Direct and indirect seasonal rainfall forecasts for East Africa using global dynamical models

Abstract

Regional‐scale seasonal climate outlooks are typically produced using forecast information either local to the region or from another area with teleconnections to the region. Dynamical global long‐range forecast (LRF) systems can provide both types of information, and these two approaches are compared in the context of seasonal rainfall forecasts for two adjoining areas in the Greater Horn of Africa region in tropical East Africa. The direct method utilizes the unprocessed LRF outputs for the region. For the “indirect” method, canonical correlation analysis is used and works by identifying patterns in historical LRF predictions over large tropical domains that relate to observed variability in the East Africa regions. For a given year, projections of the LRF forecasts onto those patterns can then be employed to construct the regional forecasts. This approach takes advantage of the tendency for LRF systems to have greater skill for large‐scale variability than for smaller regional‐scale features. Using case studies, it is found the two approaches contain complementary information: the indirect approach can provide notable skill benefits in years with strong large‐scale forcing while in some years, particularly when large‐scale signals are weak, the “direct” forecast are superior. Skill comparisons over many years found that, although results are region/season dependent, in general the indirect approach has higher skill overall—with improvements equal to or greater than those afforded by a 1‐month reduction in lead time with the direct approach. Results for both methods used separately and in combination are provided for the March–April–May and September‐to‐December seasons in the two regions, using data from currently operational dynamical LRF systems. Skill was best for the September‐to‐December season in the southern region, using “indirect” forecasts. The “direct” approach was better than “indirect” for the March‐to‐May season in the northern region. In general, combination did not produce a substantial benefit.

Skilful long‐lead hybrid predictions of the East African short‐rains season

“…From these results, we can conclude the skill from May from the hybrid approach exceeds or matches the average performance of direct model predictions initialized in August but is lower than that of best performing models and the statistical method. The relatively good performance of the hybrid models is consistent with Colman et al (2020) who found that combining statistical and dynamical prediction methods for East Africa could enhance skill.…”

“…The relatively good performance of the hybrid models is consistent with Colman et al . (2020) who found that combining statistical and dynamical prediction methods for East Africa could enhance skill.…”

Peramalan Langsung Dan Tidak Langsung Market Share Mobil Menggunakan Arimax Dengan Efek Variasi Kalender

“…Peramalan tidak langsung jarang dilakukan padahal dalam beberapa kasus dapat menghasilkan nilai ramalan dengan tingkat kesalahan yang kecil. Beberapa penelitian yang menggunakan konsep peramalan tidak langsung diantaranya adalah Petrorius & Botha (2010), Komara (2012), Shi et al (2011), Amendola & Candila (2017), Moosa & Vaz (2018), Colman et al (2020).…”

A comparison of seasonal rainfall forecasts over Central America using dynamic and hybrid approaches from Copernicus Climate Change Service seasonal forecasting system and the North American Multimodel Ensemble

“…Hybrid methods then combine dynamic and statistical methods (Slater et al ., 2022). Some hybrid methods generate indirect rainfall forecasts by extracting related variables (e.g., SST) from the GCMs and then statistically translate those values to target variables (e.g., Alfaro et al ., 2018; Strazzo et al ., 2019; Colman et al ., 2020). Indirect forecasts could perform well, as forecasts of large‐scale variability like SST over tropical domains are often a strength of the GCMs compared to their forecasts of more variable parameters like rainfall (Barnston et al ., 2012; 2019; Saha et al ., 2014).…”