al., 2008), LISS II (Abbas et al., 2013) and Ikonos (Dwivedi et al., 2009, Howari and Goodell, 2009). However, to our knowledge, no research has been done to evaluate the ability of different supervised classifiers for detecting salt-affected areas using VHR imagery in highly complex and dynamic irrigated areas.

The aim of this paper, which forms part of a Water Research Commission (WRC) project (WRC, 2010), is thus to evaluate a range of indirect indicators—derived from SPOT-5 imagery for detecting and classifying salt-affected areas at irrigation scheme level. Several supervised image classification techniques are evaluated, namely maximum likelihood (ML), support vector machines (SVM), nearest neighbour (NN), decision trees (DT), and random forests (RF). The techniques are evaluated on two contrasting irrigation schemes in South Africa (Vaalharts and Breede River) to determine whether any of the classifiers consistently produce superior results and to identify the indirect indicators that can reliably be used to detect salt accumulation. The results are discussed and interpreted in context of finding operational solutions for salt accumulation monitoring at irrigation scheme level.

Section snippets

Study area

The Vaalharts irrigation scheme, situated on the borders of the Northern Cape, North West and the Free State provinces, near the towns of Jan Kempdorp, Hartswater and Pampierstad, was selected as the first study area (Fig. 1). The scheme covers 29181 ha and is one of the largest irrigation schemes in South Africa (Van Rensburg et al., 2012). The area borders two plateaus on the east and west of the Harts River Valley. The valley slopes towards the south with very little topographical changes due

Spectral analysis

Fig. 2 shows the spectral profiles of salt-affected (EC > 4.0 ds/m) and unaffected (EC < 4.0 ds/m) crops in the Vaalharts irrigation scheme. Compared to healthy crops, salt-affected crops generally have higher reflectance in the blue to red region of the electromagnetic spectrum, while reflectance is lower in the near and shortwave infrared regions. This suggests that vegetation in salt-affected areas experience weaker growth (vegetation vigour) than in unaffected areas. This result is in accordance

Conclusions

This study examined a number of indirect indicators for identifying salt-affected areas in cultivated fields at irrigation scheme level. Given that such areas tend to occur in small patches within fields, 2.5 m pansharpened SPOT-5 high resolution satellite imagery was evaluated in two distinctly different South African irrigation schemes (Vaalharts and Breede River).

A series of regression analyses were carried out to evaluate the continuous relationships between the observed in situ salinity

Acknowledgements

This work forms part of a larger project titled “Methodology to monitor the status of water logging and salt affected soils on selected irrigation schemes in South Africa” which was initiated and funded by the Water Research Commission (WRC) of South Africa (contract number K5/1880//4). More information about this project is available in the 2009/2010 WRC Knowledge Review (ISBN 978-1-4312-0004-7) available at www.wrc.org.za. The authors also acknowledge and thank the project leader, Dr Piet

References (126)

  • A. Huete

    A soil-adjusted vegetation index (SAVI)

    Remote Sens. Environ.

    (1988)
  • J.J. Jenkin

    Terrain, groundwater and secondary salinity in victoria, Australia

    Agric. Water Manage.

    (1981)
  • Z. Jiang et al.

    Development of a two-band enhanced vegetation index without a blue band

    Remote Sens. Environ.

    (2008)
  • N.M. Khan et al.

    Assessment of hydrosaline land degradation by using a simple approach of remote sensing indicators

    Agric. Water Manage.

    (2005)
  • M. Kruger et al.

    Perspective on the development of stem borer resistance to Bt maize and refuge compliance at the Vaalharts irrigation scheme in South Africa

    Crop Prot.

    (2009)
  • M.P. Lenney et al.

    The status of agricultural lands in Egypt: the use of multitemporal NDVI features derived from landsat TM

    Remote Sens. Environ.

    (1996)
  • M.P. Lenney et al.

    The status of agricultural lands in Egypt: the use of multitemporal NDVI features derived from landsat TM

    Remote Sens. Environ.

    (1996)
  • A.P. Leone et al.

    A field experiment on spectrometry of crop response to soil salinity

    Agric. Water Manage.

    (2007)
  • G.I. Metternicht et al.

    Remote sensing of soil salinity: potentials and constraints

    Remote Sens. Environ.

    (2003)
  • G. Mountrakis et al.

    Support vector machines in remote sensing: a review

    ISPRS J. Photogramm. Remote Sens.

