12.1 Introduction

Of the various types of weather phenomena, thunderstorms produce some of the most immediate and impactful hazards—damaging winds and hail, frequent lightning, and intense rainfall. Resilience to high-impact weather can be attained through investment in several key areas: proper infrastructure; effective emergency management; public education; and well-informed weather forecasting services. Unfortunately, some of the most intense thunderstorms occur in developing nations that have yet to build sufficient resilience to such weather hazards. Although there is a perceived cost associated with establishing National Hydrological and Meteorological Services (NHMS), investment in these services can boost the socioeconomic well-being of a developing nation (e.g., Nepal) by a factor of 10 (Perrels 2011). Early warning services are identified as providing the greatest and most immediate socioeconomic benefit amongst other types of disaster risk reduction strategies (Hallegatte 2012).

The Hindu-Kush Himalaya (HKH) region is host to some of the most intense thunderstorms on Earth (Zipser et al. 2006), primarily during the pre-monsoon season (Cecil and Blankenship 2012), and is routinely plagued by the hazards they pose (e.g., Das et al. 2014; Bikos et al. 2015; NIRAPAD 2018). Their impact is especially felt in Bangladesh where many documented tornado events have caused hundreds of fatalities, including over 700 in one event in April of 1996, and at least 1300 in a single event on 26 April 1989 (Bikos et al. 2015). A recent notable thunderstorm event in Bangladesh occurred on 30 March 2018, causing 282 casualties (at least six of them fatal), damaging 943 houses, and destroying large swaths of crops as a result of a tornado and other strong winds and large hailstones (222 of the casualties solely due to hailstones; NIRAPAD 2018). Furthermore, lightning vulnerability has become more prominent in Bangladesh with more than 250 fatalities recorded per year from 2010 to 2016 (Dewan et al. 2017). And, recent storm events on 29–30 April 2018 and 9–10 May 2018 produced widespread frequent lightning that resulted in 74 fatalities in Bangladesh.

Even though Nepal is less densely populated than Bangladesh, it has also suffered damages due to similar weather hazards (Fig. 12.1). On 31 March 2019, the first-ever documented tornado occurred in southern Nepal, leading to 30 deaths, 1150 casualties, and destroying 2890 homes (Mallapaty 2019; Shrestha et al. 2019). [It is likely that tornadoes occur in Nepal on a more regular basis during the pre-monsoon season but have historically gone undocumented due to the lower population density and lack of effective communication tools]. More prominent in Nepal have been hailstorms, such as the event in 2006 that caused crop and livestock damages which exacerbated local food shortages (ACTI 2006).

Fig. 12.1
figure 1

Weather hazard is a major issue in the mountains during monsoon seasons. Photo by Santosh Raj Pathak

Full size image

Although many NHMS in developing regions such as the HKH have access to weather forecasting tools, they currently lack the information required to support early warning services and timely disaster response that can help mitigate the impacts of hazards that are produced by frequent and intense thunderstorms. As these services undergo modernization efforts, which include enhancements in observational infrastructure (e.g., Rigaud 2015; World Bank Group-PPCR 2015; World Bank 2017), investments in new computational tools that will support enhanced early warning services are also necessary.

12.2 Forecasting High-Impact Weather

Numerical weather prediction (NWP) is the primary tool used by NHMS to produce weather forecasts. In particular, the Weather Research and Forecasting (WRF) model (Powers et al. 2017) has become the most widely used regional forecast and simulation tool in the weather community (e.g., Kain et al. 2006; Tao et al. 2016; Schwartz et al. 2019), including by NHMS in the HKH region (e.g. Ahasan et al. 2014; Kotal et al. 2015). Operationally, the WRF model is typically configured to provide hourly regional weather forecast guidance up to three or four days. Although WRF is a powerful tool, it still cannot replace a sufficiently dense observational network when it comes to monitoring and nowcasting (i.e., 0–1 h forecasts) rapidly unfolding weather such as thunderstorms. However, in countries such as Nepal such observations are often limited due to the terrain or lack of investment in Doppler weather radar networks that are critical to assessing such high-impact storms. Thus, it is critical that NHMS possess highly relevant and skillful NWP to gain proper situational awareness of potential weather hazards during the forecast period.

