Recent changes in cropland area and productivity indicate unsustainable cropland expansion in Malawi

2.1.1. Estimating changes in cropland area from 2010 to 2019

2.1.2. Characterizing changes in cropland area

We used the commonly used satellite-derived measure of vegetation greenness, the NDVI, as a proxy for cropland productivity. NDVI is proven to be a good indicator for crop yield estimation in smallholding crop fields in SSA (Burke and Lobell 2017), including for maize which is the dominant crop type cultivated by over 97% of households in Malawi (Denning et al 2009) as well as other crop types cultivated in Malawi such as millet and sorghum (Lambert et al 2018). In addition, NDVI is a widely used proxy for monitoring cropland productivity and degradation (Wessels et al 2004, Barbier and Hochard 2018, Easdale et al 2018, Gichenje and Godinho 2018). In this study, we calculated changes in yearly mean NDVI value (MOD13Q1) (Didan 2015) during crop growing season (November – April) as a proxy of cropland productivity. To avoid any confusion of changes in NDVI caused by land cover conversion, we calculated changes in NDVI in permanent croplands that are classified as cropland in both 2010 and 2019. We conducted the linear trend analysis by applying a linear regression model of annual mean NDVI at the pixel level (Higginbottom and Symeonakis 2014, Gichenje and Godinho 2018) and then averaged the slopes of the linear regression at the district level. In this case, positive values indicate increases in cropland productivity while negative values indicate reductions in cropland productivity. Detailed analysis and spatial distribution of changes in cropland productivity can be found in the supplementary material (section S3). In addition, to further test the hypothesis that cropland productivity tends to decline in croplands with longer cultivation age, we analyzed changes in cropland productivity for croplands with different cultivation ages (as described in section 2.1.2). The time period for analyzing the changes in productivity is in line with cultivation age. For example, croplands with the cultivation age of eight years, linear regression of yearly mean NDVI value from 2010 to 2018 was analyzed. To capture the meaningful time series analysis, changes in yearly NDVI were analyzed only for croplands older than four years. Finally, we further used the household survey data on soil erosion level. The soil erosion level was estimated by smallholder farmers from the Malawi household survey and is based on four categories: (a) No Erosion, (b) Low, (c) Moderate, and (d) High. We computed changes in soil erosion level for the year 2010 and 2016 at the district level by calculating the mean value for each district for both years, based on a total of 37 000 surveys. These sample surveys were collected using a stratified sampling scheme that is representative of each district (supplementary material section S3.2). In this case, positive values indicated more severe soil erosion while negative values indicated lessened soil erosion. Changes in soil erosion and NDVI were further correlated to the ratio of cropland area change to understand their association with cropland expansion.

To identify and estimate the potentially available cropland for future expansion, we used the World Bank definition of potentially available cropland (Deininger and Byerlee 2011), defined as areas that are non-forested, non-protected, populated with less than 25 persons km⁻², and are currently not cultivated. These potentially available croplands normally correspond to grasslands, long-term fallows, and shrublands suitable for rainfed agriculture (Lambin et al 2013). This definition tends to emphasize the biophysical production potential that is based on land and climate characteristics (Chamberlin et al 2014) rather than the production potential based on economic profitability, cost of land conversion, and accessibility to roads and markets (Lambin et al 2013). Based on land cover types classified in this study for 2019 and the above definition, we estimated potentially available cropland for future expansion across Malawi. More details on the justification and results of potentially available cropland can be found in supplementary material section 4. We summarized all datasets used in this study and illustrate the framework of the analysis, listing the main outputs in figure 1. All satellite data processing and land cover classification were conducted in the Google Earth Engine cloud-based platform (Gorelick et al 2017).

