Mario Guevara

University of Delaware | Student

Subject Areas: Digital soil mapping

 Recent Activity

ABSTRACT:

Soil moisture is key for quantifying soil-atmosphere interactions and the ESA-CCI (European Space Agency-Climate Change Initiative) provides historical (>30 years) satellite soil moisture global grids with spatial resolution of ~27km. This dataset is incomplete (contains gaps) due to conditions such as dense vegetation or extremely dry surfaces. Here we provide a framework to increase the spatial resolution and fill gaps (reporting associated uncertainty) of the ESA-CCI (v4.5) soil moisture dataset. The outcome is a new dataset of gap-free global mean annual soil moisture and uncertainty
for 28 years (1991-2018) across 15km grids. We compare the performance of machine learning odels using only terrain parameters (e.g., slope, wetness index) against predictions using terrain parameters, bioclimatic information, and soil type classes. We use independent field information from the International Soil Moisture Network (ISMN, n=13376) and in-situ precipitation records (n=171) only for model evaluation purposes. Using only terrain parameters to predict soil moisture results in a parsimonious approach comparable with a more complex model that includes additional bioclimatic and soil information. The correlation between observed and predicted soil moisture values varies from r=0.69 to r=0.87 with root mean squared errors (RMSE) around 0.03 and 0.04 m3/m3. Our soil moisture predictions improve: (a) the correlation with the ISMN (when compared with the original ESA-CCI product) from r=0.30 (RMSE=0.09 m3/m3 ) to r=0.66 (RMSE=0.05 m3/m3 ); and (b) the
correlation with local precipitation records across boreal (from r=<0.3 up r=0.49) or tropical areas (from r=<0.3 to r=0.46) which are currently poorly represented in the ISMN. Temporal trends show a decline of global annual soil moisture using: a) data from the ISMN (-1.5 [-1.8, -1.24]%, b) associated locations from the original ESA-CCI dataset (- 0.87[-1.54, -0.17]%), c) associated locations from predictions based on terrain parameters (-0.85[-1.01, -0.49]%), and d)associated locations from predictions including bioclimatic and soil type classes (-0.68[-0.91, -0.45]%). Our parsimonious downscaled soil moisture predictions are independent of climate variables and vegetation indexes, to avoid potential spurious correlations in future research, and they complement information about soil moisture dynamics worldwide.

Show More

ABSTRACT:

We provide a set of 26 soil moisture predictions across 15km grids at the global scale. We modeled and predicted the ESA-CCI soil moisture values across 26 years of available data (1991-2016) using a ML based kernel method and multiple terrain parameters (e.g., slope, wetness index) as prediction factors. We used ground information from the International Soil Moisture Network (ISMN, n=13376) for evaluating soil moisture predictions. Our downscaled soil moisture predictions across 15km grids showed a statistical accuracy varying 0.69-0.87% and 0.04 m3/m3 of cross-validated explained variance and root mean squared error (RMSE). We found a negative bias (-0.01 to -0.08 m3/m3 ) underestimating the values of ISMN when comparing with the ESA-CCI and our predictions across the analyzed years and a relatively better performance between 1998 and 2016. We found no significant differences between the ESA-CCI and our predictions, but we found discrepancy between multiple evaluation metrics (e.g., bias vs efficiency) comparing the ESA-CCI with the ISMN. However, the temporal analysis as revealed by a robust trend detection strategy (e.g., Theil-Sen estimator), suggests a decline of soil moisture at the global scale that is consistent in both gridded estimates and field measurements of soil moisture varying from -0.7[-0.77, -0.62]% in the ESA-CCI product, -0.9[-1.01, -0.8]% in the downscaled predictions and -1.6 [-1.7, -1.5]% in the ISMN. These results highlight the large potential of digital terrain parameters for improving the accuracy and spatial detail of satellite soil moisture grids at the global scale. The soil moisture predictions provided here (folder: predicted-2001-2016) could be useful for quantifying long term soil moisture emergent patterns (i.e., trends) across areas with low availability of soil moisture information in the ESA-CCI. To ensure reproducible results of this study, we also provide the R code and (also in R native format *.rds) the topographic prediction factors for soil moisture across 15 km grids (file: topographic_predictors_15km_grids.rds). This site also includes the harmonized ISMN data with the ESA-CCI and the downscaled predictions based on terrain analysis in an annual basis (files: harmonizedISMNvsESACCI.rds and harmonizedISMNvsPREDICTED.rds) that we used for validating our prediction framework. The soil moisture predictions provided here could be useful for quantifying soil moisture spatial and temporal dynamics across areas with low availability of soil moisture information in the original ESA-CCI database.

Show More

ABSTRACT:

We provide 26 annual soil moisture predictions across conterminous United States for the years 1991-2016. These predictions are provided in raster files with a geographical (lat, long) projection system and a spatial resolution of 1 x 1 km grids (folder: soil_moisture_annual_grids_1991_2016). These raster files were populated with soil moisture data based on multiple kernel based machine learning models for coupling hydrologically meaningful terrain parameters (the explanatory variables) with soil moisture microwave records (the response variable) from the European Space Agency Climate Change Initiative. We provide a raster stack with the annual training data from satellite soil moisture estimates (file: annual_means_of _ESA_CCI_soil_moiture_1991_2016.tif) and the explanatory variables (terrain) calculated on SAGA GIS (System of Automated Geoscientific Analysis) using digital terrain analysis (folder: explanatory_variables_dem). The explained variance for all models-years was >70% (10-fold cross-validation). The 1 km soil moisture grids (compared to the original satellite soil moisture estimates) had higher correlations with field soil moisture observations from the North American Soil Moisture Database (n=668 locations with available data between 1991-2013; 0-5 cm depth) than soil moisture microwave records. For further information refer to our preprint in bioRxiv: https://www.biorxiv.org/content/biorxiv/early/2019/07/01/688846.full.pdf

Show More

 Contact

Resources
All 0
Collection 0
Composite Resource 0
Generic 0
Geographic Feature 0
Geographic Raster 0
HIS Referenced Time Series 0
Model Instance 0
Model Program 0
MODFLOW Model Instance Resource 0
Multidimensional (NetCDF) 0
Script Resource 0
SWAT Model Instance 0
Time Series 0
Web App 0
Composite Resource Composite Resource

ABSTRACT:

We provide 26 annual soil moisture predictions across conterminous United States for the years 1991-2016. These predictions are provided in raster files with a geographical (lat, long) projection system and a spatial resolution of 1 x 1 km grids (folder: soil_moisture_annual_grids_1991_2016). These raster files were populated with soil moisture data based on multiple kernel based machine learning models for coupling hydrologically meaningful terrain parameters (the explanatory variables) with soil moisture microwave records (the response variable) from the European Space Agency Climate Change Initiative. We provide a raster stack with the annual training data from satellite soil moisture estimates (file: annual_means_of _ESA_CCI_soil_moiture_1991_2016.tif) and the explanatory variables (terrain) calculated on SAGA GIS (System of Automated Geoscientific Analysis) using digital terrain analysis (folder: explanatory_variables_dem). The explained variance for all models-years was >70% (10-fold cross-validation). The 1 km soil moisture grids (compared to the original satellite soil moisture estimates) had higher correlations with field soil moisture observations from the North American Soil Moisture Database (n=668 locations with available data between 1991-2013; 0-5 cm depth) than soil moisture microwave records. For further information refer to our preprint in bioRxiv: https://www.biorxiv.org/content/biorxiv/early/2019/07/01/688846.full.pdf

Show More
Composite Resource Composite Resource
Gap-Free Global Annual Soil Moisture: 15km Grids for 1991-2016
Created: July 5, 2019, 6:38 p.m.
Authors: Guevara, Mario · Vargas, Rodrigo · Michela Taufer

ABSTRACT:

