Investigating Landscape History

In collaboration with Madeleine Gallop and Tate Sutter

Building on a historical and theoretical narrative of protected areas, the following analysis will investigate land cover stability and variability from 1976 to 2020 in and around Ranomafana National Park, Madagascar. We also evaluate the efficacy of the protected area in reducing deforestation, as well as restoration potential through reforestation in and near the park. 

Ultimately, we’ve found:

1. The landscape in and around Ranomafana National Park is incredibly variable over time, especially near the park’s boundaries.
2. Deforestation levels are lower within the protected area compared to the greater study region.
3. There is promising evidence of reforestation efforts, especially outside of the park and near populated areas, indicating effective local collaboration and restoration efforts.

Introduction and Theoretical Background

Due to its geographic isolation, Madagascar is a biodiversity hotspot, and has become a target for international conservation efforts. The nation has a fast-growing population that depends on the land’s abundance of natural resources, and as population increases, so does the pressure to keep these resources intact (Vuola and Pyhälä, 2016). On one hand, local communities and international governments and organizations share a common goal of preserving land and its resources. But while local communities rely on the ecosystem, governments and international organizations hope to ensure endangered natural habitats remain undisturbed by establishing new protected areas (PAs) (Ibid).

Experts commonly refer to this type of conservation as “fortress conservation,” an exclusive practice that protects land by restricting human access to it, making it unwelcome to those who rely on its resources (Brockington, 2015). The concept of fortress conservation relies on the Western conception of “wilderness” as an idealized landscape devoid of human interference (Ibid). As Dan Brockington, co-Director of Institute for Global Sustainable Development notes, “The myth of wild Africa [is] perpetrated in natural history documentaries which persistently portray landscapes without people as the proper, normal and right condition for African wild places” (Ibid). This myth draws in tourists hoping to witness these wild landscapes, and provides western conservationists with a (profitable) model to create them.

Fortress conservation purifies landscapes, a process which typically involves evictions of existing populations, and large-scale transformations of lifestyles and livelihoods. In post-colonial contexts, the creation of state-imposed protected areas and national parks can remove and modernize local communities to foster “pristine” wilderness (Ibid, Vuola and Pyhälä, 2016). 

Although some conservation efforts attempt to center local communities’ perspectives, negotiations and communications typically exclude local voices or entrench existing hierarchies within these communities (Vuola and Pyhälä, 2016). Conservation must commit to social justice and elevating local voices to ensure fair and sustainable land use (Ibid). 

Ranomafana National Park

The following analysis will explore land use change in and around Ranomafana National Park, located in southeastern Madagascar. Local communities settled in this area in the 18th and 19th centuries. These communities have for centuries relied on subsistence agriculture and forest products in and around Ranomafana. Today, there are over 160 local villages surrounding Ranomafana (Vuola and Pyhälä, 2016).

In 1991, USAID and two American universities began the Ranomafana National Park project. Now, Madagascar National Parks manages Ranomafana and attempts to collaborate with people from the local villages, but some locals do not believe in the legitimacy of the park. Some view it as “an attempt by the foreigners to take away their land” (Ibid). One nearby resident notes, “we have always protected the forest, but now the national park has taken from us the right to protect it and to take benefits from it” (Ibid). 

Nonetheless, organizations such as Centre ValBio, established in 2003, work with local populations to conserve the changing landscape. Centre ValBio works to restore tree cover by planting trees, especially outside of forest boundaries. This process relies on local communities designating sites for reforestation, and illustrates one way that collaboration can create positive change. The following analysis will evaluate the extent to which this landscape has changed, the degree of deforestation it has experienced, and how local-scale reforestation efforts might restore lost tree cover.

Preparing for Analysis

Our study region encompasses Ranomafana National Park and its immediate surroundings (Fig. 1). In order to assess land use change, we differentiated between the four major types of land cover in this region: dense forest, grassland, agriculture, and urban. Each of these classes are described in detail below.

Dense Forest

In this analysis, forest is defined as dark, dense evergreen tree cover. This type of tree cover is distinct from agricultural land use as well as lighter green and brown grassland as it is clearly darker in color, covers large swaths of land, and has much higher tree density.

Grassland

Grassland, unlike agriculture, does not always have a specific or planned shape and is typically not constrained to a single plot. It is not tree’d, making it distinct from the forest class as well. Grassland is typically not irrigated or manicured, as agricultural plots may be.

