Sensing landscape level change in soil fertility following deforestation and conversion in the highlands of Madagascar using Vis-NIR spectroscopy
Introduction
Basic data on land resources and soil quality in particular are scarce in Madagascar, particularly when considering the island's diverse landforms and soils, and the vast extent of severe soil degradation. Existing soil maps are generally based on the digitized world soil map (FAO, 1995) at 1 : 1,000,000 scale, and more detailed soil surveys have not been conducted in significant areas of land degradation in Madagascar. There are, however, exceptions such as in certain important agricultural areas like the Vakinakaratra region south of Antananarivo, where more detailed soil maps (e.g., Raunet, 1981) are available. In particular there is a lack of scientifically based information on historical extents, rates and processes of land degradation, which has led to an oversimplified understanding of present problems, often blaming the practice of slash-and-burn (shifting) agriculture for widespread degradation (Jarosz, 1993, Vågen, 2004). Two principal reasons for the limited information on land degradation are (i) limited use of scientifically-rigorous statistical and sampling designs at a landscape level, and (ii) the high costs involved in large-scale soil sampling and analysis using standard procedures.
In light of these constraints, rapid and economical soil analysis techniques are critical for farmers, land managers, local authorities and researchers to be able to utilize soil testing to its full potential in assessing and managing soil quality. Visible-near-infrared reflectance (Vis-NIR) spectroscopy is a nondestructive analytical technique where the interactions between incident light and a material’s surface are studied (Chang et al., 2001, Shepherd and Walsh, 2004). The technique is simple, rapid, needs very little sample preparation, and is widely used in pharmaceutical, petrochemical and other industries and in grain and forage quality assessments.
The Vis-NIR spectra are influenced not only by the chemistry of a material, but also by its physical structure. They are directly influenced by combinations and overtones of fundamental molecular absorptions for organic functional groups found in the mid infrared region, and their potential use in analysis of soil organic carbon (SOC) content has been demonstrated in a number of studies (Ben-Dor and Banin, 1995, Couillard et al., 1997, Shepherd and Walsh, 2002). Water, particle size, surface properties, total carbon (C), total nitrogen (N), texture, moisture content and aggregation are often considered primary soil properties due to their direct influence on Vis-NIR spectra. However, previous studies have shown that near-infrared spectra can also predict soil properties not theoretically related to near-infrared light, if these properties are correlated with any of the primary soil properties mentioned above (Fritze et al., 1994, Ben-Dor and Banin, 1995, Chang et al., 2001).
Overlaps of weak overtones and fundamental vibrational bands make Vis-NIR spectra difficult to interpret directly (Wetzel, 1983). For quantitative analysis, multivariate calibration is therefore required. A number of different calibration techniques are available and have been applied when relating measured NIR spectra to measured values of soil properties. The choice of calibration technique will depend on the application of the data. Principal components regression (PCR), partial least squares regression (PLSR), stepwise multiple-linear regression, locally weighted regression and artificial neural networks are the most common (Wise et al., 2003)
Indicators of soil quality are still widely debated among soil scientists (Doran and Jones, 1996, Gregorich and Carter, 1997, Letey et al., 2003), mainly due to the complexity involved in integrating various soil properties into indices of soil quality and differential effects of soil management on different soil properties. Several authors have questioned the viability of institutionalizing soil quality as a defined parameter in soil science (Sojka and Upchurch, 1999, Letey et al., 2003, Sanchez et al., 2003) particularly since significant questions remain, making the concept illusive and value laden. Doran and Jones (1996) proposed a minimum data set of physical, chemical, and biological indicators for screening the quality (or health) of soils, which included texture, depth of soil rooting, infiltration and bulk density, water holding capacity (physical), soil organic matter, pH, electrical conductivity, extractable N, P and K (chemical), microbial biomass C and N, potentially mineralisable N, and soil respiration (biological). Others have proposed similar indicators and systems where soils are given scores (e.g., 1–5) based on SOC content, pH and other soil properties and yet others have developed soil quality (or health) cards based on farmer–scientist collaboration.
One of the major constraints in the two former approaches is the difficulties involved in comparing different studies since all of the soil properties involved may not be available, often due to the level of costs involved as discussed earlier, and the problem of setting standard norms for soil quality indicators (Letey et al., 2003). Sanchez et al. (2003) used the term “measure everything” for these approaches, which stem largely from what Sojka and Upchurch (1999) labelled as the “Mollisol-centric” soil quality paradigm, and questioned its relevance in the tropics. The latter approach, on the other hand, is largely qualitative and has limited scientific validity. The challenge therefore remains to develop quantitative approaches to understanding dynamic changes in soil functional properties within the broader context of integrated natural resource management in the tropics on a diverse range of soil types and under a wide variety of land use systems.
Shepherd and Walsh (2002) proposed a spectral library approach, whereby the variability of soils in a study area is thoroughly sampled and spectrally characterized. Soil properties or attributes of soil quality are measured on only a selection of soils, designed to sample the variation in the spectral library, and then calibrated to soil Vis-NIR reflectance. The soil quality indicators can then be predicted for the entire library and for new samples from the study area. New samples that classify as spectral outliers to the library are characterized and added to the calibration library, thereby increasing the predictive value of the library. The approach can be extended to provide spectral indicators of soil quality.
The objective of this study was to develop and test a soil fertility index (SFI) based on Vis-NIR soil spectral data and assess its performance for predicting change in soil quality (or fertility) after deforestation and conversion of forest to agricultural land and grassland in the eastern highlands of Madagascar. A secondary objective was to test whether the SFI could be calibrated to remote sensing imagery and used to map out SFI decline so that soil degradation problem areas (hot-spots) in the highlands of Madagascar can be rapidly identified.
