77. Madagascar - temperature trend PARABOLIC

The only weather station in Madagascar with significant temperature data before 1930 is Antananarivo (Berkeley Earth ID: 156627). The temperature anomalies for this station appear to oscillate over time with a warming phase before 1930, followed by a cooling phase until 1980, and then another warming phase (see Fig. 77.1 below). It is tempting to think that this station is just an outlier, but it might actually be representative of Madagascar as a whole.

Fig. 77.1: The temperature trend for Antananarivo relative to the 1971-2000 monthly averages. The best fit is applied to the monthly mean data from 1901 to 2010 and has a negative gradient of -0.82 ± 0.07 °C per century.

Overall there are twenty weather stations in Madagascar with over 300 months of data (for a list see here), but only one of these, Antananarivo, is a long station with over 1200 months of data. Another ten are medium stations with over 480 months of data. These stations are distributed fairly evenly across Madagascar as shown in Fig. 77.2 below.

Fig. 77.2: The (approximate) locations of the weather stations in Madagascar. Those stations with a high warming trend between 1901 and 2000 are marked in red while those with a cooling or stable trend are marked in blue. Those denoted with squares are long or medium stations with over 480 months of data, while diamonds denote stations with over 300 months of data.

When it comes to calculating the temperature trend for Madagascar the biggest problem (other than insufficient data) is in deciding the time interval for calculating the monthly reference temperatures (MRTs). The aim here is to choose an interval that allows the maximum number of stations to be included in the same average of the temperature anomalies. But there is also a balance to be struck in enabling as many months as possible, over as long a time span as possible, to have adequate data in order to generate a reasonably accurate mean trend over the longest possible time frame. You see, in order to detect global warming, because of the inherent variability in the temperature data, you need long time series (usually over 100 years) in order to detect any real trends.

Fig. 77.3: The temperature trend for Madagascar relative to the 1931-1960 monthly averages based on an average of anomalies from stations with over 300 months of data. The best fit is applied to the monthly mean data from 1901 to 2010 and has a negative gradient of -1.24 ± 0.06 °C per century.

The problem with data for Madagascar is that most of the station records have no data before 1951, while most of those that have data before 1951 have none after 1962. In fact of the six records with data before 1951, four have none after, while only two station records with data after 1962 have data before 1950.

The solution to this conundrum is to analyse the data in two parts, with a different MRT interval for each. The first MRT interval chosen was 1931-1960. This yielded the mean temperature anomaly (MTA) trend shown above in Fig. 77.3. The method for calculating the MRT, and then the anomalies for each station dataset has been described previously in Post 47. The second MRT interval chosen was 1971-2000 and the resulting MTA trend is shown below in Fig. 77.4.

Fig. 77.4: The temperature trend for Madagascar relative to the 1971-2000 monthly averages based on an average of anomalies from stations with over 300 months of data. The best fit is applied to the monthly mean data from 1956 to 1985 and has a negative gradient of -0.38 ± 0.25 °C per century.

It can be seen that the trends in Fig. 77.3 and Fig. 77.4 agree reasonably well, but close inspection suggests that the data in Fig. 77.4 is less noisy (look at the data spread) after 1960 than the equivalent data in Fig. 77.3, while the reverse is true for the period 1931-1960. The reason for this can be determined by examining the number of stations in the MTA calculation in each case as the plots in Fig. 77.5 below illustrate.

Fig. 77.5: The number of station records included each month in the mean temperature anomaly (MTA) trend for Madagascar in Fig. 77.3 (blue) and Fig. 77.4 (red).

The MRT problem outlined above can be resolved by combining data from both Fig. 77.3 and Fig. 77.4 into one single MTA trend. Because the two sets of data have different MRT intervals, there will be a vertical offset in their respective MTA values. Comparing anomalies between 1910 and 1930 for both cases indicates that the data in Fig. 77.4 is offset by 0.5083°C compared to the same data in Fig. 77.3. This offset is then subtracted from the MTA data in Fig. 77.4. Then a single MTA is constructed by combing the data in Fig. 77.3 before 1955 with the offset data from Fig. 77.4 from 1955 onwards. The result is the MTA trend shown below in Fig. 77.6 which is remarkably similar to that for Antananarivo shown in Fig. 77.1 at the start of this post.

Fig. 77.6: The temperature trend for Madagascar if data before 1955 from Fig. 77.3 is combined with data after 1st January 1955 from Fig. 77.4. The best fit is applied to the monthly mean data from 1932 to 2011 and has a negative gradient of -0.15 ± 0.07 °C per century.

