The authors claim that trend lines indicate that air temperatures at the lake warmed by about 0.5–0.7 °C. However, trend lines do not reveal changes over time — gaussian averaging gives a better insight. The authors' data1 for Mbala and Bujumbura (Fig. 1) reveal that there was no change or a slight decrease in average temperature between 1952 and 1975; Bujumbura wind speed was unchanged between 1967 and 1975, and Mbala wind speed increased over that period.

Figure 1: Historical monthly weather records for Lake Tanganyika.
figure 1

Temperature data: red line, from Bujumbura; green line, from Mbala (gaussian average over six years); grey lines, monthly temperatures. Wind-speed data: blue squares, from Mbala; yellow dots, from Bujumbura (gaussian average over two years). Created from the original weather data of O'Reilly et al.1.

If climate change were the reason for the decline in productivity, we should see no such decline between 1952 and 1975, or might even see an increase due to rising wind speeds at Mbala. Nor should we see any change in water temperature, as there was no air-temperature change. However, despite steady temperatures and rising winds, the authors' values for productivity (Fig. 3 of ref. 1) show a steady decline during that time, and their water temperature (Fig. 2a of ref. 1) shows an increase of roughly 0.25 °C.

After about 1975, both the temperature and wind records become suspect, with a considerable amount of missing data. There is an abrupt 0.7 °C rise in the temperature in Mbala in 1980, accompanied by a large reduction in standard deviation (see the authors' Fig. 1a: standard deviation 1952–80, 1.22; s.d. 1980–94, 0.95). This looks as though there has been either an artificial splicing of two records or a change in thermometer location, particularly as the same pattern is not present in Bujumbura.

The later parts of both wind records are particularly poor. After a year's gap in 1976, the Mbala wind records resume sporadically (a full quarter of the post-1976 Mbala monthly data is missing), with much lower values and standard deviation (s.d. 1967–75, 2.0; s.d. 1976–92, 1.4).

Between 1967 and 1982, the Bujumbura wind data are complete; there is not a single missing month. But after a three-year hiatus (1982–85), 17% of the Bujumbura monthly data is missing, and radically lower values are reported. I suggest that the wind figures after the gaps should be used with caution unless they can be confirmed by other observations.

For the sake of argument, let us assume that the wind records of O'Reilly et al. are correct. From their data1, we can draw several conclusions. First, air temperatures were steady or dropped slightly at Lake Tanganyika between 1950 and about 1978. They rose abruptly, and then rose only slightly during 1980–90, after which only temperatures at Bujumbura rose. However, we see no sign of this pattern in either water temperature or lake productivity (Figs 2, 3 of ref. 1). In particular, water temperatures rose between 1950 and 1978, whereas air temperatures were unchanged or dropped slightly; water temperatures fell over 1993–96 (and continued to fall through 2003), whereas air temperatures rose slightly from 1993 to the end of the record in 1996. Also, all their productivity indices dropped between 1952 and 1965, when air temperatures were dropping slightly rather than rising.

Second, the wind-speed data of O'Reilly et al. have several problems, but purport to show a rise in Mbala wind speed between 1967 and 1975, a sudden drop in 1976 and a dip in 1978–84, followed by a 35% increase that lasted until 1996. The authors say that wind in Mbala is critical for productivity, but their productivity data do not reflect these wind changes. The productivity data series that extends past 1970 shows an overall productivity decrease from 1970 to 1996 (one series shows a short-lived rise in about 1990).

Third, as data for both air temperature and wind are so poor, and as there is no correlation between the reported temperature changes and wind changes on the one hand and productivity and water-temperature data on the other, the null hypothesis (that changes in productivity are not the result of climate change) cannot be rejected. Based on the authors' data, therefore, the productivity changes cannot be ascribed to climate change.