Detection of Land-Use Change and Rapid Recovery of Vegetation after Deforestation in the Congo Basin

Coralie E. Adams aCentre for Atmospheric Science, University of Manchester, Manchester United Kingdom

Search for other papers by Coralie E. Adams in
Current site
Google Scholar
PubMed
Close
https://orcid.org/0000-0003-1639-7690
and
Luis Garcia-Carreras aCentre for Atmospheric Science, University of Manchester, Manchester United Kingdom

Search for other papers by Luis Garcia-Carreras in
Current site
Google Scholar
PubMed
Close
Open access

Abstract

The Congo Basin is severely understudied compared to other tropical regions; this is partly due to the lack of meteorological stations and the ubiquitous cloudiness hampering the use of remote sensing products. Clustering of small-scale agricultural deforestation events within the basin may result in deforestation on scales that are atmospherically important. This study uses 500-m MODIS data and the Global Forest Change dataset (GFC) to detect deforestation at a monthly and subkilometer scale and to quantify how deforestation impacts vegetation proxies (VPs) within the basin, the time scales over which these changes persist, and how they are affected by the deforestation driver. Missing MODIS data meant that a new method, based on two-date image differencing, was developed to detect deforestation on a monthly scale. Evaluation against the yearly GFC data shows that the highest detection rate was 79% for clearing sizes larger than 500 m2. Recovery to predeforestation levels occurred faster than expected; analysis of postdeforestation evolution of the VPs found 66% of locations recovered within a year. Separation by land-cover type also showed unexpected regrowth, as over 50% of rural complex and plantation land recovered within a year. The fallow period in the study region was typically short; by the sixth year after the initial deforestation event, ∼88% of the locations underwent a further considerable drop. These results show the importance of fine spatial and temporal information to assess Congo Basin deforestation and highlight the large differences in the impacts of land-use change compared to other rain forests.

© 2023 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Coralie E. Adams, coralie.adams@postgrad.manchester.ac.uk

Abstract

The Congo Basin is severely understudied compared to other tropical regions; this is partly due to the lack of meteorological stations and the ubiquitous cloudiness hampering the use of remote sensing products. Clustering of small-scale agricultural deforestation events within the basin may result in deforestation on scales that are atmospherically important. This study uses 500-m MODIS data and the Global Forest Change dataset (GFC) to detect deforestation at a monthly and subkilometer scale and to quantify how deforestation impacts vegetation proxies (VPs) within the basin, the time scales over which these changes persist, and how they are affected by the deforestation driver. Missing MODIS data meant that a new method, based on two-date image differencing, was developed to detect deforestation on a monthly scale. Evaluation against the yearly GFC data shows that the highest detection rate was 79% for clearing sizes larger than 500 m2. Recovery to predeforestation levels occurred faster than expected; analysis of postdeforestation evolution of the VPs found 66% of locations recovered within a year. Separation by land-cover type also showed unexpected regrowth, as over 50% of rural complex and plantation land recovered within a year. The fallow period in the study region was typically short; by the sixth year after the initial deforestation event, ∼88% of the locations underwent a further considerable drop. These results show the importance of fine spatial and temporal information to assess Congo Basin deforestation and highlight the large differences in the impacts of land-use change compared to other rain forests.

© 2023 American Meteorological Society. For information regarding reuse of this content and general copyright information, consult the AMS Copyright Policy (www.ametsoc.org/PUBSReuseLicenses).

Corresponding author: Coralie E. Adams, coralie.adams@postgrad.manchester.ac.uk

1. Introduction

Despite its importance as one of the largest tropical forests, the Congo Basin is severely understudied compared to other tropical regions. The Congo Basin has relatively low amounts of deforestation for a tropical forest, but the rates have significantly increased in recent years, mirroring population growth; a fivefold increase in population is predicted by 2100, which may further increase deforestation rates (Tyukavina et al. 2018). Current deforestation of the Congo Basin is primarily driven by small-scale clearing for agriculture and secondarily driven by selective logging (Tyukavina et al. 2018). The drivers of deforestation are likely to change as foreign investments in agro-industrial plantations and nonrenewable resources have significantly increased over recent years (Feintrenie 2014); a GIS-based assessment predicted that logging concessions are the most sensitive variable when predicting future deforestation rates (Zhang et al. 2006). It is, therefore, likely that deforestation rates will continue to increase in conjunction with changes in drivers of deforestation. This will change the patterns of deforestation and impact how the land evolves afterward.

The small-scale agricultural clearing activity within the dense, humid forest region of the Congo Basin has resulted in the rural complex, an inhabited agricultural land-cover mosaic comprising active and fallow fields, and secondary forest regrowth (Molinario et al. 2015). The successional fallow cycle that occurs during the agricultural clearing process can be summarized as follows: the tropical mature forest is deforested and crops grow quickly, the land is cultivated for 1–3 years after deforestation and is then left for fallow, tropical forest regrowth then occurs, and the land is deforested again (Fig. 1) (Akkermans et al. 2013; Makana and Thomas 2006). The cycle is ongoing for nearly all land deforested using the traditional slash-and-burn techniques and only stops when complete degradation has been reached (van Vliet et al. 2012). Fallow duration varies across the basin; one study reported mean fallow duration varying from 4.3 to 5.2 years (Moonen et al. 2019), whereas another reports fallow duration exceeding 20 years in the basin (Makana and Thomas 2006). Analysis of successional stages found that pioneer species close the canopy within 5 years after farm abandonment (Makelele et al. 2021). On longer time scales, small-scale agricultural clearing has been shown to significantly reduce the aboveground carbon storage (Depecker et al. 2021; Makelele et al. 2021), and the species composition remains distinct from that of the old-growth forest (Depecker et al. 2021).

Fig. 1.
Fig. 1.

Successional fallow cycle, adapted from Akkermans et al. (2013).

Citation: Earth Interactions 27, 1; 10.1175/EI-D-22-0020.1

Studies of regrowth time scales in the Congo Basin have often focused on selective logging—the process of cutting down selected trees within a forest and leaving the rest intact. A study of selective logging in the basin found that vegetation recovers on time scales of ∼2–3 years after the deforestation event (Hirschmugl et al. 2013). Given the majority of deforestation in the region is small-scale agricultural clearing (Tyukavina et al. 2018), it is imperative that studies of regrowth in the Basin consider the impacts of this deforestation type. Studies focusing on the short-term impacts of deforestation are also lacking in the region, and datasets that identify the spatial location of deforestation in the Congo Basin, such as the Global Forest Change dataset (GFC) (Hansen et al. 2013), are typically only available annually. Vegetation regrowth after small-scale agricultural clearing is relatively fast and can affect spectral reflectance significantly within 1 year in the basin (Rahm et al. 2013). The rapid changes in vegetation cover associated with the fallow cycle in the basin mean that identifying the point in the year in which deforestation occurs and the recovery time scales after deforestation are essential for understanding the subsequent impact on surface properties.

Remote sensing products are vital for forest monitoring; they represent the primary source of spatially explicit data for monitoring land-cover change globally (Grogan et al. 2016) and can be used to identify the time of deforestation. Detecting deforestation over time is essential for environmental management (Appiah Mensah et al. 2019). NASA’s Moderate Resolution Imaging Spectroradiometer (MODIS) products are widely used to assess deforestation and agriculture in tropical regions, due to their high-temporal, moderate-spatial resolution coverage (Brown et al. 2013; Koltunov et al. 2009). Few studies have used MODIS products to assess Congo Basin deforestation. This is partially due to the spatial resolution of the products; the smallest-resolution MODIS product has a pixel size of 250 m, which is larger than the average clearing patch size of around 118 m in the Democratic Republic of the Congo (DRC) (Potapov et al. 2012). This issue is common for slash-and-burn clearing globally and can be alleviated by using both Landsat and MODIS data, as done in Central America (Hayes and Cohen 2007) and the Congo Basin (Hansen et al. 2008). MODIS products have been successfully used in other tropical regions to detect deforestation, and a variety of methods have been applied (Arai et al. 2011; Grogan et al. 2016; Hammer et al. 2014; Hansen et al. 2008; Morton et al. 2005).

