Comparisons to Past Work
A comparison between the Le et al. (2014) land degradation dataset and aspects of the GLADIS land degradation
database serves as a useful benchmark. While the Le et al. (2014) dataset provides only one map indicating whether areas are degraded or improved and is based entirely on terrestrial processes, the GLADIS database provides estimates for multiple categories of land degradation, including biomass, biodiversity, soil, water, economic, and social. For this comparison the biomass categories were considered most relevant. Table 5.2 illustrates the input data used in the Le et al. (2014) land degradation map as compared to the GLADIS biomass analyses. Although both land degradation maps rely on the GIMMS NDVI
and CIESIN-CIAT population
datasets, each incorporates independent ancillary information using different methods such that the two datasets may be considered separate, if not independent, estimates of land degradation processes.
Table 5.2 Inputs for global land degradation datasets
Figure 5.5 demonstrates the comparison between the Le map and the GLADIS trends in NDVI, which are identified as being anthropogenic or natural, trends in total biomass, and trends
in soil degradation or improvement. The terrestrial biomass-based datasets (panels a–c) largely agree over Canada, Northern Argentina
, Democratic Republic of the Congo, Angola, Tanzania
, Mozambique, Malawi, India, Kazakhstan
, the majority of Southeast Asia, and Australia. However, they disagree over most of the US, Europe, much of Brazil, the Sahel, South Africa, China and portions of Russia. In general, the GLADIS datasets indicates larger areas of land improvement
, arising from both natural and anthropogenic influences. Without extensive ground-based measurements it is difficult to evaluate which dataset is more accurate in the areas of disagreement, but the regions of good agreement lends confidence to both datasets in these regions.
While the GLADIS estimates of soil degradation or improvement have no direct analog to the Le dataset, the comparison between the two is useful as it may elaborate the extent to which the Le dataset captures soil degradation processes (Fig. 5.5d). The GLADIS soil degradation map (panel d) shows little resemblance to the Le land degradation data. Le et al. (2014) does develop an additional dataset identifying areas in which excessive fertilizer application may mask land degradation (data not shown), which produces similar patterns to the GLADIS map over China, Northeast India
and much of Europe. The comparison of soil degradation products reveals that the Le et al. (2014) data and the GLADIS data capture some similar soil processes in parts of Europe and Asia, but diverge widely in most other regions indicating a high level of uncertainty. These dissimilarities are to be expected provided the difficulty of measuring processes of soil degradation on global scales.
In addition to considering existing land degradation
maps, it is important to use independent datasets to evaluate new estimates of land degradation. The results of the survey-based analysis and the remote sensing
analyses provide useful insight into the value of the Le et al. (2014) land degradation dataset, and of the scale-dependent processes involved in the construction of the dataset. A total of 6 countries were analyzed to compare remote sensing analyses with survey results for selected sites. The sites represent a range of agro-ecological zones, and include both areas indicated as degraded and improved in the Le et al. (2014) dataset.
Senegal Sample Sites
Figure 5.6 illustrates the seven sample sites chosen for the evaluation analysis in
Senegal
. A comparison between the focus group discussions conducted at each site and the remote sensing analyses indicate a high level of agreement (3.5/4 or 4/4) for four sites, a moderate level of agreement (3/4 or 2.5/4) for two sites and no agreement (2/4) for one site, as indicated in Table 5.3.
Table 5.3 Agreement between the Le et al. (2014) land degradation map, the focus group discussions (FGD), and the independent remote sensing analyses for each of the seven sites in Senegal
The sites that demonstrated unanimous, or near unanimous, consensus on the state of land degradation had a number of systematic similarities. Out of the four sites with a high level of agreement, three were degraded (Bantanto, Gomone, and Niassene) and one was improved (Diakha Madina). These sites all experienced clear areas of deforestation, or reforestation in the case of Diakha Madina, which was captured in the FGD
, the Landsat
analysis, and the MODIS
analyses (see Table 5.4 for a summary of FGD results and Table 5.5 for a summary of the remote sensing
results). While there are land degrading processes that were not captured in the remote sensing analysis—such as wind/water erosion and salinization—occurring at each of the degraded sites, the remote sensing analysis already correctly categorized these sites due to the clear decrease in vegetative health. The inability of remote sensing products to capture erosion or salinization processes, however, became relevant for the improving sample site. The results of the focus group discussion at Diakha Madina, the improved site, revealed that the narrative of land improvement
is somewhat undercut by factors not captured in the remote sensing
analysis, particularly a decrease in crop yields due to erosion. Figure 5.3 depicts the remote sensing analysis for Niassene as an example of the data provided by such analyses.
