Maritime continent winter circulation as a predictor of El Niño-Southern Oscillation ( ENSO ) influence on Ethiopia summer rainfall

Summer rainfall over the cropping region of Ethiopia is related to the precursor winter circulation around the Maritime Continent and El Niño–Southern Oscillation (ENSO) development and influence. Investigation of this link reveals that sea surface temperature (SST) in the north Indian Ocean and China Sea are anomalously cold and there are low level north-westerly wind anomalies around the Maritime Continent prior to dry summers in Ethiopia. The analysis shows that wind anomalies spread into the Pacific increasing convection, and across the Indian Ocean and Africa suppressing convection. Two indices that represent Asian winter monsoon penetration near the Maritime Continent are used to predict Ethiopian summer rainfall at long-lead time. The hindcast fit of the statistical algorithm exceeds 50% during the satellite era (1981-2014).


INTRODUCTION
Climate variability in Ethiopia is influenced by the Pacific El Niño Southern Oscillation (ENSO) (Semazzi et al., 1988;Rowell et al., 1992;Janicot et al., 1996Janicot et al., , 2001;;Rowell, 2001) and associated tropical ocean thermocline (White and Tourre, 2003;Jury and Huang, 2004).During warm phase, atmospheric convection spreads across the equatorial Pacific causing upper westerly winds over the Atlantic and subsidence over much of Africa (Hoskins and Ambrizzi, 1993;Jury et al., 1994;Branstator, 2002;Yeshanew and Jury, 2007;Joly and Voldoire, 2009;Segele et al., 2009;Shaman et al., 2009).Wang and Fan (2009) demonstrate that knowledge on the evolution of climate patterns in analogue years can improve the prediction of Asian summer monsoon rainfall.During the preceding dry winter season (Dec-Mar) the circulation is governed by the Tibetan high pressure, jet stream waves that conduct cold fronts across Indo-China and circulation patterns over the west Pacific.In the *Corresponding author.E-mail: mark.jury@upr.eduAuthor(s) agree that this article remain permanently open access under the terms of the Creative Commons Attribution License 4.0 International License reviews of Chang et al. (2006) and Lau and Wang (2006), Asian monsoon winter outflow to the Maritime Continent is connected with ENSO phase and equatorially propagating Madden Julian Oscillation (MJO).Cold air outbreaks over the China Sea are more frequent and join with westerly MJO surges at the onset of Pacific El Niño and its convection.How this affects the Indian Ocean circulation and African convection is the focus of this study.
The main goals here are to understand the oceanatmosphere coupling and climatic controls on summer rainfall in Ethiopia's crop growing region.The primary scientific question is: How does Asian Monsoon winter outflow near the Maritime Continent link with ENSO and reach across the Indian Ocean to affect African rainfall?A spin-off scientific question is: how can this knowledge be exploited to offer operational seasonal forecasts at a lead time suitable for intervention?

DATA AND METHODS
The data and methods employed to develop prediction algorithms are described.The primary index is the observed June to September rainfall averaged over the crop growing region of Ethiopia 36.5°-40.5°Eand 7°-14°N.The ocean and atmospheric predictors are drawn from satellite-era reanalysis products in the preceding December to March season.This work expands on earlier studies of Korecha and Barnston (2007) and Jury (2013).