12.2.1 Challenges of Forecasting Thunderstorms

Regional models such as WRF also require input from the global NWP in order to represent the initial state and evolution of the atmosphere on a larger scale that can orchestrate high-impact weather events through their interaction with land and ocean. Hence, it is important to properly represent the initial state of the land–atmosphere–ocean system for regional modeling applications, which entails bringing observations into the NWP framework via data assimilation methods. Since in situ observations are rather limited across the globe, NWP in such regions often incorporates the geophysical variables retrieved via satellite-based remote sensing into data assimilation algorithms using radiative transfer models to relate satellite observations to NWP model state variables (Barker et al. 2012; Geer et al. 2017). However, satellite retrievals are not straightforward and introduce additional uncertainty in the model initial conditions. Furthermore, there are a variety of data assimilation techniques, some of which can provide significant improvements to the forecast (e.g., Huang et al. 2009), but they can be rather complex and computationally expensive, especially for real-time applications.

Proper forecast of thunderstorm hazards requires the use of a convection-permitting model (i.e., its horizontal grid spacing must be on the scale of 4–5 km or finer). This model must adequately depict all physical processes and interactions between the land/ocean and atmosphere that lead to cloud and precipitation development and decay. Since explicitly simulating all of these interactions and processes is not feasible, they must be parameterized (i.e., approximated) within the model (Stensrud 2007). Convection-permitting (also referred to as cloud-resolving) models often include parameterizations of sub-grid-scale motions (e.g., turbulent motions within the boundary layer), mixed-phase microphysics, and radiative processes (Guichard and Couvreux 2017). Decades of research have gone into developing cloud-resolving models and parameterizations of these physical processes (Tao 2007). As a result, numerous parameterization schemes have been incorporated into the Advanced Research WRF for simulating thunderstorms (UCAR 2020). This is promising for improving the skill of model forecasts, but it requires choosing the appropriate ones for the application at hand. There have been a multitude of sensitivity tests of parameterization schemes (e.g., Milbrandt and Yau 2006; Cohen et al. 2015; Fan et al. 2017), but no single combination of parameterizations works best for all thunderstorm hazards, weather regimes, and geographical regions.

12.2.2 Ensemble-Based, Convection-Allowing NWP

In light of this wide variety of potential configurations, model uncertainty, and differing assumptions within individual parameterizations, a solution has been to perform ensemble NWP by running a suite of different model runs. Each model run may be configured with a different combination of parameterizations and/or initial conditions. The main idea of ensemble NWP is to capture the range of possible solutions (e.g., Fig. 12.2), and thereby enable probabilistic-based forecast information that facilitates a more well-informed decision. To improve thunderstorm forecasts in the United States, regional WRF-based convection-allowing models (e.g., Kain et al. 2006, 2010) have been used in an ensemble configuration (e.g., Clark et al. 2012; Schwartz et al. 2019) to support severe weather and excessive precipitation forecast experiments in the National Oceanic and Atmospheric Administration (NOAA) annual Hazardous Weather Testbed’s Experimental Forecast Program (Clark et al. 2012). This particular forecast system included over two dozen different model simulations that produced hourly forecasts out to 30 h at convection-allowing grid spacings (1–4 km) spanning the continental U.S. Similar experimental ensemble forecast demonstrations have been undertaken by the National Center for Atmospheric Research (NCAR) and have received much fanfare from the weather community (Schwartz et al. 2019), but their relatively large computational burden hinders their implementation by operational agencies with relatively limited resources.

Fig. 12.2
figure 2

An example of the range of different weather scenarios produced at the same forecast hour. Shown is the composite reflectivity, a metric for gauging thunderstorm intensity, which was produced by the HIWAT ensemble WRF runs on 25 April 2019. Traditional approaches to operational NWP only provide a solution based on a single model run (e.g., top-left panel), which may not be the correct one given the uncertainty in the model-based representation of the atmosphere and its evolution. Hence, ensemble-based NWP provides a variety of solutions that can be combined to obtain the probability of a weather hazard at any point in the forecast cycle

Full size image

12.3 A High-Impact Weather Service for the HKH Region

In response to the need to build the resilience of the HKH region to high-impact weather, NASA SERVIR developed a new tool for use by NHMS in the region. The High-Impact Weather Assessment Toolkit (HIWAT) is a service that provides probabilistic-based thunderstorm hazard forecast guidance to NHMS forecasters and other weather-sensitive decision makers. HIWAT uses a convection-allowing ensemble modeling system similar to that developed for experimental hazardous forecast demonstrations in the U.S. (e.g., Clark et al. 2012; Schwartz et al. 2019), but it is tailored for NHMS in the HKH region.