Zoom In Zoom Out Reset image size

Cropland area in Malawi increased by 8.15% of total land area from 39.6% (with an accuracy of 80.33% (Chen et al 2017)) in 2010 to 47.75% of total land area in 2019 (with a cropland classification accuracy of 83.5% from this study). Figure 2 shows the strong spatial variability of cropland expansion in Malawi: the Central Region experienced the largest expansion area (379 880 ha), followed by the Northern Region (198 880 ha), and lastly by the Southern Region (137 000 ha). Most conversions were from grasslands (7.7% of the total land), and deforestation accounted for a smaller portion (0.2% of the total land). However, cropland expansion accounted for 31% of forest loss on average from 2001 to 2018, which showing a decline trend from 41% in 2001 to 25% in 2018 even though the total forest loss area increased steadily (figure 3(a)). Croplands converted from deforestation were located increasingly towards higher elevations and further away from rivers, roads (figure 3(b)), with trend significance of p < 0.05 (elevation), p < 0.01 (roads and river), and p = 0.17 (settlements) according to the Mann-Kendall test.

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We also estimated that cropland loss accounted for 0.2% of total land (approximately 25 285 ha). Most of these croplands were converted to either settlements or bare land (87.6% of whole cropland loss area), with the remainder to forest (7.3%) and water bodies (5.1%). We found that only 5% of the total land in Malawi (approximately 4671 000 ha) remained as potentially available cropland for future expansion (according to the definition reported in section 2.3), with higher potential in the Northern region (figure 2 and supplementary material section S4). We estimated that these potentially available croplands would only support expansion for the next five years, by adopting a simple business-as-usual scenario, i.e. assuming that the national average expansion rate observed in the past will not be altered by major changes in policies or conservation practices.

We correlated changes in cropland area to rural population growth and field size to understand the roles of smallholder farming in driving cropland expansion. We found that a higher population growth rate in rural areas was associated with a higher cropland expansion ratio (figure 4(a)) (p < 0.001). Given that the rural areas consist almost entirely of smallholder farmers (Tchale 2009), this result might indicate that smallholder farmers have contributed to cropland expansion. However, this association only held when sufficient land was available for expansion. In fact, we found that a lower expansion ratio was observed in urban areas despite the higher population growth rate by correlating cropland expansion and population growth (figure 4(a)). While a weak positive correlation between cropland loss ratio and population growth (R² = 0.38) (figure 4(b)) indicate that higher cropland loss ratio generally comes with a higher population growth rate generally, particularly in the urban areas as well.

Zoom In Zoom Out Reset image size

While our results show an overall expansion in cropland area, we found that the average field size under cultivation by smallholder farmers reduced from 0.375 ± 0.0047 ha in 2010 to 0.365 ± 0.052 ha in 2016 (figure S5(a)) based on our analysis of the Malawi household survey data (supplementary material S2.1). However, looking into the spatial variation of changes in field size at the sub-national level, we found that field size tended to increase in districts with a higher expansion ratio (i.e. Central Region) and to decrease in districts with a lower cropland expansion ratio (i.e. Northern and Southern Region) (figure 5).

Zoom In Zoom Out Reset image size

We found that most districts (19 among 31) showed reduced cropland productivity between 2010 and 2019, with a concentration in the Southern Region (figures 6(a) and S7(a)). Our analysis also showed that the soil erosion level increased in most districts (22 districts among 31) in 2010–2016 (figures 6(a), S8 and S9(a)). Reduced cropland productivity correlated to cropland cultivation age, as we found increased productivity in new croplands while reduced productivity in areas with a longer cultivation period (figure 7). Our result also indicates that on average cropland productivity starts to decline after ten years of cultivation (figure 7).

Zoom In Zoom Out Reset image size

Zoom In Zoom Out Reset image size

We found that districts with reduced productivity showed a lower rate of cropland expansion ratio by correlating changes in cropland productivity to the ratio of cropland area change, such as some districts in the Southern Region (figure 6(b)). Districts experiencing increased productivity tended to show a higher cropland expansion ratio, as seen some in the Central Region (figure 6(b)). However, the correlation between changes in cropland productivity and cropland area is week. Only 13% of changes in cropland area can be explained by changes in cropland productivity and causality is not clear from this correlation. However, a comparable pattern was also observed for soil erosion by analyzing changes in soil erosion levels between 2010 and 2016, showing that expansion ratio tended to be lower in areas with a higher soil erosion level (figure S10(a)).

