The optical trapezoid model: A novel approach to remote sensing of soil moisture applied to Sentinel-2 and Landsat-8 observations

TL;DR: In this paper, the authors proposed a novel OPtical TRApezoid Model (OPTRAM), which is based on the linear physical relationship between soil moisture and shortwave infrared transformed reflectance (STR ) and is parameterized based on pixel distribution within the STR - VI space.

About: This article is published in Remote Sensing of Environment.The article was published on 2017-09-01 and is currently open access. It has received 222 citations till now. The article focuses on the topics: Ground truth.

Summary

1. Introduction

• The existing optical indices (Table 1 ) are mostly empirical, lacking the physical foundation that is at the core of thermal-optical methods such as proposed by Moran et al. (1994) and Carlson et al. (1994) .
• To overcome the TOTRAM limitations the authors well as empiricism of optical indices, they propose a novel physically-based trapezoid model, hereinafter termed OPtical TRApezoid Model , which is based on a recently developed physical relationship between soil moisture and shortwave infrared transformed reflectance, STR (Sadeghi et al., 2015) .
• Because OPTRAM does not require thermal RS data it can be directly applied to estimate soil moisture from Sentinel-2 observations.
• In addition, because STR is used instead of LST, OPTRAM is hypothesized to only require a single universal parameterization for a given location.
• In the following, the authors introduce the theoretical basis of OPTRAM, evaluate the predictive capabilities of the universally parameterized OPTRAM with Sentinel-2 observations, and compare OPTRAM and TOTRAM based on Landsat-8 observations.

2.2. The New Optical Trapezoid Model (OPTRAM)

• It should be noted that the STR-θ relationship is only valid for partially and fully saturated soils, but not for oversaturated soils (i.e. standing surface water).
• This is because water in excess of saturated soil moisture will still increase STR, but the actual soil moisture, , cannot increase beyond the saturated soil moisture content.
• Therefore, for scenes that include many oversaturated pixels (e.g., conditions after heavy precipitation) the wet edge (saturated edge in this case) falls somewhere below the upper edge of the optical trapezoid (i.e., oversaturated edge in Fig. 1 ).
• Carlson (2013) also suggested not to incorporate pixels containing standing water in the TOTRAM trapezoid.
• As shown below, OPTRAM is more sensitive to oversaturated pixels than TOTRAM.

3.1. Test Sites and In Situ Soil Moisture Data

• The newly proposed and traditional trapezoid models, OPTRAM and TOTRAM, were evaluated for the Walnut Gulch (WG) and Little Washita (LW) watersheds in southern Arizona and in southwestern Oklahoma, respectively (Fig. 2 ).
• The sites that vastly differ in climatic conditions, surface topology and land cover are among the most densely instrumented watersheds in the world and previously served as validation sites for microwave remote sensing experiments (Cosh et al., 2006; Jackson et al., 2009 Jackson et al., , 2012)) .
• Soil moisture measured with a network of electromagnetic sensors installed in 5-cm depth were used to evaluate the RS-based surface soil moisture estimates.

3.1.1. Walnut Gulch Watershed

• The WG watershed is densely instrumented with 88 rain gauges, 19 of which are colocated with soil moisture sensors installed at a 5-cm depth.
• Soil moisture data from 15 rain-gauge stations were employed together with 5-cm soil moisture data from the Soil Climate Analysis Network (SCAN) site no.
• 2026 (Fig. 2 ) for validation of OPTRAM and TOTRAM moisture estimates.
• Note that for 4 of the 19 locations no reliable soil moisture measurements were available.

3.1.2. Little Washita Watershed

• Hydrological and meteorological measurements have been conducted in the watershed for decades, providing scientists with long-term data for studying soil and water conservation, water quality, and basin hydrology (Starks et al., 2014) .
• The watershed contains the 20-station USDA-ARS Micronet for monitoring spatial and temporal soil moisture dynamics.
• The 5-cm soil moisture data from 17 Micronet stations (Fig. 2 ) were used as ground truth for validating OPTRAM and TOTRAM estimates (for 3 stations no reliable soil moisture data were available).

3.3.1. Scenario 1

• From the resultant optical trapezoids for the LW watershed, it was apparent that the image classification in LW was only able to remove deep surface water bodies, but not shallow water ponds.
• Hence, the fitted upper edge in this case was the oversaturated edge shown in Fig. 1 and not the wet edge.
• It was observed that the oversaturated zone within the trapezoid was significantly thicker for the Sentinel-2 data than for the Landsat-8 data.
• This can be attributed to the spatial resolution difference (i.e., the coarser the resolution, the lower the chance of an entire pixel to be oversaturated due to shallow surface ponds).
• Note that the resampled images were only used to determine the wet edge in scenario 1.

3.3.2. Scenario 2

• The estimated θ at the stations were dependent on calibrations using in situ data.
• Hence, a second parameterization scenario was established to examine the validity of the physical basis of TOTRAM and OPTRAM, or in other words, the strength of the correlation between θ and LST in TOTRAM [i.e. validity of Eq. ( 2)] and that of θ and STR in OPTRAM [i.e. validity of Eq. ( 6.

4.1. Model Parameters

• A nearly trapezoidal shape is formed by the pixels in the STR-NDVI space in most cases, although the edges are not perfectly linear.
• This primarily verifies their hypothesis that soil moisture is highly correlated to STR even in densely vegetated soils (e.g. NDVI > 0.6).

