It was August 6, 2013, and the data had just been processed.
Dr Kristofer Shrestha, a research scientist, sat in his third-floor office in the environmental sciences and technology building at the Georgia Institute of Technology in Atlanta, Georgia, and opened the Indus river basin dashboard on his terminal. Two maps sprang up on the screen, both of them showing Pakistan along with some parts of its neighbouring countries.
The map on the left showed the country covered in a blue-and-green blot — evidence of a rainy weather system moving in from the east. The map on the right showed the Indus river system and its tributaries, with small blue dots to mark the location of each barrage and dam. The first map told Shrestha about the expected rainfall for that day and the second told him how much water was expected to flow through each blue dot on the same day.
Underneath the maps was a forecast slider, marked Day 1 to Day 10. As his cursor hovered over each forecast day on the slider, the maps changed colour, corresponding to the amount of rainfall expected on each day. On Day 8 and Day 9, the model he was operating showed heavy rainfall over the northern parts of the Indus. He quickly clicked on the link marked “Accumulated Precipitation”. The map changed colour showing how much water was expected to accumulate in different parts of Pakistan during those days. As he moved from Day 1 to Day 10, the map turned red, showing sharply rising levels of water accumulation across many parts of Punjab and Balochistan.
Shrestha might have been the first person in the world to see that Pakistan was just about to face a flood. After studying other data related to water flow forecasts in the rivers and water inflows at major dams, he wrote a short email to the principal investigator of the Indus river basin flood forecast project, alerting him that the model was showing “a high likelihood of elevated streamflows” 10 days down the road.
The principal investigator, Dr Peter Webster, has a quarter century of experience of working on predicting monsoonal floods in northern parts of the subcontinent. His office was down the hall from Shrestha’s. Webster opened the dashboard on his computer and, after a brief discussion with Shrestha, made a call to his contact in the World Bank to ask them to alert the Pakistani authorities that heavy rains and floods may be coming their way in 10 days’ time.
Here in Pakistan, his alert fell on deaf ears. “There is complete disinterest in our work in Pakistan,” he says in a telephone interview. The floods came as predicted. The Pakistan Meteorological Department described the August rains in 2013 as “exceptionally on higher than normal side” and the “ninth highest monthly rainfall since 1961”. Statements issued by Pakistan’s National Disaster Management Authority (NDMA), as reported in the press at the time, said more than 1.5 million people were affected by the resultant floods.
Five floods in five seasons
With flood waters having caused large-scale destruction in the first two weeks of September this year, Pakistan has just had its fifth consecutive year of monsoon-related floods. Each of the five floods was predictable with a ten-day lead time. In the case of the ones in 2012 and 2013, the forecast was actually made and an alert sent to the authorities in Pakistan by Webster’s team.
Why have there been five consecutive years of heavy rainfall followed by flooding in Pakistan? This year, Webster and his colleagues have published a large, analytical paper in which they take a close look at the storm structures that produced three consecutive years of flooding in Pakistan between 2010 and 2012.
“Striking similarities between all three floods exist,” they write, adding that the “flood-producing storms exhibited climatologically unusual structures” in all three cases. So, we had three consecutive years of highly unusual storms, each of which bore striking resemblance to each other. This suggests that the monsoon systems that have governed rainfall in northern India for millennia might be undergoing a structural change. “If these were natural phenomena, you would have seen this sort of thing occur in the past,” says Webster. “Clearly the climate has changed.”
What was so unusual about these storms? And what were the similarities between them?
Under normal conditions, weather patterns that produce rainfall in northern parts of the subcontinent differ between the eastern and western ends of the monsoon system. Over the Bay of Bengal, where the monsoon system originates, a depression sucks in high levels of moisture from the ocean air, and creates layered clouds, one on top of another, known as “stratiform clouds”, spread over a large area. The resultant storm system is “less intense, but much more widespread and productive of precipitation”. Hence, the rains in the east are gentler than they are in the west, but in both regions they cover a large area and last quite long.
In each of the three years the authors of the paper studied, large stratiform clouds “embedded with wide convective cores, rarely seen in this region” somehow travelled from the Bay of Bengal, where they are normal, across the subcontinent and unloaded their enormous cargo of moisture in a short, intense burst over Pakistan. In 2010, this system was pushed northwards, into the indentation formed by the meeting of the Himalayas and the Hindu Kush mountain ranges. Once the storm system collided with the mountains, it was pushed upwards, causing it to cool rapidly and thereby offloading its moisture in a short burst over northern Pakistan, causing flash floods.
