Authors: Daniel Runfola*, The College of William and Mary | AidData, Geeta Batra, Global Environment Facility, Anupam Anand, Global Environment Facility
Topics: Geographic Information Science and Systems, Applied Geography, Spatial Analysis & Modeling
Keywords: GIE, Impact Evaluation, Causal Inference, Causal Attribution, Machine Learning, HPC
Session Type: Paper
Start / End Time: 5:00 PM / 6:40 PM
Room: Capitol Room, Omni, East
Presentation File: No File Uploaded
In 2011, an effort was undertaken to link the Global Environmental Facility (GEF) Land Degradation Focal Area Strategy and the United Nations Convention to Combat Deforestation ten year (2008 to 2018) strategy to streamline investments in sustainable land management. One goal of this streamlining initiative was to promote understanding of the long-term impacts of GEF activities on key environmental indicators. This paper presents a novel datasets on the location of GEF activities, and uses this information in conjunction with satellite ancillary data and a novel machine learning technique to examine heterogeneity in the global impacts of GEF projects along three dimensions - vegetation productivity, forest fragmentation, and forest cover change. A four-step approach is adopted in which (a) precise geospatial data on GEF project locations is generated, (b) satellite information is used to derived long-term measurements of each outcomes, (c) the data generated is integrated with a wide set of geographically-varying ancillary data, and (d) a novel propensity score matching approach, Causal Trees (CT), are employed to attribute the impact of GEF project locations on each indicator of interest. We highlight the advantages and challenges to using Causal Trees for Geospatial Impact Evaluation in this paper, as well as discuss the state of the geostatistical field regarding causal inferential model design.