Machine learning-based integration of remotely-sensed drought factors can improve the estimation of agricultural drought in South-Eastern Australia
- Publication Type:
- Journal Article
- Agricultural Systems, 2019, 173 pp. 303 - 316
- Issue Date:
|2019-03-Machine learning-based integration of remotely-sensed drought factors can improve the estimation of agricultural drought in South-Eastern Australia.pdf||Published Version||2.5 MB|
Copyright Clearance Process
- Recently Added
- In Progress
- Closed Access
This item is closed access and not available.
© 2019 Agricultural drought is a natural hazard arising from insufficient crop water supply. Many drought indices have been developed to characterize agricultural drought, relying on either ground-based climate data or various remotely-sensed drought proxies. Ground-based drought indices are more accurate but limited in coverage, while remote sensing drought indices cover large areas but have poor precision. Application of advanced data fusion approaches based on remotely-sensed data to estimate ground-based drought indices may help fill this gap. The overall objective of this study was to determine whether various remotely-sensed drought factors could be effectively used for monitoring agricultural drought in south-eastern Australia. In this study, thirty remotely-sensed drought factors from the Tropical Rainfall Measuring Mission (TRMM) and the Moderate Resolution Imaging Spectroradiometer (MODIS) satellite sensors were used to reproduce a ground-based drought index, SPEI (Standardized Precipitation Evapotranspiration Index) during 2001–2017 for the New South Wales wheat belt in south-eastern Australia. Three advanced machine learning methods, i.e. bias-corrected random forest, support vector machine, and multi-layer perceptron neural network, were adopted as the regression models in this procedure. A station-based historical climate dataset and observed wheat yields were used as reference data to evaluate the performance of the model-predicted SPEI in reflecting agricultural drought. Results show that the bias-corrected random forest model outperformed the other two models for SPEI prediction, as quantified by the lowest root mean square error (RMSE) and the highest R 2 values (<0.28 and ~0.9, respectively). Drought distribution maps produced by the bias-corrected random forest model were then compared with the station-based drought maps, showing strong visual and statistical agreement. Furthermore, the model-predicted SPEI values were more highly correlated with observed wheat yields than the station-based SPEI. The method used in this study is effective and fast, and based on data that are readily available. It can be easily extended to other cropping areas to produce a rapid overview of drought conditions and to enhance the present capabilities of real-time drought monitoring.
Please use this identifier to cite or link to this item: