RainForest: a random forest algorithm for quantitative precipitation estimation over Switzerland

Abstract

Abstract. Quantitative precipitation estimation (QPE) is a difficult task, particularly in complex topography, and requires the adjustment of empirical relations between radar observables and precipitation quantities, as well as methods to transform observations aloft to estimations at the ground level. In this work, we tackle this classical problem with a new twist, by training a random forest (RF) regression to learn a QPE model directly from a large database comprising 4 years of combined gauge and polarimetric radar observations. This algorithm is carefully fine-tuned by optimizing its hyperparameters and then compared with MeteoSwiss' current operational non-polarimetric QPE method. The evaluation shows that the RF algorithm is able to significantly reduce the error and the bias of the predicted precipitation intensities, especially for large and solid or mixed precipitation. In weak precipitation, however, and despite a posteriori bias correction, the RF method has a tendency to overestimate. The trained RF is then adapted to run in a quasi-operational setup providing 5 min QPE estimates on a Cartesian grid, using a simple temporal disaggregation scheme. A series of six case studies reveal that the RF method creates realistic precipitation fields, with no visible radar artifacts, that appear less smooth than the original non-polarimetric QPE and offers an improved performance for five out of six events.

Calibration of X-Band Radar for Extreme Events in a Spatially Complex Precipitation Region in North Peru: Machine Learning vs. Empirical Approach

“…When comparing the RF result to corrected reflectivity images, specifically the case studies, the approach is able to reproduce the spatial variability of rainfall in more detail and precision, although its quantitative performance is lower than its counterpart because of underestimation/overestimation of stronger rainfall. Limitations on the RF calibration scheme can be attributed to the reduced number of samples used for training the model in comparison with previous studies [30,31]. Particularly high precipitation records are scarce, and thus, the RF approach lacks enough training samples for generalization.…”

“…On the other hand, the derivation of radar QPE by using the random forest algorithm has recently proven to have high potential and promising results [30][31][32], and its application has increasingly been reported in the literature [33][34][35]. This technique has many advantages in comparison with other machine learning methods and has successfully improved the performance of radar QPE data in mountain regions.…”

“…This technique has many advantages in comparison with other machine learning methods and has successfully improved the performance of radar QPE data in mountain regions. The key advantage of RF lies in the simplicity of the hyperparameter tuning [30,31]. In preliminary tests, [31] showed that without extensive fine tuning, other machine learning techniques such as ANN and gradient boosting (GB) provide relatively similar performance to RF.…”

“…The key advantage of RF lies in the simplicity of the hyperparameter tuning [30,31]. In preliminary tests, [31] showed that without extensive fine tuning, other machine learning techniques such as ANN and gradient boosting (GB) provide relatively similar performance to RF. This makes RF more appealing for transfer knowledge to practitioners, since it guarantees a less complex model update in the long term when obtaining larger datasets along the time.…”

Assessing Machine Learning Models for Gap Filling Daily Rainfall Series in a Semiarid Region of Spain

“…Other researchers have already assessed RF in rainfall with promising results [40][41][42]. For further details, the following work can be revised [42].…”