Benchmarking a random forest for predicting preferential flow in soils
Preferential flow processes are important to fully understand flow and solute transport in the vadose zone and implement adequate management practices. Physical models are difficult to use at large scales to predict soil susceptibility to preferential flow. Instead, pedotransfer functions might b...
| Autor principal: | |
|---|---|
| Formato: | H2 |
| Lenguaje: | Inglés |
| Publicado: |
SLU/Dept. of Soil and Environment
2013
|
| Materias: |
| _version_ | 1855570908486303744 |
|---|---|
| author | Jordà Guerra, Helena |
| author_browse | Jordà Guerra, Helena |
| author_facet | Jordà Guerra, Helena |
| author_sort | Jordà Guerra, Helena |
| collection | Epsilon Archive for Student Projects |
| description | Preferential flow processes are important to fully understand flow and solute transport in the vadose zone and implement adequate management practices.
Physical models are difficult to use at large scales to predict soil susceptibility to preferential flow. Instead, pedotransfer functions might be applied. The strength
of preferential flow can be measured by the relative 5% arrival time obtained from breakthrough curve (BTC) experiments. I used a database containing 560 BTC
experiments to build random forests to predict the relative 5% arrival time and analyse the importance of soil properties and site factors on predicting this
feature. The coefficient of determination for a 10-fold cross-validation was 70%, whereas the benchmarking process obtained a coefficient of 27%. Sand contents between 0.80 and 0.92 reached the highest importance and were strongly related to weak preferential flow. High importance was also observed in silt contents lower than 0.11, and clay contents between 0.04 and 0.08, which were strongly
correlated to high preferential flow. In addition, experimental conditions such as flow rate, column diameter, the use of fixed drippers and column venting had 20%
importance. This study revealed that texture can broadly predict soil susceptibility to preferential flow, while other site and soil factors can later refine this estimate.
However, the dataset lacked land use information and a broader range of experimental conditions. I consider that enlarging the database is a key factor to obtain better predictions and to further understand how soil and site characteristics influence soil susceptibility to preferential flow. |
| format | H2 |
| id | RepoSLU5817 |
| institution | Swedish University of Agricultural Sciences |
| language | Inglés |
| publishDate | 2013 |
| publishDateSort | 2013 |
| publisher | SLU/Dept. of Soil and Environment |
| publisherStr | SLU/Dept. of Soil and Environment |
| record_format | eprints |
| spelling | RepoSLU58172013-07-03T11:55:47Z Benchmarking a random forest for predicting preferential flow in soils Jordà Guerra, Helena preferential flow random forest benchmarking undisturbed soil soil database Preferential flow processes are important to fully understand flow and solute transport in the vadose zone and implement adequate management practices. Physical models are difficult to use at large scales to predict soil susceptibility to preferential flow. Instead, pedotransfer functions might be applied. The strength of preferential flow can be measured by the relative 5% arrival time obtained from breakthrough curve (BTC) experiments. I used a database containing 560 BTC experiments to build random forests to predict the relative 5% arrival time and analyse the importance of soil properties and site factors on predicting this feature. The coefficient of determination for a 10-fold cross-validation was 70%, whereas the benchmarking process obtained a coefficient of 27%. Sand contents between 0.80 and 0.92 reached the highest importance and were strongly related to weak preferential flow. High importance was also observed in silt contents lower than 0.11, and clay contents between 0.04 and 0.08, which were strongly correlated to high preferential flow. In addition, experimental conditions such as flow rate, column diameter, the use of fixed drippers and column venting had 20% importance. This study revealed that texture can broadly predict soil susceptibility to preferential flow, while other site and soil factors can later refine this estimate. However, the dataset lacked land use information and a broader range of experimental conditions. I consider that enlarging the database is a key factor to obtain better predictions and to further understand how soil and site characteristics influence soil susceptibility to preferential flow. SLU/Dept. of Soil and Environment 2013 H2 eng https://stud.epsilon.slu.se/5817/ |
| spellingShingle | preferential flow random forest benchmarking undisturbed soil soil database Jordà Guerra, Helena Benchmarking a random forest for predicting preferential flow in soils |
| title | Benchmarking a random forest for predicting preferential flow in soils |
| title_full | Benchmarking a random forest for predicting preferential flow in soils |
| title_fullStr | Benchmarking a random forest for predicting preferential flow in soils |
| title_full_unstemmed | Benchmarking a random forest for predicting preferential flow in soils |
| title_short | Benchmarking a random forest for predicting preferential flow in soils |
| title_sort | benchmarking a random forest for predicting preferential flow in soils |
| topic | preferential flow random forest benchmarking undisturbed soil soil database |