Ross Maps: 2010

Fall 2010 Geography 169

Dr. Thomas Gillespie

Final Project - Haiti Vegetation Change in the Artibonite Valley

Abstract

    Haiti occupies the western third of the island of Hispaniola and shares its only border with the Dominican Republic. The nation gained independence from its French colonial masters in 1804, but has been plagued by political instability coupled with environmental degradation ever since. The nation faces severe problems with deforestation that lead to and compound other problems such as watershed stability, soil degradation, and erosion. Reforestation efforts have been met with limited success due to the extreme demand for wood for charcoal based cooking which is the main cooking source for the majority of the population. Mountain communities face some of the most serious road blocks for reforestation due to lack of irrigation, limited sites for agriculture, and complete dependence on the rain of the 6 month wet season followed by an often crippling 6 month dry season. A tree planting organization called The Haitian Timber Re-Introduction Project (HTRIP) based in the central Artibonite Valley works with local communities to plant trees and crops in an effort known as agroforestry. Satellite imagery available for the area since 1973 allow environmental change to be tracked and monitored. This report will look at change in the region to monitor trends, determine if reforestation efforts can be detected by satellites, and make predictions for theoretical locations for reforestation efforts.

Introduction

     Haiti has had a difficult time recovering from years as a plantation-based French colony from the 17th century to the beginning of the 19th century. In addition to the political instability of the country since its independence in 1804, the country has suffered greatly from environmental degradation due to a large population, small country size, lack of political restrictions, and a positive feedback loop related to deforestation, soil erosion, landslides, and hurricanes. The Library of Congress - Research Division estimates that the country suffered 40% deforestation by 1923, and a further 98% by 2006. This problem is based in the great demand for wood because charcoal is still the primary fuel source for the bulk of Haiti’s rural and urban population.
     This study will look at vegetation cover in Haiti’s central region, the Artibonite Valley: 18º 83' North: 72º 31' West of Greenwich. The region is known as Haiti’s breadbasket and represents a large part of Haiti's agricultural production. The agriculture is supported at the valley's base by the Artibonite River that flows steadily throughout the region.
      The Hôpital Albert Schweitzer located in Deschapelles along the Artibonite River and about 25 km inland from the coast helps support an agroforestry and reforestation effort called the Haiti Timber Re-Introduction Project (HTRIP) and was started in 2006. Although the base of the valley has access to water and some irrigation from the Artibonite River, the surrounding mountain villages have much more limited access to water, and often have difficulty because their agriculture is based entirely on seasonal rains.

Methods

    Public access to Landsat imagery provided for free by the USGS makes analyzing vegetation cover in most regions of the world possible with basic remote sensing techniques. The bulk of the data I use is provided by Landsat 7's sensor the Enhanced Thematic Mapper Plus (ETM+) which was launched in 1999. The most useful bands for looking at vegetation cover are bands 1-4 provided by the ETM+ at 30 meter resolution with a 16 day repeat interval.
    The entire study area of the central Artibonite Valley is captured by a single tile of Landsat 7 - WRS-2 path 9, row 47, latitude: 18.8, longitude: -72.9. The long history of the Landsat program makes hundreds of images available due to the frequent pass time and number of Landsat Satellites. The first satellite image of Haiti is available from Landsat 1 MSS from December 1973. Although this project will focus on detecting change due to the presence of HTRIP in the past 5 years from 2006 - 2010, historical records are useful.
    Elevation data is also helpful when looking into reforestation efforts due to the importance of slope and aspect for many tree species (Lin, 2008). The elevation data is available worldwide from the USGS Eros Data Center (EDC) at 30 arc-second (approximately 1 km) Digital Elevation Model (DEM). This data is sufficient to reveal slope and aspect through the study area.
      The data was processed using the remote sensing program ENVI 4.7. A powerful tool for analyzing vegetation is the NDVI (Normalized Difference Vegetation Index). The NDVI is calculated by comparing Bands 3 (.63 µm - .69 µm) and band 4 (.75 µm - .90 µm) which corresponds to the end of the visible (VIS) and the near infrared(NIR) bands. Plants strongly absorb light for photosynthesis from .4 µm - .7 µm, but reflect near infrared light from .7µm to .11µm. We can reveal this difference using the calculation for NDVI = (NIR-VIS) / (NIR+VIS). Thus a normalized output (-1.0 - 1.0) reveals when there is much more reflection of NIR than absorption of VIS, and a monochromatic NDVI map shows vegetation density and resulting species richness throughout the study region. The NDVI is an included feature of the ENVI tool-set under ‘Transformation,’ and is a straightforward process that allows the NDVI to be completed for a large number of images to be used for comparison. You start by inputting band 3 into the red and blue bands and band 4 into the green band, creating a new image, and then transforming using the NDVI function.
     One further step was needed before the NDVI could be completed.   Images acquired by the ETM+ after 2003 have a common problem associated with the malfunctioning of the SLC (scan line corrector). The result of this sensor failure is the loss of approximately 25% of collected data and the problem is amplified at the borders of images. Fortuitously the Artibonite Valley is located at the center of the image and the study area is not prohibitively disrupted. Additionally, there is a correction tool provided by ENVI to counter this problem. The ‘Replace Bad Data’ tool under the ‘Topographic’ header allows ENVI to average nearby values and cover up the missing values so processing can take place, albeit with some loss of accuracy.
     The elevation data (DEM) was processed with another included feature of ENVI - ‘Topographic Modeling’. Topographic modeling allows for the input of the DEM and produces aspect, slope, convexity, and a multitude of other options I did not include in this analysis.
Because the two data-sets are of distinct spatial scale, the use of ‘Layer Stacking’ was necessary to combine the data into an equivalent scale that could be processed by ENVI. A ‘Basic Tool’ of ENVI, layer stacking combines images to a common spatial scale so further analysis can be used.
   Because the two data sets are of distinct spatial scale, the use of ‘Layer Stacking’ was necessary to combine the data into an equivalent scale that could be processed by ENVI. A ‘Basic Tool’ of ENVI, layer stacking combines images to a common spatial scale so further analysis can be used.
    The final tool used for looking into vegetation cover is Google Earth. Google Earth compiles the highest resolution and most up to date satellite images and elevation data into a user friendly tool for terrain analysis. My own observation coupled with Google Earth revealed the tendency for vegetation cover to be greater on north facing slopes (figures 1-3). This is possibly due to decreased evaporate demand on the sun-shaded slopes. This suggests future restoration efforts should look into this effect of slope aspect as a predictor of tree growth success.

