Evaluation of the trend detection methods used in this study

Overall, the different trend techniques applied to the AVHRR NDVI3g time-series demonstrated similar spatial patterns of change in vegetation productivity over SSA. While these results were not validated using ground observations, the replication of results across methods provides some confidence in the trends in vegetation over the last 34 years. In addition, the analysis presented here demonstrated that the use of a range of different approaches is required to obtain a complete understanding of the spatial and temporal variability of vegetation productivity and seasonality.

Since the linear trend analyses are based on a limited time-series it is often difficult to detect vegetation change occurring within the first or last two years of the time series (Wessels et al. 2007a). This implies that the period of assessment also has a large influence on the detection of linear trends where the addition of years to the time-series could change the trend direction. The piecewise linear trends technique (BFAST) added considerable value in this study by demonstrating that trends in vegetation are often more complex with abrupt changes and shifts in vegetation productivity occurring in the NDVI3g time-series. The detection of breakpoints and trends in multiple shorter segments allowed for an improved understanding of temporal changes in NDVI3g as well as the potential drivers of these shifts in vegetation.

A challenge still remains in attempting to understand change is to distinguish between statistically significant changes and changes that are practically significant at the local scale (Forkel et al. 2013). Field-based validation of shifts in vegetation over a 34-year period across SSA remains unfeasible. A common approach, similar to that taken in this paper, is to use regional or global related publications to validate trend analyses but these are often insufficient since many of these studies utilise the same NDVI data and similar trend detection approaches. Future research is required to conduct smaller scale assessments over the regions, such as Central African Republic, which demonstrate substantial shifts in vegetation productivity and seasonality in order to assist in the interpretation of finer scale changes in vegetation dynamics and associated climatic drivers. The MODIS EVI time-series could also be used as a comparative time series to further understanding changes in both vegetation productivity and seasonality over SSA.

5.5

Conclusion

The purpose of this chapter was to assess changes in vegetation productivity and seasonality over Africa using the longest continuous temporal record available and a range of range previously published trend detection techniques. The research extends the results identified by previous studies and presents an important synthesis of changes in vegetation productivity over Africa over the last 32 years.

Evidence for a number of important spatial and temporal patterns of change in vegetation productivity over SSA is presented, which are generally consistent with independently reported long-term trends. Insights from the analysis include:

• Spatial patterns of trends provide evidence for latitudinal changes in productivity with significant increases in the northern hemisphere tropics and sub-tropics of both the southern hemisphere and significant decreases over the southern hemisphere tropics.

• Greening accounts for nearly half of the observed changes in vegetation productivity over SSA.

• In areas that have experienced significant increases in productivity tend to be associated with increases in the length of the season as a result of an earlier start and later end date of the season. This finding could potentially provide a first indication of the impact of recent climate change over these regions.

• Over the Sahel a consistent greening trend provides evidence for the recovery of vegetation from drought conditions that affected the region in the early to mid-1980s. • While acknowledging that NDVI tends to saturate in densely vegetated areas,

patterns of increasing productivity and length of season were observed over the parts of West and central Africa suggesting that forests are expanding into the surrounding savanna regions. This expansion is interpreted as a response to increases in rainfall combined with increases in atmospheric CO2 concentration as well as reforestation efforts in some countries.

• Patterns of gradually increasing productivity were detected over the south western Cape, eastern coastline, and north-east extent of South Africa which are consistent with the observed patterns of bush encroachment and expansion of alien invasive species in these regions.

• Declines in vegetation productivity were observed over the majority of southern Africa most likely due to land-use changes and reduction in woody vegetation as a result of non-climatic factors including the clearing for agriculture and population expansions. • Differing patterns of greening and browning were observed over East Africa

suggesting the role of local-scale factors driving vegetation change.

While it is possible, to some extent, to attribute an underlying process or driver of change to a particular type of trend in this study, simulations of DGVM will assist in understand the influence of changing rainfall patterns and increases in atmospheric CO2 in the context of rapidly changing land-use patterns in the region. Further analysis is needed to understand and identify the climatic mechanisms behind the observed trends in the content of local scale land-use change.

Chapter 6: