Open AccessEthiopian Highlands Crop-Climate Prediction: 1979–2009
Author(s) -
Mark R. Jury
Publication year - 2013
Publication title -
journal of applied meteorology and climatology
Language(s) - English
Resource type - Journals
SCImago Journal Rank - 1.079
H-Index - 134
eISSN - 1558-8432
pISSN - 1558-8424
DOI - 10.1175/jamc-d-12-0139.1
Subject(s) - climatology , subtropics , sea surface temperature , environmental science , latitude , principal component analysis , multivariate statistics , tropics , wind speed , atmospheric sciences , geography , meteorology , geology , mathematics , statistics , fishery , biology , geodesy
This study compares different methods of predicting crop-related climate in the Ethiopian highlands for the period 1979–2009. A target index (ETH4) is developed as an average of four variables in the June–September season—rainfall, rainfall minus evaporation, estimated latent heat flux, and vegetation, following correlation with crop yields at Melkassa, Ethiopia (8.4°N, 39.3°E, 1550 m elevation). Predictors are drawn from gridded near-global fields of surface temperature, surface air pressure, and 200-hPa zonal wind in the preceding December–March season. Prediction algorithms are formulated by stepwise multivariate regression. The first set of predictors derive from objective principal component (PC) time scores with tropical loading patterns, and the second set is based on key areas determined from correlation with the target index. The second PC of upper zonal wind reveals a tropical–subtropical dipole that is correlated with ETH4 at two-season lead time (correlation coefficient r = −0.53). Point-to-field regression maps show high-latitude signals in surface temperature (positive in North America and negative in Eurasia) and air pressure (negative in the North Pacific Ocean and positive in the South Pacific). Upper zonal winds are most strongly related with ETH4 over the tropical Pacific and Africa at two-season lead time. The multivariate algorithm that is based on PC predictors has an adjusted r 2 fit of 0.23, and the algorithm using key-area predictors achieves r 2 = 0.37. In comparison, numerical model forecasts reach r 2 = 0.33 for ECMWF simulations but are low for other models. The statistical results are specific to the ETH4 index, which is a climate proxy for crop yields in the Ethiopian highlands.