We know that poorly ventilated humidity measurements from pyschrometers have a moist bias compared to well ventilated humidity measurements from pyschrometers. Poorly ventilated instruments are considered to be those from screens (S), ship's screens (SN) or ventilated screens (VS). The ventilation in this case refers to slatted sides rather than active ventilation. Well ventilated instruments are considered to be whirled hygrometers (W), slings (SL), ship's slings (SG), aspirated instruments (A) or unscreened instruments (US). Obviously this will be exacerbated in low wind situations.
Previously we have applied a 3.4% reduction in specific humidity to all poorly ventilated observations and a 3.4*0.3333 reduction in specific humidity to all unknown ship observations, following previous work by Josey et al. 1999 and Berry and Kent, 2011. Only 30% of the reduction is applied to account for the fact that approximately 30% of all ship observations are listed as being from poorly ventilated instruments, following Josey et al. (1999). Actually, Berry and Kent (2011) use a proportional value based on the percentage of obs within a 10degree area that are believed to be poorly ventilated.
The screen adjustments result in quite large changes to the global relative humidity record - compare figure 1 with figure 2. In particular the 1973-1977 period, 2005-2012 period and year 2015 are moister. There are considerable changes in the amount of metadata reported over time - see figures 3, 4 and 5. There is quite an increase in the amount of metadata (black lines, figure 3) available from 1973 to 1985 with a complete drop off in 2015. The majority of observations come from poorly ventilated instruments after 1977 but there is some variation over time (blue lines, figure 3). The green solid and dashed lines in figure 4 show that the number of poorly ventilated ship observations is increasing generally over the period. To complicate things further, as shown in figure 5, the percentage of observations made from pyschrometers has decreased over time from near 100% to around 50%, with electric and capacitance sensors making up the difference. This information is only for ships so does not include the increase in capacitance sensors due to increasing buoys. The capacitance and electric sensors are likely not to be as badly affected by poor ventilation as pyschrometers as they do not require evaporation to occur for an accurate measurement.
It
is clear that these changes can make a contribution to the features in
the global relative humidity series compared to the unadjusted series -
especially the 1973-1977 period and the complete lack of metadata in 2015. This is clear to see from the black dotted line in figure 4. This shows (or attempts to show) the mean affect of the adjustments applied over time. It is calculated by:
(No. Poorly Ventilated Ship Obs * 3.4) + (No. Unknown Ship Obs * (3.4*0.3333)
Total No. Ship Obs
This amount is presented as the mean percentage by which specific humidity is reduced - so a peak in the dotted lines would correspond to a lower specific humidity (and likely relative humidity) and vice versa. This is clearly seen in the big dip down in 2015 in the dotted lines that corresponds to a moister 2015 in the bias corrected relative humidity series (figure 2 compared with figure 1).
It is difficult to decide whether to have a flat rate percentage adjustment for the unknowns which is almost certainly wrong in most cases, an annually varying adjustment that is vulnerable to fluctuations in metadata, or a linearly increasing adjustment which accounts for the change over time but not annual fluctuations. I have explored this in Figure 4 with the black, dark grey and light grey dotted lines respectively.
The black dotted line (figure 4) shows the mean annual adjustment if I use a flat rate expectation that 30% of ship obs come from poorly ventilated instruments. This is based on Josey et al., 1999. (black line). Actually, although it could be argued that around 30% of all ship observations are from poorly ventilated observations (green line in figure 4) this does not account for the unknowns. Figure 3 shows that around 50% of ship obs that have metadata are from poorly ventilated instruments.
A flat assumption of percentage of ships that are poorly ventilated could be justified because although in the early years the proportions suggest that most observations were well ventilated, it is quite plausible that it is the well ventilated (maintained?) ships that are more likely to have reported their metadata. So we could interpret the low percentage of poorly ventilated screens as an underestimate. However, it is also likely that there was a reduction in hand-held psychrometers and an increase in those housed in screens over the period with a general move to more time efficient observing practices.
It might be better to assume a 50% value, assuming that the proportion of ship obs without metadata have a similar instrument type/exposure composition to those that have metadata. The pink dotted line (figure 4) shows the mean annual adjustment if I use a flat rate expectation that 50% of ship obs come from poorly ventilated instruments. This is almost identical to the 30% line in shape although the year-to-year jumps are slightly muted - note the 1973-1977 jump and the 2015 drop down when comparing the black (30%) and pink (50%) lines. Either way, this flat rate adjustment has quite large effects with the drop in and out of metadata.
The dark grey dotted line (figure 4) shows the mean annual adjustment if I use an annually varying expectation of the percentage of ship obs from poorly ventilated screens. In this case I have used the percentage of ship obs with metadata and assumed that the unknowns have similar proportions. This has very large effects at the early and late points of the record where the amount of metadata, and percentage of poorly ventilated observations, fluctuates quite strongly. My gut feeling is that this is less desirable than a flat rate expectation of either 30% or 50%.
