Flood risk forecasting at weather to medium range incorporating weather model, topography, socio-economic information and land use exposure

Highlights

• Inadequacy of weather forecasts to identify the flood hotspots.

• Development of a new framework for weather scale flood risk forecasting.

• New definition and forecasts of Hazard at a weather scale using the hindcasts.

• Incorporation of socio-economic, geomorphologic and exposure information in flood forecasting.

• Forecasting and identification of high risk zones using the developed methodology.

Abstract

Non-structural mitigation measures to the globally increasing flood events include forecast based alert generation. However, the extreme rainfall forecasts are associated with low hit rate, high false alarm, and spatiotemporal bias; which makes it difficult to rely on them. Further, the losses due to flood in a region not only depend on rainfall severity but also on topography, socioeconomic conditions and exposure of the region to floods. Here, we introduce a new concept of spatial flood risk mapping and forecasting at weather to medium range based on forecasted hazard, embedded with vulnerability (topographic and socioeconomic) and exposure information. We define hazard as the probability of extreme rainfall event during upcoming days given an available weather forecast for the same days. As hindcast is used for computation of probabilities, hazard contains prior information about the false alarm, hit rate and spatiotemporal bias of the forecast. Vulnerability is calculated by averaging the topographic and socioeconomic indicators, and exposure is calculated using a land use land cover map. Topographic vulnerability is computed with digital elevation model using Height Above the Nearest Drainage method, and Data Envelopment Analysis is performed to derive the socioeconomic vulnerability based on the demographic census data. For a specific region and a specific event, the relative flood risk maps are generated at an administrative level (e.g., district, subdistrict or village level for India) and the high-risk areas can be identified from those maps for mitigation. The methodology is demonstrated for a very recent extremely severe flood event that happened in Kerala, India in August 2018. It is evident from the results that the high-risk areas forecasted well in advance (as high lead time as 15 days) match fairly well with the areas, which suffered maximum losses because of direct flood.

Introduction

Climate change has caused a considerable impact to the global water cycle which lead to changes in seasonal patterns as well as an increase in the frequency of extreme rainfall events (Oki and Kanae, 2006). Intergovernmental Panel on Climate Change (IPCC, 2012) reports that this increase in extreme rainfall is statistically significant in many parts of the globe. It also states with medium confidence, that these changes in extremes are attributed to anthropogenic influences. In the Indian sub-continent as well, both empirical methods and model projections have shown an increase in the frequency and magnitude of extreme rainfall events (Goswami et al., 2006; Preethi et al., 2017; Roxy et al., 2017). A decrease in the total rainfall and intensification of extremes in the tropics for 21st century has been projected in the IPCC reports (Seneviratne et al., 2012). Irrespective of climate change, the importance of forecast and understanding its uncertainty have a very high societal relevance. Increasing extremes in a changing climate further increases its importance. Extreme rainfall events not only affect our day to day lives but also result in floods and landslides which cause huge loss of lives and property (Dottori et al., 2018; Fowler et al., 2010). Every year a billion people are affected and thousands die because of these extreme events (IFRC/RCS, 2011).

According to the National Disaster Management Authority (NDMA, 2008), floods have become a cause of concern because of an increasing trend in flood related losses in India. Some examples of such extreme events include the heavy rainfall event in Mumbai, India on 26th July 2005, which recorded 944 mm rainfall in 24 h (Jenamani et al., 2006). The rainfall and resulting flood caused the death of almost 1000 people and an economic loss of about US$100 million (Kumar et al., 2008). Uttarakhand in India received almost 340 mm rainfall in a day (375% above daily mean) on 17th June 2013, which resulted in severe flash floods (Dube et al., 2014). Chennai city experienced a terrible flood during November-December 2015, which caused at least 400 deaths, economic loss of US$ 1120 million (Narasimhan et al., 2016; Seenirajan et al., 2017).

A significant amount of research has been conducted in the past decades to understand these events. A large proportion of these events could not have been predicted accurately and hence resulted in devastation (Coumou and Rahmstorf, 2012). The commonly used alert generation and warning system for flood are mostly based on the rainfall forecast. Hence, an accurate prediction of extreme rainfall at an administrative level is very important for the stakeholders and decision-makers. However, the complex multiscale atmospheric processes responsible for the occurrence of any extreme rainfall event and their inherent variability makes them difficult to predict (Fritsch et al., 1998). Coarse resolution dynamical models often fail to predict these extreme rainfall events with accuracy as these have high false alarm, low hit rate and spatio-temporal biases (Březková et al., 2010; Khaladkar et al., 2007; Selvam, 2011; Shastri et al., 2017). Prediction skills can be improved using regional models at a high spatiotemporal resolution, but these simulations are computationally very intensive thus are difficult to perform on real-time (Dodla and Ratna, 2010).

Very heavy rainfall in a short time span often results in floods when it exceeds the ground absorption capacity and the runoff exceeds the capacity of river system (Neuendorf et al., 2005). This makes mitigation planning difficult. Flood risk associated with these events are difficult to predict as it not only depends on the complexity of processes related to extreme rainfall but also on the interaction between these events and the geography, population, infrastructure, the preparedness of the region (Balk et al., 2012).

Flood risk can be defined as a product of hazard, vulnerability and exposure (IPCC, 2012; Kron, 2005; Karmakar et al., 2010), which is used for climatological projections. In the risk framework for climate applications, hazard is defined in IPCC (2012) as "The potential occurrence of a natural or human induced physical event that may cause loss of life, injury, or other health impacts, as well as damage and loss to property, infrastructure, livelihoods, service provision, and environmental resources". Calculating flood risk also requires vulnerability, which works as a proxy of human-environment relationship (Turner et al., 2003). IPCC (2012) defines vulnerability as the “propensity or predisposition to be adversely affected”. In order to get the overall vulnerability of a region, various classes of vulnerability are combined together (Karmakar et al., 2010). The exposure (IPCC, 2012) component of risk is defined as “the presence (location) of people, livelihoods, environmental services and resources, infrastructure, or economic, social, or cultural assets in places that could be adversely affected by physical events and which, thereby, are subject to potential future harm, loss, or damage”. In the changing climate, along with an increase in the amount and frequency of extreme rainfall events, the exposure of humans to flood is also increasing (Hirabayashi et al., 2013), thus it is considered for flood risk quantification. However, this approach of defining climatological flood risk cannot be used for event specific flood mitigation.

Here, we propose a new methodology to use this concept of flood risk at weather to medium range scale to generate event specific risk maps. Traditionally the concept of hazard is used at a climate scale, where it is defined as the probability of extreme rainfall above a specific threshold and computed from the long-term data. Here we have introduced a concept of hazard, which is computed at weather to medium range in the forecast system. Hazard is computed as the probability of occurrence of an extreme rainfall in any grid, given a forecast value. Flood risk maps are generated by combining hazard with socioeconomic and topographic vulnerability and, exposure of a region. To demonstrate and evaluate this methodology, it is applied to the flood event in Kerala, India, which occurred during August 2018. The following section discusses the case study and the data used. The limitations of the state of art weather forecast system is discussed in the Section 3. Sections 4 and 5 contains the methodology and results respectively and the paper is summarised in the last section.