Small-scale deforestation can impact local climate by altering surface fluxes of heat, energy, and moisture. Atmospheric impacts are typically only observed when surface features are large enough to be detected when averaged across scales larger than the boundary layer depth (∼1–2 km), as boundary layer mixing will act to homogenize the effects of smaller-scale features (Baidya Roy and Avissar 2002; Khanna et al. 2017). The rural complex comprises deforestation events that are substantially smaller than the kilometer scale; however, the nature of the rural complex means that deforestation events typically occur close together spatially (Molinario et al. 2015) and are, therefore, likely to aggregate to scales that are atmospherically important. The use of coarser products, such as those from MODIS, allows us to specifically identify features that are likely to be climatically relevant. Furthermore, these products have a substantial added benefit of providing much higher temporal resolution, which also enables assessment of deforestation in the context of seasonal variations.

Vegetation proxies are often used to assess the coupling strength between vegetation and the atmosphere. Back-trajectory analysis in tropical regions found a strong relationship between leaf area index (LAI) and subsequent rainfall, dependent on the water-recycling ratio (Spracklen et al. 2012). Another study found a negative relationship between the enhanced vegetation index (EVI) and land surface temperature (LST) in the Horn of Africa and used this research to investigate the impact of forest loss; it resulted in average local heating of up to 6°C (Abera et al. 2018). Assessment of the impact of deforestation on vegetation proxies can, therefore, help us understand the potential atmospheric impacts. Studies focused on the climatic impacts of realistic deforestation scenarios in the Congo Basin are lacking as research has typically focused on idealized scenarios of severe deforestation in the basin (Bell et al. 2015; Nogherotto et al. 2013; Shem 2006). Consequently, there is a research gap surrounding realistic deforestation patterns, which are climatically important.

In this study, we assess whether subkilometer-scale Congo Basin deforestation can be observed on scales that are relevant for atmospheric impacts, and over what time scales these changes persist. The objectives of this study are the following: (i) use remote sensing vegetation products to identify deforestation events at a monthly and kilometer scale; (ii) assess or quantify the evolution of vegetation products after deforestation, including time scales of recovery and recurrence of deforestation; and (iii) assess how the evolution of vegetation products after deforestation is affected by land-cover type. Two-date image differencing was applied to MODIS vegetation data to pinpoint the exact month of deforestation from the GFC annual data. Regrowth time scales and patterns were then assessed using the MODIS vegetation products and were separated by land-cover type.

2. Materials and methods

a. Study area

The study area is in the northern region of the Congo Basin, to the north of the Congo River and the town of Bumba (Fig. 2a). This region was chosen due to the large deforestation patches present in the region, which include both commercial plantations and rural complex land (Fig. 2b).

Fig. 2.
Fig. 2.

(a) Land-cover type map of the Congo Basin from the Congo Basin vegetation-types map (Verhegghen et al. 2012). The study region is denoted by the red rectangle; white patches are areas outside of the map’s domain. (b) Cumulative forest loss during 2000–19, calculated as the percentage of the 500-m MODIS pixel that has been deforested according to the Global Forest Change 2000–19 (Hansen et al. 2013). White patches are water bodies; magenta patches are commercial plantation land.

Citation: Earth Interactions 27, 1; 10.1175/EI-D-22-0020.1

b. Data sources

All data sources used in this study are detailed in Table 1. The dates for this study cover the data period shared by all the used datasets: 2002–20.

Table 1.

Datasets used in this study and their spatial and temporal scales/resolution, time period, and high-quality data coverage for the study area.

Table 1.

Forest loss and forest cover data were a subset from the GFC, which is derived from decision-tree analysis of spectral metrics derived from Landsat-8 Operational Land Imager (OLI) data at a pixel resolution of 30 m (Hansen et al. 2013). The OLI is a pushbroom sensor that views a 185-km, across-track ground swath and retrieves image data for nine spectral bands (Irons et al. 2012). This dataset contains the following variables, which were utilized in this study: tree canopy cover for the year 2000 (treecover2000) and year of gross forest cover-loss event (loss-year). Forest loss was classified as the complete removal of the tree cover canopy of the pixel or as a stand-replacement disturbance (Hansen et al. 2013).

The normalized vegetation difference index (NDVI) and the EVI, referred to as vegetation indices (VIs), are proxies for vegetation activity globally and are produced by MODIS, a satellite-borne instrument (Didan et al. 2015). This study uses data from both the Terra and Aqua satellites. VIs are derived from analysis of spectral reflectance; green leaves have a unique spectral signature and have very low reflectance in the visible spectrum, due to high absorption in the red and blue wavelengths, and have high scattering in near-infrared radiation, which corresponds with the canopy’s structural properties (Didan et al. 2015). The contrast between the red and near-infrared absorption properties can, therefore, be used as a precise measure of vegetation amount and results in the NDVI. However, the red band tends to saturate for high vegetation amounts, due to chlorophyll absorption, and, therefore, the resultant VI is less accurate for high vegetation amounts. To counteract this issue, the EVI product was developed, which has improved sensitivity to high vegetation amounts and a reduction of atmospheric influences; the EVI product is recommended for areas with dense vegetation (Didan et al. 2015). The MODIS LAI 8-day 500-m product was also used for deforestation detection. The LAI is a unitless quantity defined as the leaf area per unit of ground area. This product utilizes an algorithm that analyzes the vegetation structure, sun-sensor geometry, bidirectional reflectance factor at red and near-infrared spectral bands, and associated uncertainties (Myneni 2020).

The study area includes commercial plantations that were not differentiated in the Congo Basin vegetation types map. The Global Forest Watch map (Harris et al. 2019) and a deforestation investigation of the region (Earthsight 2018) were used to identify plantations in the region that were then verified by visual interpretation of high-resolution Google Earth imagery (Google Earth 2020).

c. Data processing

The associated data-quality flags were applied to all MODIS data to ensure that only the highest quality data were used. The GFC has a finer pixel resolution than the MODIS 500-m products and was resampled to match the MODIS resolution. The treecover2000 dataset was resampled using linear interpolation. Loss-year data were resampled by calculating the percentage of pixels within the resampled area that were lost for each year. Land-type data were interpolated from 300-m pixel resolution to 500 m using the nearest-neighbor method.

d. Data quality

The Congo Basin has some of the highest cloud coverage in the world (Wilson and Jetz 2016); the frequent cloudiness greatly reduces the number of successful retrievals of land surface properties. The lack of high-quality remote sensing and ground station data in the Congo Basin is a key constraint for studies of the area. Overall data coverage for the study area for all data sources is shown in Table 1. The seasonal cycle of MODIS VI retrievals (Fig. 3) has the most consistent high-quality data for both Terra and Aqua in November, December, and January, which partially aligns with the dry season [December–February (DJF)]. However, the other dry season (JJA) has relatively low amounts of high-quality data as it has the highest proportions of pixels contaminated due to both cloudiness and aerosols, compared to the other seasons. The Terra retrievals have consistently higher proportions of high-quality data than the Aqua retrievals on similar dates; this is to be expected as the Aqua satellite traverses the region in the early afternoon when convection, and cloudiness, is maximized.

Fig. 3.
Fig. 3.

MODIS 500-m VI data quality aggregated for each time stamp from 2000 to 2021 (Terra) and from 2002 to 2021 (Aqua).

Citation: Earth Interactions 27, 1; 10.1175/EI-D-22-0020.1

e. Change-detection algorithm

The GFC is only available annually; the rapid regrowth in the basin means that identifying the point in the year when deforestation occurs is important for understanding the subsequent impact on surface properties. Change-detection methods are used to detect deforestation and typically fit into two categories: either searching for anomalies from stable conditions by using annual or biannual time scales or through trend analysis of the entire time series, which requires denser time series and enables more subtle trend detection (Grogan et al. 2016). Due to the high amounts of missing data (Fig. 3), trend analysis is difficult or currently impossible; methods based on change-detection algorithms were tested but did not work well because of this problem. To overcome this issue, we developed a method based on two-date image differencing that was better able to deal with the irregular data availability in the region. We applied this method to identify the exact month of deforestation. The MODIS vegetation products (LAI, EVI, and NDVI) were used as vegetation proxies (VPs) and were each cross-referenced with the GFC loss-year data to assess performance. The deforested locations detected by the VP with the best performance were then used to assess how deforestation affects vegetation amounts in the Congo Basin.

Pixels were initially categorized as forested or deforested using the yearly forest cover data from GFC. Deforested MODIS pixels had to have at least 60% of the finer GFC pixels over the deforestation threshold, and treecover2000 had to have more than 70%. Forested MODIS pixels had to have at least 75% of the GFC pixels, and treecover2000 had to have more than 70% and cumulative deforestation lower than 5%. Deforested locations were only analyzed if: the surrounding 20 km of forest had more than 80% high-quality data, the sample size for the standard deviation reached the minimum required, and the deforested location had high-quality data for the year before, year of, and year after deforestation.

