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

==== 3.2.1.1 Global scale ====

<div id="section-3-2-1-1-global-scale-block-1"></div>

Complex human–environment interactions, coupled with biophysical, social, economic and political factors unique to any given location, render desertification difficult to map at a global scale (Cherlet et al. 2018 <sup>[[#fn:r240|240]]</sup> ). Early attempts to assess desertification focused on expert knowledge in order to obtain global coverage in a cost-effective manner. '''Expert judgement''' continues to play an important role because degradation remains a subjective feature whose indicators are different from place to place (Sonneveld and Dent 2007 <sup>[[#fn:r241|241]]</sup> ). GLASOD (Global Assessment of Human-induced Soil Degradation) estimated nearly 2000 million hectares (Mha) (15.3% of the total land area) had been degraded by the early 1990s since the mid-20th century. GLASOD was criticised for perceived subjectiveness and exaggeration (Helldén and Tottrup 2008 <sup>[[#fn:r242|242]]</sup> ; Sonneveld and Dent 2007 <sup>[[#fn:r243|243]]</sup> ). Dregne and Chou (1992) <sup>[[#fn:r244|244]]</sup> found 3000 Mha in drylands (i.e. about 50% of drylands) were undergoing degradation. Significant improvements have been made through the efforts of WOCAT (World Overview of Conservation Approaches and Technologies), LADA (Land Degradation Assessment in Drylands) and DESIRE (Desertification Mitigation and Remediation of Land) who jointly developed a mapping tool for participatory expert assessment, with which land experts can estimate current area coverage, type and trends of land degradation (Reed et al. 2011 <sup>[[#fn:r245|245]]</sup> ).

A number of studies have used '''satellite-based remote sensing''' to investigate long-term changes in the vegetation and thus identify parts of the drylands undergoing desertification. Satellite data provides information at the resolution of the sensor, which can be relatively coarse (up to 25 km), and interpretations of the data at sub-pixel levels are challenging. The most widely used remotely sensed vegetation index is the NDVI, providing a measure of canopy greenness that is related to the quantity of standing biomass (Bai et al. 2008 <sup>[[#fn:r246|246]]</sup> ; de Jong et al. 2011 <sup>[[#fn:r247|247]]</sup> ; Fensholt et al. 2012 <sup>[[#fn:r248|248]]</sup> ; Andela et al. 2013 <sup>[[#fn:r249|249]]</sup> ; Fensholt et al. 2015 <sup>[[#fn:r250|250]]</sup> ; Le et al. 2016 <sup>[[#fn:r251|251]]</sup> ) (Figure 3.5). A main challenge associated with NDVI is that although biomass and productivity are closely related in some systems, they can differ widely when looking across land uses and ecosystem types, giving a false positive in some instances (Pattison et al. 2015 <sup>[[#fn:r252|252]]</sup> ; Aynekulu et al. 2017 <sup>[[#fn:r253|253]]</sup> ). For example, bush encroachment in rangelands and intensive monocropping with high fertiliser application gives an indication of increased productivity in satellite data though these could be considered as land degradation. According to this measure there are regions undergoing desertification, however the drylands are greening on average (Figure 3.6).

<div id="section-3-2-1-1-global-scale-block-2"></div>

<span id="figure-3.6"></span>
<!-- START IMG -->
<!-- IMG TITLE -->
'''Figure 3.6'''

<span id="trend-in-the-annual-maximum-ndvi-19822015-global-inventory-modelling-and-mapping-studies-ndvi3g-v1-calculated-using-the-theilsen-estimator-which-is-a-median-based-estimator-and-is-robust-to-outliers.-non-dryland-regions-aridity-index-0.65-are-masked-in-grey"></span>
<!-- IMG CAPTION -->
'''Trend in the annual maximum NDVI 1982–2015 (Global Inventory Modelling and Mapping Studies NDVI3g v1) calculated using the Theil–Sen estimator which is a median based estimator, and is robust to outliers. Non-dryland regions (aridity index &gt;0.65) are masked in grey'''
<!-- IMG FILE -->
[[File:8db6a9b42f7bb9dc07a519c001b7e633 Figure-3.6.png]]

