ESA CCI Soil Moisture product version comparison (version 3.2 and version 4.4)

Overview

This document describes and analyze the difference between the ESA CCI Soil moisture product version 4.4, in comparison with the version used in our research. We acknowledge that the latest version has improvements in the retrieval of soil moisture estimates, but this version is also affected by other factors that generate more gaps in our region of interest.

Version 4.4 is generated gathering active and passive retrievals by means of a weighted mean, being proportional to the signal-to-noise ratio (SNR) (Chung et al. 2018). These ratios are estimated using Triple Collocation Analysis, which is a method that estimates random error variances of three collocated datasets of soil moisture estimates (Gruber et al. 2017). In areas, where no triple collocation analysis estimates are available, soil moisture values are estimated using a polynomial regression between the signal-to-noise ratios (Chung et al. 2018). The last version of the ESA CCI product (Version 4.4) even offers ancillary layers of information used to diminish the uncertainty carried out by the mentioned factors.

In our general gap-filling approach, we used ESA CCi soil moisture product version 3.2, which is still affected by the specific circumstances such as vegetation density and frozen soils (Dorigo et al. 2015).

As result of the comparison we performed showed that using this version, we can better reproduce the reference spatial patterns of soil moisture offered by field measurement. Thus, our methodology offers an alternative approach to improve the spatial representation of soil moisture values in ESA CCI version 3.2, in addition to other methods such as triple collocation analysis and polynomial regression between signal-to-noise ratios, proposed by (Gruber et al. 2017) in the development of ESA CCI version 4.4.

NA Counting

According to the product description, unreliable retrievals are masked. Although, these improvements have been applied in the last version, gaps are not successfully filled, and due to the new algorithm characteristics, more gaps are present in the newest version. In our region of interest, for instance, the total number of information gaps is even higher.

#Counting of NA values along the study period (153 months) 

##Version 3.2
setwd("E:/ESA_CCI_SM_products/Monthly_metrics/Oklahoma/version_3.2")

files <- list.files(pattern = 'region_interest')
mean_stack <- stack()

for (i in 1:length(files)) {
  layer <- stack(files[i])
  layer <- layer[[1]]
  
  mean_stack <- stack(mean_stack, layer)  
  print(i)
}

final_count <- mean_stack[[1]]
reclass1 <- c(-Inf, Inf, 0, NA, NA, 0)
rclass_matrix1 <- matrix(reclass1, ncol=3, byrow=TRUE)
final_count <- reclassify(final_count, rclass_matrix1)
 
for (j in 1:nlayers(mean_stack)) {
  x <- mean_stack[[j]] 
  n <- c(-Inf, Inf, 0, NA, NA, 1)
  rclass_matrix2 <- matrix(n, ncol=3, byrow=TRUE)
  x <- reclassify(x, rclass_matrix2)

  final_count <- final_count + x
  print(j)
}

writeRaster(final_count, 'region_interest_NaNpixels_199101_201806_monthly_counting_esacci_sm_version_44.tif')

##Version 4.4
setwd("E:/ESA_CCI_SM_products/Monthly_metrics/Oklahoma/version_3.2")

files <- list.files(pattern = 'region_interest')
mean_stack <- stack()

for (i in 1:length(files)) {
  layer <- stack(files[i])
  layer <- layer[[1]]
  
  mean_stack <- stack(mean_stack, layer)  
  print(i)
}

final_count <- mean_stack[[1]]
reclass1 <- c(-Inf, Inf, 0, NA, NA, 0)
rclass_matrix1 <- matrix(reclass1, ncol=3, byrow=TRUE)
final_count <- reclassify(final_count, rclass_matrix1)
 
for (j in 1:nlayers(mean_stack)) {
  x <- mean_stack[[j]] 
  n <- c(-Inf, Inf, 0, NA, NA, 1)
  rclass_matrix2 <- matrix(n, ncol=3, byrow=TRUE)
  x <- reclassify(x, rclass_matrix2)

  final_count <- final_count + x
  print(j)
}

writeRaster(final_count, 'region_interest_NaNpixels_199101_201806_monthly_counting_esacci_sm_version_44.tif')

Figure 1. The figure below shows the difference in the amount of gaps along the study period, regarding our 153 monthly layers.

Montlhy gaps by pixel in the region of interest, 153 months (January 2000 -September 2012). ESA CCI soil moisture v 3.2 (left), Right.- ESA CCI soil moisture v 3.2 (right).

Montlhy gaps by pixel in the region of interest, 153 months (January 2000 -September 2012). ESA CCI soil moisture v 3.2 (left), Right.- ESA CCI soil moisture v 3.2 (right).

ESA CCI version 3.2 might be more suitable to be tested in our gap-filling approach, as these product provides more valid soil moisture estimates, diminishing uncertainty carried out by the lack of data in spatial interpolation methods. However, we tested our methodology and its complete workflow with the data of ESA CCI version 4.4. We applied created soil moisture monthly layers using the new version, and applied the same parametrization to Ordinary Kriging and Generalized Linear Model (supplementary material S2), using the same meteorological data monthly layers (Thornton et al. 2018).

Cross-Validation

We performed cross-validation using our models oupust with ESA CCi version 4.4, in the same way with did with version 3.2 As shown in the table below, correlation coefficients are consistently low using the new version, and root mean square errors are higher in each case, compared with the modeling outputs using version 3.2.