The MTA trend in Fig. 77.6 above differs markedly from both the official Berkeley Earth version (shown in Fig. 77.8 below) and the MTA trend for the Indian Ocean that was discussed in the previous post. A comparison of the 5-year moving averages for Madagascar and the Indian Ocean shows fairly good agreement from 1975 onwards but little before that (see Fig. 77.7 below). Even so, the level of agreement between the two data sets seen after 1975 is still not comparable to that seen for temperature trends of neighbouring countries in central Europe as demonstrated in Post 57 despite Madagascar being surrounded by most of the islands whose temperature data its data is being compared with.

Fig. 77.7: A comparison of the 5-year moving average MTA trends for Madagascar and the Indian Ocean.

A comparison of the MTA trend in Fig. 77.6 with the official Berkeley Earth (BE) regional trend shows a similar disparity between temperature data before 1975 and the data after (see Fig. 77.8 below). While the behaviour of the two trends is again similar after 1975, before this point they diverge markedly.

Fig. 77.8: The temperature trend for Madagascar since 1750 according to Berkeley Earth.

Just as I have shown in previous posts, it is possible to construct a temperature profile that is very similar to the official BE regional trend in Fig. 77.8 simply by averaging the adjusted anomaly data from each station. This data is easy to find as it is recorded alongside the raw temperature data in each BE station data file on the BE website. If we perform this average for the stations in Madagascar we obtain the temperature trend shown below in Fig. 77.9.

What is noticeable is how well the curves in Fig. 77.9 agree with those in Fig. 77.8. As I have pointed out many times before, Berkeley Earth use gridding and homogenization in their averaging, yet omitting these techniques appears to produce very similar results. The condition for this to happen, is of course the usually one, namely, the weather stations need to be fairly evenly distributed. But as Fig. 77.2 shows, in Madagascar, as in most other places, on a local level they are.

Fig. 77.9: Temperature trends for Madagascar based on Berkeley Earth adjusted data. The average is for anomalies from all stations with over 300 months of data. The best fit linear trend line (in red) is for the period 1911-2010 and has a gradient of +1.07 ± 0.03°C/century.

The difference between the Berkeley Earth (BE) trend in Fig. 77.9 and the MTA trend based on the raw unadjusted data in Fig. 77.6 is shown in Fig. 77.10 below. It shows that the adjustments made by BE to the temperature data reduced the temperature values of all the data from before 1930 by about 0.8°C, but more significantly added 1.48°C per century of warming between about 1930 and 1980, thus completely reversing the direction of the temperature trend in that time interval. The result is that the parabolic trend seen in Fig. 77.6 is turned into the almost linear trend shown in Fig. 77.9. This completely changes the nature of the trend.

Fig. 77.10: The contribution of Berkeley Earth (BE) adjustments to the anomaly data in Fig. 77.9 after smoothing with a 12-month moving average. The blue curve represents the total BE adjustments including those from homogenization. The linear best fit (red line) to these adjustments for the period 1933-1977 has a positive gradient of +1.48 ± 0.03 °C per century. The orange curve shows the contribution just from breakpoint adjustments.

Summary

There is clearly strong warming of almost 1°C in Madagascar after 1980 (see Fig. 77.6 and Fig. 77.4). This conclusion is supported by data from up to fifteen different weather stations (see Fig. 77.5).

Between 1930 and 1980 there appears to be equally strong cooling occurring (see Fig. 77.6 and Fig. 77.3). This trend is seen in all of the seven different weather stations with data in this period (see Fig. 77.5).

The overall land temperature in Madagascar before 1930 is not known with any precision due to the lack of data. Only one station (Antananarivo, Berkeley Earth ID: 156627) has significant data before 1930, but four others have fragments (see Fig. 77.5) that may partially support the trend seen for Antananarivo in Fig. 77.1.

It appears that the warming seen in Madagascar after 1980 is just a part of a natural temperature variability. It cannot be attributed to global warming without more data.

Acronyms

BE = Berkeley Earth.

MRT = monthly reference temperature (see Post 47).

MTA = mean temperature anomaly.

76. Indian Ocean - temperature trend WARMING 1°C

There are sixteen weather stations in the Indian Ocean region with over 480 months of data, most of which are situated on the islands surrounding Madagascar (see Fig. 76.1 below). The exceptions are the stations on Diego Garcia (Berkeley Earth ID: 173559) which is in the eastern half of the ocean, and Minicoy (Berkeley Earth ID: 155480) which is off the coast of India north of the Maldives (Berkeley Earth ID: 156700). The rest are located on, or near, the islands of Comoros, Mayotte, Seychelles, Agalega, Mauritius, Réunion, Ile Tromelin, Ile Juan de Nova and Ile Europa.

