Tropical Cyclone Jasper: FloodMapp ForeCast's Performance Capability

Tropical Cyclone Jasper (TC Jasper) kicked off Australia’s 2023-2024 cyclone season when it formed on December 5, 2023, west of the Solomon Islands. TC Jasper moved southwest toward mainland Australia, strengthening into a Category 4 system on the Australian Tropical Cyclone Intensity Scale, with maximum wind speeds of 225 kilometers per hour (kph) (or 140 miles per hour [mph] on December 7. By the time TC Jasper made landfall about 80 kilometers (50 miles) north of Port Douglas on the northern tropical coast of the State of Queensland on December 13, it had weakened into a Category 2 system, with maximum wind speeds of 115 kph (70 mph). The system stalled over land just north of the town of Cairns for about one week after making landfall, dumping extreme amounts of precipitation in the area - about 600 millimeters (2 feet) in the town of Cairns and over 2000 millimeters (6.5 feet) in more isolated locations over 7 days (Figure 1). The slow and continuous rainfall quickly filled rivers around the region to levels rarely seen before, with gauges along the Barron and Daintree rivers each breaking water level height records over 50 and 100 years old, respectively. Water level heights in the Barron River at the town of Mareeba peaked at 11.5 meters (38 feet), surpassing the major flood stage category of 9 meters (29.5 feet), and the previous maximum record of the Daintree River in Daintree Village was smashed by about 2.5 meters (8.2 feet), a record set over 100 years ago. In Cairns, the international airport was closed for multiple days and a wake of destruction tallied up damage costs of about AUD 200 million (USD 130 million). The cyclone-driven flooding caused substantial damage to both property and life: over 8,000 insurance claims were submitted for houses, motor vehicles, and various items while one person was tragically reported as missing and dead. 

Figure 1. 7-day rainfall totals for TC Jasper from December 13-19, 2023.

Background on FloodMapp's ForeCast Product

FloodMapp’s ForeCast is produced via a proprietary hydrologic model that leverages machine learning, historical data, and physical catchment characteristics to predict river heights 6-72 hours before a flood happens. The best available terrain data, real-time river height data, catchment characteristics, and rainfall predictions are ingested to predict and map flood inundation extent and depth. Forecasted flood maps provide extra lead time and show localized flood impacts, updating every hour to reflect changes in precipitation forecasts and river heights.

ForeCast is delivered as a live mapping feed via a web map service (WMS) or Web feature service (WFS) to integrate directly within existing GIS systems. This facilitates live, dynamic situational awareness and a common operating platform to support rapid communication and response across agencies and jurisdictions.

Emergency managers and asset owners use FloodMapp ForeCast to proactively identify asset and community flood impacts at the neighborhood scale to support emergency planning, exercises, and training. This product enables client-driven analytics to identify which roads, bridges, or other relevant assets may be impacted to create a risk profile and support evidence-based decision-making for operational planning and effective loss prevention.

ForeCast Performance During TC Jasper

FloodMapp provided emergency management agencies in Queensland with forecast flood intelligence before and during TC Jasper. The scope of response by emergency managers, in terms of geography, time, and severity, widely varied – posing extreme challenges to logistics and resources. Flood intelligence provided by FloodMapp automated and centralized critical situational awareness to expedite decision-making for emergency responders.

Two specific case studies are examined in detail, below, to explore the performance of FloodMapp ForeCast during TC Jasper. The first area of focus is a rural village along the Mossman River, while the second is the town of Mareeba along the Barron River, upstream of Cairns.

Mossman River

Water spilled out of the banks of the Mossman River on December 13, 2023, rising an outstanding 6 meters (20 feet) in 10 hours to a height of 8.5 meters (28 feet), causing several dozen homes in nearby areas to be flooded. The rapid rise in water levels was captured by a nearby gauge shown in Figure 2. Flooded roadways and houses were captured by an on-the-ground photographer, shown in Figure 3. Detailed flood maps from 2017 identify this neighborhood outside of the 1 in 500-year floodplain, giving it a likelihood of flooding each year less than 0.2%. Local news crews broadcasted resident grievances about the lack of flood warnings while emergency responders had to evacuate a dozen people during the night. Residents said they had never witnessed flooding on this scale before, with extensive damage to cars, bridges, and houses occurring in and around the village of Mossman.

