Chapter 4 An Integrated Surface Parameter Inversion Scheme over Agricultural Fields
4.5 Validation and Analysis
4.5.3 Surface Roughness Validation
The validation of the surface roughness is performed on two aspects: the first is its variation over time, and the other is the estimated surface roughness compared with the measured one. Specifically, the C2 and S2 fields in study area 2014 are selected for the variation analysis. The histograms of the surface roughness in their fields are shown in Figure 4-14, and the comparison between the estimated roughness and measured roughness is shown in Figure 4-15.
(a) (b) (c) (d)
(e)
Figure 4-14. Surface roughness histograms for the corn and soybean fields, from left to right, on different days: (a) May 5th (b) May 18th (c) June 11th (d) June 21st (e) variation of roughness over time in corn and soybean fields.
(c) (d)
(e) (f)
(i)
Figure 4-15. Measured and estimated surface roughness on different dates: (a) Corn field on May 4th (b) Soybean field on May 4th (c) Corn field on May 5th (d) Soybean field on May 5th (e) Corn field on May 18th (f) Soybean field on May 18th (g) Corn field on June 11th (h) Soybean field on June 11th (i) entire field on all days.
Before the beginning of May, the corn field had been ploughed showing mainly bare soils, and many smooth large size clods were left in the field, which makes the corn field very rough from the beginning of May to the middle of May. The soybean field also appear rough due to the fluctuation scattering caused by the corn residues. However, the seedbed preparation of both fields occurred at the end of May, resulted in a relatively smooth surface for both fields. That means the KS will be changing from a high value to a low value from May to June. Figure 4-14(c) and Figure 4-14(d) show this change, with average KS values of 1.22 and 1.32 in the corn fields and 1.43 and 1.41 in the soybean fields. This change has also been depicted in Figure 4-14(e), which shows the change of roughness from May to June, before and after the crop planting. The same as for the estimated soil moisture, the surface roughness is also obtained by averaging the pixels. It should also be noted that the peaks in the histograms of Figure 4-14(a) and Figure 4-
14(b) on May 5th and May 18th are observed in both the corn and soybean fields. In the
corn fields, there are around 30% and 15% pixels having values of 2.5 and 2.1 on May 5th
and May 18th, respectively. It is primarily caused by the relatively large roughness during
the ploughed stage as the ploughed field had large clods according to our measurements. Other studies have also reported that the majority of the averaged RMS heights are
approximately 2.6 cm, or as high as 4 cm (Alvarez-Mozos et al., 2006; Baghdadi et al., 2008), which are consistent with our measurements. For the soybean field, the peaks in the histograms have approximately 30% and 40% pixels with their roughness being approximately 2.5 and 2.1 respectively. The high peak is likely caused by the corn residues that were left in the soybean fields as shown in Figure 4-2(b), and the corn residues can cause fluctuated scatterings. The similar histograms observed on May 5th
and May 18th in the soybean field can also demonstrate the consistent performance of the
ISPIS. This is because the high peaks are observed on both dates except that the
roughness on May 18th is less than that on May 5th due to the flattened residues caused by
the human activities.
Figure 4-15 shows the KS derived by different methods on different dates. For both the ISPIS and Y-CIEM, the RMSE in the corn fields is lower than that in the soybean fields, which is primarily due to the crop stalks left in the corn fields, which caused the scattering to fluctuate. Specifically, on May 4th and May 5th in the corn fields, the surface roughness derived by the ISPIS and Y-CIEM did not change much on either date because they were both bare soils. We also know that the RADARSAT-2 data on the two dates had different orbits: one is ascending and the other is descending. However the orbit difference is not the primary reason causing the variation, as there were no prominent roughness patterns. In addition, both the ISPIS and the Y-CIEM have the issue of underestimation. This is because to avoid the speckle noise, a window size averaging process is adopted for the estimation of the soil roughness. This can influence the estimated results, because the roughness often shows little spatial dependency, which means that the surface roughness taken at one position often poorly represents its surrounding areas. Therefore, the averaging process for the estimation of surface roughness could lead to an underestimation.
We also observed that the estimated roughness of both Y-CIEM and ISPIS has no strong correlation with the ground truth, with their R2 values of 0.184 and 0.185, respectively, which is perhaps caused by the small range of the roughness between 1.5 and 2.5, resulting in a biased correlation coefficients calculation. Both the ISPIS and Y-CIEM
have RMSEs less than 0.75 on different days in the corn and soybean fields. The overall RMSE of the ISPIS is around 0.48, which is very similar to that of the Y-CIEM with its value of 0.50. This similar RMSE is because the sample sites in the corn or soybean fields with bare soils that are dominated by surface scattering are also considered for the overall RMSE calculation. For the surface scattering dominant regions, the CIEM is employed by both the ISPIS and Y-CIEM for surface parameter inversion, because the H
and 𝛼 threshold to distinguish the surface and volume scattering employed in this chapter
is both adopted by the Y-CIEM and ISPIS. Therefore, to invert the surface parameters of these surface scattering dominated pixels, their results will be almost the same as shown in Figure 4-15(c), Figure 4-15(e) and Figure 4-15(h). However, for fields covered with corn residues or short corn plants, the volume scattering is dominant. To invert surface parameters for these fields, the difference between the ISPIS and the Y-CIEM becomes larger compared with fields with bare soils as depicted in Figure 4-15(d) and Figure 4- 15(g). From this perspective, we conclude that the ISPIS can describe more complex situations than the Y-CIEM, as the ISPIS can vary with the RVI, whereas the RVI of the Yamaguchi volume scattering model stays constant.