Synergy for Enhanced Crop Productivity — A Technological Paradigm for Improved Smallholder Farming
Gabriel Eze, Prince C. Peter, AGRA+NAERL
Abstract: Food insecurity is a fast growing problem with global dimension. Smallholder farmers account for the bulk of the agricultural produce in Nigeria but are threatened by dwindling soil nutrients. Traditionally, the remediation of soil nutrient deficiency involves a long process of destructive soil sampling and laboratory analysis. However, remote sensing and related technologies have proven to be more efficient in soil nutrient management strategies in terms of less time and greater accuracy. But the knowledge and adoption of such technology is limited among smallholder farmers in Nigeria. Therefore, a preliminary study was conducted across an agricultural belt of North-central Nigeria, where selected soil nutrients were determined by both laboratory sample analysis and remote sensing. The aim was to establish the possibility for substitutability of destructive soil sampling and laboratory process for in situ time saving remote sensing approach. A transect which cuts across several smallholder farms was established, and soil samples collected for laboratory analysis, while satellite imagery of the areas were obtained for remote sensing data.
Keywords: geospatial variation; soil nutrients; multispectral data, remote sensing
1. Introduction
Sub-Saharan Africa is characterised by high number of food insecure people, a sizable part of the population is constantly hungry and malnourished (FAO, IFAD, WFP, 2014); (Croft et al., 2016), a situation that is exacerbated by climate change and most recently global pandemic. (Kisaka-Lwayo & Obi, 2016) noted that food and nutritional security are major global concerns, but predominant in developing countries. Smallholder farmers constitute the majority in the farming population of the tropics and are mostly subsistence (Gelaw et al., 2014) (Powlson et al., 2016). According to (Nwafor & van der Westhuizen, 2020) smallholder farming has potential to improve food security and enhance rural livelihood. However, the performance of smallholder farming has been low due to inadequate soil nutrient management systems. Crop yields in Nigeria (particularly for grains) have been declining for 3 decades (FAOSTAT, 2020); the average productivity of maize in Africa is less than half the global average. In Asia, it was 2.5 metric-tonne per ha, and 4.1 metric-tonne per ha in Europe.
The process of determining soil nutrient status for proper monitoring and appropriate policy formulation is usually lengthy and highly technical; from field sampling through laboratory analysis to result interpretation. The stages involved in soil analysis do not make for quick assessment and timely intervention especially across a broad area. According to (Mokhahlane & Ajuruchukwu, 2011) there is need for up–to–date information to enhance smallholder farming.
Enhancing total farm productivity in situ in developing countries is unarguably a brilliant alternative. To this end, expanding the capacity of farmers to produce more food for increased income at low cost, using easily accessible technologies, without further damage to the environment should be prioritised. (Song et al., 2018) noted that spatial distribution of soil nutrients is essential for higher soil productivity, but determination of soil nutrients and their spatial distribution has usually been by field sampling and laboratory analysis, which are characterised with many inevitable sources of error, high cost and inefficiencies (due to handling of samples) as well as time-consuming. Hence, multispectral satellite data has been considered as a possible alternative for mapping and predicting soil nutrient status at a higher level of accuracy, relatively lower cost and shorter time duration (Seutloali et al., 2016).
According to (Song et al., 2018) there could be individual similarities existing in a single sample among soil nutrients and environmental parameters, which can lead to a gradual change/difference in continuous space for linear and nonlinear relationships between remote sensing, vegetation, topography and climate. Using hyperspectral remote sensing images (115 bands) of the Chinese Environmental 1A satellite (Song et al., 2018) were able to map and monitor soil nutrients at a regional scale. In the study, the most important bands for predicting soil nutrients were identified as those which exhibited the best response and sensitivity according to the models used. The researchers were able to establish best fitting ability between hyperspectral remote sensing bands and soil nutrients for spatial mapping of soil nutrient concentrations. Other researchers (Yu et al., 2016); (Cao et al., 2017); (Jia et al., 2017); (Pullanagari et al., 2017) also used different models successfully for image classification, land surface mapping, as well as plant and soil properties mapping with remarkable success. However, such an approach to agriculture is either lacking or very few in Nigeria. Therefore, this pilot study was conducted across an agricultural belt of North Central Nigeria, to determine the feasibility of using remote sensing to investigate the spatial variation of selected soil nutrients. The specific objectives were:
- determine selected soil physicochemical properties across different farms in the same agricultural belt,
- determine comparative multispectral variables across the agricultural belt for spatial prediction, and
- determine possible nexus and synergy in the data obtained from objectives 1 and 2.