Results

    It is uncertain whether the results of the NDVI for the January 2010 showing vegetation cover near some of the HTRIP plots are a result of HTRIP tree planting efforts or not. As must be common with efforts in reforestation, more time might be needed to view a significant difference. Tree planting only started in 2007, so the oldest trees are now only three years old. While this might not seem old enough to possibly be detected, some of the trees are already over 8 meters. Within the next few years, more change should be measurable by even the the 30 meter resolution of the ETM+.
    The second part of the review yields inconclusive results based on the time series NDVI comparison. Certainly there are regions that have a constant amount of vegetation cover throughout the years, but there is a great variability from year to year and even consecutive months. There can be confounding factors in the NDVI, such as atmospheric effects, temperature, and precipitation, so a more thorough analysis will be necessary to gain information from this time series.
    The results of the NDVI coupled with the change detection tool suggest much the same as the NDVI time series analysis. The resulting image suggest some areas are losing vegetation and some areas are gaining. Nothing for certain could be revealed about the study area.
   The concept of vegetation modelling prediction could be more useful. Elevation derived aspect data coupled with the NDVI could be useful in predicting efforts for future planting. The output map which shows aspect in the red and green bands and NDVI in the blue bands shows where there is already vegetation and where the north-facing slopes might be more favorable to future plantings.

Discussion

   The 2008 Lin study uses a more complex algorithm to produce predictive tree planting model. I think if I were to incorporate more factors and a similar equation-based model, I could produce more useful results. As his study suggests, there are many factors; biotic, abiotic, edaphic, ad infinitum that satellite images simply can not measure at the moment. Determining the importance of these factors is a difficult, but important step for making headway in this area, and studies like these that can determine significance even from a limited number of factors can help produce more predictive models in the future.
    Analyzing the NDVI time series, it has been difficult to measure a difference because the high degree of variability in the data set from image to image. Some images, particularly images taken from the same month and year should theoretically have much more similar NDVI results than I often found was the case. Atmospheric and other effects cause a difference in the readings of the sensor, and my analysis did not employ a method to try to counter some of these differences, which is likely the cause of my inability to determine significance.
    The oldest of the HTRIP plots now contain trees only three years old. While some of these trees do have significant growth, it is altogether likely that there is not enough widespread growth to be detected as of yet based on the limited spatial scale of the ETM+ in the study region. I predict that as these tree plots grow and age, it will become possible to detect them using remote sensing and get a larger understanding of their significance on a greater scale that satellite imagery makes available.
    Furthermore, the ability to create models based on factors that benefit trees was demonstrated, albeit with a limited number of factors. There do not appear to be many studies of this kind in the literature, and although overly simplistic, such models can reveal better regions to plant trees based on demonstrated factors known to benefit tree growth.

References

Lin, G., Xia, B., Zeng, Z., & Huang, W. (2008). The Relationship between NDVI, Stand Age and Terrain Factors of Pinus elliottii Forest. In 2008 International Workshop on Education Technology and Training & 2008 International Workshop on Geoscience and Remote Sensing (pp. 232-236).

Sprenkle, S. (2008). Community-Based Agroforestry as Restoration: The Haiti Timber Reintroduction Project Methods and Framework. Ecological Restoration. 26(3), (p. 201).

Wilson, J. S., Brothers, T. S., & Marcano, E. J. (2001). Remote Sensing of Spatial and Temporal Vegetation Dynamics in Hispaniola: A Comparison of Haiti and the Dominican Republic. Geocarto International, 16(2), 7–18.

Figures