Table 5.4 Summary of the focus group discussion results for sample sites in Senegal
Table 5.5 Summary of remote sensing analyses for sample sites in Senegal
The sample sites that demonstrated intermediate agreement, Guiro Yoro Mandou and Missira, also contain systematic similarities in that both sites demonstrated a dynamic of cultivation that the remote sensing analysis was unable to capture (see Guiro Yoro Mandou and Missira in Tables 5.4 and 5.5). In Guiro Yoro Mandou the MODIS
analysis indicates an increase in wooded cover in some areas and a loss of natural vegetation in other areas due to cropland
expansion. However, the FGD
results clarify that this site has seen considerable planting of mango and cashew trees. The site is therefore clearly degraded due to expanding croplands, declining yields, and loss of natural vegetation, but the remote sensing estimates have difficulty capturing these dynamics due to the tree-crops planted. In Missira, on the other hand, the MODIS and FGD analysis agree on a decrease in forested area, but show opposite trends
in cropland
: MODIS indicates a loss while the FGD
indicates increased cropland
extent. This difference is likely due to patterns of fallowing and regeneration, as indicated by the FGD. This same pattern of regenerating fallowed fields may account, at least in part, for the demonstrated increases in NDVI
measured by Landsat. There may, in fact, be a number of competing processes at work as increasing cropland
area is causing deforestation, but regeneration of fallowed fields is increasing natural vegetation cover. Similarly the site has experienced water erosion and a perceived decrease in crop values but also reports an increased yields in recent years.
In Talibdji, the site that demonstrated the lowest level of agreement, uncertainties in the remote sensing
dataset compound with dynamic local processes to confound agreement on the status of the site (see Talibdji in Tables 5.4 and 5.5). The long-term NDVI
trend for Talibdji showed no coherent pattern, although it demonstrated degradation in recent years (2000–2005). Figure 5.4 demonstrates the data derived from the remote sensing analysis for Talibdji. The GLS Landsat data showed no coherent trend for NDVI, MODIS land cover showed many shifts in cropland
extent in both directions, while the FGD
indicated stable cropland. These discrepancies likely indicate a system of fallowing or rotating fields that is not captured well in the remote sensing analyses. The FGD and MODIS analyses also disagree on wooded cover, which is likely because the MODIS
land cover classes that dominate Talibdji have poor user accuracy (many below 50 %, see Table 5.6), meaning that misclassifications are likely (Friedl et al. 2010). The FGD, meanwhile, revealed that erosion has decreased yields and therefore the value of crops, which would not have shown up in the NDVI analysis.
Table 5.6 User’s accuracy for MODIS MCD12Q1 land cover classifications
Niger Sample Sites
In addition to Senegal
, six sites were chosen in Niger
to compare the FGD
and remote sensing
results. Only two of the sites chosen have FGD results due to data collecting issues. Sites containing only remote sensing data
were still analyzed to see whether agreements existed. Figure 5.7 illustrates the six sites chosen, which have a range of agro-ecological zones, and include both areas of degraded and improved land from the Le et al. (2014) dataset. Of the two sites with FGD
results, Tiguey had high agreement (3.5/4) and Koné Béri had low agreement (2/4). Out of the four sites without FGD results, three showed a moderate to high level of agreement between the remote sensing datasets (2/3–3/3) and one showed a high level of disagreement (0/3), as shown in Table 5.7.
Table 5.7 Agreement between the Le et al. (2014) land degradation map, the focus group discussions (FGD), and the independent remote sensing analyses for each of the six sites in Niger
Table 5.8 Summary of the focus group discussion results for the two sample sites in Niger
Similar to the Senegal results, sites showing a high level of agreement were mostly degraded. Tiguey for example, one of the two sites with FGD
results, showed a conversion of grassland to cropland
in the MODIS
analysis and a recent decrease in NDVI
(although historically NDVI increased from 1975 to 2000) (Table 5.9). Another change contributing to degradation was deforestation indicated by both the MODIS and FGD data (Table 5.8). Sites showing a moderate to high level of agreement without FGD results are Babaye, Bazaga, and Béla Bérim, all of which are mostly degraded. Béla Bérim showed 100 % agreement. The MODIS
results indicate a decrease in wooden cover here where conversion to grassland is occurring. Babaye and Bazaga had somewhat mixed results, and without the FGD results it is more difficult to interpret what is actually occurring.
Table 5.9 Summary of remote sensing analyses for sample sites in Niger
Two sites showed less agreement, Ndjibri with 0/3 agreement and Koné Béri with 2/4 agreement. The FGD
results indicate that Koné Béri had a decrease in crop land but an increase in shrub land and bare soil. Results also showed a severe forest loss and an increase in cropping intensity but a decrease in yields. Another issue that the FGD results showed was water erosion, which would not be an observable change viewed by remote sensing
. MODIS land-cover changes and NDVI
however show an increase in vegetation. MODIS shows a slight increase in forest
area and a change from barren to grassland. Landsat
NDVI trends show a recent increase.
