Climate Change Impact - Part 2 - Southern Bangladesh

Climate Change Impact

Part 2: Example – Southern Bangladesh


This posting is based on a study of the impact of climate change on Southern Bangladesh. It examines current climate related problems which include storm rainfall, drought duration and sea level rise. It concludes that the main changes in climate in the region will be increases in temperature, storm rainfall, sea level and drought severity. Average rainfall will remain similar to the present.


The project area is shown in the following map. It covered 13 Upazilas in southern Bangladesh. The total population of the project areas was 17.6 million. In some Upazalis the population density was more than 1000 people per square kilometre; about the average for Bangladesh.

Figure 1 Map of project area

The project tackled three areas of vulnerability: roads, cyclone shelters and markets. In terms of climate vulnerability, a significant factor in the project area was the fact that much of the it was only a few metres above mean sea level.

Current climate

Climate data were obtained from three main sources shown in the following table.

Table 1Sources of climate data used in report
Data source
National Climatic Data Centre (part of the National Oceanographic and Atmospheric Administration of the USA)
Daily values of rainfall, maximum and minimum temperature.
Climate Explorer, a site run by the Netherland Meteorological Service.
Monthly values of temperature and rainfall.
TuTiempo, a weather site run by a Spanish company.
Daily values of temperature, rainfall, wind speed, atmospheric pressure, relative humidity.

Other data analysed included sea levels (from the Permanent Service for Mean Sea Level (PSMSL)) and Cyclone intensity (from the ‘Joint Typhoon Warning Center’ web site.)

The observed climate data were analysed and some of the conclusions were:
  •         Annual rainfall in the project area is reducing. For the period 1947 to 2015 the annual average is 2072 mm/year and it is falling by 1.3 mm each year.
  •         The monsoon rain, defined as the rainfall in the three wettest consecutive months in a year, is not decreasing.
  •         The monsoon season starts earlier with the wettest month being June in recent years but July in earlier years.
  •         The average temperature in the project area, based on the period 1890 to 2015, is 26.1 °C. It is increasing slowly and over the 125 years rose by 0.35 °C.
  •          Sea level data were obtained for 6 stations in the project area or from nearby sites in the Bay of Bengal. The data were downloaded from Most sites had around 20 years of data but one had 60 years. The conclusion was that sea levels were rising at 4 mm a year but in places the settlement of the sediments in the delta gave an apparent increase of more than 10 mm a year.
  •      Data on cyclones were analysed for the Northern Indian Ocean, an area which encompasses the Bay of Bengal, using the Accumulated Cyclone Index. The data covered the period 1972 to 2014. It showed that cyclone energy was increasing by 1% a year – though there was a lot of inter-annual variability. Additional analysis of this data showed that the increase in energy was mainly related to an increasing number of storms – the energy in individual storms showed little increase. Cyclones are a major problem in the study area and in one incident it is estimated that 300,000 people were killed.

Climate projections

In many countries, you are given a lead as to where to go for climate projections. For example, certain climate models might have been shown to perform well or special models might have been developed. In the case of Southern Bangladesh, I could find no such guidance. I therefore decided to evaluate the accuracy of different climate models. In this case I established a set of criteria related to both temperature and precipitation. For the period of available data, the simulation of the climate models was compared to observed values. The criteria included absolute values (some models are biased having values which are consistently higher or lower than observed) and relative values (based on trend and seasonal variation). There were 8 criteria in all; temperature and precipitation, monthly and annual, bias corrected and uncorrected.

From this analysis, a few points stood out:
  •         Average monthly values were more accurately simulated than annual time series.
  •         Temperature was more accurately simulated than precipitation.
  •         No model scored consistently higher than others on all criteria.
  •         Some models were highly rated for temperature or precipitation but poorly for the other one.

Four models were selected for more detailed analysis. These were:

  •         NOAA GFDL-CM3, had the best average accuracy when all criteria were considered.
  •         MIROC-ESM-CHEM, was ranked 2nd overall.
  •         CESM1-CAM5, was ranked 5th overall. The model ranked 3rd was from the same source as model 1 and it was decided against using two models from the same source.
  •         MPI-ESM-LR, was ranked 6th based on all criteria but it was never lower then 11thon any criteria. This means it was less likely to produce a ‘bad’ projection.

The following chart shows the monthly simulated and observed average temperature. Value for five models are shown – the four retained as the best overall and the model which was best for this criterion.  As can be seen all models performed well on this criterion.

Figure 2 Observed and projected monthly temperature

The next chart shows the simulated and observed annual precipitation for the study area. This supports the statement above the annual values and precipitation are not as accurately simulated as temperature.

Figure 4 Observed and projected annul precipitation

For the projections, the highest value of RCP (RCP 8.5) was used. There were two reasons for this. Firstly, as the value associated with the maximum change in climate, it represents an outer envelope of projections. Secondly, the lower values of RCP assume a more stringent agreement on emissions than the ones achieved so far.

The next chart shows the projected change in average monthly temperature for the period 2040-2060 relative to 1985-2015. Whilst there is a range of projections, about 0.6 °C, the model are generally consistent in projecting a lower temperature increase during the rainy season, May to September, than for the rest of the year.
Figure 4 Projection of monthly temperature in 2050

The chart for annual precipitation shows much less consistency among the models. For each model, there are two traces – a faint one for annual values and a heavy one for the 10-year moving average. The range of value suggest the change could be from minus 150 mm to plus 150 mm. For comparison, the observed annual rainfall is around 2000 mm/year.

Figure 5 Projection of annual rainfall

The projection of other parameters was also considered.