Deforested MODIS pixels were compared to the surrounding (<10 km away) forested MODIS pixels to isolate the effects of land-use change from broader changes caused by changing environmental conditions. For each deforested MODIS pixel at location s and time t, the change in VP relative to its surroundings, ΔVP(s, t), was calculated by subtracting the mean of a 10-km buffer of the surrounding forested MODIS pixels, VPF(s ± 10 km, t), from each deforested MODIS pixel, VPD(s, t), that is,
ΔVP(s,t)=VPD(s,t)VPF(s±10km,t).
Two-date images were created for every time stamp available for all locations flagged as deforested by the GFC. A lack of available high-quality data hindered assessing every time stamp; when high-quality data meeting the requirements stated previously were not available, the time stamp was omitted. Two-date image differencing was applied to the ΔVP(s, t) values to compare the year of and year before deforestation and assess whether the difference was larger than two standard deviations of the predeforestation mean (Fig. 4). This was repeated for each VP. Deforestation may occur late in the year and, due to missing data, the impact may not be observed in the year of deforestation; the analysis considers this by assessing whether the drop is observed in the ensuing year. The predeforestation mean is calculated for the same date for all years before deforestation, although we also included the time steps before and after the deforestation date to increase the sample size. We cycle through all the dates in the deforestation year until a drop in ΔVP(s, t) larger than two standard deviations of the predeforestation mean is identified. The analysis was run for 500-m pixels at a 60% deforestation threshold. The results were tested for a range of deforestation thresholds (30%, 40%, 50%, and 60%) to allow detection comparison. Cohen’s kappa, a statistical coefficient that represents the degree of accuracy and reliability in a statistical classification, was used to assess the statistical significance of deforestation detection.
Fig. 4.
Fig. 4.

Diagram of the application of two-date image differencing for deforestation detection.

Citation: Earth Interactions 27, 1; 10.1175/EI-D-22-0020.1

To improve confidence in our results, we calculated the minimum sample size (n) needed for the predeforestation standard deviation calculation, using the z value for a confidence level of 95%, the population’s standard deviation (σ), and a margin of error (MOE), which was adjusted for each VP:
n(zσMOE)2.
The minimum sample size was calculated using Eq. (2). The analysis looks at deforestation from 2010 to 2017 to allow for a large enough sample for the predeforestation mean calculation and to allow for analysis of postdeforestation regrowth. The MOE was derived by testing how different MOEs affected the resultant kappa coefficient. The NDVI had an MOE of 0.027, the EVI had an MOE of 0.025, and the LAI had an MOE of 0.4. The minimum sample size was calculated for each time stamp assessed.

3. Results

a. Deforestation detection

Cumulative forest loss over the study region shows large patches of deforestation (Fig. 2). The most common land types in the study area are dense moist forest (63%), rural complex (22%), and edaphic forest (11%); dense moist forest and edaphic forest were grouped to form the forest group. The mean clearing size was 0.0038 km2 [95% confidence interval (CI) = ±4.97 × 10−5], and the median was 0.0018 km2, thus showing a positive skew toward larger clearing sizes. These values are smaller than the average clearing size of DRC, 0.014 km2 between 2005 and 2010 (Molinario et al. 2015), which may reflect the area or the time chosen.

We tested the performance of MODIS VPs at detecting deforestation and used the results to determine the exact month of deforestation (Table 2). Of the three VPs we used, NDVI had the highest detection rate. All VPs had similar kappa coefficients (0.87–0.90), indicating a substantial level of agreement (Table 2). The MODIS NDVI product has been used extensively to detect deforestation; analysis in the Amazon has resulted in a 0.62 kappa coefficient (Hammer et al. 2014) and 0.70–0.94 kappa coefficient (Arai et al. 2011). The deforestation detection algorithm using the NDVI performed well; its 79% detection rate, at 500-m pixel size and 0.89 kappa coefficient, is comparable to other methods of forest monitoring.

Table 2.

Change detection for deforested and forested locations for MODIS variables and their corresponding κ coefficient. The number in parentheses signifies the number of locations.

Table 2.

To understand the disparity in deforestation detection between NDVI and EVI, we looked at the distribution of their values (Fig. 5). The NDVI peak is narrower and close to the maximum possible value for the index, with a sharp right-hand shoulder, thus indicating that the saturation effect typically observed over humid forests (Huete et al. 2002) can also be observed in this study area. The EVI peak is relatively broad and encompasses a larger range than NDVI. This difference can explain the disparity in deforestation detection; the NDVI has a much smaller standard deviation and, therefore, drops due to deforestation are more obvious and easier to detect. On the other hand, the broader distribution of EVI values may mean the EVI is better suited for quantifying changes in vegetation properties in the region, as it can detect subtle differences in vegetation amount. For this reason, for the rest of the analysis, we use the NDVI results for deforestation detection and the EVI and LAI, referred to as VPs, to look at subsequent vegetation changes.

Fig. 5.
Fig. 5.

Normalized histogram of all EVI and NDVI results for forested locations from 2000 to 2019, both Aqua and Terra satellite data.

Citation: Earth Interactions 27, 1; 10.1175/EI-D-22-0020.1

b. Subsequent regrowth

We then investigated the VP trends for the years before and after the identified deforestation event to assess how deforestation affects vegetation properties and how these subsequently evolve in the following years (Fig. 6). The changes in VPs due to land-use change is likely to result in climate impacts and, therefore, analysis of subsequent regrowth can indicate the longevity of atmospheric impacts. Regrowth and recovery are discussed in terms of the VPs; assessment of the vegetation type after the deforestation event is not within the scope of this study.

Fig. 6.
Fig. 6.

Boxplots of (a) EVI and (b) LAI from 3 years before deforestation to 8 years after. The horizontal dashed line indicates the median value of the 3-yr predeforestation, and notches on the boxplot indicate the 95% CI of the median. Outliers have a 0.5 opacity and so appear darker where there is a high density of outliers. Colored circles denote the statistical significance by which the forested and deforested distributions are different.

Citation: Earth Interactions 27, 1; 10.1175/EI-D-22-0020.1

The predeforestation years consistently have a relatively small interquartile range for both VPs; this is consistent with the land being forested before deforestation, as we would not expect large variations in an established forest. However, the EVI and LAI are both slightly higher for the deforested land (before deforestation occurs) than the surrounding forested land (p < 0.01, Welch’s t test), suggesting that predeforestation land is not the same as the surrounding forest. Distinguishing between primary and secondary forests is beyond the land-type input data and the scope of this study, but predeforestation values likely comprise both primary and secondary forests, as the area contains a large amount of historic rural complex land. This may explain why the predeforestation land is significantly different from surrounding forest land.

A comparison of the VP distributions for the years before and year of deforestation shows a clear significant drop (p value < 0.01), with most of the distribution falling below the average predeforestation values for both the EVI and the LAI (Fig. 6). The EVI recovered to predeforestation values swiftly, as 66% of deforested locations recovered to their predeforestation levels within a year. The median of all deforested locations also recovers rapidly (Fig. 6) and has a higher median than the predeforestation median within 2 years after deforestation for the EVI (Fig. 6a) and the LAI (Fig. 6b). The LAI was slower to recover, as 45% of locations recovered to their predeforestation levels within a year. The rapid recovery of VPs suggests that the downstream impacts on LST, and other climatic variables, may also only be impacted for a short time.

We tested the LAI sample to assess whether the same sample for EVI was representative of the whole population. The samples are significantly different (p < 0.01, Welch’s t test) and, therefore, this must be considered when analyzing our results. To test whether the slower recovery of the LAI was due to the differences in the sample size of the two VPs, the analysis was repeated for EVI using the same locations as the LAI analysis. The only change to the results observed was a small reduction in the percentage of recovered locations within a year (62%), and so this difference does not appear to be related to differences in the sample size.

The swift recovery of both VPs indicates that the vegetation type present in the year after deforestation often has a VP value comparable to the predeforestation vegetation type; however, VPs cannot be used to distinguish vegetation types. In line with the typical fallow cycle in the region, it is likely that the subsequent vegetation type after deforestation is either tropical young or old cropland, or ensuing pioneer species. For both VPs, the median peaks in the third year and then decreases and stays at similar values for the remaining years. The range of the data decreases with time after deforestation, thus indicating a larger variability in surface properties after deforestation, which becomes less variable with time.