Trend in the annual maximum NDVI 1982–2015 (Global Inventory Modelling and Mapping Studies NDVI3g v1) calculated using the Theil–Sen estimator which is a median based estimator, and is robust to outliers. Non-dryland regions (aridity index &gt;0.65) are masked in grey

<!-- END IMG -->
<div id="section-3-2-1-1-global-scale-block-3"></div>

A simple linear trend in NDVI is an unsuitable measure for dryland degradation for several reasons (Wessels et al. 2012 <sup>[[#fn:r254|254]]</sup> ; de Jong et al. 2013 <sup>[[#fn:r255|255]]</sup> ; Higginbottom and Symeonakis 2014 <sup>[[#fn:r256|256]]</sup> ; Le et al. 2016 <sup>[[#fn:r257|257]]</sup> ). NDVI is strongly coupled to precipitation in drylands where precipitation has high inter-annual variability. This means that NDVI trend can be dominated by any precipitation trend and is sensitive to wet or dry periods, particularly if they fall near the beginning or end of the time series. Degradation may only occur during part of the time series, while NDVI is stable or even improving during the rest of the time series. This reduces the strength and representativeness of a linear trend. Other factors such as CO <sub>2</sub> fertilisation also influence the NDVI trend. Various techniques have been proposed to address these issues, including the residual trends (RESTREND) method to account for rainfall variability (Evans and Geerken 2004 <sup>[[#fn:r258|258]]</sup> ), time-series break point identification methods to find major shifts in the vegetation trends (de Jong et al. 2013 <sup>[[#fn:r259|259]]</sup> ; Verbesselt et al. 2010a <sup>[[#fn:r260|260]]</sup> ), and methods to explicitly account for the effect of CO <sub>2</sub> fertilisation (Le et al. 2016 <sup>[[#fn:r261|261]]</sup> ).

Using the RESTREND method, Andela et al. (2013) <sup>[[#fn:r262|262]]</sup> found that human activity contributed to a mixture of improving and degrading regions in drylands. In some locations these regions differed substantially from those identified using the NDVI trend alone, including an increase in the area being desertified in southern Africa and northern Australia, and a decrease in southeast and western Australia and Mongolia. De Jong et al. (2013) <sup>[[#fn:r263|263]]</sup> examined the NDVI time series for major shifts in vegetation activity and found that 74% of drylands experienced such a shift between 1981 and 2011. This suggests that monotonic linear trends are unsuitable for accurately capturing the changes that have occurred in the majority of the drylands. Le et al. (2016) <sup>[[#fn:r264|264]]</sup> explicitly accounted for CO <sub>2</sub> fertilisation effect and found that the extent of degraded areas in the world is 3% larger when compared to the linear NDVI trend.

Besides NDVI, there are many vegetation indices derived from satellite data in the optical and infrared wavelengths. Each of these datasets has been derived to overcome some limitation in existing indices. Studies have compared vegetation indices globally (Zhang et al. 2017 <sup>[[#fn:r265|265]]</sup> ) and specifically over drylands (Wu 2014 <sup>[[#fn:r266|266]]</sup> ). In general, the data from these vegetation indices are available only since around 2000, while NDVI data is available since 1982. With less than 20 years of data, the trend analysis remains problematic with vegetation indices other than NDVI. However, given the various advantages in terms of resolution and other characteristics, these newer vegetation indices will become more useful in the future as more data accumulates.