Section snippets
Materials and methods
The study was conducted in an area to the north-east and east of the town of Ambositra, in Fianarantsoa province (47° 25′ 15″ East and 20° 34′ 11″ South to 47° 16′ 04″ East and 20° 27′ 02″ South). Average annual rainfall varies considerably, but generally increases from about 1300 mm (Ambositra) to over 2000 mm (rainfall data from Fandriana) from west to east, at the eastern escarpment. The topography is hilly as are most of the highland plateau areas of Madagascar, with the steep eastern
Spectral prediction of soil properties
The soils (Oxisols) in the study area had low pH and (generally) low contents of exchangeable cations (Table 1). Most individual soil properties varied widely despite the relatively small size of the study area (482 km2) (Table 1). The shape of the raw spectral curves was similar to that described by Ben-Dor et al., (1999) and Shepherd and Walsh (2002), with prominent absorption features at 1400, 1900 and 2200 nm (Fig. 1). These features are associated with clay minerals (Hunt, 1982, Ben-Dor et
Conclusions
We have demonstrated the ability of multivariate calibration techniques commonly used in chemometrics and other disciplines to predict basic soil physicochemical properties from soil Vis-NIR reflectance data. Stable calibration models were developed for SOC, TN, CEC and clay contents for an area under a wide variety of land use types and landscape forms in the highlands of Madagascar. These rapid and cost effective methods permit the use of statistical sampling frames in landscapes. They
Acknowledgements
We would like to thank the staff of the FOFIFA soil laboratory in Antananarivo and the ICRAF soil laboratory in Nairobi. We also thank FTM (Foiben-Taosarintanin'i Madagasikara) for providing access to maps and aerial photos used in correction of satellite imagery and land use mapping, and to Masy and Salmata Andrianorofanomezana for invaluable assistance.
References (58)
- et al.
Linear Probability, Logit, and Probit Models. Series: Quantitative Applications in the Social Sciences
(1984) FieldSpec™ User's guide
(1997)- et al.
Near infrared analysis (NIRA) as a rapid method to simultaneously evaluate several soil properties
Soil Science Society of America Journal
(1995) - et al.
Soil reflectance
- Bilmes, J.A., 1998. A gentle tutorial of the EM algorithm and its application to parameter estimation for Gaussian...
- et al.
An analysis of transformations
Journal of the Royal Statistical Society
(1964) - et al.
Determination of total, organic and available forms of phosphorus in soils
Soil Science
(1945) - Bremmer, J.M., Sulvaney, C.S., 1982. Total nitrogen. In: Page, A.L. et al. (Eds.), Methods of Soil Analysis, Part 2. p....
- et al.
Logistic modeling to spatially predict the probability of soil drainage classes
Soil Science Society of America Journal
(2002) - et al.
Near-infrared reflectance spectroscopy — principal components regression analyses of soil properties
Soil Science Society of America Journal
(2001)
Near infrared reflectance spectroscopy for analysis of turf soil profiles
Crop Science
SIMPLS: an alternative approach to partial least squares regression
Chemometrics and Intelligent Laboratory Systems. Laboratory Information Management
Canonical partial least squares and continuum power regression
Journal of Chemometrics
The Expectation Maximization Algorithm
Mixtures densities, maximum likelihood from incomplete data via the EM algorithm
Journal of the Royal Statistical Society
Development of a soil quality index for the Chalmers soil series in the midwestern USA. CD-ROM. West Lafayette
Introduction to Graphical Modeling
Digital Soil Map of the World and Derived Soil Properties
Misuse of ridge regression in the calibration of a near infrared reflectance instrument
Applied Statistics
Near-infrared characteristics of forest humus are correlated with soil respiration and microbial biomass in burnt soil
Biology and Fertility of Soils
Particle-size analysis
The design library for S-Plus version 6.x
A comparison of the discrimination of discriminant analysis and logistic regression under multivariate normality
Amelioration of Al toxicity and P deficiency in acid soils by additions of organic residues: a critical review of the phenomenon and the mechanisms involved
Nutrient Cycling in Agroecosystems
A Textbook of Soil Chemical Analysis
Cited by (107)
Lateritic processes in Madagascar and the link with agricultural and socioeconomic conditions
2022, Journal of African Earth SciencesCitation Excerpt :More advanced economies have allowed farmers to mitigate these physiochemical disadvantages via judicious use of lime, fertiliser, and mechanisation, e.g. in parts of Brazil and eastern India (e.g. Brahmachari et al., 2019). Vagen et al. (2006) introduced a visible-near-infrared reflectance spectroscopic technique for rapid characterisation of soils (including oxisols) across Madagascar, in terms of key soil properties such as soil organic carbon, cation exchange capacity, and total nitrogen. While research data on Malagasy soil quality are scarce (in spite of the country's significant soil degradation and low agricultural productivity), they defined 10 key agronomic indicators of soil fertility across Madagascar, including pH, organic carbon, and silt and clay content.
Airborne imaging spectroscopy for assessing land-use effect on soil quality in drylands
2022, ISPRS Journal of Photogrammetry and Remote SensingComparing CalReg performance with other multivariate methods for estimating selected soil properties from Moroccan agricultural regions using NIR spectroscopy
2021, Chemometrics and Intelligent Laboratory SystemsSoil spectroscopy with the use of chemometrics, machine learning and pre-processing techniques in soil diagnosis: Recent advances–A review
2021, TrAC - Trends in Analytical Chemistry