(ii)

• In both the WG and LW watersheds, the S2-based and L8-based trapezoids are generally similar in shape [e.g., as evident from comparison of the resampled S2 and L8 data for the LW watershed (yellow trapezoids) or from OPTRAM's wet edge parameters for LW in scenario 1].
• This similarity leads to the conclusion that universal parameterization of OPTRAM is achievable because S2 and L8 data were acquired on different dates (Table 2 ). (iii).
• For S2 in WG, scenario 1 led to a positively-sloped wet edge, while scenario 2 led to a negativelysloped wet edge yielding a triangular geometry.
• This discrepancy is obviously due to the fact that θ values at the time of L8 passage were well below the maximum θ values measured at the stations during the entire study period (2015 and 2016), which were considered in scenario 2.
• Future laboratory, greenhouse and field research is required to explore to what extent and under what conditions this assumption is valid.

(iv)

• The integrated LST-NDVI trapezoid used to parameterize TOTRAM consists of several separate smaller trapezoids each corresponding to a specific date.
• This is because the LST depends on ambient environmental factors besides soil moisture and implies that universal parameterization of TOTRAM is not possible.
• This behavior was not observed for OPTRAM.

4.4. Other Optical Models

• One noticeable point in the α-NDVI trapezoid is that there is significantly less scattering around the edges when compared to the STR-NDVI trapezoid.
• This point is considered as an advantage of the α-NDVI trapezoid model when compared to OPTRAM, as the oversaturated pixels are not an issue in this model.

5.1. Conclusions

• The disadvantage of OPTRAM when compared to TOTRAM is its higher sensitivity to oversaturated and shadowed pixels.
• When the optical trapezoid consists of too many oversaturated pixels, solving for the wet edge needs some refinements.
• This, however, may not be a significant limitation because of the feasibility of a single universal model parameterization.

Figures

Table 2. Satellite images used in this study.

Figure 3. Flowcharts illustrating the sequence of Sentinel-2 and Landsat-8 data analyses steps for mapping surface soil moisture with TOTRAM, Eq. (5), and OPTRAM, Eq. (10).

Table 3. TOTRAM [Eq. (5)] and OPTRAM [Eq. (10)] parameters obtained for the Walnut Gulch (WG) and Little Washita (LW) watersheds based on Sentinel-2 (S2) and Landsat-8 (L8) satellite data.

Figure 9. Estimated versus measured normalized soil moisture [W = (θ – θd)/(θw – θd)] from 481 OPTRAM (blue circles) and TOTRAM (red x’s) parameterized based on scenario 2 for 482 Landsat-8 imagery at various dates during 2015 and 2016. 483 484

Figure 5. OPTRAM and TOTRAM soil moisture estimates (parameterized based on scenario 405 1) compared to in situ soil moisture measurements for the Walnut Gulch (WG) and Little 406 Washita (LW) watersheds. 407 408

Figure 1. Sketch illustrating parameters of the traditional thermal-optical trapezoid model [Eq. (5), TOTRAM] and the new optical trapezoid model [Eq. (10), OPTRAM]. TOTRAM and OPTRAM are parameterized based on the pixel distributions within the LST-NDVI space and STR-NDVI space, respectively, NDVI is the normalized difference vegetation index, LST is the land surface temperature, and STR is the SWIR transformed reflectance [see Eq. (7)].

Figure 11. Soil moisture estimates based on the Albedo-NDVI trapezoid model (parameterized based on the 4 different scenarios shown in Fig. 10) compared with in situ soil moisture measurements for the Little Washita watershed.

Figure 8. Soil moisture maps generated with OPTRAM and TOTRAM based on scenario 2 parameterization for the Little Washita watershed. White pixels represent masked pixels due to water bodies, shadows, and rural/urban areas.

Table 1. Soil moisture and drought indices derived from optical RS observations.

Figure 10. Pixel distributions within the Albedo-NDVI space for Landsat-8 images from the Little Washita watershed from the dates listed in Table 2. Four different scenarios were considered to determine the edges; (1) both edges determined visually; (2) dry edge determined visually and wet edge determined with least-square regression; (3) wet edge determined visually and dry edge determined with least-square regression; (4) both edges determined with least-square regression.

Figure 2. The Walnut Gulch and Little Washita watersheds with marked locations of soil moisture sensors used for validation of OPTRAM and TOTRAM. Note that the sensor installation depth is 5-cm.

Figure 7. Soil moisture maps generated with OPTRAM and TOTRAM based on scenario 2 parameterization for the Walnut Gulch watershed. White pixels represent masked pixels due to water bodies, shadows, and rural/urban areas.

Figure 6. OPTRAM and TOTRAM soil moisture estimates (parameterized based on scenario 433 2) compared to in situ soil moisture measurements for the Walnut Gulch (WG) and Little 434 Washita (LW) watersheds. 435 436

Figure 4. Pixel distributions within the STR-NDVI (OPTRAM) and LST-NDVI (TOTRAM) spaces for all images listed in Table 2 (red dots). The yellow dots shown for the optical trapezoids in the Little Washita watershed are from the images resampled to 120-m resolution and were used to determine the wet edge, which falls below the upper edge for this case due to the existence of oversaturated pixels. Thermal imaging required by TOTRAM is not available on the Sentinal-2 satellite.

Frequently asked questions

Q1. What contributions have the authors mentioned in the paper "The optical trapezoid model: a novel approach to remote sensing of soil" ?

To overcome these 18 restrictions the authors propose a novel OPtical TRApezoid Model ( OPTRAM ), which is based on the 19 linear physical relationship between soil moisture and shortwave infrared transformed 20 reflectance ( STR ) and is parameterized based on the pixel distribution within the STR-VI space. The authors also demonstrate that OPTRAM only requires a single universal 28 parameterization for a given location, which is a significant advancement that opens a new 29 avenue for remote sensing of soil moisture.