Forecasts, long term trends and return period of floods

This webpage is a summary of what type of predictions can be made about floods, how floods have changed in the past century and how often they occur. Further information and references can be found in 'Floods and Water Management in Chiang Mai and the Upper Ping Catchment (Pirard, 2025)'.

Go directly to:

Previous Page: Public opinion about floods

Next Page: Climate change in Chiang Mai
FAQ

The prediction of floods concerns various aspects of forecasting which include seasonal trends, rainfall and runoff models and the behaviour of the Ping river during floods using complex models and historical records to assess how, when, where and why floods occurred, etc. as well as statistical analysis of flood return rates.

Flood forecasts

Seasonal forecasts are mostly based on meteorological predictions. This type of forecast can be made a few months ahead and are based on large scale weather patterns, regional data, the status of the ENSO, etc. It results in probabilities of intense rainfall over Northern Thailand mountains and an estimation of the risk of flooding or the absence of rain for extended period of times leading to a drought.

For floods in particular, this type of modeling can give a mildly accurate prediction on the occurrence of floods but not as much on the size of floods which are due to unpredictable intense rainfall events. Modeling of drought is mildly accurate and is supported by 78% of ground truth when adequate local calibration is done.

Figure 1. Comparison between maximum and minimum flow in the Ping river with the status of the El Nino - Southern Oscillation over the past 100+ years.

Rainfall-Runoff forecasts are directly linked with meteorology. There are several models that transform rainfall into effective rainfall and flood risk based on historical data and calibration with river gauges. In recent years, neural networks are also capable of assessing data from rainfall and the behaviour of river gauges to produce a quantitative model of floods which is similar in quality to what is done by experts in predicting floods. Another significant aspect of neural network approach and data intensive programs is that weather radar reflectivity, directly associated with rainfall covers a very large area with high resolution, providing information on rain reaching the ground with a lot more accuracy than the scattered rain gauge stations that fail to exactly represent the variability of precipitation in the mountainous terrain of Northern Thailand. In the coming years, it is probable that accuracy of floods with 1 or even 2-days warning will increase.

Specifics of flood behaviour can be modeled by St Venant equations, which are a specific variant in shallow conditions of the famous Navier-Stokes equations for fluid flow. In very simple terms, topography, the characteristics of the input and output flow and a term called the Manning coefficient are the main variables. The Manning coefficient is a numerical representation of how rough the land surface is. A road will be very smooth, grass a bit less while a dense forest has a very high Manning coefficient that significantly slow down the flow of water.

Using these inputs, 2D or 3D models can be produced for risk forecast and flood maps and can also explain some simple flood observations. For example, two opposite banks in Chiang Mai showing a couple of meters of flooding on one side, while having 200-300 m on the other side just due to flow dynamics in a meander or the time required for a certain level of water to reach a specific point inland in a shallow slope or horizontal plain (it’s not instantaneous and can take a few hours to be at equilibrium).

Flood models are also useful for comparison with reality to see the impact of some features on flood waters. The simple 20 cm ridge that separate rice paddies has been demonstrated through this modeling to have a significant impact on water storage and reducing floods. Buildings and human structures also significantly channel flood waters and slowing down the flow (higher Manning coefficient), creating higher flood levels. Elevated roads, such as the ring highways, which stands 2 to 3 m above the floodplain or perimeter walls of housing estates create significant obstruction and diversion to natural flow causing flood in historically shielded area. Higher in the catchment, roads in hilly terrains have been identified as being extremely more efficient water channels and producing high peak flows than extreme deforestation cases.

Figure 2. Maximum and minimum annual water level in the Ping river since records began in 1921. The high anomalies in 2005, 2011, 2022 and 2024 are noticeable.

Long term trends in the Ping basin

Chiang Mai has around one hundred years of continuous weather and river records which provide some baseline to observe trends in the past century and how it would compare with climate change models for the next century (see climate change).

While temperatures have decadal fluctuations, rainfall interannual variability is very high, masking anys clear trend in the past century. Relative humidity on the other hand, is quite stable and show no trend. The length of wet periods show that 1-day rainfall and the number of rainy days per year has increased significantly since 1921 but not the 7-day rainfall which remains the same. The length of the wet monsoon is high variable with no discernible trend (see Meteorology).