The same models were used for the projection on maximum daily rainfall in each year (RX1 projection from the ETCCDI set).  This showed that intense rainfall would increase. All four models were consistent in projecting this increase. As the precipitation is that for the whole area of the model cell which most closely corresponded the study area, the difference in absolute values is likely to be due to difference in model cell size.
Figure 6 Projection of annual maximum daily rainfall

The study area is a rural area where agriculture is important. To study the impact of climate change on agricultural productivity the projections of projections of variables such as temperature, humidity and others were used to calculate changes in potential evapotranspiration. These values along with precipitation projections were used with a simple irrigation model to show the change in water demand. This shows that annual water demand is likely to increase from 600 mm/year to 800 mm/year for the mid-century
Figure 7 Projection of unsatisfied crop water demand


The main changes in climate in the region will be increases in temperature, storm rainfall, sea level and drought severity. Average rainfall will remain similar to the present.


Calculating the Impact of Climate Change - Part 1 - Introduction

Climate Change Impact

Part 1. Background



It is widely accepted that the climate is changing, and will change more in the future, as a result of human activity. I have carried out many studies where I have quantified the impact of changes to climate. These have been in Europe, Asia, the Pacific and Africa. This posting is an introduction. Other postings will examine specific studies.


There is widespread acceptance the climate is changing and that humans are driving, at last part of, the change. As a consequence, it is normal for infrastructure projects to examine the potential impact of climate change and then adjust the design to take account of it. This involves developing a quantified timeline of the changes.

So, in this and a series of following posts I am going to describe how to quantify the impact of climate change based on my experience in many parts of the world: Europe, Asia, the Pacific and Africa.

The purpose of these posts is two-fold:

  •         Firstly, to pass on my experience to others who are required to quantify climate change.
  •          Secondly, and unashamedly, to advertise my skills and experience.

This post is introductory. Following posts will be more detailed and specific.

Impact of Climate Change

NASA lists the projected impacts of climate change[1]as:
  •         Change will continue through this century and beyond
  •         Temperatures will continue to rise
  •         Frost-free season (and growing season) will lengthen
  •         Changes in precipitation patterns
  •         More droughts and heat waves
  •         Hurricanes will become stronger and more intense
  •         Sea level will rise 1-4 feet [0.3 to 1.2 m] by 2100

I have examined all these types of impact – and a few more.

Climate Models

A climate model represents the earth as a series of cells (or boxes).

  •          These cells are of the order of 100 km by 150 km horizontally and have around ten levels of atmosphere and a similar number of levels of the ocean.
  •          The models simulate the interaction between each of the model cells about once every hour.
  •         The execution time of climate models is of the order of 1 minute of computer simulation for one day of simulation. Typically, a model will simulate the climate for a period of more than 200 years and the execution time will be a few months.

The above figures are a generalisation for global climate models and individual models will have different values for the above parameters. In particular, regional models, which on represent part of the earth’s area, will have a finer grid.

Representative Concentration Pathways (RCPs)

The whole purpose of climate models is to calculate the changes in climate due to human activity and if these are found to have negative consequences, to evaluate mitigation options. The changes in human activity can lead to an energy imbalance – with more energy being absorbed by the earth than is radiated back into space. The best-known factor is the production of Carbon Dioxide, which allows more energy in to the earth’s atmosphere than out of it, but others include the effect of soot particles in the atmosphere and changes to the reflection of radiation.

Exactly what humans will do to the atmosphere in the coming century is unknowable so four possible trends have been considered. These are known as Representative Concentration Pathways (RCPs). They are labelled by the associated energy imbalance in watts per square metre at the end of this century: RCP 2.6, RCP 4.5, RCP6.0 and RCP 8.5. The first of these would occur if humans severely curtailed their emission of greenhouse gases. The last of the four assumes a future with little or any limitation of emissions.

The use of RCP values was introduced in 2013. Before that the equivalent was SRES (Special Report on Emissions Scenarios) values. Some of the studies I worked on used SRES values.


As stated above, global climate models work at a grid size of the order of 100 km side. (The phrase ‘of the order of’ is used as there is variety of scales between different models.) However, it is sometimes necessary to consider areas that are smaller than this, for example a specific length of proposed road. Going from a model cell to the specific area is known as ‘downscaling’. In theory, there are two methods: dynamic and statistical. However, the ‘dynamic method’ effectively requires a climate model with a reduced grid size developed for a specific study which in all but a few cases is impracticable.

The alternative, known as the ‘statistical’ method or the ‘delta method’, assumes that the changes in climate projected for a model cell apply uniformly over the whole cell. For example, assume that a model cell projects a temperature increase of 2°C but that the observed temperature within the area of the cell is from 9°C to 15°C. The projection will be that that in all locations the increase will be 2°C. 
This method places reliance on observed climate data. Source of such data will be discussed later.

Source of climate projections

For climate projections, I use almost exclusively the Climate Explorer site ( run by the Netherland’s Meteorological Service. The only exception has been in a few cases when ‘pre-digested’ projections were provided by the client.

Use of the web site is free and if you sign up it facilitates use by ‘remembering’ your previous selections.

In terms of projections I use mostly two sets of projections:
  •         Monthly CMIP5 scenario runs
  •         Annual CMIP5 extremes

The acronym ‘CMIP5’ refers to the ‘Coupled Model Inter-comparison Project Phase 5’.
The ‘scenario runs’ part of the site has output from climate models under four groups: Surface variables, Radiation variables, Ocean, Ice & Upper Air variables, and Emissions. In most cases for impact analysis it is the variables in the first group that are important. These include temperature and precipitation.