12.3.1 Model Configuration

The WRF model is configured in HIWAT using an outer domain with 12-km grid spacing centered on South Asia, a 4-km inner domain that includes Nepal, Bangladesh, Bhutan, and north-eastern India (Fig. 12.3), and 42 terrain-following vertical levels ranging from the ground to a barometric altitude of 20 hPa.

Fig. 12.3
figure 3

Map of HIWAT domains over Asia

Full size image

The larger 12-km domain is primarily used for regional synoptic-scale situational awareness and, more importantly, to downscale the boundary and initial conditions obtained from the larger global-scale model to the higher-resolution nested grid. The 12-km resolution outer domain parameterizes convection using the Kain-Fritsch scheme (Kain 2004). On the 4-km domain, the cumulus parameterization is turned off to explicitly simulate convective storms (i.e., convection-allowing). Although a higher resolution would likely result in a more accurate representation of the convective processes and subsequent thunderstorm intensity (e.g., Potvin and Flora 2015), the resolution used in HIWAT is a trade-off between reducing the computational burden while retaining the capability to forecast the hazards posed by mesoscale weather systems (e.g., Weisman et al. 1997; Bryan and Morrison 2012).

For simulating thunderstorms, the representation of the planetary boundary layer (PBL) processes that lead to convective initiation and the microphysical processes that result in precipitation development are key. Hence, the HIWAT ensemble forecasting system has diversity in the parameterization of the PBL and microphysics (Table 12.1). Each of the microphysical schemes account for riming (i.e., formation of graupel/hail)—a critical process in the development of deep convective storms and lightning (see review by Williams 2001). To include additional diversity and spread in the HIWAT ensemble, the initial conditions are varied. The operational run of the Global Forecast System (GFS; Zhou et al. 2019) and members from the Global Ensemble Forecast System (GEFS; e.g., Guan et al. 2015; Zhou et al. 2017), which are run every 6 h by the U.S. National Centers for Environmental Prediction (NCEP) Environmental Modeling Center’s (EMC) and freely obtained from them, are used to initialize the WRF model employed by HIWAT. Also, HIWAT ingests a 2-km resolution, northern-hemispheric sea surface temperature (SST) composite product derived from NASA’s MODIS and VIIRS satellite measurements (Zavodsky et al. 2017), which improves upon the considerably coarser SST analysis in the GFS/GEFS models. To execute the ensemble WRF model runs, HIWAT employs the Unified Environmental Modeling System (UEMS; Rozumalski 2019), which greatly simplifies the complex workflow involved in NWP by managing data acquisition, initialization, pre- and post-processing, and generating derived fields.

Table 12.1 Ensemble configuration of the 4-km resolution domain of the WRF-based probabilistic forecasting system used in HIWAT-HKH
Full size table

12.3.2 Probabilistic Forecast Products

The primary guidance products generated by HIWAT are probabilistic forecasts of the thunderstorm hazards associated with tornadoes, damaging wind and hail, frequent lightning, and intense rainfall. These are provided at each hour of the forecast cycle and temporally composited into daily summary probability maps that can readily facilitate the convective outlooks of thunderstorm hazards (Fig. 12.4), which is a capability recommended in the report on the Bara-Parsa (Nepal) tornado of 31 March 2019 (Shrestha et al. 2019). Furthermore, products such as daily summaries provide a weather-sensitive decision maker the information in a concise manner to quickly assess future thunderstorm impact(s). Because of uncertainties in model forecast accuracy, combined with convective features that typically have relatively small footprints (e.g., Kain et al. 2010), a 20-km spatial neighborhood (i.e., grid points within 20 km) is first applied to the individual ensemble member thunderstorm hazard fields prior to computing the probabilities. Additionally, a Gaussian smoother is used to produce smoother and more visually appealing probability maps. Such neighborhood approaches can result in more skillful probabilistic forecasts (Schwartz and Sobash 2017).

Fig. 12.4
figure 4

HIWAT probabilistic forecast of thunderstorm-related weather hazards for the Day-2 time period. Intense storms are defined as composite reflectivity exceeding 50 dBZ. Damaging winds are defined as the maximum 10-m wind speed in the previous hour (Kain et al. 2010) exceeding 40 kts. Rotating storms are defined as the maximum 2–5 km updraft helicity (Kain et al. 2008) in the previous hour exceeding 200 m2s−2. Intense rainfall is defined as the 3-hour accumulated precipitation exceeding 75 mm. Damaging hail is defined as the maximum column-integrated graupel in the previous hour (Kain et al. 2010) exceeding 30 kg m−2. Frequent lightning is defined as one lightning flash per minute produced by the diagnostic WRF Lightning Forecast Algorithm (McCaul et al. 2009, 2020)

Full size image

An example of the Day-2 summary forecast from HIWAT (i.e., a 25–48 h forecast period) is shown in Fig. 12.4. This gives the probability of the indicated weather hazard within 20 km of a point at any time of the Day-2 forecast period. Using composite reflectivity exceeding 50 dBz as a proxy for storms with significant precipitation rates (i.e., intense storms), we see there is greater than a 90% probability of intense storms occurring from southeastern Nepal across northern and eastern Bangladesh into southern Bhutan and Northeast India. A Day-2 convective outlook would highlight this region. The next question is the likelihood of specific hazards within this region. We see intense rainfall that might lead to flash-flooding should not be a major concern, but there are greater probabilities of damaging winds, hail, rotating storms, and frequent lightning across this region of intense storms. Furthermore, the higher probability of rotating storms in northern Bangladesh being aligned with the relatively enhanced probability of damaging wind and hail suggest that tornadoes are more probable in northern Bangladesh than in eastern Bangladesh. This analysis of Fig. 12.4 is akin to the forecast process used by NOAA’s Storm Prediction Center in the U.S. to determine future convective hazards (Jirak et al. 2014).

12.4 HIWAT Forecast Demonstrations

Two forecast demonstrations of HIWAT took place during the pre-monsoon (March-May) and monsoon seasons (June–August) of 2018 and 2019, with the Bangladesh Meteorological Department (BMD) and Nepal’s Department of Hydrology and Meteorology (DHM). For these demonstrations, HIWAT used the public release of WRF version 3.7.1 (included in UEMS version 15.99.1), while NASA’s SERVIR program provided the virtual computing cluster needed to run it. This virtual cluster consisted of 13 nodes, each with 16 dual-core Intel Xenon 2.10 GHz processors and 128 GB of RAM. It produced 0–48 h HIWAT-HKH forecast products within 6 h of HIWAT initialization. This resulted in a probabilistic forecast guidance covering two diurnal maximums in the convective activity in the HKH region (e.g., Romatschke et al. 2010; Mäkelä et al. 2014; Dewan et al. 2018).

During these demonstrations, HIWAT trainings were held in Kathmandu and Dhaka to familiarize the operational forecasters at DHM and BMD with HIWAT and its probabilistic-based thunderstorm hazard forecasting products. The products were made available to forecasters via two web-based display tools. An image viewer, which was contributed and hosted by NASA’s Short-term Prediction Research and Transition Center (SPoRT; Jedlovec 2013), provided quick looks and animations of all current and archived HIWAT forecast products. Also, select HIWAT products were packaged (i.e., separate from the full model output suite) into a netCDF format for an interactive (i.e., pan, zoom, query) web-based application, which was produced within the Tethys platform (Nelson et al. 2019), to enable efficient interrogation of the forecasts within a web-mapping service. The Tethys-based HIWAT Model Viewer App (Fig. 12.5) provides users with the capability to quickly determine where, when, and how likely a thunderstorm hazard may occur. It also has the capability to provide the spatiotemporal statistics of a forecast variable (e.g., as to when maximum rainfall will occur within the selected area). Figure 12.5 shows the HIWAT forecast run from 29 March 2018 which indicates that a high-impact weather event is possible across the northern half of Bangladesh. The time-series plots of the forecast across this area show a greater than 60% chance of rotating thunderstorms, with a 45–55% chance of damaging winds and hail at the selected points between 11:00 and 15:00 UTC on 30 March 2018.

