Forecasting vegetation condition with a Bayesian auto-regressive distributed lags (BARDL) model

Abstract. Droughts form a large part of climate- or weather-related disasters reported globally. In Africa, pastoralists living in the arid and semi-arid lands (ASALs) are worse affected. Prolonged dry spells that cause vegetation stress these regions have resulted loss income livelihoods. To curb this, global initiatives like Paris Agreement United Nations recognised need to establish early warning systems (EWSs) save lives Existing EWSs use combination satellite earth observation (EO)-based biophysical indicators condition index (VCI) socio-economic factors measure monitor droughts. Most rely on expert knowledge estimating upcoming drought conditions without using forecast models. Recent research has shown robust algorithms auto-regression, Gaussian processes, artificial neural networks can provide very skilled models for forecasting at short- medium-range lead times. However, enable preparedness action, forecasts with longer time needed. previous paper, process model an auto-regression were used VCI pastoral communities Kenya. The objective this was build work by developing improved premise is controlled precipitation soil moisture; thus, we Bayesian auto-regressive distributed lag (BARDL) modelling approach, which enabled us include effects lagged information from moisture improve forecasting. results showed ?2-week gain range compared univariate as baseline. R2 scores ARDL 0.94, 0.85, 0.74, model's 0.88, 0.77, 0.65 6-, 8-, 10-week time, respectively.