Exploring improvements for landslide risk and flood monitoring products in Copernicus Emergency Management Service (CEMS) mapping: the validation studies for cyclones in Honduras

Tropical Cyclones Eta and Iota, Honduras
CEMS Risk and Recovery (RRM) was activated by DG-ECHO ERCC upon request of the Centre for Prevention of Disasters in Central America (CEPREDENAC) to assess the effects of the Eta and Iota hurricanes that hit Central America in November 2020.  
   
Honduras was severely affected by historic level floods, destructive winds and devastating landslides. The aim of the RRM activation was to evaluate the risk of future landslides and to carry out a temporal analysis of the flooding.

Landslide risk estimation 
The RRM standard product P17 supports decision-makers by localizing areas of high landslide risk and defining prevention and mitigation measures. The CEPREDENAC request the Rapid Response Mapping team at CEMS received was to identify critical regions in the El Cajon reservoir surroundings after landslides caused by Eta and Iota cyclones.

The RRM landslide risk product was computed using the Landslides Susceptibility Index (LSI), using the following parameters: slope angle, lithology, elevation, distance from rivers and land cover. The LSI is the sum of the weighted parameters, which are generated based on the relative importance of the input variables, using the Analytical Hierarchical Process approach.
 
CEMS Validation investigated and proposed an advanced model that includes precipitation, as it was the causal factor in these landslides.
 
A distinction was made between Landslide Susceptibility (LS) and Landslide Hazard (LH). LS models were suggested to include slope angle, geology/lithology and land cover, together with triggering variables such as precipitation or seismicity. 
 
This proposed Landslide Hazard Index (LHI) model was the result a balanced compromise between the required ease and speed of map production and an accurate representation of the most relevant variables in order to identify the areas with the highest landslide hazard.
  
Figure 1 Subdivision of variables into relevant classes
  
When selecting the input variables, the relative contribution of the variables in landslide occurrence must be considered. 
 
Figure 2 Extreme monthly rainfall RP100y. Precipitation data can be derived from satellite data, for example from GPCC and CHIRPS (used here).
 
The results comparison shows that Very High hazard class increases its area from 2,450ha in P17 to 7,939ha. Its spatial distribution suggests that it is directly related to the addition of the extreme monthly rainfall Return Period=100 years. 
 
 
 
Figure 3 An area showing the impact of the inclusion of the extreme precipitation triggering variable: which significantly increases the estimate of landslide hazard.
 
Temporal analysis of the floods 
Monitoring floods using earth observation satellites can be challenging. Why? Because optical sensors cannot provide images through the cloud cover associated with floods. SAR image data is also limited in forested and urban areas. 
 Due to these difficulties encountered in flood monitoring by satellite, hydraulic modelling can provide what users need: accurate estimates of flood extent and its full temporal evolution, including the maximum flood extent that satellite sensors often miss. 
This standard RRM product P06 uses hydraulic modelling to provide maps of maximum flood extent, maximum water depth and maximum flood retention for the two cyclones. It requires elevation data covering the entire catchment, precipitation and land cover data.
 CEMS Validation proposed a faster and more robust modelling of the flooding in Honduras that used 30-minute precipitation data instead of the daily precipitation estimates used by RRM. This allows a more detailed representation of basin hydrography in terms of cumulative precipitation and runoff, resulting in improved flow simulation. Detailed flood mapping at the peak runoff was generated, together with water level depths, both before and after the peak rainfall. 
 
 
 
 
 
 
 
Figure 4 - 3D representation of the flooding at the peak of the simulation. Red arrows show velocities > 1 m/s; graded blue shows water depths; buildings are as
 
The value added of this validation is highlighted by some of the key methodological comparisons between these models. Specifically, in terms of the hydrological representation of the entire Ulua and Chamelecon catchments, the advantages of using 30-minute precipitation data, model calibration and the capabilities of the hydraulic modelling software resulted in a system that provides: 
· State-of-the-art-operational flood forecasting, early warning and emergency management in real-time.
· Highly improved computation time: using proprietary software and processing on a CUDA graphic card makes the processing time 20-30 times faster, which allows for robust simulation of varying flows.
· Flow rates that are closer to published (reference) event flow information.
 Conclusions 
Aiming to the continuous improvement of CEMS, the two studies identified and tested advanced models in a real case. 
The models presented provide improvements in both computational speed and robustness in terms of the temporal resolution of variables and model parameterization that can be implemented within the cost and time constraints of a RRM standard activation.
The CEMS mapping team reviewed the results of the studies and considered the inclusion of these models in the next phase of RRM.