SLM Intervention Impact Assessment Using Remotely Sensed Data: Impact Assessment of the Tana-Beles Integrated Water Resources Development Project WB Land and Poverty Conference March 21-23, 2017 Daniel Monchuk, Daniel Ali, Klaus Deininger, and Marguerite Duponchel The World Bank
Tana-Beles Integrated Water Resources Development (TBIWRD) Project 2009-15 Four key components: Sub-basin Resources Planning and Management B. Natural Resource Management Investments Growth-Oriented Investment Facilitation Project Management
5 critical watersheds - 85,700 ha, 163 micro-watersheds – 526ha average
Component B Activities/Interventions Reported in M&E Output 1.2 - Soil and water conservation measures undertaken on cultivated lands Output 1.3 - Gully Treatment and Rehabilitation Output 1.4 - Degraded land (hillside, grazing and forestry land) treated Output 1.5 - Existing natural and planted community forests protected and sustainably managed Output 1.6 - New area planted by community forestry and agro-forestry systems to stabilize landscape and produce fuel wood and timber
Soil and water conservation activities/interventions on cultivated land has largest footprint.
Impact on the ground? –> Change in Normalized Difference Vegetation Index (NDVI)?
Side – Can NDVI (using LS7) pick-up meaningful change?
Processing imagery The old way… The new way… Download 11 years worth of LS7 imagery -> ~2 MONTHS from country office Batch routines Filter out clouds, shadows, etc. Merging raster files The new way… Google Earth Engine developer platform -> some javascript -> sit back, relax… -> 22 images/ year (16 day pass), 4 images span area of interest, 11 years, 250 MB, 4GB/day from country office
Sample ~30m pixel resolution Control – 5km buffer ~3 million pixels -> ~99 million obs in panel (11 years, 3seasons/year) 10% sample of pixels 98,182 TBIWRD 201,140 control
Remote sensed data NDVI - LS7 imagery used to compute seasonal average Rainfall - Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) Elevation - The Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM) Rainfall - Climate Hazards Group InfraRed Precipitation with Station data (CHIRPS) which incorporates 0.05° (~5km) resolution satellite imagery with ground-based weather station data to create gridded rainfall time series Elevation - The Shuttle Radar Topography Mission (SRTM) digital elevation model (DEM)
Physical characteristics
Identification Panel with pixel fixed-effects Overall impact of TBIWRD ( )
Treatment the M&E data are used to construct three treatment variables constructed as follows: Tr1 - an indicator variable identifying micro-watersheds if reporting any activity under any of the outputs listed above (=1) and 0 otherwise; Tr2 - the ratio of (cumulative) area of cultivated land treated for soil and water conservation (Output 1.2) as a share of total micro-watershed. Tr3 - the ratio of (cumulative) area of all area treated (Output 1.2 - Output 1.6) to total micro-watershed.
Discussion/Conclusions Strong empirical evidence of positive impact of TBIWRD on NDVI Magnitude? Confounding factors/ other considerations Meaningfully interpreting NDVI (i.e. carbon sequestered) and alternatives (EVI, SAVI, TCT, etc.) Ex-post evaluation possible “Easy” to process imagery Scalable – anywhere, anytime, any size … but need good intervention data (where, what, when)