Figure 2. Hydrograph of the Mossman River at Mossman during TC Jasper

Figure 3. Observed flooding around Mossman

The ForeCast model extent predicted severe flood impacts along Mossman Street, resulting in several dozen homes being inundated - shown in pink in Figure 4. The houses shown in Figure 3 were geolocated and used as “ground truth” data to compare to the ForeCast results, indicated by the red and blue stars in both Figure 3 and Figure 4. This validation of ForeCast model results to on-the-ground photos provides evidence of agreement between the model and what actually happened during TC Jasper, with both showing specific homes and Mossman Street as flooded.

Figure 4. Forecasted flooding around Mossman

FloodMapp’s ForeCast extent can be generated up to 72 hours before flooding occurs but increases in accuracy approaching a 6-hour timeframe due to compounding errors from precipitation forecasts. This 6-hour lead time offers emergency managers the opportunity to evacuate residents, close roads, and set up temporary flood diversion structures. The accuracy of predictive models is crucial because end-users need to be confident in results and understand implicit uncertainty to make the best decisions possible with available information.

Barron River at Mareeba

The Barron River burst its banks in the town of Mareeba during the afternoon of December 17, flooding neighborhoods and turning streets into rivers. Roads and bridges were impacted around the town and one bridge was washed away completely, leaving residents entirely encircled by water and isolated for several days. Flooded streets and neighborhoods were captured by on-the-ground photographers (Figure 5) , validating the accuracy of our ForeCast results.

Figure 5. Observed flooding in Moreeba

The Barron River rose rapidly on December 17, its water levels surging by about 9 meters (30 feet) in 24 hours. The quickest rise (3 meters, or 10 feet) occurred 6 hours before the river reached its peak height of 11.5 meters (38 feet), shown in Figure 6. The rapid rise in water levels caused flooding in nearby neighborhoods, captured by an on-the-ground photographer in Figure 5. The extent of observed flood waters was compared to the predicted water extent by ForeCast in Figure 7 which shows a slight underprediction of flood waters. Modeling results do not always exactly match observed data when flood forecasting due to uncertainties in the model inputs, such as the precipitation forecast, which propagate through the model and may produce slightly inaccurate results. Despite the inherent uncertainty of predictive data, it can still provide valuable predictions on flood impacts within a range of specified confidence levels.  If input data confidence is defined, emergency managers can account for the uncertainty in their response, such as by adding a “buffer zone” around the provided flood intelligence to make depth- and area-defined decisions. Although the flooded street and houses of ForeCast did not exactly align with the observed results, the prediction still informed emergency responders that larger-scale flooding was unlikely and that the type of response necessary is likely to be limited to a vehicle response, rather than a swiftwater or airlift response. To communicate modeling and data uncertainty, FloodMapp provides confidence ratings of each model and product to enable users to understand the accuracy of the input data used to generate flood extents.

Figure 6. Hydrograph of the Barron River at Mareeba during TC Jasper

Figure 7. ForeCasted and observed flooding in Mareeba

Conclusion

Emergency responders in Queensland utilized FloodMapp’s ForeCast data to inform personnel of flooding severity across a wide geographical extent and also centralize the accuracy of disparate data quality critical to flood response. Flooding examples along the Mossman and Barron Rivers were examined to highlight how quickly flooding can occur and how predictive flood intelligence can improve emergency response efforts. ForeCast results will always vary due to uncertainty in data and unique storm conditions, as showcased in the case studies, but they can still provide valuable information about the scale of future flooding, the impact to people, property and critical infrastructure, and the level of effort required to respond. FloodMapp’s operational impact-based forecasting supported emergency managers to enhance public safety and streamline their emergency response.