Impact of deforestation on regional surface temperatures and moisture in the Maya lowlands of Guatemala



[1] The Petén basin of Guatemala is a region in Mesoamerica experiencing extensive deforestation. Twelve 30 × 30 km2 forested, deforested and partially deforested habitats are selected to compare changes in surface temperature, Normalized Difference Vegetation Index (NDVI) and soil moisture parameters derived from MODIS satellite data during the dry and wet seasons from 2000 to 2008. During the wet season surface temperatures, soil moisture, and NDVI values have similar values in forested and deforested regions. During the dry season, deforested regions tend to be 4–8°C warmer than nearby forested regions, along with significantly lower values of NDVI and soil moisture. From 2000 to 2008, the partially deforested regions became progressively warmer during the dry season, and both NDVI and soil moisture values approached those of the deforested regions. A supervised classification revealed that three partially deforested regions contained 17.7%, 12.9% and 13.4% forests in 2000, decreasing to 2.9%, 8.3% and 4.7%, respectively, in 2008. Overall, partially forested regions in the Petén region decreased in forest cover by 9.3% from 2000 to 2008, while there has been a 0.33% per year deforestation rate over the entire Petén basin.

1. Introduction

[2] Central America provides habitat for many of the world's endangered species in an area of now highly fragmented forest. Due to the rapid ecosystem fragmentation and habitat loss, a consortium has proposed to establish the Mesoamerican Biological Corridor, a network connecting protected regions throughout Central America to promote the survival of species. Effective establishment of corridors connecting these forest fragments largely depends on the recovery of currently deforested regions. However, large-scale deforestation in Central America significantly impacts the regional climate. Land use change impacts the local climate primarily by changing the surface energy budget which is dependent on the vegetation type, surface roughness, soil type, soil moisture, albedo and land surface temperature [Pielke, 2001]. Land use heterogeneity alters regional climatic parameters: surface temperature, soil moisture, the depth of the boundary layer, cloudiness, cloud optical properties and local precipitation [Nair et al., 2003; Ray et al., 2003]. These regional climatic changes may, in turn, adversely influence the regeneration of forests. Indeed, DeFries et al. [2002] suggest a system of feedbacks in which land use change may alter the local climate which then enhances and sustains these changes. The issue of the impact of land use change on surface temperatures was also raised in a study by O'Brien [1998] over the tropical forests of Chiapas in Mexico, and by Pielke et al. [2007]. Thus the continued alteration of landscape with high rates of deforestation in Northern Central America may impact 1) the regional climate, 2) the stability of existing protected regions and 3) the regeneration of now-deforested regions.

[3] This study focuses on the low-lands of Guatemala. The Petén basin is characterized by low and uniform elevation, thus eliminating complications arising due to varied topography. The period of study is from 2000 to 2008 using MODerate resolution Imaging Spectroradiometer (MODIS) satellite imagery. Samples of forested habitats, deforested habitats and partially forested and/or fragmented forested regions in the low lands of the Petén basin are chosen for the study. The primary objective is to examine trends in climatic variables in the targeted regions for the 2000 to 2008 time period for both the dry season and the wet season as a function of deforestation. This paper is structured as follows: section 2 provides details of the study area and the data used for the investigation; section 3 contains the results and discussion; and the conclusions are found in section 4.

2. Study Area and Data

[4] The Petén basin is located in the northern portion of Guatemala, roughly bounded from 15°N to 18°N and 94°W to 86°W (Figure 1a). The study area is often referred to as the Maya Lowlands. Figure 1a shows the study region, which includes a variety of land covers, including evergreen forests, deciduous forests, woodlands, croplands and small urban areas. The topography in this area is characterized by low relief (less than 300 m). Figure 1b shows the locations of twelve 30 × 30 km2 boxes of the forested, deforested and partially forested regions used in this study. Each of these individual boxes was chosen to provide an area of fairly homogeneous habitat, as determined by detailed examination of high resolution satellite images, using Google Earth. Boxes 1–6 include mature forests, apparently little disturbed since pre-Columbian times (Figure 1c). These forested areas were chosen to avoid the large seasonal wetlands (“bajos”), which are widespread in the Petén.; Boxes 7–9 represent locations that have been largely deforested during the past two to three decades (Figure 1d); and areas 10–12 are partially deforested, i.e., regions that are in the process of conversion from forests to pastures and crop lands (Figure 1e). The chosen forested locations all fall within park reserves, extractive reserves for conservation, or undeclared protected areas. The deforested areas primarily include land converted from forest to crop lands and urban regions.

Figure 1.