The ‘extremes’ part of the site has a second set of projections.  These were developed by the Expert Team on Climate Change Detection and Indices (ETCCDI). Values are provided for 31 variables. These include maximum daily precipitation, number of frost days (when the minimum was zero or below), number of ice days (when the maximum was zero or below) and growing season length.

Selection of climate projections

The climate explorer site has projections for more than 20 climate models. In addition, some models are run for multiple ‘experiments’ in which slightly different but credible model parameters are used. So, which one to use?
 In some cases, there might be guidance on the choice of climate models, for example from previous studies. Often however a decision has to be made on which models to use. What I have often done is to compare the simulated climate model output with observed values. This is rarely simple. For example how do you choose between a model which is biased (with values consistently higher or lower than observed) but which represent the inter-annual variation with a different model which is less biased but does not represent annual variations?

Sources of observed climate data

The best source, if available, is from the meteorological and hydrological services in the country you are working in.  For various reasons that is not always possible. Sometimes, for example, the meteorological service requires payment which the project has no funds for. Other sources of data include:
  •         The Climate Explorer site ( mentioned above. This has monthly data on precipitation and temperature.
  •         The National Climatic Data Center ( This site has daily data on precipitation and temperature.
  •         The Climatic Research Unit ( This has a range of data including monthly temperature and average values of several meteorological variables on a 10’ grid.

Climate change impact

Quantifying how the climate will change is but the first step to estimating the impact of climate stage. For example, for the impact on water resources it necessary to run a hydrological model with, firstly, observed climate data and, secondly, projected climate data.

Climate change impact studies

The following is a list of the climate change impact studies to be covered in other posts.
  •         Southern Bangladesh. The impact of climate change on rural communities including temperature and rainfall changes and the effect of sea level rise.
  •         Tonle Sap is a shallow lake/wetland in Cambodia. The hydrology is complicated as at times the lake receives water from the Mekong river and at times discharges to the river. A model of lake levels was developed which calculated changes in level due to climate change.
  •         The Mekong River Basin. A hydrological model was developed for the whole of the Mekong basin from the Himalayas in China down to the final flow measuring station in Cambodia. A hydrological model was used to estimate changes in flow due to climate change.
  •         Great African Lakes. The three ‘Great’ lakes (Lakes Victoria, Malawi and Tanganyika) are important for their fisheries. Data on lake temperature was decoded and the impact of climate change on water temperature was estimated.
  •         Hydrology of the Tagus river basin. The Tagus (Tejo/Teju) is one the most developed major river basins in Europe. A water resources/hydrological model of the basin was developed and the impact of climate change evaluated.
  •         Road flooding in Vanuatu. The impact of climate change on road flooding and rural economy was studied.
  •         Road flooding in Samoa. Data from different sources were combined to estimate flooding at different elevations. The impact of climate change was also studied.
  •         Road flooding in Kyrgyzstan. In this case flooding was but one of the potential problems the other one being icing during winter months. Again, the impact of climate change was studied.
  •         Variation of climate change in Zambia.
  •         The Yesilirmak Basin in Northern Turkey is highly developed for hydropower and irrigation. It was projected that average flows would decrease and, equally importantly, the seasonal distribution would change. At present, as a result of snow melt, the peak flow is in early summer at the start of the irrigation season; in future the peak flow will be in December.
  •      The Kagera Basin flows through 4 countries (Rwanda, Burundi, Uganda and Tanzania) before entering Lake Victoria. An extensive data base of flow, rainfall and climate was available this was sufficient for a hydrological model, HYSIM, to be calibrated. It was concluded that the increase in evaporation and in precipitation would to some extent cancel each other out.


The Canary in the Coal Mine

In days of yore, coal miners would take a caged canary into the mine with them as the birds were more sensitive to poisonous gases than humans; if the canary died then the miners got out – alive.

‘Climate sceptics’ have long accused ‘climate activists’ of (to continue the metaphor) breeding highly sensitive canaries and looking for dangerous coal mines. Up to now I’ve studiously respected this site's motto as being a place where ‘numbers count’ and stayed out of debate. After a recent paper on ‘vanishing islands’ in the Solomon Islands archipelago I felt I had to comment. I was partly spurred on to do this by a guest post by David Middleton on the web site.
The paper in question is “Interactions between sea-level rise and wave exposure on reef island dynamics in the Solomon Islands” (Albert et al, Environmental Research Letters, Volume 11, Number 5). The headline message of the paper was

“..we present the first analysis of coastal dynamics from a sea-level rise hotspot in the Solomon Islands [and] have identified five vegetated reef islands that have vanished.”

That message got widespread coverage. At breakfast this morning in my hotel in Dhaka (Bangladesh) a fellow guest (a curriculum development specialist – nothing to do with climate) asked me if I had heard of the seven (sic) islands which had disappeared.

The total land area of the Solomon Islands is 27,990 km2 (World Bank figure). The area of 5 islands which have disappeared is given in the paper as 160,310 m2. Why did the authors use square metres? Why not hectares or square kilometres? More usual surely for an island? Perhaps it was because 160,310 m2 is 0.16 km2; that is 0.0006 % of the total area of the Solomon Islands.

OK. We are talking about canaries in coal mines so perhaps they are justified in a little sleight of hand. Let’s look further.

Their introduction starts:

“How islands and the communities that inhabit them respond to climate change and particularly sea-level rise is a critical issue for the coming century. Small remote islands are viewed as particularly vulnerable.” The authors do acknowledge a role for wave action but this is seen as secondary to sea level rise.

The following table is taken from the paper.

