Editing IPCC:AR6/SRCCL/Chapter-3 (section)

=== 3.2.2 Attribution of desertification ===

<div id="section-3-2-2-attribution-of-desertification-block-1"></div>

Desertification is a result of complex interactions within coupled social-ecological systems. Thus, the relative contributions of climatic, anthropogenic and other drivers of desertification vary depending on specific socio-economic and ecological contexts. The high natural climate variability in dryland regions is a major cause of vegetation changes but does not necessarily imply degradation. Drought is not degradation as the land productivity may return entirely once the drought ends (Kassas 1995 <sup>[[#fn:r410|410]]</sup> ). However, if droughts increase in frequency, intensity and/or duration they may overwhelm the vegetation’s ability to recover (ecosystem resilience, Prince et al. 2018), causing degradation. Assuming a stationary climate and no human influence, rainfall variability results in fluctuations in vegetation dynamics which can be considered temporary, as the ecosystem tends to recover with rainfall, and desertification does not occur (Ellis 1995 <sup>[[#fn:r411|411]]</sup> ; Vetter 2005 <sup>[[#fn:r412|412]]</sup> ; von Wehrden et al. 2012 <sup>[[#fn:r413|413]]</sup> ). Climate change on the other hand, exemplified by a non-stationary climate, can gradually cause a persistent change in the ecosystem through aridification and CO <sub>2</sub> changes. Assuming no human influence, this ‘natural’ climatic version of desertification may take place rapidly, especially when thresholds are reached (Prince et al. 2018 <sup>[[#fn:r414|414]]</sup> ), or over longer periods of time as the ecosystems slowly adjust to a new climatic norm through progressive changes in the plant community composition. Accounting for this climatic variability is required before attributions to other causes of desertification can be made.

For attributing vegetation changes to climate versus other causes, rain use efficiency (RUE – the change in net primary productivity (NPP) per unit of precipitation) and its variations in time have been used (Prince et al. 1998 <sup>[[#fn:r415|415]]</sup> ). Global applications of RUE trends to attribute degradation to climate or other (largely human) causes have been performed by Bai et al. (2008) <sup>[[#fn:r416|416]]</sup> and Le et al. (2016) <sup>[[#fn:r417|417]]</sup> (Section 3.2.1.1). The RESTREND (residual trend) method analyses the correlation between annual maximum NDVI (or other vegetation index as a proxy for NPP) and precipitation by testing accumulation and lag periods for the precipitation (Evans and Geerken 2004 <sup>[[#fn:r418|418]]</sup> ). The identified relationship with the highest correlation represents the maximum amount of vegetation variability that can be explained by the precipitation, and corresponding RUE values can be calculated. Using this relationship, the climate component of the NDVI time series can be reconstructed, and the difference between this and the original time series (the residual) is attributed to anthropogenic and other causes.

The RESTREND method, or minor variations of it, have been applied extensively. Herrmann and Hutchinson (2005) <sup>[[#fn:r419|419]]</sup> concluded that climate was the dominant causative factor for widespread greening in the Sahel region from 1982–2003, and anthropogenic and other factors were mostly producing land improvements or no change. However, pockets of desertification were identified in Nigeria and Sudan. Similar results were also found from 1982–2007 by Huber et al. (2011) <sup>[[#fn:r420|420]]</sup> . Wessels et al. (2007) <sup>[[#fn:r421|421]]</sup> applied RESTREND to South Africa and showed that RESTREND produced a more accurate identification of degraded land than RUE alone. RESTREND identified a smaller area undergoing desertification due to non-climate causes compared to the NDVI trends. Liu et al. (2013) <sup>[[#fn:r430|430]]</sup> extended the climate component of RESTREND to include temperature and applied this to VOD observations of the cold drylands of Mongolia. They found the area undergoing desertification due to non-climatic causes is much smaller than the area with negative VOD trends. RESTREND has also been applied in several other studies to the Sahel (Leroux et al. 2017 <sup>[[#fn:r422|422]]</sup> ), Somalia (Omuto et al. 2010) <sup>[[#fn:r423|423]]</sup> , West Africa (Ibrahim et al. 2015) <sup>[[#fn:r424|424]]</sup> , China (Li et al. 2012 <sup>[[#fn:r425|425]]</sup> ; Yin et al. 2014 <sup>[[#fn:r426|426]]</sup> ), Central Asia (Jiang et al. 2017 <sup>[[#fn:r427|427]]</sup> ), Australia (Burrell et al. 2017 <sup>[[#fn:r428|428]]</sup> ) and globally (Andela et al. 2013 <sup>[[#fn:r429|429]]</sup> ). In each of these studies the extent to which desertification can be attributed to climate versus other causes varies across the landscape.

These studies represent the best regional, remote-sensing based attribution studies to date, noting that RESTREND and RUE have some limitations (Higginbottom and Symeonakis 2014 <sup>[[#fn:r431|431]]</sup> ). Vegetation growth (NPP) changes slowly compared to rainfall variations and may be sensitive to rainfall over extended periods (years), depending on vegetation type. Detection of lags and the use of weighted antecedent rainfall can partially address this problem, though most studies do not do this. The method addresses changes since the start of the time series; it cannot identify whether an area is already degraded at the start time. It is assumed that climate, particularly rainfall, is a principal factor in vegetation change which may not be true in more humid regions.