Fig. 12.5
figure 5

The Tethys-based HIWAT model viewer app

Full size image

12.4.1 Hailstorm: 30 March 2018

The severe thunderstorms that occurred across the eastern HKH on 30 March 2018 caused significant damages in Nepal, Bangladesh, and Northeast India (Fig. 12.6). In Bangladesh alone, over 274 casualties, numerous acres of crops, and several hundred houses were damaged, mostly due to large hail but also due to strong winds, and  a tornado in Nepal’s south-central districts of Bara and Parsa (Shrestha et al. 2019). The Day-1 probabilistic forecasts from the HIWAT-HKH runs, initialized at 18:00 UTC on 29 March 2018, are shown in Fig. 12.6a–c. They clearly indicate more than an 80% chance of hailstorms in northern and central Bangladesh and Meghalaya in north-east India (Fig. 12.6a). The forecast hotspots of frequent lightning overlap with this region (Fig. 12.6b), but they also extend further east, including Assam in Northeast India and Sylhet division in northeastern Bangladesh. It also shows a greater than 80–90% chance of 10-m AGL winds exceeding 50 kts (~93 km h−1) in Sylhet (Fig. 12.6c).

Fig. 12.6
figure 6

Performance of the HIWAT forecasts for the 30 March 2018 high impact weather event. a–c The 29 March 2018 forecasts from HIWAT-HKH indicate that thunderstorm-related hazards are highly likely to occur across parts of the eastern HKH on 30 March 2018. d, e Precipitation and lightning observations from the GPM core satellite radiometer and Lightning Imaging Sensor onboard the International Space Station, respectively, during overpasses of the eastern HKH region on 30 March 2018. f Location of damaging wind, hail, and lightning due to a high impact weather event in the eastern HKH region on 30 March 2018

Full size image

Direct, near real-time observations of storm intensity are sparse in the rural areas of the HKH region; hence, HIWAT also consists of a satellite-based observational component to facilitate assessment of its forecasts. On 30 March 2018, at around 08:11 UTC, there was an overpass of the HKH region by the GPM Microwave Imager (GMI) onboard the GPM core satellite (Skofronik-Jackson et al. 2017). Passive microwave observations provided a measure of storm intensity through their sensitivity to the upwelling radiation at 37 GHz being scattered by precipitating ice such as hail (Spencer et al. 1987; Cecil 2009). Applying the empirical fits of Bang and Cecil (2019) that relate the polarization-corrected brightness temperature at 37 GHz to the probability of hail observed at the ground, the GMI observations indicated three intense storms over the region with a greater than 75–95% chance of producing damaging hail at the ground, especially in north-western Bangladesh and west of Cherrapunjee, India (Fig. 12.6d). Similar damaging hail probabilities had been found for these storms with measurements from the Advanced Microwave Scanning Radiometer 2 (AMSR2) during its overpass about 30 min earlier (not shown). These patterns aligned with the northern-most local maxima of the hail probability forecast by HIWAT (Fig. 12.6a). Also, observations collected with the Lightning Imaging Sensor onboard the International Space Station (ISS) (Blakeslee et al. 2020) during an overpass at 10:30 UTC depict three individual thunderstorms with relatively higher lightning flash activity over central Bangladesh where the HIWAT forecasts suggested nearly a 100% chance of frequent lightning (Fig. 12.6e). Finally, the reports of thunderstorm-related damage, although likely incomplete and affected by subjective reporting, largely corroborated the HIWAT Day-1 forecasts that strongly suggested a high-impact weather event unfolding from eastern Nepal to Northeast India and central Bangladesh (Fig. 12.6f).

12.4.2 Lightning

Lightning-related casualties have increased in Bangladesh in recent years (Dewan et al. 2017; Holle et al. 2018), so much so that the government declared lightning as a national disaster in 2016. During the 2018 pre-monsoon season, there were 275 reported casualties (215 of them fatal) due to lightning in Bangladesh, with 181 of these occurring during a two-week period in late April and early May (Fig. 12.7a). Although agricultural practices in Bangladesh during the annual peak lightning months of April and May contribute to the vulnerabilities (Dewan et al. 2017; Holle et al. 2019), a lack of capacity to forecast lightning is also a factor.