(a) Map of Northern Mesoamerica subset from the Google Earth, (b) map of the Peten basin, (c) 30 × 30 sq. km high resolution images of the forested region (box 1 in Figure 1b), (d) partially forested region (box 10 in Figure 1b) and (e) deforested region (box 7 in Figure 1b).

[5] MODIS Level 1B calibrated radiance data for the period March to September, 2000 and 2008 are used to derive Normalized Difference Vegetation Indices (NDVI), surface temperatures and soil moisture values. March to April is the peak of the dry season [Ray et al., 2006]. The rainy season begins in late May and peaks in August and September. MODIS channels 1 (0.645 μm – visible red) and 2 (0.858 μm – near infra red) are used to derive NDVI values, and channel 31 (11.3 μm) is used for the land surface temperature retrievals.

[6] Soil moisture is estimated by the triangle method of Gillies et al. [1997] using temperature and NDVI values. Gillies et al. [1997] proposed that the triangle shape observed in the NDVI and surface radiant temperature scatter plot is due to the physical characteristics imposed by the soil moisture and fractional vegetation cover. At low NDVI values variation in soil moisture causes wide variation in temperature, but there is less soil moisture induced variation in surface radiant temperature. A soil moisture value of zero represents completely dry soil, while a value of unity represents completely saturated soil, with an accuracy of approximately 2 to 3%.

[7] To estimate the rate of deforestation, a supervised Maximum Likelihood classification [Berendes and Welch, 1993] is performed using the MODIS-derived NDVI values and Channel 1 (0.645 μm), Channel 2 (0.858 μm) data. Figures 2a–2c show clusters in the scatter plots of NDVI versus Channel 1, NDVI versus Channel 2 and Channel 1 versus Channel 2, respectively. Pixels are broadly classified into two classes as forested and deforested regions, and these classes are well-separated. All forested regions defined in the University of Maryland ecosystem map [Hansen et al., 2000], including evergreen needle leaf forests, evergreen broadleaf forests, deciduous broadleaf forests and mixed forests, are grouped together as “forest”. Woodlands, grasslands, croplands, bare ground and urban regions are grouped together as “deforested”. From an overlay of the University of Maryland ecosystem map on the MODIS imagery, a total of 625 pixel samples are selected for supervised training of the deforested areas and 774 pixel samples are selected for the forested areas.

Figure 2.

(a) Scatter plot of MODIS derived normalized difference vegetation index versus MODIS Channel 1 (0.645 μm) data over the Peten basin for March 11, 2000. The two colors blue and red represents the two clusters, forested and deforested regions respectively, (b) same as Figure 2a but depicts for MODIS Channel 2 (0.858 μm), (c) scatter plot of MODIS Channel 1 versus Channel 2 data over the Peten basin.

3. Results and Discussion

3.1. Seasonal Variations of Climate Parameters

[8] The temperature, soil moisture and NDVI values are analyzed on available clear days during 2000 through 2008 over the twelve regions of study. Table 1 shows temperature, soil moisture and NDVI values for the peak of the dry season (March–April) and the wet season (August–September) during 2000 and 2008 for the forested, partially forested and deforested regions. Figure 3a shows the average temperatures for the twelve boxes on 11 March 2000, representing the dry season, and on 21 September 2000, representing the wet season. First, note that surface temperatures are within about 1°C of each other during the wet season for the forested, partially forested and deforested regions, and that values are nearly the same in 2000 and 2008. Clearly, surface temperatures are similar throughout the study area during the wet season regardless of habitat or year. Forested regions are 3–4°C warmer in the dry season than the wet season, and these temperatures are nearly the same for 2000 and 2008. We conclude that the forested regions have not seen notable changes during this decade. In contrast, during the dry season the deforested regions were about 8°C on average warmer than the forested regions in 2000 and about 4°C warmer in 2008 and similarly warmer than deforested areas in the wet season. The temperature of the partially deforested region falls between the extremes of the forested and deforested regions in 2000. However, in 2008 the temperatures in the partially deforested regions are nearly equal to the deforested regions, suggesting that significant deforestation occurred during this eight-year period in boxes 10–12 (see below). These are very large differences which have an affect upon a large number of climatic variables, such as latent and sensible heat fluxes. The combination of the dry bulb temperature and the absolute moisture into moist enthalpy could be an additional way to assess the differences between the forested, partially deforested and deforested regions [e.g., Pielke et al., 2004; Davey et al., 2006].

Figure 3.