Another assumption in RESTREND is that any trend is linear throughout the period examined. That is, there are no discontinuities (break points) in the trend. Browning et al. (2017) <sup>[[#fn:r432|432]]</sup> have shown that break points in NDVI time series reflect vegetation changes based on long-term field sites. To overcome this limitation, Burrell et al. (2017) <sup>[[#fn:r433|433]]</sup> introduced the Time Series Segmentation-RESTREND (TSS-RESTREND) which allows a breakpoint or turning point within the period examined (Figure 3.7). Using TSS-RESTREND over Australia they identified more than double the degrading area than could be identified with a standard RESTREND analysis. The occurrence and drivers of abrupt change (turning points) in ecosystem functioning were also examined by Horion et al. (2016) <sup>[[#fn:r434|434]]</sup> over the semi-arid Northern Eurasian agricultural frontier. They combined trend shifts in RUE, field data and expert knowledge, to map environmental hotspots of change and attribute them to climate and human activities. One-third of the area showed significant change in RUE, mainly occurring around the fall of the Soviet Union (1991) or as the result of major droughts. Recent human-induced turning points in ecosystem functioning were uncovered around Volgograd (Russia) and around Lake Balkhash (Kazakhstan), attributed to recultivation, increased salinisation, and increased grazing.

Attribution of vegetation changes to human activity has also been done within modelling frameworks. In these methods ecosystem models are used to simulate potential natural vegetation dynamics, and this is compared to the observed state. The difference is attributed to human activities. Applied to the Sahel region during the period of 1982–2002, it showed that people had a minor influence on vegetation changes (Seaquist et al. 2009 <sup>[[#fn:r435|435]]</sup> ). Similar model/observation comparisons performed globally found that CO <sub>2</sub> fertilisation was the strongest forcing at global scales, with climate having regionally varying effects (Mao et al. 2013 <sup>[[#fn:r436|436]]</sup> ; Zhu et al. 2016 <sup>[[#fn:r437|437]]</sup> ). Land-use/ land-cover change was a dominant forcing in localised areas. The use of this method to examine vegetation changes in China (1982–2009) attributed most of the greening trend to CO <sub>2</sub> fertilisation and nitrogen (N) deposition (Piao et al. 2015). However in some parts of northern and western China, which includes large areas of drylands, Piao et al. (2015) <sup>[[#fn:r438|438]]</sup> found climate changes could be the dominant forcing. In the northern extratropical land surface, the observed greening was consistent with increases in greenhouse gases (notably CO <sub>2</sub> ) and the related climate change, and not consistent with a natural climate that does not include anthropogenic increase in greenhouse gases (Mao et al. 2016 <sup>[[#fn:r439|439]]</sup> ). While many studies found widespread influence of CO <sub>2</sub> fertilisation, it is not ubiquitous; for example, Lévesque et al. (2014) found little response to CO <sub>2</sub> fertilisation in some tree species in Switzerland/northern Italy.

Using multiple extreme-event attribution methodologies, Uhe et al. (2018) <sup>[[#fn:r440|440]]</sup> shows that the dominant influence for droughts in eastern Africa during the October–December ‘short rains’ season is the prevailing tropical SST patterns, although temperature trends mean that the current drought conditions are hotter than they would have been without climate change. Similarly, Funk et al. (2019) <sup>[[#fn:r441|441]]</sup> found that 2017 March–June East African drought was influenced by Western Pacific SST, with high SST conditions attributed to climate change.

There are numerous local case studies on attribution of desertification, which use different periods, focus on different land uses and covers, and consider different desertification processes. For example, two-thirds of the observed expansion of the Sahara Desert from 1920–2003 has been attributed to natural climate cycles (the cold phase of Atlantic Multi-Decadal Oscillation and Pacific Decadal Oscillation) (Thomas and Nigam 2018 <sup>[[#fn:r442|442]]</sup> ). Some studies consider drought to be the main driver of desertification in Africa (e.g., Masih et al. 2014 <sup>[[#fn:r443|443]]</sup> ). However, other studies suggest that although droughts may contribute to desertification, the underlying causes are human activities (Kouba et al. 2018 <sup>[[#fn:r444|444]]</sup> ). Brandt et al. (2016a) found that woody vegetation trends are negatively correlated with human population density. Changes in land use, water pumping and flow diversion have enhanced drying of wetlands and salinisation of freshwater aquifers in Israel (Inbar 2007 <sup>[[#fn:r445|445]]</sup> ). The dryland territory of China has been found to be very sensitive to both climatic variations and land-use/land-cover changes (Fu et al. 2000 <sup>[[#fn:r446|446]]</sup> ; Liu and Tian 2010 <sup>[[#fn:r447|447]]</sup> ; Zhao et al. 2013, 2006 <sup>[[#fn:r448|448]]</sup> ). Feng et al. (2015) shows that socio-economic factors were dominant in causing desertification in north Shanxi, China, between 1983 and 2012, accounting for about 80% of desertification expansion. Successful grass establishment has been impeded by overgrazing and nutrient depletion leading to the encroachment of shrubs into the northern Chihuahuan Desert (USA) since the mid-19th century (Kidron and Gutschick 2017 <sup>[[#fn:r449|449]]</sup> ). Human activities led to rangeland degradation in Pakistan and Mongolia during 2000–2011 (Lei et al. 2011 <sup>[[#fn:r450|450]]</sup> ). More equal shares of climatic (temperature and precipitation trends) and human factors were attributed for changes in rangeland condition in China (Yang et al. 2016 <sup>[[#fn:r451|451]]</sup> ).

