The estimation of soil moisture is approached by various remote sensing technologies. Initially the main research focused on the Synthetic Aperture RADAR (SAR) images. SAR are widely used for this purpose because of the relatively deep penetration of the microwave radiation, up to one meter deep. Despite this advantage SAR imagery suffer from high noise levels (speckle) and the surface roughness derived from the backscatter coefficient can significantly alter the results.

Hyperspectral remote sensing (350nm-2500nm) has great potentials for the estimation of soil surface moisture content. Although the penetration depth is only superficial, about 50μm, the measured spectrum is highly affected by the presence of water. More analytically, in the full range of the electromagnetic spectrum (350nm - 2500nm) the reflected radiance is decreased as the soil moisture content rises until the soil water capacity (SWC). Water also causes distinct absorption features at the Visible Near Infrared (VNIR) and the Short Wave Infrared (SWIR) at 900nm, 1400nm, 1900nm, 2700nm and 2800nm. Haubrock et al. (2008) developed a soil moisture estimation index, the normalized difference soil moisture index (NSMI). NSMI is a normalized ratio of the reflectance values at 1800nm and 2119nm. It has been tested over heterogeneous soils with various gravimetric moisture levels ranging from the driest moisture to the SWC moisture level.

The soil spectral signature is also affected, among others (moisture content, organic matter, iron oxides, minerals, CaCO3, etc), by the soil texture, the soil mechanical composition and the soil surface roughness. Soil texture is strongly related to the soil mechanical composition if the samples are under laboratory conditions (dried and sieved). More analytically the texture of soil samples that have been sieved is in accordance with their mechanical composition. However, soils under natural conditions may not preserve the laboratory texture. For example very fine particle sized soils (clayey) can form under the proper weather conditions, compactions or gravels (particle size over 2mm diameter), resulting to changes in texture. On the other hand soil roughness refers to the degree of anomaly of the soil surface. Surface roughness is strongly related to the wavelength, according to the Rayleigh criteria. Hyperspectral data use wavelengths in the VNIR and SWIR of the spectrum (μm). According to height variations greater than 2μm result to a rough surface for VNIR and SWIR wavelengths. Surface roughness affects the reflection directions of incident light, while soil texture and soil mechanical composition affects the volumetric radiance scattering of incident light. Thus for soil studies based on hyperspectral imagery, soil roughness is the most apparent parameter that affects the recorded signatures. It is a major component that can alter the results of any algorithm which estimates soil parameters.

Soil moisture estimation

Soil moisture content estimation algorithms using hyperspectral data have been extensively investigated. Four main categories can be discriminated:

  1. single band reflectance approaches
  2. multi-band reflectance approaches
  3. spectrum modeling approaches
  4. multivariate statistical analysis approaches.

The normalized difference soil moisture index (NSMI), which relies on a multi-band reflectance approach, is used in this study. This index was selected because it is simple, fast implemented and proved to be robust against heterogeneous soil types. NSMI takes advantage of two absorption features of water at 1800nm and 2119nm. NSMI uses the soil reflectance values (R) at 1800nm and 2119nm in a normalized ratio as follows:

NSMI is a dimensionless parameter that is linearly related to soil moisture content. Soil moisture content values refer to gravimetric moisture. In this article the moisture values refer to volumetric moisture, because they result from the DECAGON 5TE soil moisture sensor. Volumetric and gravimetric moisture are linearly correlated. Thus NSMI can also be used with volumetric moisture content values.

GEO Premium

Access our ENTIRE content with a yearly subscription. ONLY 8$/month (billed annually)

1 Year Access

Spectral Signature Intensity

Soil spectral signature intensity is an important index to determine the amount of energy that is reflected from the soil surface. It is known that moisture content and surface roughness affect the whole spectrum. In this study the combined effect of moisture content and surface roughness at the soil spectral intensity is being investigated. The spectral intensity (SI) index is derived from the following equation:

where SI is the soil spectral intensity and R1, R2,…, RL are the reflectance values of soil for each spectral band.

Data and Experiments

Study area and soil samples

The study area is located at the rural part of Thessaloniki, northern Greece, (Fig. 1). The study area is agricultural, i.e. cereal, corn, rice, tomatoes and several vegetables are cultivated. According to an edaphological study performed in 2004 the sampled soils are classified as: entisols, inceptisols and alfisols. The soils in this area result from the combined effect of the soil genesis factors, i.e. parent material, climate, microorganisms, topography, time. The main factors involved in the soil formation are the parent material and the topography.

Fig. 1 presents the study area and the locations of the 7 soil samples (red points) that were used for the experiments of the current study. The soil samples were extracted from the field and no further processes were applied, i.e. drying or shivering. The surface area of each soil sample is 20cmX20cm and has 7cm height. The sampling took place at 15th April 2011 and the Landsat image (Fig. 1) was acquired on the 20th April 2011. The samples used in this study are part of a larger sampling campaign (100 soils sampled) that took place at the same time. The 7 soil samples were selected according to the highest soil parameter variation. Such parameters are: mechanical composition (MC) (clay, silt, sand), MC and Texture consistency, ph, Electrical conductivity (Ec), CaCO3, organic matter, NO3, P, K, Ca2+, Mg2+, Na+, K+, Fe, Zn, Mn, total iron content. MC and Texture consistency refers to the textural changes in the appearance of the MC due to various natural procedures in field. The value “Yes” in the “MC and Texture consistency” field implies that the texture matches the MC. The value “No” implies the opposite, i.e. the soil particles have formed gravels. The soil parameters for the 7 samples are presented in table I.