Agriculture

Planned agricultural plots are distinct from the forest or grassland class they can be nestled within. These areas have a specific use (production) and are typically rectangular or tiered steps. Agricultural areas can range in color depending on geography and time of year.

Urban

Urban areas are clearly distinct from vegetative land cover types and have enough of a presence in the landscape to constitute its own independent class. This land cover is characterized by buildings, roads, parking lots and developed areas, typically outside the boundaries of protected areas.

Temporal Specifications

As we prepared to conduct remote sensing analysis, we decided to examine this area in the dry season, to best differentiate forested land and irrigated rice fields from surrounding grassland. The climograph below (Fig. 2) summarizes the region’s annual temperature and precipitation trends. The target study period extends from May to the end of October, when precipitation and temperature are at their lowest.

Analysis of Land Use Stability and Change, 1990-2020

Figure 3 summarizes the analysis steps taken to create maps of land use change from 1990 to 2020 in and around Ranomafana National Park. In a broad sense, we used spectral and temporal characteristics of Landsat imagery as inputs for a random forest classifier, a type of supervised machine learning algorithm. The output was a custom land cover classification for the region that we applied to four different time periods, allowing us to subsequently examine the changes in classifications over time.

The first step was to create dry season composite images from ten-year intervals over our study period (Fig. 4). These images display noticeable areas of deforestation within the park from 1990 to 2020, particularly near the eastern boundary. However, other forms and areas of land conversion are difficult to discern in a set of simple true color images.

The classified images (Fig. 5) more clearly illuminate areas of land use change. In particular, we can see that the core area of forest remains relatively undisturbed, while the edge of the forest and the surrounding region display a shifting array of land cover types. However, some of the land classified as urban in the western portion of the study region appears to be misclassified. In the true color composite images (Fig. 4), it can be seen that much of this land is not in fact urban, but rather a mix of dryer agriculture and cleared land. When working with land cover classifications, there is some margin of error in the process due to the inherent limitations in using moderate-resolution remotely sensed imagery to infer on-the-ground characteristics. Fortunately, visual inspection indicates that dense evergreen forest cover, the most important subject of our analysis, is correctly classified in most cases.

Figure 6 summarizes the landscape’s stability (and lack thereof) seen over time in the four classified images by differentiating between points that maintained the same land cover from 1990 to 2020 and those that experienced some kind of conversion. Again, it is clear that the core area of forest within the park is able to maintain some degree of stability, while the surrounding region is an ever-changing mosaic of land conversion. 

Accuracy Assessment

Next, we performed an accuracy assessment on the land cover stability image (Fig. 6). We collected a stratified random sample of 225 validation points, then examined the spectral-temporal chart for each point to manually classify it as stable or unstable. The results of this accuracy assessment are shown in Tables 1 and 2. 

Our map of land cover change between 1990 and 2020 provides insights into landscape scale change; however, the accuracy assessment clarifies its reliability. Our map has an overall accuracy of 61.3% and a Kappa coefficient of 0.228. The producer’s accuracy for unstable pixels is 0.636; therefore, 36% of unstable training point locations were misclassified as stable (Table 2). 41% of stable training points were misclassified as unstable. While our classification has errors, it still can provide broad-scale guidance on locations that have experienced land cover change.

However, these accuracy metrics are not ideal for a land cover analysis that is meant to inform conservation efforts. Future work should revise the methodology for the supervised classification and assessment of spectral-temporal characteristics to create a land use change map with a more usable accuracy.

Deforestation and Reforestation

After assessing our land cover stability map, we conducted further analyses to identify specific areas of forest gain and loss.

From Fig. 7, we were able to calculate the total area of deforested land during the study period. Since 1990, about 21% of the study region has experienced deforestation. This loss of dense evergreen forest cover is not solely an issue of the past; 7% of the study region was deforested in the more recent years since 2010. However, the establishment of Ranomafana National Park has been effective to some extent in limiting forest loss. 15% of the park’s land cover has experienced deforestation since 1990, and about 6% of the park was deforested since 2010, demonstrating a lower rate of both overall and recent forest loss within the park relative to the region as a whole.

Much of this deforestation occurs near the borders of the park, while the interior of the park is more protected. Figure 8 illustrates an area of recent forest conversion on the park’s eastern boundary. Closer examination reveals that some of this land has been converted into small agricultural plots, while the remainder has become grassy shrubland.