2. Materials and Methods
2.1. Farms and Soil Sampling
Smallholder farmers across four local government areas (LGA) in Kaduna state, North Central Nigeria were selected for the pilot study. Forty points were determined using coordinates that captured major active smallholder farms in the different LGAs from which representative soil samples were collected at 0 to 20 cm and 20 to 40 cm depths. The soil sample collections were done using stainless steel core rings for undisturbed soil samples and soil auger for disturbed soil samples. The soil samples were collected and grouped according to common crops grown, soil depth, soil chemical and physical parameters of interest. The soil samples were placed in appropriately labelled sampling bags and containers before transporting to the Soils Science Department of the Ahmadu Bello University Zaria for laboratory analysis.
2.2. Area(s) of Interest
The investigation was carried out in Kaduna state (9°26' to 11°13' N and 7°47 to 8°42' E), which is located in the north-west of Nigeria (Pic 2). The climate belt of the area is tropical Guinea Savanna, with an annual average temperature of 25.2℃ and an annual average rainfall of 1,323 mm. The major food grown in this area includes maize, rice, and soybean. The area is characterised by plain and grasslands with a maximum elevation of 250m. The soil type in the farmlands in the study area is classified as Sandy-Clay loam; this is according to Soil texture triangle.
Prior to soil sample collection, the following comparative baseline data were collected using our remote sensing software application (Capture).
2.3. Field Sampling, Chemical and Spatial Analysis
We opted for the option of collecting soil samples in month(s) near the end of the harvest period of the rainfed calendar because it minimises errors due to
- excessive soil moisture from the rainfed window (requiring extra drying time) for samples for chemical analysis on one hand, and
- cloud cover for spatial analysis.
For the former, we used a simple grid partition method where samples are collected from 5 points of a virtual “X” marked over the actual size of the farms under study. Two (2) sets of samples were collected: mixed and unmixed samples. The mixed samples were collected in 4 local government areas (LGA) of Kaduna state distributed evenly between the northern and southern part of the state, presenting the northern and southern Guinea Savanna belts respectively; for each LGA, 3 farms were selected at random, and for each farm, 2 samples were collected per farm at depth 0–20 cm and 20–40 cm. The total mixed samples collected was 20.
The unmixed samples were collected in the same 4 LGAs of Kaduna state distributed evenly between the northern and southern part of the state; for each LGA, 1 of the 3 farms was selected at random, and for each farm, a sample was collected for each 5 points of the “X” markup at depth of 0–20 cm only. This is because most satellite data for soil properties are within the top-soil range (0–9cm); this we assume is within a reasonable margin of error. For the latter, a data streaming pipeline is used to query and download multispectral data from Sentinel-2 repository which is then processed using a proprietary algorithm. It is important to note that the data acquisition and processing techniques are not the subject under investigation, hence its specifics will not be discussed; the result from this algorithm and how it correlates with those from chemical analysis is the subject and primary objective.
The samples were air-dried naturally at room temperature, and after removing plant and other residues, all samples were passed through a standard nylon mesh sieve. The determination of micro- and macro-nutrients was done by standard laboratory procedures.
Below are the result of the standard laboratory analysis of soil samples collected:
2.7. Comparative Multispectral Variables
We selected multispectral imagery from September 2020, which coincides with the time frame soil samples were collected. We also took care to select only imagery at such time with the least cloud cover, this is to ensure the accurate correspondence between the remotely sensed data and the soil sample for chemical analysis. We made use of Copernicus data by the European Space Agency (ESA), specifically from Sentinel 2A and 2B satellites which carry multispectral radiometers and sensing in 13 different spectral bands. We used the same coordinate as the collected soil sample (geo-reference).
Below are the imagery derived after software computation of a time series satellite data. In order to derive a numerical equivalent of the imagery so as to compare with laboratory results, we divided the image into four (4) equal quadrants (representative of 4 of 5 points of the “X” markup, excluding the center), and then used a python script to count the number of pixels in each quadrant per colour type. Because pixel count does not take into consideration the variation in significance or weight of each colour, we opted for normalisation where we multiplied all the value for H pixels by 3, M pixels by 2, and L pixels by 1; where H represents the point with highest concentration of nutrient, M represents average concentration, and L represents lower concentration.
