Abstract
The annual rainfall cycle in the Caribbean is characterized by a bimodal pattern with peaks in the late spring (“early rainfall season”) and late summer (“late rainfall season”) with a mid-summer minimum (“mid-summer drought”). The time average rainfall pattern during the early rainfall season reveals a distinct southwest to northeast spatial pattern , known as the Caribbean rain-belt, that is similar to other northern hemisphere subtropical rain-belts. A series of Caribbean farmer interviews guided my decision to focus on the dynamics and evolution of the Caribbean rain-belt. Results from farmer interviews reveal that their livelihoods are more vulnerable to variability in the timing and amount of the early season rains rather than variability in the mid-summer drying. Therefore, there is a strong social and economic relevance to understand rainfall dynamics during the Caribbean early rainfall season. The atmospheric dynamics that contribute to the Caribbean rain-belt are diagnosed from the quasi-geostrophic omega equation from daily observations. Forcing for ascent at the upper troposphere is supported by positive zonal wind at 200hPa and jet streaks, while positive temperature advection from the tropics at 500hPa provides forcing for ascent in the mid-troposphere. Moisture availability for the Caribbean rain-belt is regulated by local sea surface temperature and by moisture advection from the tropics in the lower troposphere. The forcing for ascent weakens throughout the Caribbean and strengthens in the North Atlantic during the mid-summer drought period. Therefore, the mid-summer drought may be diagnosed in terms of weakened uplift dynamics. A liner inverse model (LIM) is built from OLR, u200, and u850 using 15 degree longitude space channels rather than the traditional reduced EOF space. A physical space based channel LIM to predict tropically propagating OLR anomalies shows positive OLR hindcast prediction skill relative to climatology up to 3 weeks lead time. Confidence in the construction and sub-seasonal prediction skill of the physical based channel LIM allows for the construction of a similar LIM to predict the time evolution of the dynamical parameters that shape the Caribbean rain-belt. The Caribbean LIM is trained between1982-2010 from an anomalous state vector of ingredients identified from the Caribbean rain-belt diagnostic section and includes the daily anomaly time series of SST, OLR, u200, u850, and v200. SST anomalies have the highest prediction skill over a 3 week lead time, while the positive prediction skill of OLR is muted after 5 days. Prediction skill for the wind channels of the Caribbean LIM remains positive up to one week lead time. The simultaneous and time lagged covariance statistics produce a dynamical operator matrix that describes the internal dynamics between each parameter and identifies SST anomalies to exert the most influence on OLR anomalies. In addition, SST anomalies have the highest signal to noise ratio of each Caribbean LIM state vector channel. Caribbean LIM prediction skill is dependent on how well the state vector represents the linear dynamical system diagnosed from Caribbean rain-bely conceptual model. Low LIM prediction skill may be due to underestimating the non-linear system forcing or from omission of relevant parameters in the state vector.