But, in the subsequent two years, the same storm systems veered southwards instead, due to an absence of a south-to-north wind, appearing over Punjab and Sindh. The intensity of the rains was lower in those two years and much of the rains fell outside the Indus basin. As a result, the swelling of the rivers was also not as intense as it was in 2010.
The shifting of these Bay of Bengal storm systems towards the west is one common anomaly in each of the three flood years. Another puzzling anomaly in these three years is a link between the storms in Pakistan and an intense heat wave in eastern Europe which created a high pressure trough above the Himalayas. This high pressure system, rarely seen before, served as a natural barrier, a massive atmospheric wall running from Tibet to the northern reaches of Afghanistan, that apparently deflected an otherwise important wind that always blows over Pakistan from the Afghan plateau. That wind is dry and warm, and usually caps the moist winds coming from the Arabian Sea where the western fringe of the subcontinent’s monsoon system primarily draws its moisture from. Because this moist air is capped on top by the dry and warm air from the Afghan plateau, the moisture does not coalesce into large storm structures. But in each of the three flood years between 2010 and 2012, “[w]arm air from the Afghan plateau did not flow out over Pakistan”, reads the latest paper by Webster and his team. “Rather, a deep layer of moist air flowed into the region from the Arabian Sea and the Bay of Bengal. The high pressure trough above the Himalayas, therefore, played a crucial role in the storms of all three years,” it says — first by creating the wind patterns that caused the Bay of Bengal storm system to travel westward, then by blocking the warm dry air from the western Afghan plateau which caps the moisture the seas blow into our weather. Large moisture-laden clouds, therefore, arrived over Pakistan during each year from across the Gangetic plains and freely joined with those coming from the Arabian Sea to form gigantic storm systems over the Indus basin.
The volume of rains that fell over Pakistan in a short period of time was staggering in 2010. Cumulatively, up to 6,000 millimetres of rain fell over much of northern Pakistan that year. In the next two years, this amount declined somewhat — exceeding 2,000 millimetres in some areas in 2011 and just touching 1,000 millimetres in 2012.
In their papers, dating from one written in February 2011, Webster and his colleagues have pointed out another common theme between the three flood years: The storm systems that resulted in the floods may be highly anomalous but they were all predictable with high levels of confidence up to ten days in advance, in some cases even more. In the February 2011 paper, titled Were the 2010 Pakistan Floods Predictable?, they find that “the July 28  event was predicted almost eight days in advance with a probability larger than 60 per cent”.
Their latest paper extends the scope of the analysis further. The storm systems, that produced the floods in each of the three years studied, arose from a combination of global and regional weather patterns. The global climatic patterns can be very accurately modelled because the data required for that exercise is readily available in databases like the European Centre for Medium-Range Weather Forecasts (ECMWF), an intergovernmental organisation supported by 34 countries, and located in the UK. Forecasting specific regional storms “is not possible at this time”, although “the large-scale environments conducive to the development of [regionally directed] storm systems that produce flooding in South Asia” can be forecast with “considerable accuracy”.
To underline this point, Webster built a model for the Indus river basin. The model draws more than 40 million meteorological readings from the ECMWF database every day and runs them through a series of computational processes so complex that they require a computer server with 64 cores and a processing speed of 2.3 gigahertz to operate. The model couples these computations with river flow data from Pakistan – or whatever of it is available – and computes water flows into the Indus river as well as its tributaries that will result from the rainfall being forecast, the absorptive capacity of the terrain, the spread of vegetation, the solar energy signature over the entire Indus basin and more. It takes the system four to five hours, every day, to ingest the data and process it, before yielding a detailed forecast for the next 10 days.
The model then tells you how much rain to expect where, how much accumulation of water will occur in what region and what river flow will be at each hydrological station on each forecast day. The model began providing its first operational forecasts in August 2012. In the first few weeks of its operation, it forecast large floods in Sindh. Daily forecasts made from August 31  onwards “consistently predicted main-stem flows in the Indus to peak between September 11 and 13”, says Shrestha. They alerted the authorities in Pakistan.
