SHIFTING CULTIVATION AND FOREST COVER CHANGE
IN NGHE AN PROVINCE, VIETNAM

INTRODUCTION

This paper describes the results of an analysis of the impact of shifting cultivation on forest cover change since 1973 in the uplands of Nghe An Province, Vietnam. The analysis was performed on a 110,000 ha transect located near the border with Laos at elevations between 1,000 and 1,800 m (see map). The transect covers an area with a high incidence of shifting cultivation. Shifting cultivation is a form of agriculture adapted to the uplands where slopes are steep and soils poor, whereby the forest is cut and burned to release nutrients from the forest biomass to the soil. Within a few seasons this temporary fertility is exhausted and new fields are cut. When the population density is low and the area of forest relatively large, shifting cultivation may be environmentally benign. But as population expands and the available forest area shrinks, this process can lead to a high level of habitat fragmentation and an insufficient time for cleared fields to recover before being cut again. In some areas in Nghe An, the fallow length has fallen from 10-15 to 4-5 years and the forest has become increasingly dominated by bamboo and other secondary species (CRES, 1999).

Government officials tend to blame shifting cultivation for most of the deforestation in Vietnam, an attitude that is prevalent in other countries in the region (Brown and Schreckenberg, 1998; Thrupp et al., 1997). However, empirical studies show that government-sponsored forest clearing for cash crops and settlement of migrants, and unrestricted logging by state-owned forestry enterprises, have caused more permanent deforestation than shifting cultivation (Do Dinh Sam, 1994; GOL, 1998). These conclusions are consistent with a study of forest cover change between 1952 and 1995 in a village in Hoa Binh province, which shows low levels of deforestation, but a significant change in forest composition, and a high degree of forest fragmentation (Fox, et al., 1998). Contrary to common perception, the result is not a lunar landscape devoid of all vegetation, but a heterogeneous mosaic of fields, pasture, and forest patches in various stages of secondary succession. This paper aims to complement this site-specific study (740 ha) by analyzing the impact of shifting cultivation on deforestation and forest fragmentation over a much larger area and using more frequent observations.

DATA PROCESSING

Changes in forest cover over time were analyzed using data and methods developed by the Landsat Pathfinder, which is a U.S. multi-agency project that uses Landsat images to map the world’s tropical moist forests. Pathfinder has developed image processing techniques that are generally applicable to the analysis of forest conversion and reforestation in Southeast Asia. This region, which is characterized by a complex mix of deciduous and evergreen forests, and moderate to steep topography, presents a challenge for mapping forest cover using Landsat data. Primary forest, secondary forest, and scrub fallow are difficult to distinguish for the following reasons: forest succession in the tropics can be rapid (young stands can have complete canopy closure with dense green biomass within a few years of abandonment), near-infrared and visible bands saturate rapidly due to the greenness of the regrowth and density of the vegetation canopy, and differences in slope and aspect impact the solar illumination angle and sensor viewing angle.

When analyzing forest cover change over time, low classification accuracies are compounded and reduce the reliability of the change analysis. To improve the accuracy of change detection, high accuracies of the input classifications are needed. This is achieved by using a broad classification scheme whereby the forest class contains areas of primary forest, secondary forest, and scrub fallow. The non-forest class contains areas of barren lands, grasslands, rock, and agricultural areas. Areas that were classified as non-forest in an earlier image and classified as "forest" in a later image are known to be areas of regrowth or forest fallow. New forest fallow and secondary growth are thus inferred from the change detection process, rather than being observed directly.

Change detection is performed using post-classification change detection. Since the images and, therefore, classifications are co-registered, changes in land cover can be tracked on a pixel by pixel basis by combining the individual classifications of forest and non-forest areas. By overlaying these classifications, changes in class between dates can be mapped, which provides information on gross and net changes in forest cover, as well as the spatial patterns of land use.

