Reforestation policies around 2000 in southern China led to forest densification and expansion in the 2010s

Southern China, largely deforested three decades ago, now boasts one of the world's largest tropical and subtropical forest areas. This dramatic shift, from deforestation to reforestation, mirrors similar transitions observed in Europe and the USA during the 19th and 20th centuries and more recently in Vietnam. In China, reforestation efforts intensified after 2000, with large-scale tree planting programs. While increased greenness has been observed, coarse-resolution satellite data hindered the precise identification of the greening's cause: increased density of existing forests or expansion into new areas. This study aims to address this gap by utilizing high-resolution satellite imagery to understand the spatial and temporal dynamics of forestation in Southern China, offering crucial insights into the age structure, growth rates, and fragmentation patterns of the region's forests. This information is critical for assessing biodiversity, ecosystem services, and carbon sequestration potential.

Existing studies using coarse resolution satellite data documented the extensive greening of Southern China, characterizing the transition from grasslands and croplands to tree cover. However, these studies lacked the spatial resolution to distinguish between forest densification and expansion. Upscaling from field plots and stand growth modeling offered overall indications of forest ages, but lacked the detail to capture the spatial nuances of the heterogeneous plantation landscapes. The heterogeneity is further amplified by varying elevation and slope across small areas, typical of the karst regions dominant in Southern China. Previous research focused on large-scale changes, overlooking the complexity of the mosaic of planted forests and individually managed fields, leading to uncertainties in estimates of forest age, growth rate, and potential ecosystem services.

This study utilizes the entire Landsat 5/7/8 archive (1986-2020) available in Google Earth Engine. Atmospheric correction was performed, and clouds and snow/ice were removed using the quality assurance band. Annual median images and a 3-year moving median were used to reduce noise. Vegetation indices (NDVI and NBR) were calculated for each year. A Random Forest regression model, trained with 15,991 dense forest and 158,728 non-forest pixels, was used to estimate "forest probability" (fp). This model incorporated Landsat bands, NDVI, NBR, and a digital elevation model (ASTER GDEM v3). The model's accuracy was 98% (Kappa 93%). A threshold of 50% fp was used to define a "dense forest". Forest age was calculated as the number of years a pixel remained above 50% until 2018. The densification rate was determined as the mean annual probability increment between 20% and 50% fp. Forest fragmentation analysis employed the Morphological Spatial Pattern Analysis (MSPA) method to classify forests into core and non-core areas. This approach improved spatial resolution from previous studies that used a 250m resolution. Polynomial fitting was applied to smooth the forest probability time series. The study also uses biomass data from the ESA CCI Globbiomass dataset to correlate biomass accumulation with forest age and probability.

The study reveals a continued strong forest expansion (dense forest area) from 1986 to 2018, with two pronounced peaks: one in the mid-1990s and another around 2010. Forest extent increased from 249,414 km² in 1986 to 978,954 km² in 2018 (9% to 35% of southern China). Only a small fraction (23%) of the current forest area comprises forests older than the 33-year satellite record (1986-2018), with minimal loss of these older forests. 32% of current forests reached their dense state in the last 10 years. Forests reaching the dense stage more than 10-11 years ago had slow densification rates, while those reaching the dense stage within the last decade had nearly double the rate. This faster growth of recent forests correlates with more favorable climate conditions (higher rainfall) around 2010–2013 and progressive expansion from higher to lower elevations. The increased densification rate suggests the use of fast-growing plantation species. The continued probability increment of forests after reaching a dense state also shows faster growth in recent years. Biomass density increases with forest age up to about 15 years, with older forests showing similar levels, implying growth saturation. The spatial distribution reveals a shift from fragmented forests in 1986 to a vast increase in connected core forests, representing a reversal of forest fragmentation. Core forests increased by 517% from 1986 to 2018, while the average core forest patch size increased from 0.03 km² to 0.07 km².

The findings challenge assumptions about human-forest interactions. Instead of continued deforestation pushing forests to higher elevations, the study shows forest expansion downhill into valleys, indicating a shift in human pressure. The expansion of new forests around older ones acts as a buffer, reducing pressure on the older stands from logging. The massive increase in core forests contradicts the global trend of increasing forest fragmentation, and highlights the role of successful reforestation policies. The accelerated densification rate of forests planted after 2000 is likely due to the selection of fast-growing species. While this resulted in rapid carbon sequestration, the low diversity of these plantations poses potential risks. The impact of climate variability on forest growth should also be considered for future sustainability.

This study provides a high-resolution analysis of forestation dynamics in Southern China, revealing a significant reversal of fragmentation and expansion of core forests. The rapid growth of younger forests, though driven by successful reforestation policies, highlights the need for careful species selection and monitoring to ensure the long-term sustainability and resilience of these forests in a changing climate. Future research should focus on assessing the biodiversity and long-term carbon storage capacity of these newly established forests.

The study's reliance on satellite imagery may underestimate the actual extent of forest cover, particularly for young or sparsely distributed forests. The use of a 3-year moving median and polynomial fitting to reduce noise might have smoothed out some short-term dynamics. The definition of forest age is based on the year a dense state is reached, potentially underestimating the actual planting year if growing conditions were initially unfavorable. The study focuses on dense forests, excluding short-rotation plantations or areas with frequent harvests. Lastly, the study’s definition of core forests might not capture all ecological considerations of connectivity.