The Challenge of Selecting Precipitation Products for Extreme Weather Services in a Data Scarce Environment

Abstract

Evaluation studies of global precipitation datasets invariably show poor performance over regions with limited gauge availability, including most of the African continent, where reporting rate is lowest of any region in the world. While technical advances lead to a vast increase in sensor information in many domains, functional weather stations have been deteriorating progressively since the 1980s. Recently we have seen an increase through the TAHMO network with more than 500 stations in Africa (www.tahmo.org), but their record is still short. At the same, development of reliable weather services has a growing urgency, in regions that depend on rain-fed agriculture and for cities with fast growing populations prone to extreme rainfall flooding. The question then becomes: given the urgency and poor information quality, what is the best pathway towards services that can support society to build extreme weather resilience in a changing climate?
Here, we present a methodology that supports evaluation of precipitation products across a range of rainfall features relevant for hydrological or agricultural applications: spatial distribution patterns, representation of small-scale variability and seasonality, detection of dry and wet spells, and timing and intensity of extreme rainfall. We demonstrate application of the methodology for the case of Tanzania, based on six commonly available, (near)global precipitation datasets and data from 16 rain gauges across the country.

The analysis distinguishes between performance across season and for regions dominated by different weather systems or with different topographical structure. One of the conclusions is that 5-day aggregation is the minimum time-scale that can be used for the products to reach a quality better than monthly-mean of gauge data. We also show that performance varies strongly over different regions and seasons. Timing of the precipitation was poorly estimated by all products, particularly for heavy rains.

Based on this methodology, precipitation products can be selected based on their strengths with respect to particular applications. It also, and perhaps more importantly, points out the limitations of using products at fine-scale resolutions, where their predictive performance is hardly better than using a climatological average.