As if on cue, the rains began on September 8, 2012, and turned into a deluge in a matter of days. The hardest hit part of the country was Sindh, as shown by the model 10 days earlier, where up to three million people were affected, according to the NDMA. By the end of the month, Pakistan was asking for international assistance to fight the floods.
The model can predict rain very accurately, even when it is dealing with anomalous storm patterns. But, in order to predict streamflows at precise locations, it needs river flow data from each of the hydrological structures on the Indus river system. The creators of the model, however, have found that the government of Pakistan is not willing to share this data with them. They, therefore, have developed a system to download daily reports from the Pakistan Meteorological Department website which contains some of this information.
In 2013, they updated the model further by including river flow data in it, and this is when it yielded the flood forecast in August that year, giving detailed streamflow figures as well. As a rule, the more data you can feed into the model, the more precise the results it will give. “It needs to be upgraded every year,” says Webster. As data from an outgoing year is fed into it, the model better understands the relationship between atmospheric events and the hydrology on the ground.
So how does advance alert help? The model can tell you the likelihood of a flood, its location and intensity and effects on each individual river of the Indus river system. The real game actually begins after an alert has been issued. With advance warning, embankments can be strengthened, dams can be emptied out, barrages can be reinforced, breaching priorities for embankments that lie along the path of the flood can be drawn up in time and residents can be alerted so that perishables and livestock can be moved to higher ground.
Pakistan is blessed with a highly developed river management system which can be effectively used to mitigate the full impact of a flood. For a clearer idea on how this would work, consider the floods of this year, which began due to unusually high rains over the catchment areas of the Chenab and Jhelum rivers. The first flood alert was issued by the Pakistan Meteorological Department on September 3, 2014, just over 48 hours before the flood peak arrived in Pakistan from India. When the alert was issued, water level at Mangla Dam reservoir stood at 1,227 feet (this same level had been obtaining since mid-August, at least). The Indus River System Authority (IRSA), responsible for overseeing the distribution of river water among different parts of Pakistan, had been releasing water from the dam very slowly since July 28, anticipating rains. Inflows at the dam were around 20,000 cusec during those days. Then, suddenly, on September 4, the first surge arrived, with inflows jumping to 95,000 cusec. The IRSA responded to the flood alert issued a day earlier by raising outflows only slightly — to 30,000 cusec.
The flood peak arrived on September 5, when inflows jumped to 310,000 cusec but outflows on that day were brought down to 15,000 cusec — presumably, in an attempt to arrest the floodwaters. In a single day, water level in the dam rose by eight feet – a staggering increase for a reservoir of Mangla’s size – and the water level in it touched 1,236 feet. On September 6, 2014, the inflow of water rose to 413,000 cusec and water level in the dam rose to almost 1,240 feet. Since top water level that Mangla Dam’s reservoir can reach is 1,242 feet, further increases in the water level could not be accommodated, so the dam’s spillways were opened. Outflows jumped from 15,000 cusec to 282,000 cusec on September 6, 2014.
The resultant surge in the Jhelum river combined a few days later with a similar surge travelling down the Chenab, at the confluence of these two rivers just upstream from Trimmu headworks [see diagram on pg. 57]. Trimmu is where the majority of the breaches had to be made to prevent the headworks from getting washed away. Three breaches were made, including at least one in the embankments around the barrage itself. This is where most of the flooding occurred.
With advance warning, Mangla could have been emptied out much sooner, making it possible to absorb the floodwaters surging down the Jhelum river. With no additional water flowing in from the Jhelum, the peak flood in the Chenab could have been relatively better managed at Trimmu, possibly without significant breaching of dykes. All downstream structures – Panjnad, Guddu and Sukkur – could also have easily handled the floodwaters. The flooding would not have been nearly as damaging as it has turned out to be.
Pakistan’s water managers frequently ask for more infrastructure – dams and barrages – as a flood control measure. But how will they operate this infrastructure in the absence of lead time in flood alerts, given the current state of flood forecasting in Pakistan? Without advance warning of a major flood event, along with attendant streamflow forecasts, hydrological infrastructure will only be a silent spectator to any flood, at best, and a liability to be protected by breaching embankments, at worst.
Part II of this story can be read here.