The Ping river show no significant increase in peak flow since 1921, indicating that on average, flood volume has not changed in the past 100 years. However, the variability of all flow parameters has significantly increased along with very significant decrease in minimum flows, annual and rainy season discharges. This clear decline appears in the mid-1950s and is directly due to anthropogenic changes and human activities. Fluctuations from 1950 to 1965, followed by the big variations in the 70s to 1984 and then a stabilization except for peak flow can be associated with known changes in water management in the Ping basin during that period (see Infrastructure).

Figure 3. Various meteorological parameters recorded for the past decades show no obvious change that could explain recent floodings

River flow, like rainfall, does not show obvious trends and most peak flow positive anomalies are associated with tropical storms, stationary monsoon and the status of ENSO. The corollary is however not systematic and tropical storms early in the season have generally little to no effect on the annual peak flow.

The El Nino Southern Oscillation (ENSO) only start to show a strong correlation post-1980s with warmer and dryer El Nino producing lower streamflow while La Nina, bringing more moisture has higher rainfall, streamflow and occurrence of floods. There is a possibility that the stronger relationship seen between ENSO and floods is due to anthropic modification of the catchment. The association of floods and droughts with the status of ENSO is however not systematic and many exceptions exist. There is also a tentative claim that rainfall in South East Asia follow a 15 to 30-year cycle possibly linked with ENSO long term cycles that would have brought wetter and more flood-prone conditions in 1880-1895; 1935-1957 and 1966-1980.

Return rates of floods

Historical data from 1954 to present time and equations applied to the recurrence of floods allow to provide estimation on the statistical time period between two floods of a specified volume. Floods of 370 to 450 m3/s is an annual occurrence. 10-year return period is reached for floods of 620 m3/s (i.e. 1973, 1975, 1977, 2024, 2011) and 30-year return period for 750 to 850 m3/s (i.e. 1952, 2005). Before the 1950s, the exact intensity of floods is unclear but there is little doubt that floods such as 1525 and 1831 reached extreme volumes superior to 2005.

In terms of flooded surface in the basin, 600 km2 is a biennial event, 800 km2 is reached for 10-year return rate, 880 km2 for 25 years, 935 km2 for 50 years and 1000 km2 for a 100-year return period. For water height, floods in the past couple of decades have been higher than normal but the cause is mostly found in urbanization and could be reversed with significant state involvement so it cannot be taken as a statistically meaningful natural trend since its dependent on quickly changing anthropic factors.

Figure 4. Relationship between flow volume during a flood and water level recorded in P.1. in Chiang Mai. The 2024 event, while not the largest flood in recent years (not fundamentally different from 2022, 2006; smaller than 2011 and 2005), is the highest flood, probably due to a restricted river channel.

The extreme floods of Mae Chaem in the early 20th century is an interesting case that can be considered for the Ping river. Some time in the early 1900s, a 2420 m3/s flood passed through Ob Luang gorge. It is very significantly higher than the 1030 m3/s recorded event in 1960. Based on standard return-rate models, a 2420 m3/s flood has no chance to happen in historical time since 1000 m3/s is already a 84-year return rate. There is some evidence that the standard logarithmic law applied to return rate is no longer applicable for very large flooding event and another mathematical expression (power law) would take over around the 10-year return rate mark.

Applied to the Ping catchment and tributaries, in similar conditions, a 1500 m3/s flood could hypothetically be possible in Chiang Mai, which would make it twice bigger than the largest 2005 flood and maybe similar to the largest historical floods. While the highest recorded floods (2005) affected only 10% of the Chiang Mai-Lamphun basin, it is possible that such extreme floods would cover 1/3 of the basin.

Figure 5. Return rate curve providing the frequency of a flood (in volume). the Gumbel function is the logarithmic curve used by the Thai government to assess the recurrence of a specific flood. The power-law curve, based on the Mae Chaem case, show where recent floods should sit in terms of how common they are and are tentatively closer to observations.