Area (m2) of the 5 islands which have disappeared

What the table shows is that there was significant loss of area between 1947 and 1962.  The loss was 41% in that period. Expressed in m2 per year the rate was 4100 m2/year for the period 1947 to 1962 and 1800 m2/year for the remainder of the period. I recognise that defining two years just because they have data might bias the answers but when the rate in the second period is less than half that in the first period it is hard to accept that loss of island area is due to increasing sea levels.

Let’s now take a look at sea level rise. The following chart shows sea levels from two sources. The first is from the Permanent Service for Mean Sea Level and covers the period 1975 to 2015. Levels were measured at two locations with a short, 5-month, overlap. The second record from 1992 to the present is from the University of Colorado Sea Level Research Group and is based on satellite altimetry. 

The PSMSL record has a rate of rise of 2.7 mm/year. The University of Colorado gives a rate of 5.9 mm/year much less than 7 mm/year quoted in the paper; the difference in rate is in part probably linked to the recent drop in sea levels due to the El Nino effect. One drawback of these data is that they do not cover the whole period 1947 to the present used for analysis of the area of the islands.

In January of this year I was in Samoa – looking at the impact of climate change on roads. There it is something to be concerned about. On both of the two main islands there are few inland roads but they do have roads all the way round each of the islands. In places these roads are on a narrow coastal band and barely above the current high tide level. So, a modest increase over the next decade or so could have serious consequences. While there I prepared estimates of sea levels from 1948 to 2014. These, together with the PSMSL figure for Solomon Islands are shown on the next chart.

The amplitude of the sea level estimates is higher for the Solomon Islands than for Samoa but they show a similar trend. I’ve also plotted a quadratic trend line through the Samoa data when shows that for the early period sea levels were more-or-less constant but in recent decades have been rising more rapidly.

In other words, if sea levels in the Solomon Islands have followed a similar trend to Samoa, the most rapid loss of area coincided with the least change in sea level.

I mentioned above that I am working in Bangladesh. At the northern end of the Bay of Bengal the 2 metre contour is 100 km from the coast. A typical spring tide has a range of 4 metres. In that part of the country most agricultural land is behind embanked polders and when they are overtopped the land becomes saline. So creeping sea level rise has a real impact there.

The paper that is the basis of this posting has, of course, succeeded in the author’s terms; it has got wide publicity for the potential impact of climate change. But whether describing the disappearance of five small islands, whose total area is that of 20 soccer pitches, has advanced climate science is a moot point.

Glabal Sea Level Rise

  • The average rate of sea level rise from 1880 to 2013 is 1.6 mm/year
  • The rate of sea level rise is not constant. It is increasing at 0.014 mm/year/year.
  • Superimposed on the rising sea levels is a cyclical component with a periodicity of about 50 years which is synchronous with the Atlantic Multidecadal Oscillation.
CSIRO Estimate
Sea levels have risen more than 100 m since the end of the last ice age and they are still rising. This post looks at the rate of rise over the last century or so and, based on sea level data, and answers the question "Is the rate of level rise increasing?",

The CSIRO provide one of the main estimate of global mean sea levels (Church, J. A. and N.J. White (2011), Sea-level rise from the late 19th to the early 21st Century. Surveys in Geophysics, doi:10.1007/s10712-011-9119-1.). The data run from 1880 to 2013. They are available as monthly or annual values. The annual values have been analysed here.

This chart shows the CSIRO sea level data. The data are in millimeters relative to an arbitrary datum. The data show that global sea levels have risen by just over 200 mm in the period 1880 to 2013. Plotting a trend line through the graph gives an average rate of rise of 1.6 mm/year. This is 160 mm  century, much less than most of the climate change projections.

The above chart gives just one rate of sea level rise - the one for the whole period. So what if we look at year-on-year sea level change.

Year-on-year rate of sea level rise

The next chart plots the difference between the value of sea level in the given year and the value in previous year, for each year from 1981 to 2013. So, the first value is the difference between sea level in 1881 and in 1880, and so on. Looking at the chart there is a lot of year-to-year variation, from minus 17 mm/year to plus 21 mm/year. A trend line through the data shows that the rate of sea level rise has increased, by 0.0141 mm/year/year. That means the underlying rate of sea level rise was 1.9 mm/year higher in 2015 than it was in 1880. In other words, the rate of sea level rise is increasing.

Smoothed rate of sea level rise

One way of observing underlying trends more clearly when the data have a lot of year-on-year variation is to use a moving average. This takes the average of the values of the data for a number of years before and after each point plotted. As the data are so variable a long period has been used for averaging, 31 years. The first point plotted is for 1896 and is the average of the sea level from 1881 to 1911, the next is the average from 1882 to 1912 and so on.

This chart confirms that the rate of sea level rise is increasing but not in a uniform way. From a peak of 1.66 mm/year in 1900 to fell to minimum of 0.51 mm/year in 1920. It then rose to another peak of 2.2 mm/year in 1946 before falling to to a minimum of 1.33 mm/year in 1978. The rate of sea level rise then increased again to another peak of 2.84 mm/year 1997.

The trend line on the chart gives a slightly different value for the rate at which the rate of sea level rise in increasing, 0.0117 mm/year/year, to that in the chart above. This is due to effect of the averaging.

Cyclical component to sea level rise

The final chart plots the difference between the rate of sea level rise and the trend line. This is described as the detrended rate of sea level rise. For example, the peak in 1946 was 2.20 mm/year, the value on the trend line for that year was 1.63 mm/year so the value plotted was the difference between them 0.57 mm/year.

