Comparing PERSIANN and African rainfall climatology satellite products with the ground observation
Abstract
Satellites are gaining global attention for use in operations and research. This is because of the
scarcity of ground weather and climate data. However, to use these datasets, extensive validation
is required. In this study, two widely available satellite precipitation products:- African Rainfall
Climatology version 2 (ARC2) and Precipitation Estimation from Remotely Sensed Information
Using Artificial Neural Networks (PERSIANN) were validated and intercompared on dekadal
timescales with ground observation data for the period 1990–2020, using 10 weather stations over
Uganda. The statistical methods used were correlation coefficient (CC), mean error (ME), and
Root mean square error (RMSE). Contingency table statistics were used to get an insight into the
skill of the products in estimating rainfall amounts at monthly scale at different stations in the
country. From the results, it was observed that the two products generally overestimated rainfall
for the majority of the stations. PERSIANN over estimated rainfall at Jinja, Kampala, and Kasese
while ARC2 over estimated rainfall at Arua, Kampala, and Kasese. The time series for the two
rainfall estimates analysis exhibit the same temporal variation patterns with similar amplitudes
over all of the ten stations for the period from July 1999 to December 2020. In a comparison of
the two products, ARC2 has a better representation of the observed rainfall. Results show that
PERSIANN presented largely high POD over most stations, greater than 97%. ARC2 presented
moderately high POD all over the stations with percentages greater than 60%. Jinja presented the
highest POD with the PERSIANN satellite data compared to the rest of the stations with Kasese
having the lowest. However, for the ARC2, Mbarara presented a higher percentage of POD of
about 71% while Kampala had the lowest POD of about 62%. The two products had at least 30%
of the identified rainfall days as false alarms, and the products rarely correctly identified more than
80% of the observed rainfall days. The study concluded that ARC2 has a better representation of
observation data compared to PERSSIANN.