Overall, we observed a significant cropland expansion in Malawi over the last decade, but a little cropland loss. Compared to the The Food and Agriculture Organization (FAO) dataset, which provides cropland information only at the national level (FAO 2019), this study presented spatially explicit data on cropland expansion and cropland loss on both national and sub-national scales. This more nuanced approach is critical for policymakers, given their need to develop evidence-based interventions that are appropriate for specific contexts. Our estimation of cropland expansion is higher than the FAO estimation (FAO 2019) (figure S3); this might be due to methodological differences, specifically pertaining to the satellite image classification used in this study and the questionnaire survey conducted by the National Statistical Offices and Ministries of Agriculture (FAO 2019). Although inconsistencies exist in the estimation of cropland area using different methodologies, changes in cropland area should still be captured in individual datasets. In this regard, we found that the FAO estimation had not captured any changes in cropland area since 2014 when the cropland area remained the same with 3800 000 ha in Malawi (FAO 2019) (figure S3). On the contrary, our results are in line with other estimates which have also observed cropland expansion in Malawi in recent years (Omuto and Vargas 2019). In addition, some researchers argue that in general studies tend to underestimate existing cropland area but overestimate potentially available cropland, mainly considering widespread land scarcity, land degradation and deforestation (Young 1999, Chamberlin et al 2014).

Accurately identifying cropland location and changes in time is challenging in SSA because of the high heterogeneity and fragmentation of smallholder farming landscapes. Even though high-resolution satellite data with a grid size of 30 m (Wei et al 2020) or 10 m (Gong et al 2013, 2019) may overcome this challenge by improving the accuracy of land cover classification, limitations of satellite estimation still exist for small fields, especially when the field size is smaller than satellite image pixel size. Also, it is difficult to detect scattered trees, houses, and small natural vegetation patches among and between crop fields with current satellite data (figure S4). This technical challenge can lead to an overestimation of cropland area, but this overestimation should be consistent across the analysis conducted in different years and therefore have a limited impact on estimations of changes in cropland area. Another challenge is to distinguish fallow fields from permanent cropland fields (Tong et al 2020). In this study, we classified fields that were cultivated in 2019 as crop fields but some may have been left fallow before 2019. However, these fallow fields should account for a small area as fallow practices have reduced throughout SSA (Jayne et al 2014, Binswanger-Mkhize and Savastano 2017). Multi-temporal satellite data that can capture the different seasonal cycles of vegetation types enable classifying fallow fields from crop fields (Tong et al 2020). Combining this phenology-based classification method with high-resolution satellite data (i.e. Worldview) could offer potential improvements in classifying small crop fields in SSA (McCarty et al 2017, Mohammed et al 2020), which could be a way forward for future studies aiming at improving accuracy in cropland area detection.

Rapid cropland expansion leaves less room for future expansion. We estimated that the potentially available cropland only accounts for just over 5% of the total land, which is very small compared to the current cropland proportion of 47% of the total land in 2019. As noted in section 2.3, the definition of potentially available cropland used in this study emphasizes production potential, giving less consideration to profitability and expansion constraints, including infrastructure availability, and therefore likely leads to an overestimation of potentially available cropland. However, our finding that cropland had expanded to higher-elevated regions further removed from roads, settlements, and river basins indicates that new farmlands have been established despite poor infrastructure. This firstly suggests a low level of agricultural intensification in the expanded croplands. Secondly, the findings suggest that severe land scarcity and growing land demand have pushed farmers to cultivate in areas with unfavorable conditions. The establishments of new farmlands in higher-elevated and unfavorable conditions were probably driven by smallholder farmers in Malawi, as also noted in other countries in SSA (Ordway et al 2017), suggesting that undeveloped infrastructure might not deter smallholders from seeking new farmlands. Indeed, we found that population growth in rural areas leads to cropland expansion, which has also been widely reported throughout SSA (Chapoto et al 2013, Dorosh and Rashid 2013). Malawi household survey analysis showed that there is worsening land scarcity as exemplified by reduced field size at the national scale. However, spatially analysis showed that field sizes are getting bigger in the Central region with a higher cropland expansion ratio. Recent studies evidenced a rise of medium-scale farms in Malawi (Jayne et al 2016), and this is particularly the case for districts in the Central Region (Anseeuw et al 2016). However, future research is needed to understand interactions between changes in field size and expansion. For example, further research topics could include whether increases in field size in Central Region districts were necessarily happening where cropland was expanding, who was benefiting from increased field size in Central Malawi, and how cropland expansion and changes in field size affect the equality of land acquisitions in Malawi.

