Loading...

Hindcasting and nowcasting rainfall-induced landslides – do global precipitation estimates help?

Figure: Landslide in Oita Prefecture, Japan at 11:33 on July 7, 2017, during the 2017 Fukuoka floods (mainichi.jp, accessed August 6, 2018)

In July 2017, an unusual torrential storm hit southwestern Japan in Fukuoka Prefecture triggering more than 2000 landslides within 12 hours. Unlike the seemingly large extent of the overall rainfall activity, landslides were concentrated in a small area of about 200 km2, where total rainfall accumulation exceeded 500mm. A year later, trenching rains during a larger weather system for about 5 days affected the entire southwestern Japan. This time, the week-long rainfalls triggered about 8500 landslides in an area of 3000 km2 with spatially highly variable rainfall accumulations. The contrast between these distinct events motivated us to explore the usefulness of global grid rainfall data, such as GPM IMERG (Integrated Multi-satellitE Retrievals for Global Precipitation Measurement) and ERA5 climate reanalysis data, for landslide hindcasting or nowcasting.

The quality of these global rainfall estimates is frequently questioned and they have been compared to their ground-based, but higher resoluted and more accurate counterparts, i.e., rainfall radar. Relying on the promising performance, there are already a few early models for landslide nowcasting, such as the one by NASA that evaluates GPM IMERG data in real-time to increase situational awareness for landslide hazard. However, whether these rainfall estimates are adequate for landslides nowcasting remains largely unexplored.

In a recent study we address this research gap. We developed a simple logistic regression model to hindcast rainfall-induced landslides in these two Japanese sites. We started off with a relatively simple landslide susceptibility model and investigated to which degree the model can be improved by different rainfall products. Rainfall-radar performed best, elevating the performance by 30 %-points from 65% up to 95% (measured by the metric ROC-AUC). We hoped to see a similar performance boost when using the global and grid-based rainfall estimates. Yet, for these rainfall products the performance increase was rather meager, amounting to 4-7%.

Our study shows that satellite- or reanalysis-derived rainfall products are not yet fully capable of capturing rainfall patterns in the detail for reliably hindcasting or nowcasting spatial patterns of rainfall-induced landslides. However, we emphasize the potential for these products to become valuable predictors as soon as they attain a sufficient spatial accuracy.

Reference

Ozturk, U., Saito, H., Matsushi, Y., Crisologo, I., Schwanghart, W., 2021. Can global rainfall estimates (satellite and reanalysis) aid landslide hindcasting? Landslides, in press. [DOI: 10.1007/s10346-021-01689-3]

This blog is previously published on https://topotoolbox.wordpress.com/2021/06/21/hindcasting-and-nowcasting-rainfall-induced-landslides-do-global-precipitation-estimates-help/