Title
Rangeland Condition Metrics for the Gobi Desert Derived from Stakeholder Evaluations. Arthur Rylah Institute for Environmental Research Technical Report Series No. 289
Author(s)
Otgonsuren Avirmed, Matt D. White, Khorloo Batpurev, Peter Griffioen, Canran Liu, Sergelenkhuu Jambal, Hayley Sime, Kirk Olson & Steve J.Sinclair
Published
2018
Abstract
Context: Changes in the condition of rangelands in the Gobi Desert of Mongolia must be understood and quantified. This imperative is driven by land use intensification. There is now a need to account for expenditure on incentive schemes to manage grazing intensity, and to measure the impacts and mitigation measures associated with mining projects such as Oyu Tolgoi (OT), which has an offset program with a requirement to monitor rangeland condition. There is, however, currently no accepted means of quantifying rangeland condition. This project addressed the need for robust metrics to measure rangeland condition. The work was undertaken as a collaboration between Wildlife Conservation Society, Mongolia Country Program (WCS) and the Arthur Rylah Institute for Environmental Research (ARI). It was funded by Oyu Tolgoi. Aims: We aimed to create condition metrics for five ecosystems in the Gobi Desert: True Desert, Desert Steppe, Semi Desert, Saxaul and Elm Forest. Such metrics must be able to distinguish sites of different condition, using only field-measured data for several (11-14) simple parameters (no stakeholder data is required for operation of the metrics). Methods: We used a method adapted from previously-published work on Australian ecosystems (Sinclair et al. 2015, 2018). That method assumes that the concept of condition is inherently subjective, and thus the metric algorithm is derived from human opinion. It is important to note that the use of opinion in this context is not in lieu of other empirical data; as no such data could conceivably be obtained. Ninety-four stakeholders contributed quantitative data that were used to derive the metrics. They represented four groups: nomadic pastoralists, specialists in botany, specialists in wildlife, and conservation practitioners and policymakers. They evaluated a set of hypothetical rangeland sites, providing each with a score between 100 (a desired state) and 0 (no values of the desired state retained). We used these evaluation data to train a model (an ensemble of 30 regression trees) capable of predicting the score based on the site variables. The models were converted directly into metrics for each system, and are presented as decision trees which can be implemented in a spreadsheet (as “if, then” statements). Results: The primary test of the metrics’ utility was to compare the metrics’ scores derived from field-measured test sites with the median score assigned by a group of stakeholders for the same test sites. The test sites were not used to train the model. The metrics for three of five ecosystems were tested in this way (True Desert, Semi-Desert, Desert Steppe, the other systems remain untested). All tested metrics performed well (with r2 values of 0.78, 0.82, 0.68 respectively). We also visualised the performance of the metrics using Multi Dimensional Scaling (MDS), where the model and each stakeholder was each represented as a point in a space defined by the evaluations. The metrics were positioned within the cluster defined by the stakeholders. We confirmed that the selected variable sets adequately addressed stakeholder conceptions of condition, and that the field plot method adequately measured these variables, by demonstrating a close correlation between the scores provided by stakeholders in the field, and stakeholders assessing the same sites in a workshop context, where the sites were abstracted using only the selected variables to describe them. We performed this test for True Desert, Semi-Desert, Desert Steppe (with r2 values of 0.81, 0.82, 0.53 respectively). We also checked that each variable within the metrics related to the condition score in ways that would be generally expected in conservation biology, and that the relative importance of each variable in the regression tree models reflected their perceived importance in each ecosystem. We performed these checks for all five ecosystems. Conclusions: We conclude that the data collection method and the metrics for deriving condition scores are robust and fit for purpose for the ecosystems we tested (True Desert, Semi-Desert, Desert Steppe). We suggest that the metrics for Saxaul and Elm Forest are likely to be useful, pending field testing for those metrics. We recommend that all metrics be mounted on a suitable web-based application, and be used for monitoring and reporting on rangeland condition between sites, over time and between ecosystems in the Gobi Desert.
Full Citation
Avirmed, O., M.D. White, K. Batpurev, P.A. Griffioen, C. Liu, S. Jambal, H. Sime, K. Olson, and S.J. Sinclair (2018). Rangeland Condition Metrics for the Gobi Desert Derived from Stakeholder Evaluations. Arthur Rylah Institute for Environmental Research Technical Report Series No. 289. Heidelberg, Victoria, Australia: The State of Victoria Department of Environment, Land, Water and Planning and Wildlife Conservation Society, Mongolia
Access Full Text
Back
DMX4430200000