Of all these stations, only two have temperature records with more than 1000 months of data (Seychelles Airport and Minicoy) while thirteen have over 600 months of data. Overall, twenty-two stations have at least 300 months of data.

 

Fig. 76.1: The (approximate) locations of the long and medium temperature records in the Indian Ocean. Those stations with a high warming trend between 1911 and 2010 are marked in red while those with a cooling or stable trend (Comoros) are marked in blue. Those denoted with squares are long stations with over 1200 months of data, while diamonds denote medium stations with over 480 months of data.

 

Summing the anomalies from these stations gives the mean temperature trend shown in Fig. 76.2 below. The anomalies for each station where determined relative to the monthly reference temperature (MRT) averages for the period 1941-1970 using the method outlined in Post 47. Only stations with twelve years of data in this MRT interval are included in the final calculation of the regional mean temperature trend. This means that seventeen stations were included in the calculation, including fifteen long and medium stations. The one medium station to be excluded was that at Vacoas in Mauritius (Berkeley Earth ID: 156749) which had insufficient data in the MRT interval.

 

Fig. 76.2: The temperature trend for the Indian Ocean based on an average of anomalies from stations with over 300 months of data. The best fit is applied to the monthly mean data from 1916 to 1975 and has a positive gradient of +0.33 ± 0.07 °C per century. The monthly temperature changes are defined relative to the 1941-1970 monthly averages.

 

It can be seen that the temperature trend in Fig. 76.2 has two distinct intervals. Before 1975 there is a gentle warming of about 0.33°C per century. This gives a total warming of about 0.2°C since 1920. After 1975 the warming increases significantly and this adds a further 0.8°C to the total. This suggests that the total warming since 1900 has been about 1°C.

 

 

Fig. 76.3: The number of station records included each month in the mean temperature trend for Bangladesh in Fig. 76.2.

The reliability of this trend can be estimated by considering the number of stations included in the mean. Previous analysis for central Europe (see Post 57) suggests that between fifteen and thirty stations are needed for the mean trend to be truly representative of the actual temperature trend. This is because the individual temperature records are prone to measurement errors. However, many of these errors from different stations cancel when averaged in sufficiently large numbers (Regression Towards The Mean).

The temperature trend in Fig. 76.2 is the result of averaging up to seventeen different records, but this is only true for data from the 1950s (see Fig. 76.3 above). For data after 1960 the typical number of records in the average is only about twelve, while before 1950 its is typically less than five. This suggests that the trend after 1960 is much more reliable than that before 1950. So we can be reasonably confident in the magnitude of the warming post-1960, but much less so for the data before 1950.

 

Fig. 76.4: Temperature trends for the Indian Ocean based on the average of anomalies for all long and medium stations using Berkeley Earth adjusted data. The best fit linear trend line (in red) is for the period 1911-2010 and has a gradient of +1.03 ± 0.02°C/century.

 

Finally, if we compare the result in Fig. 76.2 with the equivalent Berkeley Earth result using their adjusted data we find only a small difference as Fig. 76.4 illustrates. In fact the only significant difference is the warming seen before 1950 which is twice as large in Fig. 76.4 as it is in Fig. 76.2. But given the scarcity and hence unreliability of this data, it is impossible to ascertain which, if any, of the two results is the more accurate.

 

Summary

The islands of the Indian Ocean have warmed by about 0.8°C since 1960.

There is some evidence of a slight warming before 1950, but with insufficient data available for this period, it is difficult to quantify this with any certainty.

75. Southern Asia - overall temperature trend STABLE to 1975

In my previous four blog posts I determined the temperature trends for India, Pakistan, Sri Lanka and Bangladesh using unadjusted temperature data. The number of stations used to calculate the mean temperature each month is shown in Fig. 75.1 below. In the first three cases no warming was detectable before 1975, and only a modest temperature increase of about 0.6°C thereafter. In the case of Bangladesh there was a continuous warming that amounted to less than 0.3°C. This is significantly different from the conventional narrative on global warming, and highlights the impact that temperature adjustments have on the warming trends published by most of the main climate groups. In almost all cases the affect of these adjustments is to increase the rate of warming in the final trend as most of the regional trends I have published on this blog have also illustrated. In this post I will combine the results for India, Pakistan, Sri Lanka and Bangladesh into a temperature trend for the region.

Fig. 75.1: The number of station records included each month in the mean temperature anomaly for each of four countries in South Asia.