This chart shows that the rate of sea level rise has two components. The first is the underlying increase in the rate of sea level rise, this is 0.0141 mm/year/year as seen in the first chart. The second is a cyclical component with an amplitude of plus or minus 0.6 mm/year and a periodicity of around 50 years.

And the orange line? Climate scientists have detected a number of cycles in observed climate data. One of these is called the Atlantic Multidecadal Oscillation (AMO). It is based on sea temperatures in the northern part of the Atlantic Ocean. When the values of this oscillation are plotted along with the detrended rate of sea level rise they show a high degree of synchronicity. It cannot be argued that the AMO causes the variation in the rate of sea level rise. On other hand, it could be argued that both phenomena share an unknown forcing agent.



Sea Levels - Pacific Islands

There is concern that sea level rise might threaten the existence of some small island communities.

Since the early 1990s the Australian Bureau of Meteorology has been running the Pacific Sea Level Project. The continually monitor sea level, air temperature and water temperature among other parameters. Given the motto of this site “Where numbers count” this is something of which we fully approve.

Figure 1 shows the location of the monitoring sites.

Figure 2 shows a schematic layout of a typical station.

In a recent update of our web site
... we plot the values of sea level for a twelve stations in the network. The data of these stations were summarised by the following figure 3.

Two factors are very evident. Firstly sea levels are rising: a trend line through the average of all stations gives a rate of rise of 5 mm/year. The second very noticeable feature is the way in which sea levels were influenced by the strong El Nino of 1997.

Since I since first set up the web site I have looked at the impact of climate change on rural roads in Vanuatu. This was one of the photos I took – on the island of Ambae. It shows clear signs of coastal erosion with dead tree stumps up to 50 metres out to sea. Such erosion is common on the south coast of that island and the sea was encroaching by about 3 metres every year. However given that the problem is localised to one side of the island the reason is unlikely to due to sea level rise.

The above photo was on the south-east side of the island. This one was on the north coast. Here there is no sign of erosion – indeed vegetation seems to moving close to the sea.

The National Geographic web site recently carried an article with the “a growing body of evidence amassed by New Zealand coastal geomorphologist Paul Kench, of the University of Auckland's School of Environment, and colleagues in Australia and Fiji, who have been studying how reef islands in the Pacific and Indian Oceans respond to rising sea levels. They found that reef islands change shape and move around in response to shifting sediments, and that many of them are growing in size, not shrinking, as sea level inches upward. The implication is that many islands—especially less developed ones with few permanent structures—may cope with rising seas well into the next century.”

Figure 6 show the equivalent of figure 3 but for sea temperature. They are plotted as variation about the mean to show trends more clearly. This plot also shows the influence of the El Nino with a drop in sea temperature. A trend line through the average sea temperature shows an increase of 0.011 C per year.

The average sea temperature for the twelve islands is given in the table below. The range is from 25.4 °C to 30.5 °C.

Mean Sea Temperature - °C
Cook Islands
Marshall Islands
Papua New Guinea
Solomon Islands
Federated States of Micronesia

Figure 7 is complementary to figure 6 and shows air temperature for the 12 islands and the moving average for the mean of all twelve islands.

This shows that, as expected, islands further from the equator have larger seasonal variation in air temperature. They also show a very low rate of increase in temperature for islands; the annual rate is 0.018 C per year.



Tonle Sap wetland and the influence of climate change

Model of Tonle Sap

Tonle Sap is the largest lake in South-East Asia and is a wetland of international importance and is recognised by the Ramsar convention. Like most wetlands its area varies significantly through the year, from 2000 km2at its lowest to ten times that figure at its largest.  The bed of lake is close to sea level and its maximum level is normally only 10 m above sea level. The channel from the lake to the Mekong can flow in either direction. When levels in the lake are higher than those in the Mekong water flows out of the lake toward the Mekong (generally from October to April) and for the rest of the year it flows in the opposite direction. 

The following map shows three significant locations for level and/or flow measurement. Levels in the lake are recorded at Kampong Loung. 

Figure 1 - Important measuring sites related to the model of Tonle Sap
Levels and flows in the Mekong are measured at Kampong Cham and in the channel connecting the lake to the Mekong at Prek Kdam.

Figure 2 - Level measurement on the Mekong at Kampong Cham

Figure 3 - Level measurement on Tonle Sap River at Prek Kdam

The next chart shows the level at Kampong Cham and in Tonle Sap Lake. Two features are worth noting:
  • there is approximate synchronicity in the timing of the two sets  of levels but with peaks in the Mekong generally being a bit earlier than those in Tonle Sap
  • the range of levels in the Mekong, about 15 m, is higher than in the Lake, about 7 m.

As the water levels are recorded relative to local data it is not possible to know from this graph the relative levels between the two measuring locations. 
Figure 4 - Water levels in Tonle Sap Lake and the Mekong at Kampong Cham

The next chart shows the flow at Prek Dam in the Tonle Sap River. During the period of October to April water flows from the lake to the Mekong. During the rest of the year the flow is toward the lake from the Mekong.

Figure 5 - Flow in Tonle Sap River at Prek Kdam

The above data sets, levels in the lake and the Mekong and flow via the Tonle Sap channel give us many of the important elements for a model of Tonle Sap.  However there are a number of other important factors. These are:
  •          Flow into the lake from surrounding rivers
  •          The relationship between depth, volume and area of the lake
  •          Precipitation on the lake
  •         Evaporation from the lake
There are four usable records of flow into the lake. There are shown on the following map and comprise he flows records at Battombong (Stung Sangker), Kampong Kdey (Stung Chikriang), Kampong Chen (Stung Staung) and Pursat (Stung Sen). There are other level records but they do not have an accurate rating curve linking levels and flows.