Further clearing after the initial deforestation event was expected as repetition of the fallow cycle is common in the region. We applied the deforestation detection algorithm to the deforested location’s NDVI for the years after the initial event to assess the occurrence of subsequent drops (Fig. 7). Locations were split by the minimum number of successive available years after the initial deforestation event to ensure unavailable data do not skew our results. The proportion deforested in each year typically stays at similar amounts regardless of the minimum number of available years; this adds confidence to our results. The largest mean and median proportion of drops occurred in year 4 (17.5% and 17.7%) and the smallest proportion of drops occurred in years 5 and 6. By the sixth year after the deforestation event, ∼88% of deforested locations have undergone a further substantial drop after the initial event. These results indicate that the fallow period in the region is typically very short in duration and, when subsequent drops occur in years 1–3, potentially does not occur at all. Other studies have reported longer periods before the subsequent clearing; one study reports that fallow periods can be up to and over 20 years (Makana and Thomas 2006), and another found that the fallow duration varies between the study area and ranged from 1.5 to 5 median years (Moonen et al. 2019). This suggests that this region has a much shorter fallow cycle than is typical for the Congo Basin, which could affect regrowth.

Fig. 7.
Fig. 7.

Cumulative bar plot of NDVI detected drops after the initial deforestation event, grouped by the minimum number of available years after the initial deforestation event. Bars are split by the year of deforestation; the small gray numbers on top represent the total number of drops/the total sample size.

Citation: Earth Interactions 27, 1; 10.1175/EI-D-22-0020.1

The few existing studies quantifying regrowth in the region have focused on selective logging and found recovery of the green vegetation fraction within 2–3 yr after the deforestation event (Hirschmugl et al. 2013). Results from selective logging studies in the Amazon using VIs were broadly consistent with those observed in the Congo Basin (Koltunov et al. 2009) (Souza et al. 2005). Therefore, our results indicate a faster VP recovery from small-scale agricultural deforestation and so, from an atmospheric impacts perspective, the effects might be relatively short-lived. The vegetation type of regrowth cannot be identified in this study, and this faster recovery may indicate a different vegetation type with a faster growing rate.

Land-cover type

We then split the data to assess how deforestation and subsequent regrowth are impacted by the predeforestation land type. All locations used were classed as forested in the GFC, but some of these locations are further classified in the land-cover-type map (Fig. 2a) and by identification of commercial plantations (Fig. 2b). The predeforestation land types of the identified deforested locations were forest (42%), plantation (42%), and rural complex (16%). The temporal evolution of VP distributions for different land types is shown in Fig. 8.

Fig. 8.
Fig. 8.

Boxplots of all vegetation parameters as a time series of 3 years before deforestation and 8 years after. The horizontal dashed line indicates the median value of the 3-yr predeforestation; notches on the boxplots indicate the 95% CI of the median. Outliers have a 0.5 opacity and so appear darker where there is a high density of outliers. (a) EVI. (b) LAI. Land-cover types: (i) all data, (ii) plantations, (iii) rural complex, (iv) forest.

Citation: Earth Interactions 27, 1; 10.1175/EI-D-22-0020.1

The rural complex undergoes rapid regrowth; median recovers within a year for both VPs. Forested land has the slowest recovery; it takes 2–3 years for the median to recover for both VPs. Full recovery to predeforestation values within a year for the rural complex land is surprising as small-scale deforestation in the region is typically followed by 1–2 years of cultivation and then the land is abandoned and left for fallow (Makana and Thomas 2006). We, therefore, expected slow regrowth during the cropland years and then faster regrowth after the land was left to fallow. The forested land’s EVI does not fully recover until the fallow years and aligns more with our expected regrowth trends. A Borneo study assessed how EVI varies for different land-cover types; the range of values for the evergreen broadleaf forest overlaps with the cropland values (Vijith and Dodge-Wan 2020). The rapid recovery of rural complex EVI, therefore, may indicate cropland growth.

The slower recovery of forested land, compared to rural complex land, is an unexpected finding as it is likely that the rural complex land has undergone multiple slash-and-burn cycles and would be less likely to have fast-growing pioneers (Makana and Thomas 2006; Moonen et al. 2019). An increase in the number of slash-and-burn fallow cycles has also been linked to declines in fallow biomass in the Congo Basin (Moonen et al. 2019) and so we expect rural complex land to recover relatively slowly. The forest mask applied to the data does not distinguish between primary and secondary forests, so the surrounding forest may also have undergone multiple fallow cycles. Furthermore, most forest clearing in the Democratic Republic of Congo is from the secondary forest and young fallows (Molinario et al. 2015; Potapov et al. 2012), which further indicates that the deforested forested land was likely secondary forest. Analysis of the surrounding land-cover type, F(s, t) [Eq. (1)], found that the forested locations are surrounded by 97% forest while the rural complex locations are surrounded by 70% forest and 29% recovered rural complex. This may also affect the recovery to predeforestation values, as the predeforestation values may be lower for recovered rural complex and forest and could lead to slower recovery of forested land. To assess this, the EVI of the surrounding forest and rural complex land was compared, using data from the two time stamps with the largest amount of high-quality data (01/01 and 17/01) to avoid seasonality bias due to missing data. The average forest EVI was 0.67 ± 1.59 × 10−4 and the average rural complex was 0.74 ± 1.03 × 10−4. The faster recovery of the rural complex land is, therefore, not due to differences in surrounding land-cover type as the recovered rural complex had higher predeforestation values.

The plantation land undergoes consistent rapid regrowth for both VPs; the median recovers to just under the predeforestation median within 1 year. Plantations in the area are typically palm or rubber (Earthsight 2018; Harris et al. 2019); visual inspection of Google Earth imagery indicates that they are primarily palm plantations and were plantations before this study’s deforestation period. Both VPs increase for the first 3 years after deforestation and then slowly decrease for the remainder of time; the EVI stays above the predeforestation median while the LAI decreases to the predeforestation median. The EVI results for the plantations differ from those of other studies on oil palm growth. For example, a study carried out in Borneo found that the EVI takes between 3 and 5 years to reach forested values (De Petris et al. 2019), so it is highly unusual for oil palm to reach forested values in 1 year. This highlights the importance of assessing vegetation and deforestation trends across multiple regions, given how variable these results can be, as shown here for the Congo Basin.

4. Summary and discussion

The fast regrowth, coupled with the lack of high-quality remote sensing data, makes assessing deforestation’s impact on vegetation properties challenging, as determining the point in the year that deforestation occurs is essential to assess the impacts on vegetation and recovery after deforestation. This represents a major uncertainty if we are to quantify the climatic impacts of land-use change in this region. The main deforestation driver in the region is from shifting cultivation (Tyukavina et al. 2018), which requires high-resolution spatial and temporal data, highlighting the need for suitable remote sensing products. Combining the GFC’s year of deforestation data with the MODIS VPs deforestation detection method resulted in the identification of the time of deforestation at a relatively high temporal resolution, due to the 16-day temporal resolution of NDVI. The consistency of deforestation detection with the GFCs adds confidence to both the GFC and the method developed in this study. This work shows that the time of deforestation in the Congo Basin can be detected using remote sensing data, despite the large data gaps in the region. Around 88% of deforested locations underwent a further substantial drop within the 6 years after the initial deforestation event, of which the largest proportion typically occurred in the fourth year. This suggests that the fallow period in the study area is typically very short and, in some cases, does not exist. Recovery to predeforestation EVI values occurred faster than expected as 66% of locations recovered within 1 year. Separation by land-cover type also showed unexpected regrowth as the EVI median of rural complex and plantation land recovered within a year, contrasting with reports in the established literature.

The rapid recovery of VPs after small-scale deforestation highlights the inherent conflict when investigating deforestation in the region: high-resolution spatial data, such as the GFC, are only available annually, whereas high-resolution temporal data, such as the MODIS VPs, are only available at coarser resolutions. Establishing the month of deforestation and the time scale of recovery is vital to understanding other impacts of land-use change; for example, whether climate variables recover at the same rate as vegetation or are affected by the season in which deforestation occurs. These results highlight the differences in the impacts of land-use change between the Congo Basin and other rain forests and so the need for an increased focus on this region despite the data challenges.

a. Limitations

The MODIS data used in this study have a data-compositing algorithm that preferentially selects the highest values to minimize the influence of undetected clouds and aerosols (Huete et al. 2002). Thus, if deforestation occurs during the compositing period, this may lead to the selection of predeforestation values (Morton et al. 2005). Another key limitation for remote sensing in this area is the lack of high-quality data due to prevalent cloudiness, particularly in the boreal summer and the wet seasons. These factors combine to limit the accuracy of the identified time period of deforestation, particularly if it occurs during periods in a season with less data availability. Despite this uncertainty, the rapid recovery in VPs identified in this study, often occurring within a year, highlight the importance of monitoring these changes at a subyearly scale.