A major shortcoming of these studies based on vegetation datasets derived from satellite sensors is that they do not account for changes in vegetation composition, thus leading to inaccuracies in the estimation of the extent of degraded areas in drylands. For example, drylands of eastern Africa currently face growing encroachment of invasive plant species, such as Prosopis juliflora (Ayanu et al. 2015 <sup>[[#fn:r267|267]]</sup> ), which constitutes land degradation since it leads to losses in economic productivity of affected areas but appears as a greening in the satellite data. Another case study in central Senegal found degradation manifested through a reduction in species richness despite satellite observed greening (Herrmann and Tappan 2013 <sup>[[#fn:r268|268]]</sup> ). A number of efforts to identify changes in vegetation composition from satellites have been made (Brandt et al. 2016a <sup>[[#fn:r269|269]]</sup> , b <sup>[[#fn:r270|270]]</sup> ; Evans and Geerken 2006 <sup>[[#fn:r271|271]]</sup> ; Geerken 2009 <sup>[[#fn:r272|272]]</sup> ; Geerken et al. 2005 <sup>[[#fn:r273|273]]</sup> ; Verbesselt et al. 2010a <sup>[[#fn:r274|274]]</sup> , b <sup>[[#fn:r275|275]]</sup> ). These depend on well-identified reference NDVI time series for particular vegetation groupings, can only differentiate vegetation types that have distinct spectral phenology signatures, and require extensive ground observations for validation. A recent alternative approach to differentiating woody from herbaceous vegetation involves the combined use of optical/infrared-based vegetation indices, indicating greenness, with microwave based Vegetation Optical Depth (VOD) which is sensitive to both woody and leafy vegetation components (Andela et al. 2013 <sup>[[#fn:r276|276]]</sup> ; Tian et al. 2017 <sup>[[#fn:r277|277]]</sup> ).

Vegetation Optical Depth (VOD) has been available since the 1980s. VOD is based on microwave measurements and is related to total above-ground biomass water content. Unlike NDVI, which is only sensitive to green canopy cover, VOD is also sensitive to water in woody parts of the vegetation and hence provides a view of vegetation changes that can be complementary to NDVI. Liu et al. (2013) <sup>[[#fn:r278|278]]</sup> used VOD trends to investigate biomass changes and found that VOD was closely related to precipitation changes in drylands. To complement their work with NDVI, Andela et al. (2013) <sup>[[#fn:r279|279]]</sup> also applied the RESTREND method to VOD. By interpreting NDVI and VOD trends together they were able to differentiate changes to the herbaceous and woody components of the biomass. They reported that many dryland regions are experiencing an increase in the woody fraction often associated with shrub encroachment and suggest that this was aided by CO <sub>2</sub> fertilisation.

'''Biophysical models''' use global datasets that describe climate patterns and soil groups, combined with observations of land use, to define classes of potential productivity and map general land degradation (Gibbs and Salmon 2015 <sup>[[#fn:r280|280]]</sup> ). All biophysical models have their own set of assumptions and limitations that contribute to their overall uncertainty, including: model structure; spatial scale; data requirements (with associated errors); spatial heterogeneities of socio-economic conditions; and agricultural technologies used. Models have been used to estimate the vegetation productivity potential of land (Cai et al. 2011 <sup>[[#fn:r281|281]]</sup> ) and to understand the causes of observed vegetation changes. Zhu et al. (2016) <sup>[[#fn:r282|282]]</sup> used an ensemble of ecosystem models to investigate causes of vegetation changes from 1982–2009, using a factorial simulation approach. They found CO <sub>2</sub> fertilisation to be the dominant effect globally, though climate and land-cover change were the dominant effects in various dryland locations. Borrelli et al. (2017) <sup>[[#fn:r283|283]]</sup> modelled that about 6.1% of the global land area experienced very high soil erosion rates (exceeding 10 Mg ha− <sup>1</sup> yr− <sup>1</sup> ) in 2012, particularly in South America, Africa, and Asia.

Overall, improved estimation and mapping of areas undergoing desertification are needed. This requires a combination of rapidly expanding sources of remotely sensed data, ground observations and new modelling approaches. This is a critical gap, especially in the context of measuring progress towards achieving the land degradation-neutrality target by 2030 in the framework of SDGs.

<div id="section-3-2-1-2-regional-scale"></div>

<span id="regional-scale"></span>