The light grey dotted line (figure 4) shows the mean annual adjustment if I use an linear annually increasing expectation of the percentage of ship obs from poorly ventilated screens. I have obtained this linear fit from a basic linear regression of the green time series showing the % of ships obs from non-ventilated instruments (ignoring the 0 in 2015) of the percentage of poorly ventilated observations. I have changed the intercept to be the percentage of ship obs with metadata rather than percentage of all ship obs by comparing this intercept with that of the intercept from all ship obs with metadata (red line). This gives me linear increasing trend in poorly ventilated observations of:
y = 0.4548x + 44.8
I could have used the percentage of ship obs with metadata that are non-ventilated but this is affected by the drop in ship obs with metadata data around 1980-1981 (see figure 3 blue lines). The year-to-year variability is very similar to the pink line (flat 50%) but the 2015 drop is slightly dampened. The trend here could introduce a very small negative trend in humidity but arguably the adjustments are accounting for a change in the observing system (increase in poorly ventilated observations).
This is quite a complex decision to make. An ensemble approach would allow full exploration of all options. Currently we have used the black line (flat 30%) approach. My preference is leaning towards either the pink line (flat 50%) or light grey (linear trend) approach which both have much less variability when the metadata drop out in 2015.
At present my preference is to use the 50% flat adjustment assumption, or even 55% given that the mean % of SHIP obs with metadata that are poorly ventilated over the 1973-2014 period is 55% (54.59%). The justification for a linear increase is quite weak although could probably be justified. The problem with using 50% or 55% is still 2015 where all metadata drop out. This results in quite a drop in the mean adjustment applied. To account for this the 2015 observation adjustment could be artificially grown to match the mean over the 2000-2014 period (working arbitrarily with decades here). The mean adjustment for 2000-2014 is 2.13 and the adjustment for 2015 is 1.87. This is a ratio of 1.14. So, I think we could justifiably grow the 2015 values by 1.14 to account for the reduction in obs with the full 3.4% adjustment applied. This now results in a mean adjustment of 2.13 which avoids the artificial drop in adjustment for 2015.
Previously we have applied a 3.4% reduction in specific humidity to all poorly ventilated observations and a 3.4*0.3333 reduction in specific humidity to all unknown ship observations, following previous work by Josey et al. 1999 and Berry and Kent, 2011. Only 30% of the reduction is applied to account for the fact that approximately 30% of all ship observations are listed as being from poorly ventilated instruments, following Josey et al. (1999). Actually, Berry and Kent (2011) use a proportional value based on the percentage of obs within a 10degree area that are believed to be poorly ventilated.
 |
| Figure 1 Global average RH anomaly time series for ship/buoy/platform observations that have been quality controlled. |
 |
| Figure 2 Global average RH anomaly time series for ship/buoy/platform observations that have been quality controlled and bias corrected for instrument exposure only. |
The screen adjustments result in quite large changes to the global relative humidity record - compare figure 1 with figure 2. In particular the 1973-1977 period, 2005-2012 period and year 2015 are moister. There are considerable changes in the amount of metadata reported over time - see figures 3, 4 and 5. There is quite an increase in the amount of metadata (black lines, figure 3) available from 1973 to 1985 with a complete drop off in 2015. The majority of observations come from poorly ventilated instruments after 1977 but there is some variation over time (blue lines, figure 3). The green solid and dashed lines in figure 4 show that the number of poorly ventilated ship observations is increasing generally over the period. To complicate things further, as shown in figure 5, the percentage of observations made from pyschrometers has decreased over time from near 100% to around 50%, with electric and capacitance sensors making up the difference. This information is only for ships so does not include the increase in capacitance sensors due to increasing buoys. The capacitance and electric sensors are likely not to be as badly affected by poor ventilation as pyschrometers as they do not require evaporation to occur for an accurate measurement.
 | ||
| Figure 3. Change in instrument exposure over time for ship observations. The solid lines are for exposure of the hygrometer and the dashed lines are for exposure of the thermometer. Percentages shown for each exposure type are shown as a percentage of those ship observations with metadata present. |
 |
| Figure 4. Change in ship instrument exposure metadata. The mean annual percentage reduction in specific humidity is shown for different estimates by the dotted lines (left Y axis). The annual mean percentage of ship observations for each category is shown by solid lines (right Y axis). Note that these are percentages of all ship observations. A linear trend has been fitted to the % Ships Non-ventilated to show that these are generally increasing in percentage over time. |
 |
| Figure 5. Change in instrument type over time for ship observations. Percentages shown for each instrument type are shown as a percentage of those ship observations with metadata present. |
(No. Poorly Ventilated Ship Obs * 3.4) + (No. Unknown Ship Obs * (3.4*0.3333)
Total No. Ship Obs
This amount is presented as the mean percentage by which specific humidity is reduced - so a peak in the dotted lines would correspond to a lower specific humidity (and likely relative humidity) and vice versa. This is clearly seen in the big dip down in 2015 in the dotted lines that corresponds to a moister 2015 in the bias corrected relative humidity series (figure 2 compared with figure 1).
