Dynamic vegetation senescence analysis with moderate spatial resolution in support of desert locust control operations in Mauritania

Renier, Cécile;Waldner, François;Jacques, Damien Christophe;Defourny, Pierre;et.al.
(2014) Global Vegetation Monitoring and Modeling International Conference — Location: Avignon (France) (3.February.2014)

Files

Renier2014.pdf
  • Open Access
  • Adobe PDF
  • 2.19 MB
Download

Details

Authors
Abstract
The Desert locust represents a major threat for agro-pastoral resources, from Northern Africa to the Arabian peninsula and India over almost 30 million km². Locust infestations and invasions repetitively jeopardizes the food security of more than 60 countries. Remote sensing plays a major part of in the preventative control strategy as it allows monitoring the locust habitats in near real-time over its entire repartition area, especially in remote or unsecured regions. Since 2010, dynamic greenness maps are provided every ten days to the Food and Agriculture Organization and the affected countries to help them analyze the current situation and plan the control operations. These maps indicate the priority areas for survey by detecting the appearance of green and fresh vegetation, i.e. locusts' potential habitats. The successful implementation and integration of the dynamic greenness maps into the daily operations led to a new need: control centers want to know where and when the habitats are becoming less attractive and thus likely to be abandoned as vegetation is drying out. This information through a dynamic dryness map would enable control centers to assess if they can remove or redirect their survey teams, leading to more efficiency in the allocation of resources and in decision making. In this context, this work aims at developing and evaluating a method to detect in near real time the senescence of vegetation. The design of the detection method relies on the temporal behavior of two indices : the NDVI (Normalized Difference Vegetation Index), depending on green vegetation, and NDTI (Normalized Difference Tillage Index), sensitive to both green and dry vegetation. The method is demonstrated in Mauritania a ever-affected country with 10-day MODIS mean composites for the years 2010 and 2011. The Mauritanian field survey database provided the validation samples necessary to distinguish between three phenological classes , "growth", "density reduction" and "drying".The discrimination performance of these three classes were analyzed for three classification methods: a decision tree (DT), a support vector machine (SVM) and a maximum likelihood (ML) classifier. The DT and SVM, with an overall accuracy of 71.5% and 72.3% respectively, outperform the ML (61.4%). Performance of the DT and SVM is affected by several factors: (i) the spatial distribution of the error confirms the influence of the north-south climatic gradient on the detection, with an error 10% higher in northern regions, (2) the vegetation density plays an important role in detection and depends on the type of habitat and on rainfall, itself subjected to climate and seasons, (3) the relief interferes also with vegetation detection. On the other hand, the overlap of the "density reduction" and "drying" class distribution leads to an important omission (46,19% for the DT and 49,14% for SVM) of vegetation patches affected by density reduction. Application of the decision tree to the Mauritanian case highlights the diachronic consistency of the land cover classification ; the seasonal cycle is clear, despite some patches whose temporal behavior is difficult to interpret. Smoothing the NDVI and NDTI temporal profiles results in an increase of the overall accuracy of 5% for the decision tree and of 4% for the SVM. The classification performance could be enhanced using the new generation of satellites with a higher spatial resolution such as PROBA-V or Sentinel-2. The results obtained in the framework of this paper are promising to pave the way for a first operational implementation of the senescence dynamic maps and, consequently, further strengthen the capacity of the locust control management.
Affiliations

Citations

  • APA
  • Chicago
  • FWB

Renier, C., Waldner, F., Jacques, D. C., Defourny, P., & et al. (2014). Dynamic vegetation senescence analysis with moderate spatial resolution in support of desert locust control operations in Mauritania. Global Vegetation Monitoring and Modeling International Conference, Avignon (France). https://hdl.handle.net/2078.5/48773