Figure 7 - Relationship between Level and Volume in Tonle Sap Lake

The curves on the chart were fitted using Excel. In the case of the flood area the relationship is:
Area = 30.061(Level)2+ 1094.1(Level) + 716.69
Where the surface area of the lake is in square kilometres and level is in metres.
The equivalent relationship for volume is:
Volume = 0.914(Level)1.883
...where the volume of the lake is in cubic kilometres.

The final elements for a model of the lake, rainfall and evaporation were based on average values taken from climate stations around the lake. For each daily time step the volume of rainfall and evaporation were based on the amount in millimetres multiplied by the area of the lake. The model also included a further loss mechanism. As Tonle Sap contracts in size during the October to April period water evaporates from the exposed soil which, given there is little rain in this period, becomes very dry.   When, later, Tonle Sap again expands the water flows from the lake over land which has been dry for, in some cases, several months. This water then sinks into the voids in the soil. The model applied this loss cumulatively. If the area of Tonle Sap was expanding the loss was equivalent to the total evaporation during the period of expansion at that point in time. The maximum loss by this mechanism was 200 mm. Once the lake started contracting then the loss from this component was set to zero.

The basic formula for the lake model was:

Volume[t+1] = Volume [t] + Inflows – Outflows

The model operated on a daily time step and was developed as an Excel file.

The inflows were: flow from local rivers, flow via the Tonle Sap channel and rainfall on the lake. During calibration it was found the estimate of local inflows based on the above method (adjusted in proportion to the drainage area) overestimated the inflow by a factor of two. A preliminary analysis suggested two reasons for this. One is that the area used refers to the area of the Tonle Sap ecosystem which might be larger than the drainage area upstream of the level measuring point. The second is that the gauging stations receives water from the upland parts of the drainage basin and therefore exaggerates the average runoff. Since the contribution from the surrounding rivers is small compared to the contribution from the Mekong through the Tonle Sap channel this parameter is not of great importance to the overall accuracy of the model.

The flow via the Tonle Sap channel was based on the following equation:

Flow = a * (Mekong level – Tonle Sap level – b)c

If the flows were toward to the lake then this formula was used as above. If it was toward the Mekong then it adjusted by a further factor d.

The values of the four parameters a, b, c and d were obtained by using the ‘Solver’ add-in of Excel. ‘Solver’ adjusts each of the four parameters to see how they change the accuracy of the model. In this case the accuracy of the model is defined as the sum of the squares of the errors in the estimation of the levels in Tonle Sap Lake.

The outcome of Solver optimisation process is that the formula became:
Flow = 1126 * (Mekong level – Tonle Sap level – 3.98)1.32

The value of ‘d’, relating to the direction of flow,  was 0.59. In reality this parameter is compensating for some hydraulic factors not included in this model. A full solution of the equations would take account of the inertia of the water in the Tonle Sap channel; in simple terms when the relative levels in the lake and the Mekong change they first have to stop the river flowing in one direction before they can increase its flow in the opposite direction.

The values of parameter ‘b’, 3.98 m, which allows for the difference in the datum at Kampong Cham and at Prek Kdam is compatible with Figure 4of water levels at the two sites above.
 Another factor not included this model is the time delay between changes in relative water levels. To have included a hydraulic model would have required information on the channel shape and dimensions and a whole project on its own.
The next chart shows the simulated and observed flow of the Tonle Sap channel.

Figure 8 - Simulated and observed flow in Tonle Sap River

At first sight this does not look encouraging. In particular the peak inflows are not well represented.
However examination of the current meter gaugings carried out in 2008 to 2010 suggests a reason. The following chart is for the ‘out’ period only; that for flows in the other direction is very similar. It shows that for mid-range levels there is a reasonably consistent relationship between flows and levels but at low levels and high levels, when the flow direction is changing, the relationship is unclear. In the case of the following chart levels from 6 to 8 metres can be associated with either increasing flow, as in 2009, or falling flow, as in 2010. The flow associated at that level can also vary from 1,000 m3/s to almost 20,000 m3/s.

Figure 9 - Current meter gaugings in Tonle Sap River
This suggests that the calculation of actual flow values at Prek Kdam might not be consistent and that to simulate them might be as much a case of simulating peculiarities of the flow calculation as of representing the underlying flow patterns.
It should also be noted that the simulation of flows in the Tonle Sap channel is not an end in itself. The overall objective is to simulate water levels in Tonle Sap Lake. The following chart shows that very simulation.

Figure 10- Simulated and observed levels in Tonle Sap Lake

As can be seen the simulation is generally accurate. Many of the peaks of water level are slightly underestimated but otherwise it is good. The correlation between observed and simulated levels is 0.967.

It can therefore be concluded that the simulation of water levels in Tonle Sap Lake is sufficiently accurate for the model of lake levels to be used to study flooding around the lake.

Projected levels

In a separate part of the project flows of rivers within Cambodia and the whole of the Mekong  were simulated using the HYSIM rainfall/runoff model (  The hindcast values of all climate models reported on in the IPCC Technical Assessment Report of 2012 were analysed and the models rated on 4 factors: representation of observed monthly temperature, representation of observed monthly rainfall, representation of monthly temperature anomaly, representation of monthly precipitation anomaly. It was concluded that the MIROC model was most 
appropriate for Cambodia.

Using the calibrated hydrological model and the climate projections, flows were projected for a 30 year period centred on 2045.

The following chart shows the observed (or more strictly, simulated using observed meteorological data) and projected levels.  