It is difficult to decide whether to have a flat rate percentage adjustment for the unknowns which is almost certainly wrong in most cases, an annually varying adjustment that is vulnerable to fluctuations in metadata, or a linearly increasing adjustment which accounts for the change over time but not annual fluctuations. I have explored this in Figure 4 with the black, dark grey and light grey dotted lines respectively.
Applying adjustments based on 30% of observations being affected by the bias
The black dotted line (figure 4) shows the mean annual adjustment if I use a flat rate expectation that 30% of ship obs come from poorly ventilated instruments. This is based on Josey et al., 1999. (black line). Actually, although it could be argued that around 30% of all ship observations are from poorly ventilated observations (green line in figure 4) this does not account for the unknowns. Figure 3 shows that around 50% of ship obs that have metadata are from poorly ventilated instruments.
A flat assumption of percentage of ships that are poorly ventilated could be justified because although in the early years the proportions suggest that most observations were well ventilated, it is quite plausible that it is the well ventilated (maintained?) ships that are more likely to have reported their metadata. So we could interpret the low percentage of poorly ventilated screens as an underestimate. However, it is also likely that there was a reduction in hand-held psychrometers and an increase in those housed in screens over the period with a general move to more time efficient observing practices.
Applying adjustments based on 50% of observations being affected by the bias
It might be better to assume a 50% value, assuming that the proportion of ship obs without metadata have a similar instrument type/exposure composition to those that have metadata. The pink dotted line (figure 4) shows the mean annual adjustment if I use a flat rate expectation that 50% of ship obs come from poorly ventilated instruments. This is almost identical to the 30% line in shape although the year-to-year jumps are slightly muted - note the 1973-1977 jump and the 2015 drop down when comparing the black (30%) and pink (50%) lines. Either way, this flat rate adjustment has quite large effects with the drop in and out of metadata.
Applying adjustments based on an annually varying expectation of the percentage of observations being affected by the bias
The dark grey dotted line (figure 4) shows the mean annual adjustment if I use an annually varying expectation of the percentage of ship obs from poorly ventilated screens. In this case I have used the percentage of ship obs with metadata and assumed that the unknowns have similar proportions. This has very large effects at the early and late points of the record where the amount of metadata, and percentage of poorly ventilated observations, fluctuates quite strongly. My gut feeling is that this is less desirable than a flat rate expectation of either 30% or 50%.
Applying adjustments based on a linear increase in the percentage of observations being affected by the bias
The light grey dotted line (figure 4) shows the mean annual adjustment if I use an linear annually increasing expectation of the percentage of ship obs from poorly ventilated screens. I have obtained this linear fit from a basic linear regression of the green time series showing the % of ships obs from non-ventilated instruments (ignoring the 0 in 2015) of the percentage of poorly ventilated observations. I have changed the intercept to be the percentage of ship obs with metadata rather than percentage of all ship obs by comparing this intercept with that of the intercept from all ship obs with metadata (red line). This gives me linear increasing trend in poorly ventilated observations of:
y = 0.4548x + 44.8
I could have used the percentage of ship obs with metadata that are non-ventilated but this is affected by the drop in ship obs with metadata data around 1980-1981 (see figure 3 blue lines). The year-to-year variability is very similar to the pink line (flat 50%) but the 2015 drop is slightly dampened. The trend here could introduce a very small negative trend in humidity but arguably the adjustments are accounting for a change in the observing system (increase in poorly ventilated observations).
Conclusions
This is quite a complex decision to make. An ensemble approach would allow full exploration of all options. Currently we have used the black line (flat 30%) approach. My preference is leaning towards either the pink line (flat 50%) or light grey (linear trend) approach which both have much less variability when the metadata drop out in 2015.
At present my preference is to use the 50% flat adjustment assumption, or even 55% given that the mean % of SHIP obs with metadata that are poorly ventilated over the 1973-2014 period is 55% (54.59%). The justification for a linear increase is quite weak although could probably be justified. The problem with using 50% or 55% is still 2015 where all metadata drop out. This results in quite a drop in the mean adjustment applied. To account for this the 2015 observation adjustment could be artificially grown to match the mean over the 2000-2014 period (working arbitrarily with decades here). The mean adjustment for 2000-2014 is 2.13 and the adjustment for 2015 is 1.87. This is a ratio of 1.14. So, I think we could justifiably grow the 2015 values by 1.14 to account for the reduction in obs with the full 3.4% adjustment applied. This now results in a mean adjustment of 2.13 which avoids the artificial drop in adjustment for 2015.
No comments:
Post a Comment
Note: only a member of this blog may post a comment.