–Code is included in supplementary material S2–

Table 1. Cross-validation outputs for OK and GLM, all predicted and observed values along the 153 monthly layers. Versions 3.2 and 4.4 of ESA CCI Soil Moisture product respectively.

Table 1. Cross-validation outputs for OK and GLM, all predicted and observed values along the 153 monthly layers. Versions 3.2 and 4.4 of ESA CCI Soil Moisture product respectively.

Taylor Diagrams

We also generated Taylor diagrams using the models outputs derived from the version 4.4, as we did with version 3.2. Changes in correlation coefficient and RMSE values derived from both versions (3.2 and 4.4) of ESA CCI following our approaches can be observe in the figures below.

–Code is included in supplementary material S2–

Taylor diagrams showing correlation coefficients derived from cross- validation for our models outputs, as well as their root mean squared error. Standard deviation and RMSE values for each output have been normalized using the standard deviation of the observed values. Version 3.2 (left), Version 4.4 (right).

Taylor diagrams showing correlation coefficients derived from cross- validation for our models outputs, as well as their root mean squared error. Standard deviation and RMSE values for each output have been normalized using the standard deviation of the observed values. Version 3.2 (left), Version 4.4 (right).

These results seem to be promising when modeling soil moisture spatial patterns using version ESA CCI 3.2, rather than version 4.4. However, the reliability of our method relies on its capacity to reproduce the soil moisture spatial patterns on the field, in the same way they are reproduced by the satellite estimates, filling the gaps in the original product. However, as we seek to fill the gaps in the original products, we aim to best reproduce the correlation coefficients between the satellite estimates and ground-truth information.

Ground-Truth Validation

We performed ground-truth validation using ESA CCI version 4.4 and data from the North American Soil Moisture Database. In the same way as with version 3.2, we set the correlation coefficient between the ESA CCI soil moisture values and ground-truth data as reference to reproduce the spatial patterns estimated by the satellite retrievals. We acknowledge that field measurements a point-scale represent a challenge in validation satellite estimates as has been explored by Colliander et al. (2017) and described in section (ESA CCI 2010). However, as we seek to fill the gaps in the original products, we aim to best reproduce the correlation coefficients between the satellite estimates and ground-truth information.

In the table below, we observe that reference correlation coefficients are quite similar for both versions, 0.523 and 0.559, version 3.2 and 4.4 respectively and RMSE is slightly low when using new version, 0.069 m3m-3 in comparison with the previous one, 0.092 m3m-3.

–Code is included in supplementary material S2–

Table 2. Overall correlation coefficients between all ground-truth validation points and CCI soil moisture product, as well as gap-filled outputs. Percentages show the data subset used to predict soil moisture values over the region of interest.Versions 3.2 and 4.4 of ESA CCI Soil Moisture product respectively.

Table 2. Overall correlation coefficients between all ground-truth validation points and CCI soil moisture product, as well as gap-filled outputs. Percentages show the data subset used to predict soil moisture values over the region of interest.Versions 3.2 and 4.4 of ESA CCI Soil Moisture product respectively.

Nevertheless, when comparing the difference between the reference correlation coefficient and the coefficient from our models, we see that version 3.2 consistently better reproduces the soil moisture spatial patterns depicted by the ground-truth data.

Table 3. Differences between correlation coefficients derived from ESA CCI soil moisture estimates and ground-truth data of the North American Soil Moisture Data Base (NASMD) and coefficients derived from models outputs and ground-truth data from NASMD.

Table 3. Differences between correlation coefficients derived from ESA CCI soil moisture estimates and ground-truth data of the North American Soil Moisture Data Base (NASMD) and coefficients derived from models outputs and ground-truth data from NASMD.

References

Chung, D., W. Dorigo, R. De Jeu, R. Kidd, and W. Wagner. 2018. “Product Specification Document (PSD) D.1.2.1 Version 4.4.” Earth Observation Data Centre for Water Resources Monitoring (EODC).

Colliander, A., T. J. Jackson, R. Bindlish, S. Chan, N. Das, S. B. Kim, M. H. Cosh, et al. 2017. “Validation of SMAP Surface Soil Moisture Products with Core Validation Sites.” Remote Sensing of Environment 191: 215-31. https://doi.org/10.1016/j.rse.2017.01.021.

Dorigo, W. A., A. Gruber, R. A M De Jeu, W. Wagner, T. Stacke, A. Loew, C. Albergel, et al. 2015. “Evaluation of the ESA CCI Soil Moisture Product Using Ground-Based Observations.” Remote Sensing of Environment 162 (June): 380-95. https://doi.org/10.1016/j.rse.2014.07.023.

ESA CCI. 2010. “ESA Soil Moisture CCI, Validation.” 2010. https://www.esa-soilmoisture-cci.org/validation.

Gruber, Alexander, Wouter Arnoud Dorigo, Wade Crow, and Wolfgang Wagner. 2017. “Triple Collocation-Based Merging of Satellite Soil Moisture Retrievals.” IEEE Transactions on Geoscience and Remote Sensing 55 (12): 6780-92. https://doi.org/10.1109/TGRS.2017.2734070.

Thornton, M.M. M, P.E. E Thornton, Y. Wei, B.W. W Mayer, R.B. B Cook, and R.S. S Vose. 2018. “Daymet: Monthly Climate Summaries on a 1-Km Grid for North America, Version 3.” https://doi.org/10.3334/ornldaac/1345.