Can vegetation breakpoints in Eastern Mongolia rangeland be detected using Sentinel-1 coherence time series data?

Ji, Shuxin ; Gonchigsumlaa, Ganzorig; Damdindorj, Sugar; Tseren, Tserendavaa; Sharavjamts, Densmaa; Otgondemberel, Amartuvshin; Gurjav, Enkh-Amgalan; Puntsagsuren, Munguntsetseg; Tsabatshir, Batnaran; Gungaa, Tumendemberel; Batbold, Narantsetseg; Drees, Lukas; Ganbayar, Bayarchimeg; Orosoo, Dulamragchaa; Lkhamsuren, Bayartsetseg; Ganbat, Badamtsetseg; Damdinsuren, Myagmarsuren; Gombosuren, Gantogoo; Dashpurev, Batnyambuu; Phan, Thanh Noi; Dejid, Nandintsetseg; Müller, Thomas; Lehnert, Lukas (2025): Can vegetation breakpoints in Eastern Mongolia rangeland be detected using Sentinel-1 coherence time series data? GIScience & Remote Sensing, 62 (1): 2540222. ISSN 1548-1603

Abstract

Mongolian society and food production depend heavily on livestock farming, whichis usually practiced through nomadic systems. Consequently, movement patterns ofherders are crucial in respect of finding sufficient forage and sustainable use ofpastures. Since vegetation presumably changes after livestock pasture use, thisstudy hypothesizes that changes in Interferometric Synthetic Aperture Radar(InSAR) data over time are linked to herder and livestock mobility. In this study,a combination of InSAR, optical, and weather time series data has been explored asa tool for spatio-temporal grazing monitoring. To detect movement patterns, a newrandom forest-based method to detect breakpoints in vegetation condition hasbeen developed and compared to the widely used Breaks For Additive Season andTrend (BFAST) algorithm. In contrast to BFAST, the new method accounts forvegetation changes caused by weather events such as snow and rainfall. The resultshave been validated using test sites spread across the entire eastern Mongoliansteppe ecosystem, covering different rangeland use intensities. The results indicatethat (1) random forest performed better than BFAST, indicating that random forestis able to separate vegetation changes caused by grazing from those caused bynatural events. However, the detection was challenging especially for winter move-ments (for summer camps, random forest and BFAST detected 44% and 28% ofmovements, respectively). (2) Breakpoints in summer pastures mainly occurred fromApril to June, while on winter pastures, they emerged in October, November, andthe following February and March. The breakpoints in October and November canbe explained by increasing grazing pressure as the herders moved to the wintercamps while those occurring in spring are associated with enhanced vegetationgrowth after herders left for summer camps. (3) From a spatial perspective, therandom forest model predicts summer and winter pastures with homogeneouspatterns. In areas with higher productivity and higher grazing pressure, the summerpastures are located along the rivers while the winter pastures are in the surround-ing mountainous areas. This is in agreement with the general movement patterns.In drier and less intensively used areas, the predicted pattern agrees less with theknown movements. Consequently, there is insufficient evidence to definitively

• Dokumententyp: Artikel (LMU)
• Organisationseinheit (Fakultäten): 20 Geowissenschaften > Department für Geographie > Geographie und geographische Fernerkundung
• DFG-Fachsystematik der Wissenschaftsbereiche: Multidisziplinär
• DFG: Gefördert durch die Deutsche Forschungsgemeinschaft (DFG) - 491502892