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

Abstract

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