Table I


After the soil sampling, laboratory spectral measurements were carried out. The full spectrum (350nm-2500nm) ASD FieldSpecPro spectroradiometer was used with artificial illumination (2, 50Watt tungsten halogen lamps). The zenith reflected radiance of soil samples was measured at 30cm distance, using a stand in order to stabilize the measuring distance. A 5o for-optic was adapted at the instrument and a calibrated white panel (Spectralon) was used to convert radiance to reflectance.

For each soil sample the reflectance was measured for 2 or 3 moisture levels and for two surface roughness states (smooth and rough). The first level is the natural VMC of the soil samples. The other two moisture levels were created by adding 0.3 and 0.6 liters of water in the first level, respectively. For samples with large interstitial air spaces, water cannot be held and gravity forces it to go at lower levels. Consequently, by adding 0.3 liters in the samples 2, 3, and 7, their surface remains intact. As a result for these samples the impact of water on the soil signature is almost negligible and for this reason only the first and the third moisture levels were taken into consideration. Thus, three soil samples were measured at 2 moisture levels and the other four at 3 moisture levels. The VMC was measured using a DECAGON 5TE probe. The third VMC level is in all samples greater than the SWC. This is a real case scenario of the soil state, which can be encountered in nature (eg. after strong rainfall) and can significantly affect soil spectral signature. Depending on the nature of the soil samples, they react differently at the third VMC level. The soils with 2 VMC levels have staggering water on the surface, significantly affecting their spectral signatures. On the other hand soils with 3 VMC levels are viscous and muddy, at the third level.

Each spectral measurement was repeated 5 times and the mean spectral signature was calculated in the spectral region from 526nm to 2200nm. The regions from 350nm to 525nm and from 2201nm to 2500nm were rejected due to low signal to noise ratio. Fig. 2 shows the soil mean spectral signatures with smooth surface and for the lowest moisture content. The surface state was artificially changed.

Figure 2: Mean soil spectral signatures with smooth surfaces and the lowest moisture content.


Spectral signatures

A first approach to understand the interactions between VMC and surface roughness is to study the spectral signatures before any metric is computed. It is well known that the values of soil spectral signature lower as the moisture content rises (either gravimetric or volumetric moisture content) until the SWC. By studying the spectral signatures of all soil samples, the values of the soil spectral signatures lower from the first VMC level to the second, but rise from the second to the third (especially in the VIS). Theoretically, water absorbs almost all incident radiation in the SWIR spectrum. Thus the spectral signatures of soils with staggering water, and muddy soils should have significantly lower values than the other VMC levels in the SWIR spectrum, but they do not. This inconsistency is due to specular reflections that occur during the measurements. In case of soils with 3 VMC levels and muddy appearance specular reflections are due to a film of water on the soil surface. Specular reflections may lead to false interpretation results. This can be a realistic scenario when treating airborne or satellite hyperspectral images. However, it is observed that for wavelengths greater than 1800nm specular effects are significantly lower (Figures 3, 4 and 5). For wavelengths that correspond to water absorption lines specular reflections are minimized. In figures 3, 4 and 5 the spectral signatures of samples 1, 4 and 5 which present three VMC levels are shown. In those figures it is clear that spectral signatures change with both VMC and roughness.

Spectral signatures of sample 1 as function of measured VMC values with rough surface

Spectral signatures of sample 1 as function of measured VMC values with smooth surface.

Spectral signatures of sample 4 as function of measured VMC values with rough surface.

Spectral signatures of sample 4 as function of measured VMC values with smooth surface.

In all cases, the spectral values of measurements with rough surface are significantly lower in comparison to those with smooth surface. Using the ENVI Spectral Analyst for each soil sample the spectral similarity between its smooth and rough appearance was calculated. It was concluded that for the same VMC only 61.1% of the smooth spectral signatures matched the respective rough spectral signature. The three VMC levels are equally distributed in this percentage. The above indicate that surface roughness is one of the main parameters that affect the soil spectral signature.

To further understand the effect of surface roughness on the spectral signatures with variating VMC, a new metric, the relative difference between the spectral signatures with the same roughness per band is calculated. Due to the fact that in most cases the driest soils have higher reflectance values, it is defined as a base signature for the computations. The relative difference per band is derived from the following equation:

where RDL is the relative difference per band, RDRY,L and RVMC,L are the reflectance values of the drier level and the other VMC levels respectively. Figures 6, 7 and 8 present the relative difference for samples 1, 4 and 5 respectively. It is observed that changes in spectral signatures are not the same for all bands in case that VMC changes. Moreover, rough and smooth soils do not have the same variations in their signatures when VMC changes. No linear relationships between roughness and VMC levels can be derived.