Target and predictor data
The CHIRPS 5 km resolution global land-only rainfall dataset forms the basis of this study.Monthly data from over 150 Ethiopian National Meteorological Agency gauges are blended with satellite observations, as described in Funk et al. (2014) over the period 1981 to 2014.The interpolation procedure gives primary influence to in-situ observations and gauge climatology, and secondary influence to satellite and model data (Janowiak et al., 2001;Huffman et al., 2007Huffman et al., , 2011;;Saha et al., 2010;Knapp et al., 2011).
The target area is defined by crop reports from the Ethiopian Central Statistical Agency (www.csa.gov.et) and the US Dept of Agriculture Famine Early Warning System (FEWS) that show production in an eastern highlands zone: 7°-14°N and 36.5°-40.5°E(Figure 1a and b) including the states of Amhara and Oromia.This zone overlaps with the leading cluster of satellite vegetation fraction (Tucker et al 2005) which has a uni-modal peak from July to October that lags 1-month behind rainfall.The mean annual cycle of rainfall (Figure 1c) rises above 100 mm/month from June to September, peaking above 200 mm/month in July and August.Hence the season of interest extends across these four months.
Summer rainfall forecasts in April would be most useful for planning purposes, so predictors are drawn from the preceding December to March (winter) season.The CHIRPS monthly rainfall is area-averaged to create a time series and then applied to search for key predictors in fields of sea surface temperature from the Hadley Centre (Kennedy et al., 2011) and 850 mb winds from the European Centre for Medium-range Weather Forecasts (ECMWF) reanalysis (Dee et al., 2011).Exploratory analysis determined that the eastern hemisphere held most of the climate signal: 20°S-35°N and 0°-180°E.All time series were converted to standardized departures and linearly detrended.

Methods of analysis
Predictors were assembled from correlation maps analyzed for December-March in respect of detrended June-September rainfall, with values < 80% significance masked.For extraction of predictor time series, the area for averaging should exceed 15° latitude × 20° longitude in size.The candidate predictor pool was four over a training period of 33 years: 1981 to 2014.A statistical algorithm was formulated via backward stepwise linear regression onto the target time series.Initially all predictors were included and their partial correlation was evaluated.Those with lower significance (or colinearity) were screened out and the algorithm was re-calculated from the remaining variables.An optimal fit was reached with three predictors, however when repeated for the second half of the time (1998-2014), one predictor dropped out (Indian SST 45°-145°E, 5°S-25°N), leaving only the low level winds near the Maritime Continent as contributors: Indian meridional wind 45°-135°E, 7°S-12°N and Pacific zonal wind 120°-170°E, 15°S-15°N.
The performance of the multivariate linear algorithm was evaluated by r 2 fit, adjusted for the number of predictors.Conservatively assuming 16 degrees of freedom, the Pearson product-moment r 2 fit should exceed 35% for statistical significance at 99% confidence.Predicted vs observed scatterplots were analyzed for slope, tercile hits and outliers.The algorithm stability was validated by removing the first and last 7 years and comparing differences in r 2 fit and predictor coefficients.The algorithm is developed from detrended time series to minimize the effect of climate change across the 33 year record.In an operational situation, detrending is believed to have little influence.
To study the interaction of the Asian winter monsoon and Pacific ENSO over the Maritime Continent, precursor and simultaneous composites were analyzed for Ethiopian dry and wet seasons corresponding with the 850 hPa wind predictors.Dry seasons include 1982, 1987, 1997, 2002, 2014, and wet seasons include: 1988, 1996, 1998, 1999, 2000.Composites of dry minus wet were analyzed for NCEPv2 winds, vertical motion and specific humidity (Kanamitsu et al., 2002), and satellite GPCPv2 rainfall (Adler et al., 2003) and SST.These datasets are used because NCEP winds are available in vertical section and the GPCP rainfall covers both land and ocean.Similarly composites for the upper ocean were analyzed longitudinally in the 5°-15°N band using NOAA ocean reanalysis (Behringer, 2007).To quantify the local response of Ethiopian rainfall to ENSO, correlation maps (5°-15°N, 35°-42°E) were analyzed with respect to June-September Nino3.4Pacific SST index (170°-120°W, 5°S-5°N) and with Jul-Oct satellite vegetation fraction (Tucker et al., 2005).Lag-correlations were calculated based on the time series of the Ethiopian summer rainfall and Pacific Nino3.4 SST index from -6 (December-March) to +4 months.