In Post 70 I performed a similar task for data from the different countries in South-East Asia using two separate methods. One method just involved a simple average of temperature anomalies from all the different stations in the region, while the second used a weighting process that was used to average the mean anomalies for the different countries based on their land areas. If all the countries have similar densities of stations, then both methods should yield the same result. In the case of South-East Asia that was broadly the case for most countries other than Burma, but the differences in the two methods still led to a difference in the temperature trend gradients of almost 0.1°C per century. In the case of South Asia there are large differences in station density between countries, and these differences can also change over time, as shown in Fig. 75.2 below. For this reason, in this post I have chosen to adopt the area weighted method to determine the regional temperature trend.

Fig. 75.2: The station density each month for each of four countries in South Asia.

By comparing Fig. 75.1 and Fig. 75.2 it can be seen that India clearly has the most sets of station data, but it is Sri Lanka that has the highest density of stations. However, the temperature anomaly for Sri Lanka will also be subject to greater uncertainty as it is based on only a handful of stations (eleven at most). Then again, the contribution of the Sri Lanka stations to the final regional trend will be small due to the much smaller area of Sri Lanka compared to both India and Pakistan.

Fig. 75.3: The temperature trend for South Asia based on an average of anomalies from all long and medium stations. The best fit is applied to the monthly mean data from 1876 to 1975 and has a positive gradient of +0.19 ± 0.06 °C per century. The monthly temperature changes are defined relative to the 1951-1980 monthly averages.

Applying an area weighted approach to the calculation results in the temperature anomaly time series shown in Fig. 75.3 above. This is calculated by multiplying the mean anomaly data for each country (e.g. the monthly data in Fig. 74.2 for Bangladesh) by the area of that country, and then summing the resulting products for all four countries in the region. Then the result is divided by the total area of the four countries.

Like the equivalent anomaly time series for the individual countries, the regional anomaly exhibits very little warming before 1975 with about 0.5°C of warming occurring thereafter (see Fig. 75.3). To reiterate, this is the result that we get when we use the actual raw unadjusted temperature data for each station and not the adjusted/homogenized data that is generally favoured by climate scientists.

Fig. 75.4: Temperature trends for South Asia based on an average of Berkeley Earth adjusted data from all long and medium stations. The best fit linear trend line (in red) is for the period 1876-2005 and has a gradient of +0.66 ± 0.02°C/century.

If, however, we perform the same calculation with adjusted data (which is available in the same data file as the unadjusted data on the Berkeley Earth site) we get a quite different result as is shown in Fig. 75.4 above. There is now a strong and continuous warming trend from 1875 onwards. The total warming is claimed to be 1.25°C, with 0.5°C of this occurring before 1975 (see 10-year average in Fig. 75.4). This is still less than that claimed by Berkeley Earth and shown in Fig. 75.5 below. However, this is likely to be because Berkeley Earth included both Iran and Afghanistan in the Southern Asia region, and according to Berkeley Earth the regional temperature trends for both Iran and Afghanistan exhibit over 1.5°C of warming after 1970. That would help to explain the larger temperature rise post-1970 seen in Fig. 75.5 (almost 1°C) than is seen in Fig. 75.4 (only 0.7°C). What is harder to explain is why there is so much warming before 1900 in Fig. 75.5 when there is a) so little data with almost all being due to one or two stations in India, and b) very little increase in atmospheric carbon dioxide levels to cause such a temperature increase.

Fig. 75.5: The temperature trend for South Asia since 1790 according to Berkeley Earth.

Finally, if we compare the temperature trends for the four countries of South Asia we see that while there are broad similarities in their general trends over timescales of decades, there is only moderate correlation of more short term features and fluctuations (see fig. 75.6 below). The main reason for this is distance. The principal cities of Bangladesh (Dhaka), Sri Lanka (Colombo) and Pakistan (Karachi) are all at least 2000 km apart. As I demonstrated in Post 11, temperature anomaly time series from stations that are more than 1500 km apart are very poorly correlated as Fig. 11.2(a) in that post illustrates.

Fig. 75.6: A comparison of the temperature trends of Bangladesh, Pakistan and Sri Lanka with that of neighbour India. For clarity the trends for Pakistan and Bangladesh are offset by +1°C and -1°C respectively.

Summary

The temperature trend for Southern Asia shows no warming before 1975 and only about 0.5°C thereafter (see Fig. 75.3).

The trend based on Berkeley Earth adjusted data shows significantly more warming (about 1.1°C in total), including significant warming (about 0.5°C) before 1975 (see Fig. 75.4).