While there is substantial evidence of deforestation in and around Ranomafana, there is also evidence of reforestation (Fig. 9). As we’ve learned, organizations such as Centre ValBio seek to restore tree cover by planting trees, especially outside forest boundaries. Such organizations often rely on local communities to put forth potential sites for reforestation.

Upon inspection at a more specific scale, small tree plots characterize the landscape under the green classifications in the image on the right (Fig. 10), suggesting collaboration between local communities and preservation organizations. Our depiction of reforestation paints a more hopeful picture for conservation in Madagascar.

Corona Imagery

Interpreting historical land cover information can be helpful to deepen the temporal range of what we know about land use change in and around Ranomafana National Park. For instance, a 2012 study by geographer and professor Christian Kull analyzed highland Madagascar using aerial photos from 1949–1950 and 1991–1994 to deepen the temporal depth of the land use study. Ultimately, the study identified large-scale conversion of land to anthropogenic land cover, such as the conversion of grassland to cropland and the loss of 60% of wetlands between the 1950s and the 1990s. 

In our study, visually interpreting imagery from the Corona Imagery Program provides evidence of historic change before the advent of high resolution Landsat images. Consistent with the results of the Kull study, we found evidence of deforestation and increased anthropogenic land use. 

We analyzed three images covering our study site from the Corona Imagery Program from July 19, 1976. This date falls within the dry months that we used to create our Landsat composite images. The entity IDs for the images are D3C1212-100076A003, D3C1212-100076A002, and D3C1212-100076A001. Though a paywall blocked us from analyzing the high spatial resolution images with a resolution of 2-4 feet per pixel, a cursory analysis of the low resolution sample images provided insight. This provides suggestions of land cover change; however, it should not be understood as a definitive analysis of change between 1976 and 1990.

Concordant with Kull’s results, we found numerous locations exhibiting characteristics of anthropogenic land cover change. In the grasslands and agricultural lands near Ampasina (Fig. 11, site A), land cover change can be seen with dense evergreen forest being cleared. The southwest corner of the images illustrates the deforestation of dense evergreen forest. Similar cases of converting dense evergreen forest to human uses can be seen south of the town of Ranomafana (Fig. 11, site B) and south of Ambinanyndranofotaka (Fig. 11, site C). 

Conclusion

Ranomafana National Park and its surrounding region are incredibly dynamic landscapes, and have been for all of recent history. Visual analysis of Corona imagery highlights the conversion of forest land to anthropogenic land cover prior to 1990. More recent and high resolution imagery exposes both high levels of deforestation as well as promising recent examples of reforestation. Ranomafana’s “protected area” status is somewhat effective in reducing deforestation inside Ranomafana’s boundaries, and evidence of small-plot reforestation outside the park’s boundaries points to the potential of organizations such as Centre ValBio in restoring the landscape both in and around the park. As Madagascar’s population increases and conservation efforts become more widespread, balancing preservation of natural resources with expanding towns and farmland will likely become more difficult. Collaborative conservation efforts will become critical in protecting the diverse landscape for future generations.

References

Brockington, Dan. “The Enduring Power of Fortress Conservation in Africa.” The University of Manchester (2015): 2-12. https://centredestudisafricans.org/wp-content/uploads/2015/11/Brockingtonfinal.pdf

Kull, Christian A. “Air photo evidence of historical land cover change in the highlands: Wetlands and grasslands give way to crops and woodlots.” Madagascar Conservation & Development 7.3 (2012): 144-152.

Madagascar : a world out of time / photographs and text by Frans Lanting ; introduction by Gerald Durrell ; essays by Alison Jolly and John Mack. MADAGASCAR – Images | Frans Lanting Studio. (n.d.). Retrieved December 11, 2022, from https://franslanting.photoshelter.com/gallery/MADAGASCAR/G000089E6PfGpxCs

Ranomafana Weather & Climate | Temperature & Weather By Month—Climate-Data.org. (n.d.). Retrieved December 11, 2022, from https://en.climate-data.org/africa/madagascar/ranomafana/ranomafana-1069257/#climate-graph

Vuola, Marketta, and Aili Pyhälä. “Local community perceptions of conservation policy: Rights, recognition and reactions.” Madagascar Conservation & Development 11.2 (2016): 77-86.https://www.ajol.info/index.php/mcd/article/view/149657

Some code used in this analysis was adapted with permission from He Yin, Kent State University.