Retrieval of Precipitation from Satellites
269
Finally, the PIP-l Steering Commitee made an attempt to conclude on a broad ranking of the
algorithms based on the results of 1. to 3. As to be expected all compared algorithms did
reproduce the known structures of precipitation, namely high values over the ITCZ and the
storm tracks connected to the cyclones of the midlatitudes, and the low values related to the
subtropical highs. But there is large quantitative disagreement. Even for the· monthly and
zonally averaged precipitation differences of up to 100% for the ITCZ are not unusual. Some
algorithms predict substantial rainfall almost everywhere while others display at the same time
large areas with no rainfall at all. Key results of PIP-1 were (Barrett et aI., 1994; 1995):
• No single satellite method of global rainfall estimation was found to be always better than
all others, but several methods were each generally better over some major regions of the
world.
• Scattering/differencing microwave techniques perform better over land and emission based
algorithms perform better over ocean areas.
• Algorithms seem to perform better in regions and climatic conditions for which they have
been calibrated, which suggests that re-calibration and especially better ground-truth
data would lead to further improvements.
• Differences between different algorithms were particularly large in centres of trade-wind
anticyclones, over land but especially over hot and cold deserts, and in oceanic zones
peripheral to the Arctic and Antarctic ice sheets.
• Solely SSM/I-based techniques achieved results at least comparable to any other method
tested including IR-based techniques (despite their much higher temporal sampling) and
weather forecast models.
11.7.2 A remark
Interestingly, both comparison programmes concluded that precipitation estimated by methods,
which combine microwave with infrared data, do not in general score better, or did even worse
than algorithms only based on SSM/I measurements. It is certainly premature and wrong to
conclude from these results, that the combination of different spectral ranges will not increase
the quality of precipitation estimates from satellites. It must be kept in mind, that the data
sets used in combination originated from different sensors with different scanning geometries,
different resolutions, or even from different satellites. To my opinion, these results merely give
the indication, that much more work is necessary to develop schemes to combine measurements
of such diverse characteristics, or to build satellites with sensor specifications, which make these
technically difficult problems obsolete.
11.8 Remaining problems
There are a number of reasons which are responsible for the large observed scatter of the results
of the various algorithms discussed above. The most important ones will be dicussed in the
following.
269
Finally, the PIP-l Steering Commitee made an attempt to conclude on a broad ranking of the
algorithms based on the results of 1. to 3. As to be expected all compared algorithms did
reproduce the known structures of precipitation, namely high values over the ITCZ and the
storm tracks connected to the cyclones of the midlatitudes, and the low values related to the
subtropical highs. But there is large quantitative disagreement. Even for the· monthly and
zonally averaged precipitation differences of up to 100% for the ITCZ are not unusual. Some
algorithms predict substantial rainfall almost everywhere while others display at the same time
large areas with no rainfall at all. Key results of PIP-1 were (Barrett et aI., 1994; 1995):
• No single satellite method of global rainfall estimation was found to be always better than
all others, but several methods were each generally better over some major regions of the
world.
• Scattering/differencing microwave techniques perform better over land and emission based
algorithms perform better over ocean areas.
• Algorithms seem to perform better in regions and climatic conditions for which they have
been calibrated, which suggests that re-calibration and especially better ground-truth
data would lead to further improvements.
• Differences between different algorithms were particularly large in centres of trade-wind
anticyclones, over land but especially over hot and cold deserts, and in oceanic zones
peripheral to the Arctic and Antarctic ice sheets.
• Solely SSM/I-based techniques achieved results at least comparable to any other method
tested including IR-based techniques (despite their much higher temporal sampling) and
weather forecast models.
11.7.2 A remark
Interestingly, both comparison programmes concluded that precipitation estimated by methods,
which combine microwave with infrared data, do not in general score better, or did even worse
than algorithms only based on SSM/I measurements. It is certainly premature and wrong to
conclude from these results, that the combination of different spectral ranges will not increase
the quality of precipitation estimates from satellites. It must be kept in mind, that the data
sets used in combination originated from different sensors with different scanning geometries,
different resolutions, or even from different satellites. To my opinion, these results merely give
the indication, that much more work is necessary to develop schemes to combine measurements
of such diverse characteristics, or to build satellites with sensor specifications, which make these
technically difficult problems obsolete.
11.8 Remaining problems
There are a number of reasons which are responsible for the large observed scatter of the results
of the various algorithms discussed above. The most important ones will be dicussed in the
following.