The average clearing size in the region is much smaller than the pixel resolution of MODIS VPs, and it is, therefore, possible that the results are not representative of deforestation in other areas of the basin as they are from unusually large deforestation patches. As Congo Basin deforestation expands, the impact on VPs may be different as deforestation of primary forests may differ from the secondary forests. Furthermore, the regrowth rate in the basin is greatly affected by the species that grow after deforestation; deforested land located nearby primary forest benefits from regular seed sources and is exposed to more diverse species (Mukul and Herbohn 2016), which affects the regrowth rate. Lack of knowledge surrounding land-cover types is a substantial limiting factor in this study. the ability to distinguish between primary and secondary forests, the number of fallow cycles land has undergone, and the ability to distinguish between vegetation types after deforestation would be highly beneficial to this research. Land-cover types in the Lao tropical forest have been accurately distinguished using analysis of the intra-annual variations of EVI and land surface temperature (Phompila et al. 2015). This work could be applied to the Congo Basin to aid regrowth analysis; however, the severe data challenges and lack of consistent monthly data may mean this is not currently possible for this region.

b. Implications

Remote sensing data are vital for deforestation studies of the Congo Basin, due to the remoteness of the region. The rapid evolution of land-cover type after deforestation observed in this study highlights the need for high-resolution temporal data of the region. The pervasive cloudiness throughout most of the year prevents traditional remote sensing products, such as the MODIS products, from retrieving high-quality data for large parts of the year. This problem needs to be addressed to improve upon the accuracy of the identified deforestation time of these results. This would also lead to improved analysis of regrowth as we could look at regrowth straight after deforestation instead of comparing the years after, and it would allow for seasonality analysis. These results show the need for more ground observations in the region to understand what the full recovery of MODIS VPs means in terms of land-cover type and regrowth. Ground observations would also aid in understanding the subsequent drops detected after deforestation.

MODIS VPs have not been used before to detect deforestation and assess subsequent regrowth in the Congo Basin, and these results highlight the value of these products. This is particularly impressive when considering the coarseness of the MODIS spatial resolution compared to the average deforestation clearing size. This study has also highlighted how different Congo Basin deforestation is from that of other regions; this also applies to the climatic impacts of deforestation, as they must be considered separately from other regions. Our results show that small-scale agricultural deforestation can be detected on scales relevant for atmospheric impacts and, therefore, may already be affecting the climate.

Acknowledgments.

Adams is supported by the Manchester Environment Research Institute, and Garcia-Carreras is supported by the Natural Environment Research Council (Grant NE/V012681/1). We thank David Schultz for his review, which greatly enhanced the coherence and accessibility of the paper.

Data availability statement.

All MODIS data used during this study are openly available and were obtained using the Application for Extracting and Exploring Analysis Ready Samples (AppEEARS) (https://appeears.earthdatacloud.nasa.gov/). The Global Forest Change dataset was obtained from Hansen et al. (2013) (https://glad.earthengine.app/view/global-forest-change), and the Land Cover Types map used was obtained from Verhegghen et al. (2012) (http://www.uclouvain.be/eli-maps).

REFERENCES

  • Abera, T. A., J. Heiskanen., P. Pellikka, and E. E. Maeda, 2018: Rainfall–vegetation interaction regulates temperature anomalies during extreme dry events in the Horn of Africa. Global Planet. Change, 167, 3545, https://doi.org/10.1016/j.gloplacha.2018.05.002.

    • Search Google Scholar
    • Export Citation
  • Akkermans, T., A. Van Rompaey, N. Van Lipzig, P. Moonen, and B. Verbist, 2013: Quantifying successional land cover after clearing of tropical rainforest along forest frontiers in the Congo Basin. Phys. Geogr., 34, 417440, https://doi.org/10.1080/02723646.2013.855698.

    • Search Google Scholar
    • Export Citation
  • Appiah Mensah, A., D. Akoto Sarfo, and S. T. Partey, 2019: Assessment of vegetation dynamics using remote sensing and GIS: A case of Bosomtwe Range Forest Reserve, Ghana. Egypt. J. Remote Sens. Space Sci., 22, 145154, https://doi.org/10.1016/j.ejrs.2018.04.004.

    • Search Google Scholar
    • Export Citation
  • Arai, E., Y. E. Shimabukuro, G. Pereira, and N. L. Vijaykumar, 2011: A multi-resolution multi-temporal technique for detecting and mapping deforestation in the Brazilian Amazon rainforest. Remote Sens., 3, 19431956, https://doi.org/10.3390/rs3091943.

    • Search Google Scholar
    • Export Citation
  • Baidya Roy, S., and R. Avissar, 2002: Impact of land use/land cover change on regional hydrometeorology in Amazonia. J. Geophys. Res. Atmos., 107, 8037, https://doi.org/10.1029/2000JD000266.

    • Search Google Scholar
    • Export Citation
  • Bell, J. P., A. M. Tompkins., C. Bouka-Biona, and I. S. Sanda, 2015: A process-based investigation into the impact of the Congo basin deforestation on surface climate. J. Geophys. Res., 120, 57215739, https://doi.org/10.1002/2014JD022586.

    • Search Google Scholar
    • Export Citation
  • Brown, J. C., J. H. Kastens, A. C. Coutinho, D. de Castro Victoria, C. R. Bishop, 2013: Classifying multiyear agricultural land use data from Mato Grosso using time-series MODIS vegetation index data. Remote Sens. Environ., 130, 3950, https://doi.org/10.1016/j.rse.2012.11.009.

    • Search Google Scholar
    • Export Citation
  • Depecker, J., and Coauthors, 2021: Effects of forest disturbance and regeneration on tree species composition and traits in DR Congo. bioRxiv, 2021.10.11.463162, https://doi.org/10.1101/2021.10.11.463162.

  • De Petris, S., P. Boccardo, and E. Borgogno-Mondino, 2019: Detection and characterization of oil palm plantations through MODIS EVI time series. Int. J. Remote Sens., 40, 72977311, https://doi.org/10.1080/01431161.2019.1584689.

    • Search Google Scholar
    • Export Citation
  • Didan, K., A. B. Munoz, R. Solano, and A. Huete, 2015: MODIS Vegetation Index User’s Guide (MOD13 series), version 3. The University of Arizona, 35 pp., https://vip.arizona.edu/documents/MODIS/MODIS_VI_UsersGuide_June_2015_C6.pdf.

  • Earthsight, 2018: The Coming Storm. How Secrecy and Collusion in Industrial Agriculture Spell Disaster for the Congo Basin’s Forests. Earthsight, 20 pp.

  • Feintrenie, L., 2014: Agro-industrial plantations in Central Africa, risks and opportunities. Biodivers. Conserv., 23, 15771589, https://doi.org/10.1007/s10531-014-0687-5.

    • Search Google Scholar
    • Export Citation
  • Google Earth, 2020: Mongala region, Democratic Republic of the Congo, 2.783503°, 21.863694°. Accessed 9 May 2023, https://earth.google.com/web/.

    • Search Google Scholar
    • Export Citation
  • Grogan, K., D. Pflugmacher, P. Hostert, J. Verbesselt, and R. Fensholt, 2016: Mapping clearances in tropical dry forests using breakpoints, trend, and seasonal components from MODIS time series: Does forest type matter? Remote Sens., 8, 657, https://doi.org/10.3390/rs8080657.

    • Search Google Scholar
    • Export Citation
  • Hammer, D., R. Kraft, and D. Wheeler, 2014: Alerts of forest disturbance from MODIS imagery. Int. J. Appl. Earth Obs. Geoinf., 33 (1), 19, https://doi.org/10.1016/j.jag.2014.04.011.

    • Search Google Scholar
    • Export Citation
  • Hansen, M. C., D. P. Roy, E. Lindquist, B. Adusei, C. O. Justice, and A. Altstatt, 2008: A method for integrating MODIS and Landsat data for systematic monitoring of forest cover and change in the Congo Basin. Remote Sens. Environ., 112, 24952513, https://doi.org/10.1016/j.rse.2007.11.012.

