Abstract Satellite rainfall products have been available for many years (since '90) with an increasing spatial/temporal resolution and accuracy. Their global scale coverage and near real-time products perfectly fit the need of an early warning landslide system. Notwithstanding these characteristics, the number of studies employing satellite rainfall estimates for predicting landslide events is quite limited. In this study, we propose a procedure that allows us to evaluate the capability of different rainfall products to forecast the spatial-temporal occurrence of rainfall-induced landslides using rainfall thresholds. Specifically, the assessment is carried out in terms of skill scores, and receiver operating characteristic (ROC) analysis. The procedure is applied to ground observations and four different satellite rainfall estimates: 1) the Tropical Rainfall Measurement Mission Multi-satellite Precipitation Analysis, TMPA, real time product (3B42-RT), 2) the SM2RASC product obtained from the application of SM2RAIN algorithm to the Advanced SCATterometer (ASCAT) derived satellite soil moisture (SM) data, 3) the Precipitation Estimation from Remotely Sensed Information using Artificial Neural Network (PERSIANN), and 4) the Climate Prediction Center (CPC) Morphing Technique (CMORPH). As case study, we consider the Italian territory for which a catalogue listing 1414 rainfall-induced landslides in the period 2008–2014 is available. Results show that satellite products underestimate rainfall with respect to ground observations. However, by adjusting the rainfall thresholds, satellite products are able to identify landslide occurrence, even though with less accuracy than ground-based rainfall observations. Among the four satellite rainfall products, CMORPH and SM2RASC are performing the best, even though differences are small. This result is to be attributed to the high spatial/temporal resolution of CMORPH, and the good accuracy of SM2RSC. Overall, we believe that satellite rainfall estimates might be an important additional data source for developing continental or global landslide warning systems.

中文翻译：

---

我们离在滑坡预测中使用卫星降雨产品还有多远？

摘要 卫星降雨产品已经可用多年（自 90 年以来），其空间/时间分辨率和精度不断提高。他们的全球范围覆盖和近乎实时的产品完美地满足了预警滑坡系统的需要。尽管有这些特点，使用卫星降雨估计来预测滑坡事件的研究数量非常有限。在这项研究中，我们提出了一个程序，使我们能够评估不同降雨产品的能力，以使用降雨阈值预测降雨诱发的滑坡的时空发生。具体而言，评估是根据技能分数和接收器操作特征 (ROC) 分析进行的。该程序适用于地面观测和四种不同的卫星降雨估计：1) Tropical Rainfall Measurement Mission Multi-satellite Precipitation Analysis, TMPA, real time product (3B42-RT), 2) SM2RAIN算法应用于Advanced SCATterometer (ASCAT)推导卫星土壤水分(SM)得到的SM2RASC产品数据，3) 使用人工神经网络 (PERSIANN) 的遥感信息的降水估计，以及 4) 气候预测中心 (CPC) 变形技术 (CMORPH)。作为案例研究，我们考虑了意大利领土，该领土的目录列出了 2008-2014 年期间 1414 次降雨诱发的滑坡。结果表明，卫星产品低估了地面观测的降雨量。然而，通过调整降雨阈值，卫星产品能够识别滑坡的发生，尽管精度不如地面降雨观测。在四个卫星降雨产品中，CMORPH 和 SM2RASC 表现最好，尽管差异很小。这一结果要归功于 CMORPH 的高空间/时间分辨率和 SM2RSC 的良好精度。总的来说，我们认为卫星降雨估计可能是开发大陆或全球滑坡预警系统的重要附加数据来源。