Land scarcity was compounded by cropland degradation in Malawi, as demonstrated by reduced cropland productivity and high rates of soil erosion. Defining and evaluating land degradation has been subject to considerable debate in the literature as these require a specific temporal and spatial context and can be indicated by multiple proxy data (Vogt et al 2011, Tully et al 2015, Barbier and Hochard 2018). Our evaluation focused on a national scale and used two independent indicators including satellite NDVI data (served as a proxy for cropland productivity) and farmer perception on soil erosion based on the Malawi household survey (figure 6(a)). The correlation between soil erosion and reduced cropland productivity is not clear based on the results of this study, however, both of these datasets pointed towards a trend of cropland degradation over the last ten years in Malawi, although with limitations of using the two indicators.

Firstly, NDVI from satellite data carries limitations in evaluating cropland degradation. Changes in NDVI only provide a proxy of changes in cropland productivity. Studies across SSA have found that vegetation index such as NDVI could explain yield variation in different crop types only with reasonable success (R²:0.3–0.6) (Burke and Lobell 2017, Jin et al 2017, Lambert et al 2018, Jain et al 2019). In addition, changes in NDVI at spatial resolution of 250 m generally provide estimation in changes of cropland productivity at the regional scale rather than at the field scale in SSA. Nevertheless, satellite data offers longer-term (ten years in this study) monitoring which is critical for cropland degradation monitoring and overcome any inter-annual changes in cropland productivity resulting from droughts or management practices. We found that the persistent declines in NDVI trend in the Southern region (figures 6(a) and S7) potentially indicates chronic cropland productivity reduction and cropland degradation. Further studies on examining crop yield at field scale and improve yield estimation accuracy by using crop biophysical variables such as leaf area index; implementing physically-based crop growth models and machine learning algorithms are needed (Chivasa et al 2017, Cai et al 2019, Zhao et al 2020), to support more reliable monitoring of cropland degradation.

Secondly, soil erosion level from household surveys provided limited accuracy as it was based on farmers' perception on the severity of soil erosion in Malawi. However, farmers' perception on soil erosion is critical for land degradation evaluation in the data-spare regions such as SSA countries (Nigussie et al 2017). Studies have found that farmers' estimates on soil erosion generally match empirical and theoretical findings. For example, many studies demonstrated farmers' considerable knowledge in categorizing their land and evaluating soil erosion severity, as farmers who observe their fields and experience soil degradation on a daily basis (Boardman 2006, Nigussie et al 2017). Future studies combining local knowledge based on qualitative response and quantitative models (e.g. Universal Soil Loss Equation) could provide a more robust estimation on the magnitude and spatial distribution of soil erosion (Sonneveld et al 2011).

Based on our weak positive correlation between cropland expansion and changes in productivity (figure 6(b)), a tentative interpretation we can draw is that even with cropland degradation, farmers were still establishing new farmlands when land was available, although this weak correlation does not necessarily indicate causality. However, crop expansion ratio is lower in areas with reduced productivity than areas with improved productivity, which might indicate that reduced productivity and soil erosion discourage farming activities considering profitability is a major concern in these land-use decisions (Barbier and Hochard 2018). Moreover, it might be incorrect to simply assume that improving cropland productivity would slow down expansion or even save lands, as we observed a higher expansion ratio in areas with improved productivity in Malawi (figure 6(b)). Even though the main drivers of improved productivity in some districts of Malawi—such as those in the Northern Region—were not fully clear, the enhanced productivity we observed was more strongly related to expansion than intensification, since we found that expanded croplands with younger age showed increased productivity. These findings carries important implications for cropland management and food security policies in SSA countries as cropland degradation and low agricultural intensification level lead to severer food insecurities issues (Sayer and Cassman 2013, Zingore et al 2015).