Additional Sample Sites
In addition to Niger
and Senegal
, study sites within the countries of India
, Uzbekistan
, Tanzania, and Ethiopia
were chosen to compare survey results with remote sensing
datasets. The study sites within these countries differ in that they were not chosen to represent degraded and improved areas of the Le et al. (2014) dataset but rather areas that were most successful for conducting interviews with local stakeholders. Study sites that did not intersect with either degraded or improved pixels of the Le et al. (2014) dataset were considered to have mixed results. A benefit to this analysis is the comparison of improved or degraded areas not identified by the Le et al. (2014) dataset with the survey and other remote sensing results. Following are comparisons of the survey results with the remote sensing datasets.
India Sample Sites
Figure 5.8 illustrates the eight sites chosen in India
. The site with the highest agreement was Hivrebajar (3.5/4). Other sites with high agreement were Sangoha Jeur and Bayjabaiche both 3/4 agreement.
In general, study sites in India
showed many mixed results. The highest agreement was in Hivrebajar at 3.5/4. Unlike the previous countries, this site showed both a high level of agreement and mostly improvement. Almost all sites fluctuated between improvement and degradation, with many sites showing mixed categories. The main reason for this is that many of these sites had fluctuating agriculture between years. The MODIS
land cover data showed changes within cropland
but the cropland area itself was mostly static. Changes in agricultural intensity and crops grown in this area could have wide ranging effects on both the Landsat NDVI
and the Le et al. (2014) data. This alone can cause discrepancies between datasets. Another issue with agreement was between the FGD
results and the NDVI and MODIS land cover data. In most cases the FGD results had different results than NDVI and/or MODIS land cover, although tended to agree more with the Le et al. (2014) dataset (Table 5.10).
Table 5.10 Agreement between the Le et al. (2014) land degradation map, the focus group discussions (FGD), and the independent remote sensing analyses for each of the eight sites in India
Uzbekistan Sample Sites
Figure 5.9 illustrates the six sites chosen in Uzbekistan
. The site with the highest agreement was Khorezm (4/4). Other sites with high agreement were Shirin KFI (3/4) and Fazli (3.5/4).
Similar to Senegal the highest agreement was in areas of degradation. Khorezm showed 4/4 agreement, followed closely by Fazli, which had 3.5/4 agreement due to the FGD
data being unavailable for this village. A lot of the FGD results showed mixed degradation in Uzbekistan
. Similar to India, many of the villages had rotating cropland
areas. Almost all sites showed a resent decrease in Landsat
NDVI
, much of this decrease could be due to changing agricultural practices. In many cases the MODIS
land cover showed some areas being converted to agriculture while other areas nearby were reverted to natural land. This could be due to the nature of agricultural practice rather than permanent or long-term land-cover conversion (Table 5.11).
Table 5.11 Agreement between the Le et al. (2014) land degradation map, the focus group discussions (FGD), and the independent remote sensing analyses for each of the eight sites in Uzbekistan
Tanzania Sample Sites
Figure 5.10 illustrates the eight sites chosen in Tanzania
. The sites with the highest agreement were Mazingara and Mamba (4/4). These sites also appeared to be the sites with the most degradation. Other sites with higher agreement that were mostly degraded were Sejeli, Maya Maya, Zuzu, and Dakawa (3/4).
Similar to Senegal
the highest agreement was in areas of degradation. Mamba and Mazingara both showed 4/4 agreement. The Le et al. (2014) dataset and the FGD
results showed a near consensus. MODIS
land-cover change also showed a high level of agreement. Most of the disagreement occurred with the Landsat NDVI
data which showed improvement for many villages that the other datasets showed degradation. Some of the sites, such as Sejeli, Zombo, and Zuzu, that showed recent Landsat NDVI improvement, had mostly declining NDVI values from 1975 to 2000 indicating degradation (Table 5.12).
Table 5.12 Agreement between the Le et al. (2014) land degradation map, the focus group discussions (FGD), and the independent remote sensing analyses for each of the eight sites in Tanzania
Ethiopia Sample Sites
Figure 5.11 illustrates the eight sites chosen in Ethiopia. The site with the highest agreement was Kawo (4/4). The rest of the sites showed mostly poor agreement at 2/4 or less.
The highest agreement was in an area of degradation for the site Kawo. The rest of the sites in Ethiopia
showed poor agreement at 2/4 or less. The Le et al. (2014) dataset and the FGD results showed either mixed or degradation results, however the Landsat NDVI
and MODIS land-cover change datasets also showed some sites having improvement. One issue was that all of the sites except Kawo and Koka Negewo did not intersect the Le et al. (2014) data so they were given a mixed result. This was only a small part of the lower agreement between datasets since most of the sites also have confusion within the remote sensing
datasets (Table 5.13).
Table 5.13 Agreement between the Le et al. (2014) land degradation map, the focus group discussions (FGD), and the independent remote sensing analyses for each of the eight sites in Ethiopia