Figure 11 - Observed and projected water levels in Tonle Sap Lake

The following table shows the change in the level of Tonle Sap Lake for different return periods.
Return period
Current conditions
Projected 2045


The work described above was performed while the author was working for SweRoad under a contract providing support to the Ministry of Rural Development of Cambodia, financed by the Nordic Development Fund and supervised by the Asian Development Bank. Any views expressed or those of the author and do not necessarily represent those of the other parties.




Sea levels have been rising since the maximum of the last ice 20,000 years ago. The rate of sea level rise is regarded as an indicator of climate change. The change in sea levels is driven by two factors: the thermal expansion of the sea water as it warms and the melting of ice over land.

Long Term Sea Level Change

During an ice age, ice covers are large areas around both poles. The amount of water in the ice caps is such that sea levels are markedly reduced. Levels 20,000 years ago, at the maximum of the last ice age, were 140 m lower than they are today. Until about 7,000 years ago the rate of rise was about 100 mm/decade. Since then rate of rise has averaged 10 mm/decade.

Estimation of Sea Level Change

Global sea levels have traditionally been estimated from tide gauges. As can be imagined they show fluctuations of several meters due to tide and wave action. Identifying sea level changes of a few millimetres a year against this background “noise” is problematic. Since 1993, data are available from satellites. There are two other factors which add to the difficulty of estimate changes in sea level. The first is the way the earth has reacted to the melting of the ice caps. Where major ice melt has taken place, in northern Europe and North America for example, land levels have risen; the post glacial rebound (PGR). Conversely where sea levels have risen and encroached on previously dry areas, land levels have fallen under the increased weight of the oceans; glacial isostatic adjustment (GIA). (Some sources use the two terms interchangeably) These changes typically average around 4 mm/decade but can be higher in some locations. The second factor is the influence of atmospheric pressure. The changes in pressure can be seasonal and modify levels by 1 metre; often an allowance is made for these pressure difference by applying what is called “an inverted barometer.” As can be seen the adjustments to be made to sea level are of a similar order of magnitude to change in sea level itself. It is generally considered that the rate of change of sea level cannot be accurately estimated for periods of less than 10 years.

Sea Level Change

Figure 1 shows the sea level changes from 1807 to 2001 using two estimates based on tide gauges (Jevrejeva et all and Church at al). There is broad agreement between the two estimates. The Jevrejeva record show that sea levels fell for the first half of the 19th century. This suggests that the low temperatures recorded in Europe in this period may have been representative of global temperatures. It also follows the Dalton minimum of sunspot activity.

Figure 1

Figure 2 shows a composite record from the two tide gauge estimates and satellite based data from the TOPEX/JASON satellite system. To harmonise the two data sets the satellite data were adjusted to give the same average for the period of overlap. The graph also shows the rate of rise per decade. This was based on subtracting the difference in level for pairs of months 10 years apart. Over the last century or so the rate of rise has fluctuated from -20 mm/decade up to 40 mm/decade. The increase since 1880 has been around 250 mm.

Figure 2

Whilst at first sight the rise in sea level seems almost constant looking at the estimates of the rate of sea level rise shows that this does fluctuate. To clarify this the following graph, figure 3, shows the rate of rise in twent-year periods.

Figure 3

This appears to suggests that there is an underlying increase in the rate of sea level rise of about 0.015 mm/year/year and a fluctuation about this trend of ± 1 mm year.


In a recent posting I said I would be commenting on a paper by Zhou and Tung (Zhou, J., and K. Tung, 2012: Deducing Multi-decadal Anthropogenic Global Warming Trends Using Multiple Regression Analysis. J. Atmos. Sci.doi:10.1175/JAS-D-12-0208.1, in press.)

When I came across this paper I had mixed feelings. The paper says very similar things to those have I have been saying since January 2012: that the underlying rate of temperature increase is less than IPCC models assume, due to the influence of the Atlantic Multidecadal Oscillation (AMO). I was pleased to get further corroboration in a peer reviewed paper. On the other hand I was peeved as a paper I submitted earlier this year was not accepted.

Their approach is similar to that of Foster and Rahmstorf.

Forster and Rahmstorf developed a multiple linear regression model using total solar radiation, aerosols, ENSO and a linear trend as independent variables and 5 alternate temperature records as the dependent variable. The period analysed was 1979 to 2010; the only period common to all 5 temperature series. They concluded that for that the period the underlying temperature trend was 0.014 to 0.18 °C per year. The paper was welcomed in many quarters as countering the claim that the rate of temperature increase had had been falling off or even stationary for the last decade or more of that period.

The Zhou and Tung paper adopts a similar approach but they have substituted the ENSO with the AMO. They conclude that the rate of temperature increase since the start of the 20th century, which they ascribe to anthropogenic effects, has been less that that estimated by Foster and Rahmstorf. They give 0.0068 °C for the 100 year trend, 0.0080 °C for the 75 year trend, 0.0083 °C for the 50 year trend and 0.0070 °C for the 25 year trend. These figures are about half of those of Foster and Rahmstorf.

They consider the suggestion of Booth et al that the AMO is anthropogenic and reject it.

My own equivalent figures are 0.0050 °C per year from 1856 to the present, 0.0067 °C for the 100 year rate and 0.011 for the 30 year rate. These values are similar to those of Zhou and Tung with one exception: I get an accelerating rate of increase which reflects the growing concentration of GHGs.