    • Search Google Scholar
    • Export Citation
  • Hansen, M. C., and Coauthors, 2013: High-resolution global maps of 21st-century forest cover change. Science, 342, 850854, https://doi.org/10.1126/science.1244693.

    • Search Google Scholar
    • Export Citation
  • Harris, N., E. Goldman, and S. Gibbes, 2019: Spatial Database of Planted Trees (SDPT) version 1.0. Global Forest Watch, accessed 8 December 2021, https://www.wri.org/research/spatial-database-planted-trees-sdpt-version-10.

  • Hayes, D. J., and W. B. Cohen, 2007: Spatial, spectral and temporal patterns of tropical forest cover change as observed with multiple scales of optical satellite data. Remote Sens. Environ., 106 (1), 116, https://doi.org/10.1016/j.rse.2006.07.002.

    • Search Google Scholar
    • Export Citation
  • Hirschmugl, M., M. Steinegger, H. Gallaun, and M. Schardt, 2013: Mapping forest degradation due to selective logging by means of time series analysis: Case studies in Central Africa. Remote Sens., 6, 756775, https://doi.org/10.3390/rs6010756.

    • Search Google Scholar
    • Export Citation
  • Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira, 2002: Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ., 83, 195213, https://doi.org/10.1016/S0034-4257(02)00096-2.

    • Search Google Scholar
    • Export Citation
  • Irons, J. R., J. L. Dwyer, and J. A. Barsi, 2012: The next Landsat satellite: The Landsat Data Continuity Mission. Remote Sens. Environ., 122, 1121, https://doi.org/10.1016/j.rse.2011.08.026.

    • Search Google Scholar
    • Export Citation
  • Khanna, J., D. Medvigy., S. Fueglistaler, and R. Walko, 2017: Regional dry-season climate changes due to three decades of Amazonian deforestation. Nat. Climate Change, 7, 200204, https://doi.org/10.1038/nclimate3226.

    • Search Google Scholar
    • Export Citation
  • Koltunov, A., S. L. Ustin, G. P. Asner, and I. Fung, 2009: Selective logging changes forest phenology in the Brazilian Amazon: Evidence from MODIS image time series analysis. Remote Sens. Environ., 113, 24312440, https://doi.org/10.1016/j.rse.2009.07.005.

    • Search Google Scholar
    • Export Citation
  • Makana, J.-R., and S. C. Thomas, 2006: Impacts of selective logging and agricultural clearing on forest structure, floristic composition and diversity, and timber tree regeneration in the Ituri Forest, Democratic Republic of Congo. Forest Diversity and Management, D. L. Hawksworth and A. T. Bull, Eds., Springer, 315–337.

  • Makelele, I. A., K. Verheyen, P. Boeckx, L. C. Ntaboba, B. M. Bazirake, C. Ewango, and M. Bauters, 2021: Afrotropical secondary forests exhibit fast diversity and functional recovery, but slow compositional and carbon recovery after shifting cultivation. J. Veg. Sci., 32, e13071, https://doi.org/10.1111/jvs.13071.

    • Search Google Scholar
    • Export Citation
  • Molinario, G., M. C. Hansen, and P. V. Potapov, 2015: Forest cover dynamics of shifting cultivation in the Democratic Republic of the Congo 2000–2010. Environ. Res. Lett., 10, 094009, https://doi.org/10.1088/1748-9326/10/9/094009.

    • Search Google Scholar
    • Export Citation
  • Moonen, P. C. J., and Coauthors, 2019: Disentangling how management affects biomass stock and productivity of tropical secondary forests fallows. Sci. Total Environ., 659, 101114, https://doi.org/10.1016/j.scitotenv.2018.12.138.

    • Search Google Scholar
    • Export Citation
  • Morton, D. C., R. S. DeFries, Y. E. Shimabukuro, L. O. Anderson, F. Del Bon Espírito-Santo, M. Hansen, and M. Carroll, 2005: Rapid assessment of annual deforestation in the Brazilian Amazon using MODIS data. Earth Interact., 9, https://doi.org/10.1175/EI139.1.

    • Search Google Scholar
    • Export Citation
  • Mukul, S. A., and J. Herbohn, 2016: The impacts of shifting cultivation on secondary forests dynamics in tropics: A synthesis of the key findings and spatio temporal distribution of research. Environ. Sci. Policy, 55, 167177, https://doi.org/10.1016/j.envsci.2015.10.005.

    • Search Google Scholar
    • Export Citation
  • Myneni, R., 2020: MODIS Collection 6 (C6) LAI/FPAR product user’s guide. USGS, 13 pp., https://lpdaac.usgs.gov/documents/2/mod15_user_guide.pdf.

  • Nogherotto, R., E. Coppola., F. Giorgi, and L. Mariotti, 2013: Impact of Congo Basin deforestation on the African monsoon. Atmos. Sci. Lett., 14, 4551, https://doi.org/10.1002/asl2.416.

    • Search Google Scholar
    • Export Citation
  • Phompila, C., M. Lewis, B. Ostendorf, and K. Clarke, 2015: MODIS EVI and LST temporal response for discrimination of tropical land covers. Remote Sens., 7, 60266040, https://doi.org/10.3390/rs70506026.

    • Search Google Scholar
    • Export Citation
  • Potapov, P. V., S. A. Turubanova, M. C. Hansen, B. Adusei, M. Broich, A. Altstatt, L. Mane, and C. O. Justice, 2012: Quantifying forest cover loss in Democratic Republic of the Congo, 2000–2010, with Landsat ETM+ data. Remote Sens. Environ., 122, 106116, https://doi.org/10.1016/j.rse.2011.08.027.

    • Search Google Scholar
    • Export Citation
  • Rahm, M., J. Van Wolvelaer, A. Vrieling, and B. Mertens, 2013: Detecting forest degradation in the Congo Basin by optical remote sensing. Proc. ESA Living Planet Symp. 2013 (ESA SP-722), Edinburgh, United Kingdom, European Space Agency, https://ftp.space.dtu.dk/pub/Ioana/papers/s144_3rahm.pdf.

  • Shem, W. O., 2006: Biosphere-atmosphere interaction over the Congo basin and its influence on the regional hydrological cycle. Ph.D. thesis, Georgia Institute of Technology, 141 pp., http://hdl.handle.net/1853/11558.

    • Search Google Scholar
    • Export Citation
  • Souza, C. M., D. A. Roberts, and A. L. Monteiro, 2005: Multitemporal analysis of degraded forests in the southern Brazilian Amazon. Earth Interact., 9, https://doi.org/10.1175/EI132.1.

    • Search Google Scholar
    • Export Citation
  • Spracklen, D. V., S. R. Arnold, and C. M. Taylor, 2012: Observations of increased tropical rainfall preceded by air passage over forests. Nature, 489, 282285, https://doi.org/10.1038/nature11390.

    • Search Google Scholar
    • Export Citation
  • Tyukavina, A., M. C. Hansen, P. Potapov, D. Parker, C. Okpa, S. V. Stehman, I. Kommareddy, and S. Turubanova, 2018: Congo Basin forest loss dominated by increasing smallholder clearing. Sci. Adv., 4, eaat2993, https://doi.org/10.1126/sciadv.aat2993.

    • Search Google Scholar
    • Export Citation
  • van Vliet, N., and Coauthors, 2012: Trends, drivers and impacts of changes in swidden cultivation in tropical forest-agriculture frontiers: A global assessment. Global Environ. Change, 22, 418429, https://doi.org/10.1016/j.gloenvcha.2011.10.009.

    • Search Google Scholar
    • Export Citation
  • Verhegghen, A., P. Mayaux, and P. Defourny, 2012: Mapping Congo Basin vegetation types from 300 m and 1 km multi-sensor time series for carbon stocks and forest areas estimation. Biogeosciences, 9, 50615079, https://doi.org/10.5194/bg-9-5061-2012.

    • Search Google Scholar
    • Export Citation
  • Vijith, H., and D. Dodge-Wan, 2020: Applicability of MODIS land cover and enhanced vegetation index (EVI) for the assessment of spatial and temporal changes in strength of vegetation in tropical rainforest region of Borneo. Remote Sens. Appl., 18, 100311, https://doi.org/10.1016/j.rsase.2020.100311.