Targets and correlation maps
The correlation of June-September rain onto regional SST fields in the preceding December-March (Figure 2a) reflects anomalous warm conditions in the north Indian Ocean and northwest Pacific (China Sea), encompassing the zone east of the target area.The corresponding 850 hPa V wind correlation field (Figure 2b) shows significant values over the equatorial Indian Ocean and Maritime respect of wet conditions in Ethiopia six months later.
The time series of observed and statistically predicted rainfall is given in Figure 3a, scatterplots are provided in Figure 3b and c To check for stability of the forecast algorithm, the same 2-predictor model is fitted to the observed rainfall time series after removing the first and last 7 years.The r 2 fit remains constant: 56%, but the coefficients change.The Ind V coefficient is +.37 recent +.57 past, while the Pac U is -.41 recent -.27 past.Hence the Pacific (Indian) influence grows (shrinks) with time, but predictability is steady.In the section below, we switch our analysis from patterns favouring above normal rainfall to those bringing drought conditions to Ethiopia.Composite maps of dry (1982,1987,1997,2002,2014) minus wet (1988,1996,1998,1999,2000) December-March and June-September 850 hPa winds, satellite rainfall and sea temperature and vertical motion are analyzed.The atmospheric fields (Figure 4a to d) in antecedent and simultaneous times show that zonal winds east of the Maritime Continent constitute a significant and stable signal dividing a convective Pacific (El Niño) zone from a dry Indian Ocean.The dry minus wet composite in December-March (year -1) exhibits cooler SST over the northwest Pacific (Figure 5a and c), extending to the NW Indian Ocean and SE Atlantic.The cooler SST correspond with atmospheric subsidence over the China Sea that shifts to the Maritime Continent by summer (Figure 5b and d) and spreads westward across the Indian Ocean to Africa.The analysis is extended by composite height sections averaged 5°-15°N that show upper westerlies over Africa subside toward a region of low humidity over the Maritime Continent in the preceding winter (Figure 6a to b).Meridional winds are from north (< -1 m/s) from 120°E to 140°E in the 850 to 600 hPa layer, and constitute an enhanced Asian winter monsoon outflow.The dry minus wet scenario evolves to summer with a low level westerly anomaly over the west Pacific, consistent with a developing El Niño.A corresponding dipole develops in the humidity field (Figure 6c and d): moist-Pacific, dry-Indian.Easterly winds on the east African escarpment derive from the divergent ENSO circulation over the Indian Ocean, and correspond with expansion of the dry zone from the Maritime Continent to North Africa, that inhibits moisture transport from the Congo Basin to Ethiopia (Viste and Sorteberg, 2013).

Composites of maritime continent ENSO circulation
A complementary depth section of ocean temperatures and zonal circulation is given in Figure 7a and b

Figure 1 .cFigure 2 .
Figure 1.(a) Loading pattern of first mode of vegetation fraction 1981-2013 shaded light to dark green from zero to one standard deviation with state boundaries and target area.(b) Grain crop growing areas from FEWS shaded yellow to hatch orange with increasing yield, with key cities and rivers.(c) Mean annual cycle of CHIRPS rainfall over the target area (bold) and quintiles.

Figure 3 .
Figure 3. (a) Time series of observed Jun-Sep target rainfall and 2-predictor model rainfall; scatterplot of linear multi-variate regression of predicted and observed rainfall using (b) 3 predictor model, and (c) two predictor model, with trend fit.

Figure 4 .
Figure 4. Composite maps of dry minus wet: (a) preceding Dec-Mar 850 hPa winds and (b) satellite GPCP rainfall, and (c, d) same for Jun-Sep rainy season.Vectors in (a) highlight Asian winter monsoon outflow that anticipates drought.Shading in (a, c) and (b, d) is consistent, and areas with minor differences are unshaded.

Figure 5 .Figure 6 .Figure 7 .Figure 8 .
Figure 5. Composite maps of dry minus wet: (a) preceding Dec-Mar satellite SST and (b) 500 hPa omega, and (c, d) same for Jun-Sep rainy season.Shading in (a, c) and (b, d) is consistent, and areas with minor differences are unshaded.