Eleven Landsat images covering the transect were acquired from the satellite receiving station in Bangkok (see Table 1). They were georeferenced to the 1992 image using an affine shift or first-order polynomial transformation, and resampled to 30 m. The rectification RMS error ranged from 0.96 to 1.06 pixels. In principle, the low RMS error and high spatial resolution allow inter-annual changes in forest cover to be accurately mapped. Each image was then classified using unsupervised classification techniques, combined with GIS analysis, into five classes: forest, non-forest, water, cloud, and cloud shadow. Forest is defined as land with a tree canopy cover of at least 30 percent. Each image was classified separately and then merged into a change detection (matrix) image. The matrix image was filtered to remove clusters smaller than three pixels (0.27 ha) possibly caused by misregistration and system noise, and then clumped to assign a unique value to each cluster. The classification, filtering, and clumping were carried out using Erdas Imagine image processing software.

Table 1. Input data characteristics

DEFORESTATION

The matrix image was used to calculate net and gross deforestation figures for each time period. The results are shown in Table 2. Annual percent changes were calculated relative to the area of forest in the first year of the time period.

Table 2. Net and gross deforestation, 1973-98

These results show periods of net increase and net decrease in forest cover since 1973, with no clear trend emerging. The low levels of net deforestation are due to rapid forest regrowth after fields have been abandoned into fallow. However, the results show high levels of gross deforestation, and a general increase in the rate of gross deforestation since 1973, particularly since 1989. This trend may be explained by policies introduced in the late 1980s, which transferred land management from the communes to households and allowed farmers to expand cropland by clearing previously intact forest. These policies, which were aimed at increasing food self-sufficiency, appear to have triggered a wave of deforestation.

The impact of these policy shifts may be inferred from Table 3, which shows a doubling in the average annual rate of net deforestation and a 4-fold increase in the average annual rate of gross deforestation between 1973-89 and 1989-98.

Table 3. Net and gross deforestation rates, 1973-89 and 1989-98

FOREST FRAGMENTATION

The number and size of forest patches were calculated for 1973, 1984, 1989, 1995, and 1998. In order to focus on the fragmentation of undisturbed forest, the analysis was only applied to pixels that were classified as forest throughout the study period up to the date in question. The fragmentation analysis was thus applied to a steadily decreasing area of forest. The patch size distribution was calculated using ESRI GRID GIS software. The results are given in Table 4.

Table 4. Forest patch size distribution, 1973-98

The table shows a steady increase in the number of patches and decline in the average patch size between 1973 and 1998. Again, these changes were most pronounced after 1989.

CONCLUSIONS

The application of the Pathfinder forest cover classification and analysis methods to a transect of shifting cultivation shows low levels of net deforestation, but high levels of forest fragmentation. This finding, which is consistent with Fox, et al. (1998), suggests that shifting cultivation may be highly sustainable in terms of its impact on forest cover and associated biological and hydrological systems, and that government effort to stabilize or resettle shifting cultivators may be misguided. Low levels of net deforestation suggest that more attention should be paid to improved fallow management and its role in crop diversification and food security. The study also shows a strong correlation between rates of deforestation and forest fragmentation and changes in government policy. From a technical standpoint, the Pathfinder approach permits the rapid analysis of forest cover change over large areas, albeit using a very simplified classification scheme. As such, it may be suitable as the basis of an operational forest monitoring system.

REFERENCES

Brown, David and Kathrin Schreckenberg (1998) Shifting Cultivation as an Agent of Deforestation: Assessing the Evidence, ODI Natural Resources Perspectives, available from http://www.oneworld.org/odi/nrp/29.html

Do Dinh Sam (1994) Shifting Cultivation in Vietnam: its Economic and Environmental Values Relative to Alternative land Uses, IIED, London.

Fox, Jeff, Dao Minh Truong, Terry Rambo, Nghiem Phuong Tuyen, Le Trong Cuc, and Stephen Leisz (1998) Shifting Cultivation with Deforestation: a Case Study in the Mountains of Northwestern Vietnam, mimeo, East-West Center, Honolulu, HI.

GOL (1998) Report to the 4th Meeting of the GMS Working Group on Environment, Hanoi, November 1998.

Thrupp, Ann, Susanna Hecht, and John Browder (1997) The Diversity and Dynamics of Shifting Cultivation: Myths, Realities, and Policy Implications, World Resources Institute, Washington, DC.

CRES (1999) Ca River Basin Environmental Assessment, Center for Natural Resources and Environmental Studies and Vinh University, World Resources Institute, Washington, DC.