Input map and feature selection for soil legacy data

Flynn, Trevan; Rozanov, Andrei; Clarke, Catherine E.

doi:10.1016/j.geoderma.2020.114452

Cited by 10 publications

(1 citation statement)

References 29 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…The algorithm is capable of detecting all important co‐variables that are correlated and redundant with precipitation (Flynn et al ., 2020). In order to identify the most important co‐variables, Boruta is performing the following steps: (1) generates copies from all co‐variables named shadow attribute whose values are randomly permuted in order to eliminate the correlation with precipitation, (2) computes z scores for all data using forest band classifier, (3) compares the results with shadow attribute, (4) determine the importance of co‐variables (Ali et al ., 2019; Prasad et al ., 2019; Mollalo et al ., 2020; Ebrahimi‐Khusfi et al ., 2021).…”

Section: Methodsmentioning

confidence: 99%

Enhanced precipitation prediction using DEM‐based predictors and satellite imagery

Antal

Dumitrescu

Cheval

et al. 2023

Intl Journal of Climatology

View full text Add to dashboard Cite

Over the last decades, climatic changes have triggered considerable impacts across the globe with detrimental effects on all ecosystems. Given the complexity of topography and climate, Romania is one of the most exposed countries in the South‐Eastern Europe to extreme hydrological events. As a consequence, the spatial distribution of precipitation is of greater importance for future analysis. This study examines the performance of Empirical Bayesian Kriging Regression Prediction (EBKRP) and Geographically Weighted Regression (GWR) to predict the spatial distribution of annual and seasonal precipitation over Romania. Twelve co‐variables derived from topography, and remote sensing products data were used to improve the prediction. The co‐variable selection process was performed prior to the regression model to reduce the complexity and processing time, using the Boruta Algorithm (BA), which is an innovative approach. The performance of BA was compared with Gini Coefficient. Our findings confirmed that 10 co‐variables were relevant to predict annual precipitation and nine for seasonal. The overall prediction of precipitation is more influenced by topography (altitude, slope, surface roughness) and the distance to marine bodies (Black Sea and Adriatic Sea). Cross‐validation and five statistical metrics were applied to assess the performance of the regression models. The results show similar spatial distribution pattern of precipitation, whereas the highest annual precipitation is found in the Carpathian Mountains (EBKRP: 1,399.2 mm, GWR: 1,249.7 mm) and the lowest in the lowlands (EBKRP: 355.9 mm, GWR: 346.8 mm). For seasons, both methods predicted the highest precipitation in summer and lowest in autumn. In all seasons, the precipitation is underestimated by both methods, however, for annual, only GWR does so. Our study revealed GWR as the best method to predict annual and seasonal precipitation over Romania, as it yielded the highest correlation coefficient Spearman (S), Pearson (P), determination coefficient (R2) and lowest mean absolute error (MAE) and root mean square error (RMSE).

show abstract