(a) The seasonal temperature variation in March and September 2000 over the different boxes of forested, deforested and partial forested regions, (b) time series of NDVI from 2000 to 2008 for dry and wet seasons, (c) the correlation between NDVI and soil moisture for dry and wet seasons in 2000, (d) percentage deforestation for each of the 30 × 30 km2 boxes classified as partially deforested (10–12 in Figure 1b) and averaged over the three boxes, (e) the percentage of deforestation estimated from the supervised classification versus the percentage of deforestation estimated from the NDVI as (NDVIpartial—NDVIdeforested)/(NDVIforest—NDVIdeforested).

Table 1. Climate Variables Over the Different Habitats in the Peten Basin for the Dry and the Wet Seasons of 2000 and 2008 for N Cloud Free Daysa
 TemperatureSoil MoistNDVI
  • a

    Dry season, March–April; wet season, August–September. N cloud free days: N = 5 for dry2000, N = 2 for wet2000, N = 6 for dry2008 and N = 3 for wet2008. NDVI is the Normalized Difference Vegetation Index on a scale 0 to 1; temperature is in degrees Kelvin; soil moisture is on a scale 0 to 1 (saturated). The error value in each box represents the standard deviation of the pixel values from the cloud free days over each habitat (forested, deforested and partially forested).

Forest - 2000301.70 ± 0.24296.67 ± 1.010.69 ± 0.080.89 ± 0.090.70 ± 0.020.74 ± 0.01
Partial–2000305.44 ± 0.88297.22 ± 0.930.44 ± 0.190.73 ± 0.180.53 ± 0.030.71 ± 0.04
Deforest-2000310.03 ± 0.57296.55 ± 1.310.29 ± 0.120.76 ± 0.190.43 ± 0.020.68 ± 0.01
Forest - 2008300.23 ± 2.00297.03 ± 0.310.70 ± 0.160.88 ± 0.090.73 ± 0.000.78 ± 0.02
Partial–2008304.44 ± 4.22296.66 ± 0.880.36 ± 0.250.70 ± 0.070.57 ± 0.080.71 ± 0.02
Deforest-2008304.56 ± 3.20296.75 ± 0.550.36 ± 0.090.65 ± 0.030.60 ± 0.050.67 ± 0.04

[9] During the wet season, soil moisture values are very high (0.88; saturation is a value of 1.0) in the forested regions in both 2000 and 2008, once again suggesting that these regions are relatively stable climatically. However, soil moisture is noticeably lower in the deforested and partially deforested regions even in the wet season for both 2000 and 2008. Soil moisture is lower in the dry season forests than during the wet season and much lower in the deforested and partially deforested regions in the dry season. Note that the same value of 0.36 is found for soil moisture in both deforested and partially forested regions in 2008 stating that the once partially deforested regions have become largely deforested (see below). Also note that soil moisture values are significantly lower in the deforested and partially deforested regions in 2000 than in 2008. Along with the large differences in temperatures between these two years, this suggests that 2000 was much drier than 2008 in deforested regions. The larger differences in the soil moisture values between the forested and deforested regions in 2000 suggest that these partially deforested regions were still significantly forested at that time.

[10] Three way Analysis of Variance (ANOVA) tests were conducted to determine whether the temperature and NDVI are significantly different for different habitats (forested, deforested and partially forested), for different seasons and in different years. The three way interaction of temperature and NDVI showed significant t-values (p < 0.05) implying that the temperature and NDVI varies among different habitats, among seasons and among years.

[11] Table 1 shows that NDVI values are higher in 2008 than in 2000 for the forested regions both in the dry and wet seasons. NDVI values in the wet season are lower in the deforested regions than in forested ones in both years. NDVI of deforested areas differs between wet and dry seasons more markedly than is the case for forested areas. Indeed, Figure 3b shows dry season NDVI values from 2000 to 2008, with the exclusion of 2002 which had heavy cloud cover during the dry season. The deforested and partially deforested regions show much higher values of NDVI during the dry seasons of 2007 and 2008, suggesting much wetter conditions.

3.2. Trends in Deforestation

[12] Large tracts in Northern Guatemala became accessible when the road connecting Central Petén to Guatemala City was constructed. Change detection analysis using high spatial resolution satellite data between 1988 and 1991 showed massive deforestation of the Petén's tropical forests due to the influx of settlers [Sader et al., 1994].

[13] A dry-season Maximum-Likelihood supervised classifier was constructed (at 1 km spatial resolution) to differentiate two classes: forested and deforested, as discussed previously. The classifier was applied only under cloud-free conditions (using the cloud MOD06 product) and for days without aerosol loading (using the aerosol MOD04 product). The classifier was tested using a total of 217 independent and randomly selected pixels. A total of 76 out of 79 forested pixels were identified correctly (96% accuracy) and 128 out of 138 deforested pixels (93% accuracy), for an overall accuracy of 94%.

