Mapping and monitoring the impacts of climate variability on rainfed agriculture in Semi-arid North Darfur, Sudan

Abstract

Rainfed agriculture is vital to food security and income in most parts of the world. However, one-third of the population of developing countries population lives in the less favoured rainfed agricultural regions. Around 75-82% of the total cropland areas in the world are under rainfed agriculture and produce more than 60% of the globe’s cereal grains. However, rainfed agriculture is most prominent in some regions of Africa, such as Sub-Saharan Africa, where more than 95% of the cropland is rainfed. This crucial agriculture sector usually depends on the physical environment and, most importantly, the variability and distribution of rainfall. Therefore, rainfed farming is vulnerable to climate-related hazards, and the crop yield is unreliable and difficult to predict. For instance, the spatio-temporal variability of precipitation extreme events often subjects crops to short-term water deficits, causing crop losses. Sudan heavily depends on rainfed agriculture—about 90% of arable land dominates rainfed cultivation, contributing one-third of the country’s gross domestic product (GDP). Rainfed agriculture is the primary source of livelihood for 65% of the population. Unfortunately, agriculture in North Darfur of the west Sudan is characterised by environmental hazards, e.g., frequent droughts and unpredictable low, poorly distributed, and highly variable monthly/seasonal rainfall. Therefore, using various Earth observation data, this study aimed to monitor the impacts of rainfall variability on rainfed agriculture in North Darfur State in Sudan. Firstly, the study aimed to determine the feasibility of estimating rainfall variability across North Darfur State at daily, monthly and annual timescales using six satellite precipitation products (SPPs), i.e., the Tropical Rainfall Measuring Mission (TRMM) Multi-satellite Precipitation Analysis (TMPA), African Rainfall Climatology (ARC), and Climate Hazards Group Infrared Precipitation with Station Data (CHIRPS) were evaluated using four categorical indices, i.e., probability of detection (POD), probability of false alarm (POFA), bias in detection (BID) and Heidke skill score (HSS), and four continuous indices, i.e., Pearson correlation coefficient (r), root mean square error (RMSE), per cent bias (Pbias), and Nash-Sutcliffe model efficiency coefficient (NSE) against ground rain-gauge observations. The other SPPs were Integrated Multi-satellitE Retrievals for Global Precipitation Measurements (GPM) Final Run (GPMIMERG), Precipitation Estimation from Remote Sensing Information using Artificial Neural Networks-Climate Data Record (PERSIANN-CDR), and the Tropical Applications of Meteorology using SATellite and ground-based observations (TAMSAT). Results of the statistical analysis demonstrated that 1) at the daily timescale, the SPPs underestimate daily rainfall by 6.53–17.61%, and CHIRPS was the best for detecting rainy days, while PERSIANN-CDR performed poorly; 2) monthly and annual scales performed better than daily timescale, and TAMSAT and CHIRPS portrayed better performance than the ther SPPs. Secondly, the study assessed the capability of optical Earth Observation Data (EOD), i.e., Sentinel-2 multispectral dataset, to map crop types in the heterogeneous semi-arid environment of North Darfur using machine learning classifiers in Google Earth Engine (GEE) platform. Five datasets were compared against random forest (RF) and support vector machine (SVM) classification algorithms: (1) 10 Sentinel-2 bands (comprising visible, near-infrared and shortwave infrared bands), (2) Sentinel-2 (10 bands) + 8 vegetation indices, (3) visible bands and near-infrared bands only, (4) visible and shortwave infrared bands only, and (5) 8 vegetation indices. The eight vegetation indices were normalised difference vegetation index (NDVI), enhanced vegetation index (EVI), soil-adjusted vegetation index (SAVI), green normalised difference vegetation index (GNDVI, weighted difference vegetation index (WDVI), red edge NDVI (NDVIre), ratio-vegetation index (RVI) and normalised difference infrared index (NDII). Results showed that the RF algorithm produced the highest classification overall accuracy (OA), i.e., 97% and Kappa coefficient (κ), 0.96, using 10 Sentinel-2 bands dataset. Producer’s (PA) and user’s accuracies (UA) were in the range of 40-97% and 40-100%, respectively. Thirdly, the spatiotemporal trend of drought events and their impact on millet production in North Darfur from 1981 to 2020 was analyzed using standardized precipitation index (SPI) and reconnaissance drought index (RDI) by employing different timescales, i.e., 3- month (June-August), 6-month (June-November), and 9-month (June-February) timescales. Drought-yield relationships were assessed using Pearson correlation coefficients (r). Results indicated that RDI is more sensitive to rainfall variabilities than SPI in detecting drought trends. Results revealed that drought events affected North Darfur over broad spatial extents, particularly in 1989, 1990, 1992, 1999, and 2001—an extreme drought event was in 2003. Correlation analysis between the SPI and RDI and the standardized variable of crop yield (SVCY) for millet grain yield showed a strong agreement between them. Moderate to extreme reductions in millet crop yield occurred in 1992, 1999, 2001, and 2003, corresponding to the moderate to extreme drought indicated by RDI. Severe crop losses were in Kabkabiya and Umm Kadadda. Fourthly, this study aimed to map and monitor spatio-temporal dynamics of rainfed agriculture in North Darfur State from 1984 to 2019 using multitemporal Landsat observation data using random forest (RF) classification algorithm. Overall, Landsat Operational Landsat Imageries (OLI) outperformed Landsat Multispectral Scanner (MSS), Landsat Thematic Mapper (TM), Landsat Enhanced Thematic Mapper Plus (ETM+) in monitoring change in agricultural land and other land use land cover (LULC) classes. Overall accuracies ranged between 94.7% and 96.9%, while kappa statistics were greater than 0.90. Results showed that Goz land used for rainfed agriculture increased by 889,622.46 ha between 1994 and 999, while it decreased by 658,568.61 ha between 2004 and 2009. Rainfed cultivation of wadi lands expanded significantly by 580,515.03 ha over the 2014–2019 period and decreased by 182,701.8 ha over the 1994–1999 period. Overall, this study enhances the understanding of spatio-temporal rainfall patterns and current drought trends, aiding in developing more effective policies and resource management strategies. Additionally, it offers crucial spatial data that is currently scarce due to ongoing conflicts, empowering decision-makers to establish sustainable land use monitoring systems. The methodologies used in this study have proved successful in mapping crop types in a fragmented highly heterogeneous fine agricultural semi-arid landscape; such mapping approaches can be applied in other environments with similar characteristics.