Investigator:

Beth Tellman

Advisor: James Saiers

Description:

Climate change promises to increase disasters in El Salvador’s future in a region that is already highly vulnerable to earthquakes, droughts, floods, and landslides (Ibarra and Balmore 2007).  Additionally, the sociopolitical context of El Salvador has led to an increase in urbanization with 58% of El Salvador’s poor now living in cities (FLASCO 2010).   This trend is troubling in the context of disaster vulnerability as effects of urbanization on surface runoff in San Salvador from 1999-2009 indicate that a 20% land cover change from forest to concrete urban surface significantly increases flooding (Erazo 2010).

In the past 5-10 years, according to hydrologists at the Salvadoran Ministry of the Environment (MARN) this urbanization has been increasing most dramatically around the headwaters of the San Antonio watershed, converting forest, agriculture, and ranching land cover to urban surfaces, and purportedly increasing flooding into rural areas downstream. This small watershed (~40km2,) has a mixture of humid and dry tropical forests on mostly andosol volcanic soil, which has high infiltration rates and capacity for flood mitigation.  The cool climate (16-20C), high elevation, forest cover, and proximity to the capital make this land attractive for upscale housing and shopping. While community leaders, NGOs, and hydrologists I spoke with at MARN have observed this phenomenon and attempted to prevent more development in the area, no study has been done to measure and model the impacts of land use/land cover change (LULC) on flooding. A rigorous study quantifying hydrological impacts of deforestation on the San Antonio watershed could be an important tool for NGOs and MARN to lobby for stricter regulations and a zoning law to prevent further degradation of the ecosystem service of soil infiltration. For my Master’s Thesis, I am using Rainfall Runoff model  HEC-HMS and flood inundation extent model HEC-RAS to determine how changing land use types, area, and location in the San Antonio watershed affects  stream flow response in storms.

Computer modeling helps quantify and simulate watershed complexity.  The advent of remote sensing techniques makes available previously unavailable high resolution data on evapotranspiration, land use change, flood footprints, stream geomorphology, and elevation in developing countries. Better data allows for new methodology in modeling watershed process and behavior, illuminating the complex relationship between land use and flooding, a controversial topic in tropical forest hydrology   (Winsemuis et al 2008, Khan et al 2011, Wohl et al 2012, Lakshmi 2004). 

Remote sensing is now routinely applied to aid prediction precision for modeling (Gorte 2000 Dubayah 2000).  The use of remote sensing to parameterize hydrological models is particularly useful in ungauged basins, or basins with no river gauging station, and thus no river discharge data upon which to calibrate model predictions.  San Antonio, my watershed of interest, is an ungauged river, and in order to calibrate the model for the San Antonio basin; I must first run HEC-HMS on appropriate donor catchments with a long history of stream gauge data with similar hydrological properties and relatively static land use over time (Sarhadi et al 2012).  Remote sensing can help me choose static donor catchments through a process of change detection over my period of interest (1992-present). In addition, high resolution analysis of how and where land use in the San Antonio watershed changes over time allows me to make predictions of stream flow and flood footprints.  Remote sensing will allow me to analyze the extent and nature of land cover change in the San Antonio watershed, in order to confirm and constrain claims of “rampant urbanization” made by MARN.

In this project, I make use of remote sensing to classify land in my watershed of interest and surrounding areas which contain potential donor catchments using supervised land classification from Landsat Imagery from 1992 and 2011. I used supervised classification with training regions defined with the aid of a Google Earth, my knowledge of the area living in San Salvador for the past three years, and a DEM. I delineate 5 major land classes (agriculture, forest, and water) and 12 subclasses defined by hydrologic remote sensing methodology from Su (2000) to classify both images. I measured my success by quantifying an accuracy assessment ground truth image of 2010 land use vector data from MARN.  I compare land cover in 1992 and 2011 with post classification change detection. Finally, I will export my land classification results to GIS, and assign on majority land cover to each subbasin in the watersheds in my Landsat scene as an input into my hydrological model.

This will be the start of a much larger project for my thesis, as I have received funding to ground truth my land classification in El Salvador. LULC is an essential parameter for my hydrologic model of this watershed, which outputs of my model will relate trends in land cover to trends in stream flow. Thus, this remote sensing project will make a large contribution to my thesis.