[14] In order to assess the overall trends in deforestation, clear days (cloud and aerosol free as defined by MODIS products MOD04 and MOD06) were used during the dry season of 2000 (March 11th, March 27th, April 3rd and April 19th) and during the dry season of 2008 (March 6th, March 31st, April 9th and April 25th). The supervised classifier was applied over the entire Petén region for these days. The results show that 49800 hectares of forests have been deforested during the time period 2000 and 2008, or approximately 0.33% deforestation per year. Mayaux et al. [2005] report a similar rate of deforestation in Central American humid forests of approximately 0.4 to 0.5% per year.

[15] Analysis of the various scenes during the dry season and results from Table 1 suggest that it might be possible to construct a very simple classifier of forest/deforested pixels based solely upon dry season NDVI values. In fact, NDVI-based land cover change and classification approaches are widely used in the literature. This section examines the accuracy of NDVI-based land cover change estimates for Guatemala. As discussed previously, Figure 3b shows NDVI values for “clear” dry-season days averaged over forests, deforested and partially deforested regions for each year from 2000 to 2008. It is evident that values of NDVI vary from year to year, with overall higher values in wet years and overall lower values in dry years. There is a high correlation of 0.78 between NDVI and soil moisture, as shown in Figure 3c. The values of NDVI, soil moisture and temperature in partially-deforested regions lie between these forested and deforested extreme values in all years examined, as expected.

[16] Figure 3d shows the percentage of deforestation for each of the 30 × 30 km2 boxes, 10–12 and averaged over the three boxes. The deforestation in these regions has increased from about 85% to about 95% from 2000 to 2008. Partially forested boxes, 10–12 contained 17.7%, 12.9% and 13.42% of forests, respectively, in 2000, and 2.9%, 8.3% and 4.7%, respectively, in 2008. The partially deforested regions have suffered 9.3% decrease in forest cover from 2000 to 2008. A normalized value of NDVI was computed with the relation (NDVIpartial—NDVIdeforested)/(NDVIforest—NDVIdeforested), and this value was correlated with deforestation percentages estimated from the results of classification. Figure 3e shows the cumulative percentage of deforestation between 2000 and 2008 using results of the classification method versus the NDVI approach. Accumulating the results from 2000 to 2008, we see that the total deforestation over this period was approximately 2.4% from the classification method as compared to about 1.8% from NDVI approach. Since the accuracy of the classifier is well-documented, we conclude that the NDVI approach underestimates deforestation rates.

4. Conclusions

[17] The northern region of Guatemala, known as the Petén or Mayan lowlands, has experienced high deforestation rates. A Maximum Likelihood classifier has been constructed to differentiate between dry season (March and April) forest (including all forest types) and deforested conditions (including woodlands, grasslands, croplands and urban areas), with overall accuracy of 94%. Due to a continual influx of settlers 2.64% of the overall region has been deforested from 2000 to 2008. Partially deforested regions have experienced a 9.3% decrease in forest cover over the above period.

[18] Twelve 30 × 30 km2 regions of forested, deforested and partially deforested habitats are selected to compare changes in surface temperature, soil moisture and NDVI values from 2000 to 2008, derived from MODIS satellite imagery. Forested regions have a relatively stable environment with only small variations between the dry and wet seasons over the 2000 to 2008 time period. However, the deforested and partially forested regions have high seasonal variations of temperature, soil moisture and NDVI values. During the wet season they show similar characteristics as the forested region. But, during the dry season the deforested and the partially forested regions become significantly drier and warmer (from 4–8°C warmer than corresponding forested regions) along with significantly lower values of soil moisture and NDVI values.

[19] A simple correlation analysis has also been performed between the different climate parameters to understand the impact of one on the other. Soil moisture and NDVI values show a strong positive correlation of 0.78, while surface temperature and soil moisture have a negative correlation of −0.73. Thus deforestation leads to much greater seasonal variations in climatic values, with higher temperatures and lower soil moisture and NDVI values which impact the regional climate.

[20] An NDVI-based approach was constructed to estimate deforestation rates. However, this approach leads to an underestimation of deforestation rates. For the period between 2000 and 2008, the NDVI approach estimated a 1.8% increase in deforested regions in the Petén region, as compared to 2.64% using the classifier.


[21] This research was supported by National Aeronautics and Space Administration grant NNX07AF14G. The data were obtained from NASA LAADS Web.