2.8. Correlation Analysis
3. Results
3.3. Soil Chemical Properties
As shown in pic 4, it can be established that it is possible to substitute soil sampling by laboratory process for an in situ approach like remote sensing which is more readily accessible (on-demand), affordable, and time-saving. There is a correlation (nearly 74%) between both approaches with the latter less prone to error and malhandling due to contactless interaction with the soil samples. It suffice to add that modeling the availability of soil nutrient using a time series data that is sensed remotely is still an area of study that will continue to improve.
7. References
Cao, F.X. et al., 2017. Linear vs. Nonlinear Extreme Learning Machine for Spectral-Spatial Classification of Hyperspectral Images. Sensors, 17, p.2603.
Chukwuji, C.O., Inoni, O.E., Ogisi, O.D. & Oyaide, W.J., 2006. A Quantitative Determination of Allocative Efficiency in Broiler Production in Delta State, Nigeria. Agriculturae Conspectus Scientificus, 71, pp.21–26.
Croft, M.M., Marshall, M.I. & Steven, H.G., 2016. Market Barriers Faced by Formal and Informal Vendors of African Leafy Vegetables in Western Kenya. Journal of Food Distribution Research, 47, pp.49–60.
FAO, IFAD, WFP, 2014. Strengthening the enabling environment for food security and nutrition. The State of Food Insecurity in the World 2014. Rome: FAO.
Gelaw, A.M., Singh, B.R. & Lal, R., 2014. Soil organic carbon and total nitrogen stocks under different land usesin a semi-arid watershed in Tigray, Northern Ethiopia. Agriculture, Ecosystems and Environment, 188, pp.256–63.
Jia, S.Y. et al., 2017. Hyperspectral Imaging Analysis for the Classification of Soil Types and the Determination of Soil Total Nitrogen. Sensors, 17, p.2252.
Kisaka-Lwayo, M. & Obi, A., 2016. Analysis of production and consumption of organic products in South Africa. In K. Etingoff, ed. Sustainable Agriculture and food supply Scientific, Economic and Policy Enhancements. Broken Sound Parkway NW: Apple Academic Press, CRC, Taylor and Francis. p.51.
Mokhahlane, M. & Ajuruchukwu, O., 2011. Institutional factors in natural resource management in the Eastern Cape Province of South Africa. In A. Obi, ed. Institutional constraints to small farmer development in Southern Africa. Wageningen: Wageningen Academic Publishers. pp.231–69.
Nwafor, C.U. & van der Westhuizen, C., 2020. Prospects for Commercialization among Smallholder Farmers in South Africa: A Case Study. Journal of Rural Social Sciences, 35, p.2.
Oguike, P.C. & Mbagwu, J.S.C., 2009. Variations in Some Physical Properties and Organic Matter Content of Soils of Coastal Plain Sand under Different Land Use Types. World Journal of Agricultural Sciences, 5, pp.63–69.
Powlson, D.S. et al., 2016. Does conservation agriculture deliver climate change mitigation through soil carbon sequestration in tropical agro-ecosystems? Agriculture, Ecosystems and Environment, 220, pp.164–74.
Pullanagari, R.R., Kereszturi, G. & Yule, I.J., 2017. Quantification of dead vegetation fraction in mixed pastures using AisaFENIX imaging spectroscopy data. Int. J. Appl. Earth Obs., 58, pp.26–35.
Song, Y.-Q. et al., 2018. Predicting Spatial Variations in Soil Nutrients with Hyperspectral Remote Sensing at Regional Scale. Sensors, 18, p.3086.
Udom, B.E. & Omovbude, S., 2019. Soil physical properties and carbon/nitrogen relationships in stable aggregates under legume and grass fallow. Acta Ecologica, 39, pp.56–62.
VandenBygaart, A.J. & Angers, D.A., 2006. Towards accurate measurements of soil organic carbon stock change in agro ecosystems. Canadian Journal of Soil Science, 86, pp.465–71.
Yu, X. et al., 2016. Evaluation of MLSR and PLSR for estimating soil element contents using visible/near-infrared spectroscopy in apple orchards on the Jiaodong peninsula. Catena, 137, pp.340–49.