The conclusion of both their work and mine is the same: climate models, which simulate all the increase in temperature as anthropogenic and driven by GHGs, are overestimating the increase in temperature by a factor of two. A corollary to both sets of ideas is that if, as seems likely, the AMO is regular then it is likely to restrict temperature increase for the next few decades while the AMO is decreasing.


It has been pointed out that the model I described in my earlier post (Climate and the Atlantic Multidecadal Oscillation) ignored anthropogenic aerosols. Here I look at the effect of adding these into the model.

The data used were downloaded from They were used in J. Hansen, et al. (2007) "Climate simulations for 1880-2003 with GISS model E", Clim Dyn, 29: 661-669 and J. Hansen, et al. (2011) "Earth's energy imbalance and implications, Atmos Chem Phys, 11, 13421-13449.

Figure 1 shows the individual components.

The individual components are:

WMGHGs – Well mixed greenhouse gases
O3 - Ozone
StrH2O – Stratospheric H­2O
ReflAer – Reflective Aerosols
AIE – Aerosol indirect effect
BC – Black carbon
SnowAlb – Snow albedo
StrAer – Stratospheric Aerosols (Volcanoes)
Solar – Solar irradiance
LandUse – Land Use.

To run a regression model with these 10 parameters plus the Atlantic Multidecadal Oscillation (AMO) would be nonsense. This would be particularly so since many of the components are highly correlated; the regression coefficient between WMGHGs and AIE is 0.98 for example. So I aggregated them into 4 groups. The first three (WMGHGs, O3 and StrH2O) I grouped as GHGs. Stratospheric aerosols and solar were treated separately. This gave 3 parameters which had exact equivalent in the original 4-parameter model. The fourth parameter was the sum of all the other components. The aggregated parameters are shown on Figure 2.

The fifth and final parameter was the AMO.

For temperature I used the HadCRUT3 global data set. I am aware that this has been superseded by the version 4 but for consistency with the earlier posting I am sticking to it.

The accuracy of the two models was almost identical. This can be seen on figure 3.

In terms of accuracy, the 4-parameter model (for the period 1880 to 2011) explained 89.3% of the variance and the 5-parameter model explained 89.5%, confirming the similar accuracy of the two models.

The comparison of observed and calculated temperatures does not describe how the values were actually calculated. Figure 4 shows the effect of each component on temperature.

In the above the use of ‘-5’ refers to 5 parameter model and ‘-4’ to the original 4 parameter model. The main differences between the models are:

  • The effect of GHGs is larger in the 5-parameter model, the difference being largely due to the effect of the anthropogenic factors.
  • Solar effects are slightly higher in the 5-parameter models.
  • The effect of volcanoes is minimal in both models.
  • The influence of the AMO is virtually identical in both models.

The coefficients for the 5 independent variables were:

Since the all parameters, except for AMO, are expressed as W/m2 and if the forcing sensitivity is close to the accepted value of 3.2 W/m2/°C then each coefficient should have the value 0.3125 °C/W/m2 (that is 1/3.2). In this case the parameters for GHGs and solar irradiance are close to that value. That for volcanoes is much lower than expected. The parameter for combined anthropogenic effects is also lower than expected but has wide error bands so the ‘expected’ figure is within the 95% range.

Considering the critical period from 1976 to 2005, which had the largest increase, the observed temperature increases were as given in the table below.

Considering the period of rapid temperature rise, the original 4-parameter model suggested that 49% of the rise was due to GHGs, the equivalent figure for the 5-parameter model is 59%.

These findings are consistent with those reported in Zhou, J., and K. Tung, 2012: Deducing Multi-decadal Anthropogenic Global Warming Trends Using Multiple Regression Analysis. J. Atmos. Sci. doi:10.1175/JAS-D-12-0208.1, in press. I will discuss this paper in another posting.

The conclusion remains as before. It is not clear whether the AMO is the ‘heart’ of the system and is driving global temperature or whether it is the ‘pulse’ and is an indicator another driver. What is clear that global temperatures reflect changes in the AMO but the AOGCMs used by the IPCC do not consider the AMO. In particular the AOGCMs simulated reasonably accurately the large temperature increase from 1976 to 2005 but attribute all the increase to GHGs. As a consequence, as pointed out in this and previous postings, and some recent peer-reviewed papers, temperature increases in the coming decades are likely to be lower than the IPCC projections.


The following comes from a press release from the University of Reading (UK). "Natural climate variations could explain up to 30% of the loss in Arctic sea ice since the 1970s, scientists have found. "Sea ice coverage at the North Pole has shrunk dramatically over the past 40 years. The ice is now more than a third smaller each September following the summer melt than it was in the 1970s. This affects wildlife, while potentially opening up new northern sea routes and controversial opportunities for oil and gas exploration. "Scientists at the University of Reading and the Japan Agency for Marine Earth Science and Technology (JAMSTEC) have found that some of the reduction in ice since 1979 - between 5% and 30% - may be linked to the Atlantic Multi-decadal Oscillation (AMO), a cycle of warming and cooling in the North Atlantic, which repeats every 65-80 years and has been in a warming phase since the mid 1970s." In my previous post "Climate and The Atlantic Multidecadal Oscillation" I argued that around 50% of the increase in temperature from the mid 1970s to around 2005 was due to the effect of the AMO. The researchers suggest that from 5 to 30% on the loss of Arctic Ice was also due to the AMO. My conjecture and their conclusions are compatible. What is important is the implication for the accuracy of climate models. AOGCMs do not represent the AMO and assume that virtually all warming comes from GHGs (plus a little bit from solar irradiance). So, both the findings of the University of Reading researchers and my model point to the same conclusion: AOGCMS overestimate the effect of GHGs.
See Older Posts...