Investigation of reanalysis data for regional landslides and climate change in the Emilia Romagna region, Italy

Abstract

Global climate change is expected to alter the intensity, frequency and duration of severe weather events, including droughts, floods, and landslides. Changes in climatic variables like rainfall are indicated to be responsible for increased landslide occurrences, but their causal relationship has not been explored thoroughly on the regional scale. The emerging reanalysis rainfall has spatial-temporal heterogeneity and temporal coverage of more than 30 years, which enables it to conduct climate change analysis. However, they suffer from problems of low accuracy and coarse resolution problems and have not yet been applied to regional landslide predictions. Conventionally, empirical thresholds rely on rain gauges to establish. The Emilia-Romagna (Italy) region has abundant precipitation records and a complete historical landslide catalogue, which enable it an ideal area to validate and foster the application of reanalysis rainfall estimates in landslide early warning systems. The established relationships and the merging approach proposed in this thesis hold significant implications for the application in areas like Africa with limited rainfall stations and incomplete records. By merging the reanalysis estimates and observed data, this thesis aims to establish the framework for the application of reanalysis rainfall products (including such as ECMWF Re-Analysis 5th Generation (ERA5) and the land component of the fifth generation of European ReAnalysis (ERA5-Land)) on regional landslide prediction and explore the impact of climate change on landslide occurrences from the meteorological perspective. Mainly, three parts constitute this thesis:
The first part (Chapter 4) intends to solve the problem of resolution and quality of ERA5 and ERA5-Land. It presents a framework centered around statistical interpolation, bias correction, and evaluations of ERA5 rainfall estimates, with a specific focus on the interpolated ERA5 rainfall product (ERA5-OK) and bias corrections using the OK method and the QM approach. Firstly, the suitable rainfall interpolation methods for gridded datasets were explored and the accuracy of interpolated ERA5, ERA5-Land rainfall data were evaluated by rain-gauge records. Based on the ERA5 gridded rainfall datasets, five interpolation techniques have been selected and compared at different time scales (annual, monthly, and annual maximum daily precipitation). Results show
ii
that all five interpolation methods have certain capabilities to improve spatial resolution. Among them, the OK generally outperforms the other four methods. After interpolations, the comprehensive evaluation of the simple interpolated ERA5 rainfall by the ordinary Kriging method (ERA5-OK), the ERA5-Land and observations is conducted and bias corrections is conducted on daily estimates, covering four topographical features, three temporal scales (annual, seasonal and daily rainfall) and five rainfall intensities (0.1mm/day-10mm/day, 10-20mm/day, 20-50 mm/day, 50-100mm/day and >100mm/day). In addition, the impact of station numbers on the performance of Quantile Mapping (QM) on the ERA5 products is carried out. Results show that the ERA5-OK can achieve nearly the same spatial-temporal performance as ERA5-Land's at all time scales, even better at the daily scale, especially in high-altitude mountainous zone and the coastal edge. A more reliable ERA5 rainfall estimates with higher accuracy and finer resolutions can be generated by this process.
The second part (Chapter 5) aims to explore the application of the reanalysis rainfall products (including ERA5 and corrected ERA5-Land) for landslide prediction on a regional scale. Specifically, the new procedure to migrate the statistical threshold from rain gauge observations to reanalysis rainfall was firstly proposed; three ways for quantitatively defining no-rain conditions are introduced and a new evaluation indicator named 𝑑20−50 is demonstrated. Correspondingly, their performances on regional landslide prediction is examined within thresholds generated by the frequentist method and two-dimensional Bayesian methods. Results show that ERA5-Land yields better prediction in terms of aggregation, while ERA5 lags behind that of the rain gauge. In addition, it is necessary to define rainy days before establishing empirical thresholds. Among three ways for defining no-rain conditions, the one determined by the rainfall event numbers performs the best.
The third part (Chapter 6) attempts to quantitatively define the influence of climate change on landslide occurrences and explore whether the effect could exacerbate or reduce landslide impact on a regional scale. In order to investigate the change rule, a trend analysis of meteorological variables, such as precipitation, is carried out over the past 70 years. Furthermore, the corresponding temporal and spatial distribution characteristics are demonstrated. By using the established threshold in Chapter 5, relationships between climate parameters and consequences, such as landslide event
iii
frequency and the magnitude of landslide-triggering rainfall events, are calculated. Results show that there is an overall upward trend in the annual rainfall and extreme rainfall indices; hilly areas are more sensitive to changes in precipitation compared with the coastal areas, which is consistent with the variation of landslides occurrence. According to the threshold baseline, the estimated induced frequency of triggering rainfall events has increased while the estimated average triggered rainfall intensity dropped from 21mm/day to 16.5mm/day.
Overall, the findings of this thesis prove the feasibility of applying reanalysis data on regional landslide predictions and deepen the understanding of how meteorological factors and landslide events respond to climate change. It sheds new light on the development of landslide warning systems with the aid of reanalysis data. In addition, it is beneficial for civil protection agencies to adjust risk mitigation measures and hazard reduction strategies in the future.

Date of Award	3 Oct 2023
Original language	English
Awarding Institution	University of Bristol
Supervisor	Dawei Han (Supervisor), Maria Pregnolato (Supervisor) & Lu Zhuo (Supervisor)