NSMI and roughness

The effect of the surface roughness on the soil moisture estimation is investigated. For this, the NSMI index is used. In order to evaluate the robustness of the NSMI index when roughness changes, the samples were divided into three categories. The first category includes the soil samples with smooth surfaces, the second includes the samples with rough surfaces, and the third all the samples regardless their surface roughness. For each category the NSMI values were computed. Then linear regressions and determination coefficients were computed. The following figure presents the results of this experiment.

Relation between NSMI and VMC for smooth and rough soils.

The yellow and blue regression lines correspond to the rough and smooth samples, while the green line corresponds to all samples. Although surface roughness is a dominant factor for soil spectral values, the regression line appears almost the same for any state of roughness. Rough and smooth soils have 0.61 and 0.67 determination coefficient respectively. A slight increase of the determination coefficient for smooth soil samples is observed. For VMC values ranging from 0.05% to 0.30% regression lines are practically identical. The mean absolute difference of the NSMI values for soil with the same VMC and different surface roughness was also computed and equals to 0.035. This is a very small value, which indicates that the index performs accurately regardless the surface roughness. Moreover, wavelengths 1800nm and 2119nm are appropriate for the VMC estimation in case that knowledge of the surface roughness is not available. For these bands, effects of VMC on the spectral signatures are much greater than those of the roughness. The combined use of the 1800nm and 2119nm in the normalized difference soil moisture index minimizes the roughness effects.

Based on the number of VMC levels, linear regressions and determination coefficients of NSMI were also calculated. The samples in this case were divided into two categories. This distinction has a physical meaning because soil samples with 2 VMC levels have staggering water on their surface, while soil sample with 3 VMC levels are muddy.

Relation between NSMI and VMC for soils with 2 and VMC levels.

In the figure above the blue regression line corresponds to the soils with 2 VMC levels and the orange line to the soils with 3 VMC levels. It is observed that the two categories produce different regression lines and their determination coefficients significantly vary. Soils with 3 VMC levels have significantly higher determination coefficient (0.72) than the other category (0.57), thus this category has more accurately estimated VMC values. The staggering water on the surface of soils with 2 VMC levels seriously affects the spectral signatures which cannot serve as a tool for soil parameter estimation. For this reason the estimation accuracy of NSMI is decreased. On the other hand, soils with 3 VMC levels are muddy. This means that water is rather homogeneously spread around the soil surface. The reflected radiance is a reasonable mixture of soil and water radiances and the estimation accuracy of NSMI is increased.


The objectives of this work are a) to help analysts of hyperspectral imagery to a) study soils given that roughness is a significant parameter in soil spectral signatures, b) understand that soil roughness can play an important role in quantitative studies and c) highlight possible impacts of specular reflections of staggering water at the soil spectral signatures. For these purposes, the combined effect of soil surface roughness and VMC on the spectral signature, and the estimation of VMC using the normalized difference soil moisture index have been investigated. Soil roughness parameter was considered different from soil texture, which has a similar meaning to MC for laboratory experiments. The seven samples that were used were unprepared in order to simulate more realistic field conditions, mostly encountered in airborne or satellite acquisitions.

Initially, by applying signature spectral matching on the measured samples, it was found that surface roughness and VMC are dominant parameters that affect the soil spectral signatures and can lead to misclassification/mismatching of soil types. Then, a new metric, the relative difference, and the spectral intensity were used to underline the combined effect of surface roughness and VMC on the spectral signatures. It was observed that rough and smooth soils do not have the same variations in their signatures when VMC changes. Moreover, surface roughness does not affect every band equally. Soil spectral intensity is lower for rough surfaces than for smooth surfaces. For both rough and smooth surfaces, it decreases as VMC values rise until the SWC, whereas it increases when VMC values become greater than the SWC due to secular reflections of the surface. The last occurs when staggering water is not present on the surface.

The Normalized Difference Soil Moisture index was used for estimating the VMC. It was proved robust, yielding accurate results for soils that do not present staggering water on their surface. Soil roughness does not affect its performance since the combined use of the 1800nm and 2119nm in the NSMI minimizes roughness effects.

Written by

I'm a Remote Sensing and a Surveying Engineer. I received my degree from NTUA in 2010, where I also received my Ph.D. in hyperspectral remote sensing in 2016. From graduation in 2010, my career started as a Researcher Associate and Teaching Associate in the Laboratory of Remote Sensing of NTUA. From that time I also worked at several private companies as a Remote Sensing Expert and Geospatial Analyst. From the beginning of 2015 I was positioned as Senior Earth Observation Expert. During these years, I have participated in more than 20 funded European Commission and European Space Agency projects, have over 16 peer reviewed scientific publications in the field of Remote Sensing, and have an international patent in hyperspectral data compression.My main research and professional interests are in the optical remote sensing area, where I specialize in data (images, point measurements) processing and algorithm design and development. Some of the software tools that I operate to accomplish my research and business dreams are SNAP, ENVI, IDL, QGIS, ERDAS Imagine, ArcGIS, and Python. I have been working with these tools since 2008.

Dimitris Sykas

Remote Sensing Expert

Dimitris Sykas