Fig. 12.7
figure 7

Impact of lightning activity on Bangladesh and select forecasts from HIWAT-HKH during April and May 2018. a Lightning-related casualties reported in Bangladesh between April and May 2018 (Source NIRAPAD 2018). b–d The mean 24-hour (i.e., Day-1) forecast flash rate density from the HIWAT-HKH ensemble for three different periods centered on the peak number of lightning-related casualties in panel a

Full size image

One of the tools used in HIWAT is the Lightning Forecast Algorithm (LFA), which uses the convection-allowing WRF-simulated fields important to the storm electrification process as inputs, and computes a calibrated total lightning (both cloud-to-ground and cloud flashes) flash rate density (McCaul et al. 2009, 2020). This total lightning prognostic field is produced for each of the HIWAT ensemble members to produce the hourly and daily probabilistic lightning forecasts across the model domain (e.g., Figs. 12.5 and 12.6). Figure 12.7b–d provides a daily summary of the HIWAT lightning forecasts from 17 April to 22 May 2018. The amount of lightning being forecast increases from mid-April to early May, then decreases by mid-May, especially across northeastern Bangladesh and Northeast India. This trend compares very well with the trend in lightning casualties reported in Bangladesh during this time period (Fig. 12.7a). Also, Holle et al. (2019) found that the farming districts in northeastern Bangladesh are especially susceptible to lighting-related fatalities during April and May. Lightning safety education, especially related to agricultural practices (e.g., best times of day to tend or harvest crops), can greatly help reduce the adverse impacts of thunderstorms in the region. However, it is not every day that thunderstorms will occur in the same location at the same time, and it is not practical to halt work every afternoon during the peak harvesting months (April and May) of Boro rice, potato, and wheat. Hence, access to more informative weather forecasts such as those provided by HIWAT can facilitate better planning of day-to-day activities.

12.4.3 High-Intensity Rainfall Forecasting

Another reason for using ensemble NWP is to enable the forecasting of flash floods in the regions affected by localized, intense rainfall associated with thunderstorms (Fig. 12.1). In an ensemble system such as HIWAT, hydrologists can access the information provided by the system’s multiple precipitation forecasts in such a manner to concisely convey the expected outcome with a higher degree of confidence. This can be in the form of probabilities of rainfall exceeding some threshold in a given basin, which may not be readily known, or in the form of an envelope of streamflow forecasts for each basin. Most streamflow models require a single deterministic input of precipitation instead of a probability of precipitation. However, simply averaging the precipitation forecasts of each ensemble member will simultaneously over-predict the areal coverage of rainfall and reduce the amplitude of more intense rainfall events that are often the cause of the flash flooding in the smaller basins and urban areas. Hence, HIWAT employs a statistical technique known as the probability matched mean (PMM; Clark 2017) to provide a single precipitation forecast that represents a “most probable solution” from all its ensemble forecast members. The PMM not only retains the higher-amplitude precipitation intensity from the individual ensemble members but also retains the more spatially accurate pattern of the ensemble mean, with the bias at any precipitation threshold being about the same as the average bias of the individual members (Ebert 2001; Clark 2017). The PMM precipitation product reaps the benefit of an ensemble forecast and also provides the accumulations required by hydrologic applications.

Although HIWAT-HKH is largely focused on severe storms that occur in the HKH region during the pre-monsoon season, the 2018 and 2019 demonstrations were expanded to the wet monsoon season (June –September) at the request of the end users (i.e., DHM and BMD) and primarily for the purpose of flash-flood forecasting. A few events from 2018 to 2019 are highlighted in Fig. 12.8 to demonstrate the utility of the HIWAT PMM forecasts of precipitation during heavy rainfall episodes. The top panel of Fig. 12.8 shows the PMM-based forecast of precipitation exceeding 200 mm in several highly localized areas within the Himalayan foothills and mountains. Flash flooding was reported in several small basins in western Nepal between 15 and 16 August 2018, including a particularly devastating one along the Jhupra River in Nalgad of Jajarkot district (The Kathmandu Post 2018). The bottom panels of Fig. 12.8 show other heavy rainfall events captured by HIWAT-HKH. On 25 July 2018, the HIWAT PMM-based forecast indicated precipitation accumulation would exceed 300 mm within a 24-h period in southeastern Bangladesh, near Cox’s Bazar (Fig. 12.8b). The BMD’s automated weather station in Cox’s Bazar recorded a 24-h rainfall accumulation of 315 mm on 26 July 2018. The third example (Fig. 12.8c) is the 48-h PMM-based precipitation forecast of the tropical cyclone Fani, which developed in the Bay of Bengal and made its landfall southwest of Kolkata, India, on 2 May 2019. The spread of the 12-member ensemble runs indicated that after landfall, Fani would move north-east along the coastline and significantly weaken as it crossed into Bangladesh. Typically, cyclones bring a significant flooding threat to Bangladesh, but according to the HIWAT forecast Fani carried a low risk of flooding as the heaviest corridor of rainfall was forecast to stay west of Bangladesh and in the Northeast Indian state of Meghalaya. The 48-h forecast predicted over 250–300 mm of rainfall around Cherrapunjee in Northeast India (Fig. 12.8c), where an automated weather station recorded 276 mm of rainfall between 3 and 4 May 2018.