We provide a set of 26 soil moisture predictions across 15km grids at the global scale. We modeled and predicted the ESA-CCI soil moisture values across 26 years of available data (1991-2016) using a ML based kernel method and multiple terrain parameters (e.g., slope, wetness index) as prediction factors. We used ground information from the International Soil Moisture Network (ISMN, n=13376) for evaluating soil moisture predictions. Our downscaled soil moisture predictions across 15km grids showed a statistical accuracy varying 0.69-0.87% and 0.04 m3/m3 of cross-validated explained variance and root mean squared error (RMSE). We found a negative bias (-0.01 to -0.08 m3/m3 ) underestimating the values of ISMN when comparing with the ESA-CCI and our predictions across the analyzed years and a relatively better performance between 1998 and 2016. We found no significant differences between the ESA-CCI and our predictions, but we found discrepancy between multiple evaluation metrics (e.g., bias vs efficiency) comparing the ESA-CCI with the ISMN. However, the temporal analysis as revealed by a robust trend detection strategy (e.g., Theil-Sen estimator), suggests a decline of soil moisture at the global scale that is consistent in both gridded estimates and field measurements of soil moisture varying from -0.7[-0.77, -0.62]% in the ESA-CCI product, -0.9[-1.01, -0.8]% in the downscaled predictions and -1.6 [-1.7, -1.5]% in the ISMN. These results highlight the large potential of digital terrain parameters for improving the accuracy and spatial detail of satellite soil moisture grids at the global scale. The soil moisture predictions provided here (folder: predicted-2001-2016) could be useful for quantifying long term soil moisture emergent patterns (i.e., trends) across areas with low availability of soil moisture information in the ESA-CCI. To ensure reproducible results of this study, we also provide the R code and (also in R native format *.rds) the topographic prediction factors for soil moisture across 15 km grids (file: topographic_predictors_15km_grids.rds). This site also includes the harmonized ISMN data with the ESA-CCI and the downscaled predictions based on terrain analysis in an annual basis (files: harmonizedISMNvsESACCI.rds and harmonizedISMNvsPREDICTED.rds) that we used for validating our prediction framework. The soil moisture predictions provided here could be useful for quantifying soil moisture spatial and temporal dynamics across areas with low availability of soil moisture information in the original ESA-CCI database.

Show More
Composite Resource Composite Resource
Gap-Free Global Annual Soil Moisture: 15km Grids for 1991-2018
Created: May 6, 2020, 1:22 p.m.
Authors: Guevara, Mario · Vargas, Rodrigo · Michela Taufer

ABSTRACT:

Soil moisture is key for quantifying soil-atmosphere interactions and the ESA-CCI (European Space Agency-Climate Change Initiative) provides historical (>30 years) satellite soil moisture global grids with spatial resolution of ~27km. This dataset is incomplete (contains gaps) due to conditions such as dense vegetation or extremely dry surfaces. Here we provide a framework to increase the spatial resolution and fill gaps (reporting associated uncertainty) of the ESA-CCI (v4.5) soil moisture dataset. The outcome is a new dataset of gap-free global mean annual soil moisture and uncertainty
for 28 years (1991-2018) across 15km grids. We compare the performance of machine learning odels using only terrain parameters (e.g., slope, wetness index) against predictions using terrain parameters, bioclimatic information, and soil type classes. We use independent field information from the International Soil Moisture Network (ISMN, n=13376) and in-situ precipitation records (n=171) only for model evaluation purposes. Using only terrain parameters to predict soil moisture results in a parsimonious approach comparable with a more complex model that includes additional bioclimatic and soil information. The correlation between observed and predicted soil moisture values varies from r=0.69 to r=0.87 with root mean squared errors (RMSE) around 0.03 and 0.04 m3/m3. Our soil moisture predictions improve: (a) the correlation with the ISMN (when compared with the original ESA-CCI product) from r=0.30 (RMSE=0.09 m3/m3 ) to r=0.66 (RMSE=0.05 m3/m3 ); and (b) the
correlation with local precipitation records across boreal (from r=<0.3 up r=0.49) or tropical areas (from r=<0.3 to r=0.46) which are currently poorly represented in the ISMN. Temporal trends show a decline of global annual soil moisture using: a) data from the ISMN (-1.5 [-1.8, -1.24]%, b) associated locations from the original ESA-CCI dataset (- 0.87[-1.54, -0.17]%), c) associated locations from predictions based on terrain parameters (-0.85[-1.01, -0.49]%), and d)associated locations from predictions including bioclimatic and soil type classes (-0.68[-0.91, -0.45]%). Our parsimonious downscaled soil moisture predictions are independent of climate variables and vegetation indexes, to avoid potential spurious correlations in future research, and they complement information about soil moisture dynamics worldwide.

Show More