    (2011)
  • A. Abbas et al.

    Using remote sensing techniques for appraisal of irrigated soil salinity

    International Congress on Modelling and Simulation MODSIM

    (2007)
  • M. Abdelfattah et al.

    Soil salinity mapping model developed using RS and GIS-a case study from Abu-Dhabi, United Arab Emirates

    Eur. J. Sci. Res.

    (2009)
  • Abood, S., Maclean, A., Falkowski, M., 2011. Soil Salinity Detection in the Mesopotamian Agricultural Plain Utilizing...
  • Y.Y. Aldakheel

    Assessing NDVI spatial pattern as related to irrigation and soil salinity management in Al-Hassa Oasis, Saudi Arabia

    J. Indian Soc. Remote Sens.

    (2011)
  • M.S. Alhammadi et al.

    Detecting date palm trees health and vegetation greenness change on the eastern coast of the United Arab Emirates using SAVI

    Int. J. Remote Sens.

    (2008)
  • F. Al-Khaier

    Soil Salinity Detection Using Satellite Remote Sensing

    (2003)
  • Backeberg, G.R., Bembridge, T.J., Bennie, A.T.P., Groenewald, J.A., Hammers, P.S., Pullen, R., Thompson, H., 1996....
  • J.H. Barnard

    On-Farm Management of Salinity Associated with Irrigation for the Orange-Riet and Vaalharts Schemes

    (2013)
  • W.G.M. Bastiaanssen

    Remote Sensing in Water Resources Management: The State of the Art

    (1998)
  • Beuster, H., Shand, M., Carter, C.A., 2003. Breede River basin study....
  • T. Blaschke et al.

    Was ist mit den Pixeln los? Neue Entwicklungen zur Integration von Fernerkundung und GIS

    Geo-Informations-Systeme

    (2001)
  • A. Bosch et al.

    Image classification using random forests and ferns

  • M. Bouaziz et al.

    Remote mapping of susceptible areas to soil salinity, based on hyperspectral data and geochemical, in the southern part of Tunisia

  • L. Breiman

    Random forests

    Mach. Learn.

    (2001)
  • E. Bresler et al.

    Saline and sodic soils

    (1982)
  • P. Caccetta et al.

    The land monitor project

  • J. Campbell

    Introduction to Remote Sensing

    (2007)
  • W.J. Carper

    The use of intensity-hue-saturation transformations for merging SPOT panchromatic and multispectral image data

    Photogramm. Eng. Remote Sens.

    (1990)
  • C. Castaneda et al.

    Alicationof landsat and ERS imagery to the study of saline wetlands in Semiarid Agricultural Areas of Northeast Spain

  • R. Congalton et al.

    Assessing the Accuracy of Remotely Sensed Data

    (2009)
  • D.R. Cutler et al.

    Random forests for classification in ecology

    Ecology

    (2007)
  • R. Dehaan et al.

    Image-derived spectral endmembers as indicators of salinisation

    Int. J. Remote Sens.

    (2003)
  • R.J. Dekker

    Texture analysis and classification of ERS SAR images for map updating of urban areas in the Netherlands

    IEEE Trans. Geosci. Remote Sens.

    (2003)
  • A. Douaoui et al.

    Combination of remote sensing and kriging to improve soil salinity mapping in the Hmadna plain (Algeria)

    Soil Water J.

    (2015)
  • A. Dutkiewicz et al.

    The suitability of airborne hyperspectral imagery for maing surface indicators of salinity in dryland farming areas

  • R.S. Dwivedi et al.

    Generation of farm-level information on salt-affected soils using IKONOS-II multispectral data

  • R.S. Dwivedi et al.

    The utility of IRS-1C LISS-III and PAN-merged data for mapping salt-affected soils

    Photogramm. Eng. Remote Sens.

    (2001)
  • R.S. Dwivedi et al.

    Delineation of salt-affected soils and waterlogged areas in the Indo-Gangetic plains using IRS-1C LISS-III data

    Int. J. Remote Sens.

    (1998)
  • A. Eldiery et al.

    Estimating soil salinity from remote sensing data in corn fields

    Hydrol. Days

    (2005)
  • A.A. Elnaggar et al.

    Application of remote-sensing data and decision-tree analysis to mapping salt-affected soils over large areas

    Remote Sens.

    (2010)
  • Cited by (0)

    View full text