Fig. 12.8
figure 8

Examples of HIWAT-HKH ensemble precipitation forecasts using the PMM method. The circled regions indicate areas where extreme rainfall was observed within a 1–2 day period, as discussed in the text

Full size image

The HIWAT demonstrations during 2018 and 2019 also included a hydrologic component in which the PMM-based forecast precipitation accumulations were used as input to the model of Routing Application for Parallel computation of Discharge (RAPID; David et al. 2011) in order to produce streamflow forecasts for the HKH region (Nelson et al. 2019). The HIWAT forecasts were used in this system primarily for forecasting flash floods due to intense rainfall in the smaller basins that are often not captured by the coarser-resolution global weather forecast models.

12.5 Summary, Challenges, and Way Forward

A probabilistic weather forecasting system such as HIWAT demonstrates the importance of ensemble-based NWP in building resilience to high-impact weather in the HKH region. The wealth of information it provides can improve decision-making and enhance weather services. Extreme-weather forecasting is often plagued with uncertainty and cannot be captured nor effectively conveyed through a single weather model. However, a convection-allowing ensemble forecasting system depicts a range of possible scenarios of intense thunderstorm hazards and thereby can help convey the amount of confidence and/or uncertainty of these hazards in the forecast. In order to effectively carry this out, the ensemble has to be properly configured to address the weather phenomena, including the uncertainty in the physical understanding and representation of it. Additionally, the ensemble forecasts, which may include hundreds of potential visualization products, must be made easily digestible for the decision makers. Some of these products for assessing potential threats from severe weather events have been presented herein and others can be found in a review by Roberts et al. (2019).

A big challenge for NHMS, especially in the case of developing nations, is the computational resources that are required by the ensemble weather forecasting systems. The computational load depends upon the number of ensemble members, the number and geographical size of the modeling domains, their spatial resolution, and the designated number of forecast hours. The complexity of the forecast system (i.e., workflow) must also be considered. Additionally, reliable and high-speed internet access is needed to obtain the global models that initialize and provide boundary conditions to regional forecast systems such as HIWAT. Another challenge for NHMS lies in the novelty of the ensemble NWP, especially as a tool for thunderstorm forecasting. Also, continued education on advancements in meteorological forecasting techniques are needed since many meteorologists in developing nations have not received formal training on probabilistic-based weather forecasting.

The HIWAT service was designed with these challenges in mind. Although the HIWAT demonstrations consist of 12 separate WRF runs, HIWAT is fully customizable (in terms of domain size, number of ensemble members, and forecast hours) to accommodate porting the service to other regions. Also, and just as important, the myriad outputs from the individual ensemble members are combined into a handful of meaningful products relevant to addressing high-impact weather. These products are designed to be efficiently interpreted and are used to effectively convey the probability of thunderstorm hazards. HIWAT has demonstrated to NHMS in the HKH region the capability to build or enhance its capacity to provide services related to extreme-weather events. However, local NHMS may need to collaborate with scientific researchers to further customize the toolkit using thresholds that are based on more geographically-specific observations. The default thresholds used in the HIWAT demonstrations are primarily based on investigations focused on severe weather in the U.S. Furthermore, while information on familiar meteorological variables such as temperature, humidity, pressure, and wind are available from many regional observation networks operating in developing nations, more detailed information is needed to assess the storm prediction fields provided by a system such as HIWAT. Weather radars, especially those with Doppler and polarimetric capabilities, and lightning detection networks can provide critical observational evidence to evaluate the performance of this type of convection-allowing modeling system as well as enable additional refinement of the toolkit. This process of obtaining appropriate data, analyzing it, and using it to fine-tune the system to address local needs underscores the need for governments to establish a collaboration between their NHMS and the research community, either locally or abroad.