    • Search Google Scholar
    • Export Citation
  • Wilson, A. M., and W. Jetz, 2016: Remotely sensed high-resolution global cloud dynamics for predicting ecosystem and biodiversity distributions. PLOS Biol., 14, e1002415, https://doi.org/10.1371/journal.pbio.1002415.

    • Search Google Scholar
    • Export Citation
  • Zhang, Q., C. O. Justice, M. Jiang, J. Brunner, and D. S. Wilkie, 2006: A GIS-based assessment on the vulnerability and future extent of the tropical forests of the Congo Basin. Environ. Monit. Assess., 114, 107121, https://doi.org/10.1007/s10661-006-2015-3.

    • Search Google Scholar
    • Export Citation
Save
  • Abera, T. A., J. Heiskanen., P. Pellikka, and E. E. Maeda, 2018: Rainfall–vegetation interaction regulates temperature anomalies during extreme dry events in the Horn of Africa. Global Planet. Change, 167, 3545, https://doi.org/10.1016/j.gloplacha.2018.05.002.

    • Search Google Scholar
    • Export Citation
  • Akkermans, T., A. Van Rompaey, N. Van Lipzig, P. Moonen, and B. Verbist, 2013: Quantifying successional land cover after clearing of tropical rainforest along forest frontiers in the Congo Basin. Phys. Geogr., 34, 417440, https://doi.org/10.1080/02723646.2013.855698.

    • Search Google Scholar
    • Export Citation
  • Appiah Mensah, A., D. Akoto Sarfo, and S. T. Partey, 2019: Assessment of vegetation dynamics using remote sensing and GIS: A case of Bosomtwe Range Forest Reserve, Ghana. Egypt. J. Remote Sens. Space Sci., 22, 145154, https://doi.org/10.1016/j.ejrs.2018.04.004.

    • Search Google Scholar
    • Export Citation
  • Arai, E., Y. E. Shimabukuro, G. Pereira, and N. L. Vijaykumar, 2011: A multi-resolution multi-temporal technique for detecting and mapping deforestation in the Brazilian Amazon rainforest. Remote Sens., 3, 19431956, https://doi.org/10.3390/rs3091943.

    • Search Google Scholar
    • Export Citation
  • Baidya Roy, S., and R. Avissar, 2002: Impact of land use/land cover change on regional hydrometeorology in Amazonia. J. Geophys. Res. Atmos., 107, 8037, https://doi.org/10.1029/2000JD000266.

    • Search Google Scholar
    • Export Citation
  • Bell, J. P., A. M. Tompkins., C. Bouka-Biona, and I. S. Sanda, 2015: A process-based investigation into the impact of the Congo basin deforestation on surface climate. J. Geophys. Res., 120, 57215739, https://doi.org/10.1002/2014JD022586.

    • Search Google Scholar
    • Export Citation
  • Brown, J. C., J. H. Kastens, A. C. Coutinho, D. de Castro Victoria, C. R. Bishop, 2013: Classifying multiyear agricultural land use data from Mato Grosso using time-series MODIS vegetation index data. Remote Sens. Environ., 130, 3950, https://doi.org/10.1016/j.rse.2012.11.009.

    • Search Google Scholar
    • Export Citation
  • Depecker, J., and Coauthors, 2021: Effects of forest disturbance and regeneration on tree species composition and traits in DR Congo. bioRxiv, 2021.10.11.463162, https://doi.org/10.1101/2021.10.11.463162.

  • De Petris, S., P. Boccardo, and E. Borgogno-Mondino, 2019: Detection and characterization of oil palm plantations through MODIS EVI time series. Int. J. Remote Sens., 40, 72977311, https://doi.org/10.1080/01431161.2019.1584689.

    • Search Google Scholar
    • Export Citation
  • Didan, K., A. B. Munoz, R. Solano, and A. Huete, 2015: MODIS Vegetation Index User’s Guide (MOD13 series), version 3. The University of Arizona, 35 pp., https://vip.arizona.edu/documents/MODIS/MODIS_VI_UsersGuide_June_2015_C6.pdf.

  • Earthsight, 2018: The Coming Storm. How Secrecy and Collusion in Industrial Agriculture Spell Disaster for the Congo Basin’s Forests. Earthsight, 20 pp.

  • Feintrenie, L., 2014: Agro-industrial plantations in Central Africa, risks and opportunities. Biodivers. Conserv., 23, 15771589, https://doi.org/10.1007/s10531-014-0687-5.

    • Search Google Scholar
    • Export Citation
  • Google Earth, 2020: Mongala region, Democratic Republic of the Congo, 2.783503°, 21.863694°. Accessed 9 May 2023, https://earth.google.com/web/.

    • Search Google Scholar
    • Export Citation
  • Grogan, K., D. Pflugmacher, P. Hostert, J. Verbesselt, and R. Fensholt, 2016: Mapping clearances in tropical dry forests using breakpoints, trend, and seasonal components from MODIS time series: Does forest type matter? Remote Sens., 8, 657, https://doi.org/10.3390/rs8080657.

    • Search Google Scholar
    • Export Citation
  • Hammer, D., R. Kraft, and D. Wheeler, 2014: Alerts of forest disturbance from MODIS imagery. Int. J. Appl. Earth Obs. Geoinf., 33 (1), 19, https://doi.org/10.1016/j.jag.2014.04.011.

    • Search Google Scholar
    • Export Citation
  • Hansen, M. C., D. P. Roy, E. Lindquist, B. Adusei, C. O. Justice, and A. Altstatt, 2008: A method for integrating MODIS and Landsat data for systematic monitoring of forest cover and change in the Congo Basin. Remote Sens. Environ., 112, 24952513, https://doi.org/10.1016/j.rse.2007.11.012.

    • Search Google Scholar
    • Export Citation
  • Hansen, M. C., and Coauthors, 2013: High-resolution global maps of 21st-century forest cover change. Science, 342, 850854, https://doi.org/10.1126/science.1244693.

    • Search Google Scholar
    • Export Citation
  • Harris, N., E. Goldman, and S. Gibbes, 2019: Spatial Database of Planted Trees (SDPT) version 1.0. Global Forest Watch, accessed 8 December 2021, https://www.wri.org/research/spatial-database-planted-trees-sdpt-version-10.

  • Hayes, D. J., and W. B. Cohen, 2007: Spatial, spectral and temporal patterns of tropical forest cover change as observed with multiple scales of optical satellite data. Remote Sens. Environ., 106 (1), 116, https://doi.org/10.1016/j.rse.2006.07.002.

    • Search Google Scholar
    • Export Citation
  • Hirschmugl, M., M. Steinegger, H. Gallaun, and M. Schardt, 2013: Mapping forest degradation due to selective logging by means of time series analysis: Case studies in Central Africa. Remote Sens., 6, 756775, https://doi.org/10.3390/rs6010756.

    • Search Google Scholar
    • Export Citation
  • Huete, A., K. Didan, T. Miura, E. P. Rodriguez, X. Gao, and L. G. Ferreira, 2002: Overview of the radiometric and biophysical performance of the MODIS vegetation indices. Remote Sens. Environ., 83, 195213, https://doi.org/10.1016/S0034-4257(02)00096-2.

    • Search Google Scholar
    • Export Citation
  • Irons, J. R., J. L. Dwyer, and J. A. Barsi, 2012: The next Landsat satellite: The Landsat Data Continuity Mission. Remote Sens. Environ., 122, 1121, https://doi.org/10.1016/j.rse.2011.08.026.

    • Search Google Scholar
    • Export Citation
  • Khanna, J., D. Medvigy., S. Fueglistaler, and R. Walko, 2017: Regional dry-season climate changes due to three decades of Amazonian deforestation. Nat. Climate Change, 7, 200204, https://doi.org/10.1038/nclimate3226.

    • Search Google Scholar
    • Export Citation
  • Koltunov, A., S. L. Ustin, G. P. Asner, and I. Fung, 2009: Selective logging changes forest phenology in the Brazilian Amazon: Evidence from MODIS image time series analysis. Remote Sens. Environ., 113, 24312440, https://doi.org/10.1016/j.rse.2009.07.005.

    • Search Google Scholar
    • Export Citation
  • Makana, J.-R., and S. C. Thomas, 2006: Impacts of selective logging and agricultural clearing on forest structure, floristic composition and diversity, and timber tree regeneration in the Ituri Forest, Democratic Republic of Congo. Forest Diversity and Management, D. L. Hawksworth and A. T. Bull, Eds., Springer, 315–337.