This kaleidoscope of local case studies demonstrates how attribution of desertification is still challenging for several reasons. Firstly, desertification is caused by an interaction of different drivers which vary in space and time. Secondly, in drylands, vegetation reacts closely to changes in rainfall so the effect of rainfall changes on biomass needs to be ‘removed’ before attributing desertification to human activities. Thirdly, human activities and climatic drivers impact vegetation/ ecosystem changes at different rates. Finally, desertification manifests as a gradual change in ecosystem composition and structure (e.g., woody shrub invasion into grasslands). Although initiated at a limited location, ecosystem change may propagate throughout an extensive area via a series of feedback mechanisms. This complicates the attribution of desertification to human and climatic causes, as the process can develop independently once started.

<div id="section-3-2-2-attribution-of-desertification-block-2"></div>

<span id="figure-3.7"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 3.7'''

<span id="the-drivers-of-dryland-vegetation-change.-the-mean-annual-change-in-ndvimax-between-1982-and-2015-see-figure-3.6-for-total-change-using-global-inventory-modelling-and-mapping-studies-ndvi3g-v1-dataset-attributable-toaco2-fertilisationbclimate-and-c-land-use.-the-change-attributable-to-co2-fertilisation-was-calculated-using-the-co2-fertilisation-relationship-described-in-franks"></span>
<!-- IMG CAPTION -->
'''The drivers of dryland vegetation change. The mean annual change in NDVImax between 1982 and 2015 (see Figure 3.6 for total change using Global Inventory Modelling and Mapping Studies NDVI3g v1 dataset) attributable to(a)CO2 fertilisation(b)climate and (c) land use. The change attributable to CO2 fertilisation was calculated using the CO2 fertilisation relationship described in Franks […]'''
<!-- IMG FILE -->
[[File:1575a316e0081c3ad0d78e507d3e945c Figure-3.7.png]]

The drivers of dryland vegetation change. The mean annual change in NDVImax between 1982 and 2015 (see Figure 3.6 for total change using Global Inventory Modelling and Mapping Studies NDVI3g v1 dataset) attributable to(a)CO <sub>2</sub> fertilisation(b)climate and (c) land use. The change attributable to CO <sub>2</sub> fertilisation was calculated using the CO <sub>2</sub> fertilisation relationship described in Franks et al. 2013 <sup>[[#fn:r1793|1793]]</sup> . The Time Series Segmented Residual Trends (TSS-RESTREND) method (Burrell et al. 2017 <sup>[[#fn:r1794|1794]]</sup> ) applied to the CO <sub>2</sub> -adjusted NDVI was used to separate Climate and Land Use. A multi-climate dataset ensemble was used to reduce the impact of dataset errors (Burrell et al. 2018 <sup>[[#fn:r1795|1795]]</sup> ). Non-dryland regions (aridity index &gt;0.65) are masked in dark grey. Areas where the change did not meet the multi-run ensemble significance criteria, or are smaller than the error in the sensors (±0.00001) are masked in white

<!-- END IMG -->
<div id="section-3-2-2-attribution-of-desertification-block-3"></div>

Rasmussen et al. (2016) <sup>[[#fn:r452|452]]</sup> studied the reasons behind the overall lack of scientific agreement in trends of environmental changes in the Sahel, including their causes. The study indicated that these are due to differences in conceptualisations and choice of indicators, biases in study site selection, differences in methods, varying measurement accuracy, differences in time and spatial scales. High-resolution, multi-sensor airborne platforms provide a way to address some of these issues (Asner et al. 2012 <sup>[[#fn:r453|453]]</sup> ).

The major conclusion of this section is that, with all the shortcomings of individual case studies, relative roles of climatic and human drivers of desertification are location-specific and evolve over time ( ''high confidence'' ). Biophysical research on attribution and socio-economic research on drivers of land degradation have long studied the same topic, but in parallel, with little interdisciplinary integration. Interdisciplinary work to identify typical patterns, or typologies, of such interactions of biophysical and human drivers of desertification (not only of dryland vulnerability), and their relative shares, done globally in comparable ways, will help in the formulation of better informed policies to address desertification and achieve land degradation neutrality.

<span id="desertification-feedbacks-to-climate"></span>