  • Makelele, I. A., K. Verheyen, P. Boeckx, L. C. Ntaboba, B. M. Bazirake, C. Ewango, and M. Bauters, 2021: Afrotropical secondary forests exhibit fast diversity and functional recovery, but slow compositional and carbon recovery after shifting cultivation. J. Veg. Sci., 32, e13071, https://doi.org/10.1111/jvs.13071.

    • Search Google Scholar
    • Export Citation
  • Molinario, G., M. C. Hansen, and P. V. Potapov, 2015: Forest cover dynamics of shifting cultivation in the Democratic Republic of the Congo 2000–2010. Environ. Res. Lett., 10, 094009, https://doi.org/10.1088/1748-9326/10/9/094009.

    • Search Google Scholar
    • Export Citation
  • Moonen, P. C. J., and Coauthors, 2019: Disentangling how management affects biomass stock and productivity of tropical secondary forests fallows. Sci. Total Environ., 659, 101114, https://doi.org/10.1016/j.scitotenv.2018.12.138.

    • Search Google Scholar
    • Export Citation
  • Morton, D. C., R. S. DeFries, Y. E. Shimabukuro, L. O. Anderson, F. Del Bon Espírito-Santo, M. Hansen, and M. Carroll, 2005: Rapid assessment of annual deforestation in the Brazilian Amazon using MODIS data. Earth Interact., 9, https://doi.org/10.1175/EI139.1.

    • Search Google Scholar
    • Export Citation
  • Mukul, S. A., and J. Herbohn, 2016: The impacts of shifting cultivation on secondary forests dynamics in tropics: A synthesis of the key findings and spatio temporal distribution of research. Environ. Sci. Policy, 55, 167177, https://doi.org/10.1016/j.envsci.2015.10.005.

    • Search Google Scholar
    • Export Citation
  • Myneni, R., 2020: MODIS Collection 6 (C6) LAI/FPAR product user’s guide. USGS, 13 pp., https://lpdaac.usgs.gov/documents/2/mod15_user_guide.pdf.

  • Nogherotto, R., E. Coppola., F. Giorgi, and L. Mariotti, 2013: Impact of Congo Basin deforestation on the African monsoon. Atmos. Sci. Lett., 14, 4551, https://doi.org/10.1002/asl2.416.

    • Search Google Scholar
    • Export Citation
  • Phompila, C., M. Lewis, B. Ostendorf, and K. Clarke, 2015: MODIS EVI and LST temporal response for discrimination of tropical land covers. Remote Sens., 7, 60266040, https://doi.org/10.3390/rs70506026.

    • Search Google Scholar
    • Export Citation
  • Potapov, P. V., S. A. Turubanova, M. C. Hansen, B. Adusei, M. Broich, A. Altstatt, L. Mane, and C. O. Justice, 2012: Quantifying forest cover loss in Democratic Republic of the Congo, 2000–2010, with Landsat ETM+ data. Remote Sens. Environ., 122, 106116, https://doi.org/10.1016/j.rse.2011.08.027.

    • Search Google Scholar
    • Export Citation
  • Rahm, M., J. Van Wolvelaer, A. Vrieling, and B. Mertens, 2013: Detecting forest degradation in the Congo Basin by optical remote sensing. Proc. ESA Living Planet Symp. 2013 (ESA SP-722), Edinburgh, United Kingdom, European Space Agency, https://ftp.space.dtu.dk/pub/Ioana/papers/s144_3rahm.pdf.

  • Shem, W. O., 2006: Biosphere-atmosphere interaction over the Congo basin and its influence on the regional hydrological cycle. Ph.D. thesis, Georgia Institute of Technology, 141 pp., http://hdl.handle.net/1853/11558.

    • Search Google Scholar
    • Export Citation
  • Souza, C. M., D. A. Roberts, and A. L. Monteiro, 2005: Multitemporal analysis of degraded forests in the southern Brazilian Amazon. Earth Interact., 9, https://doi.org/10.1175/EI132.1.

    • Search Google Scholar
    • Export Citation
  • Spracklen, D. V., S. R. Arnold, and C. M. Taylor, 2012: Observations of increased tropical rainfall preceded by air passage over forests. Nature, 489, 282285, https://doi.org/10.1038/nature11390.

    • Search Google Scholar
    • Export Citation
  • Tyukavina, A., M. C. Hansen, P. Potapov, D. Parker, C. Okpa, S. V. Stehman, I. Kommareddy, and S. Turubanova, 2018: Congo Basin forest loss dominated by increasing smallholder clearing. Sci. Adv., 4, eaat2993, https://doi.org/10.1126/sciadv.aat2993.

    • Search Google Scholar
    • Export Citation
  • van Vliet, N., and Coauthors, 2012: Trends, drivers and impacts of changes in swidden cultivation in tropical forest-agriculture frontiers: A global assessment. Global Environ. Change, 22, 418429, https://doi.org/10.1016/j.gloenvcha.2011.10.009.

    • Search Google Scholar
    • Export Citation
  • Verhegghen, A., P. Mayaux, and P. Defourny, 2012: Mapping Congo Basin vegetation types from 300 m and 1 km multi-sensor time series for carbon stocks and forest areas estimation. Biogeosciences, 9, 50615079, https://doi.org/10.5194/bg-9-5061-2012.

    • Search Google Scholar
    • Export Citation
  • Vijith, H., and D. Dodge-Wan, 2020: Applicability of MODIS land cover and enhanced vegetation index (EVI) for the assessment of spatial and temporal changes in strength of vegetation in tropical rainforest region of Borneo. Remote Sens. Appl., 18, 100311, https://doi.org/10.1016/j.rsase.2020.100311.

    • Search Google Scholar
    • Export Citation
  • Wilson, A. M., and W. Jetz, 2016: Remotely sensed high-resolution global cloud dynamics for predicting ecosystem and biodiversity distributions. PLOS Biol., 14, e1002415, https://doi.org/10.1371/journal.pbio.1002415.

    • Search Google Scholar
    • Export Citation
  • Zhang, Q., C. O. Justice, M. Jiang, J. Brunner, and D. S. Wilkie, 2006: A GIS-based assessment on the vulnerability and future extent of the tropical forests of the Congo Basin. Environ. Monit. Assess., 114, 107121, https://doi.org/10.1007/s10661-006-2015-3.

    • Search Google Scholar
    • Export Citation
  • Fig. 1.

    Successional fallow cycle, adapted from Akkermans et al. (2013).

  • Fig. 2.

    (a) Land-cover type map of the Congo Basin from the Congo Basin vegetation-types map (Verhegghen et al. 2012). The study region is denoted by the red rectangle; white patches are areas outside of the map’s domain. (b) Cumulative forest loss during 2000–19, calculated as the percentage of the 500-m MODIS pixel that has been deforested according to the Global Forest Change 2000–19 (Hansen et al. 2013). White patches are water bodies; magenta patches are commercial plantation land.

  • Fig. 3.

    MODIS 500-m VI data quality aggregated for each time stamp from 2000 to 2021 (Terra) and from 2002 to 2021 (Aqua).

  • Fig. 4.

    Diagram of the application of two-date image differencing for deforestation detection.

  • Fig. 5.

    Normalized histogram of all EVI and NDVI results for forested locations from 2000 to 2019, both Aqua and Terra satellite data.

  • Fig. 6.

    Boxplots of (a) EVI and (b) LAI from 3 years before deforestation to 8 years after. The horizontal dashed line indicates the median value of the 3-yr predeforestation, and notches on the boxplot indicate the 95% CI of the median. Outliers have a 0.5 opacity and so appear darker where there is a high density of outliers. Colored circles denote the statistical significance by which the forested and deforested distributions are different.

  • Fig. 7.

    Cumulative bar plot of NDVI detected drops after the initial deforestation event, grouped by the minimum number of available years after the initial deforestation event. Bars are split by the year of deforestation; the small gray numbers on top represent the total number of drops/the total sample size.

  • Fig. 8.

    Boxplots of all vegetation parameters as a time series of 3 years before deforestation and 8 years after. The horizontal dashed line indicates the median value of the 3-yr predeforestation; notches on the boxplots indicate the 95% CI of the median. Outliers have a 0.5 opacity and so appear darker where there is a high density of outliers. (a) EVI. (b) LAI. Land-cover types: (i) all data, (ii) plantations, (iii) rural complex, (iv) forest.

All Time Past Year Past 30 Days
Abstract Views 0 0 0
Full Text Views 1772 1036 25
PDF Downloads 1198 576 28