Use case: “Make agro-input recommendations”

by | December 18, 2013
Category: News

Use case summary in context

The document describes the steps for recommending agro-inputs (e.g., seed, fertilizers, lime and organic matter) in smallholder farming systems, based on site, cropping system and time specific information and decision analyses. The ultimate goal is to assist individual farmers and/or farmer groups adjust their site-specific input applications to reduce their livelihood & business risks, and to achieve higher economic returns to land, labor, capital and ecosystem service investments. An overview of the main actors and associated use-cases are shown in the diagram below (Fig. 1, in UML syntax, also see: http://en.wikipedia.org/wiki/Use_case).

Figure 1: “Make (agro)-input recommendations” use case in context.

The main steps for generating site and time-specific agro-input recommendations consist of:

  1. Measure cropland areas: this use case comprises interpretations of satellite remote sensing, unmanned aerial vehicle (UAV) and/or ground survey data for measuring, mapping and monitoring cropping system types and areas across a representative sample of Africa’s croplands.
  2. Assess site condition: comprises ground surveys, laboratory and remote sensing-based data analyses to assess the biophysical condition of sites by CKW’s, supported by AfSIS staff  (e.g., soil fertility and water availability) within the context of existing cropping systems. It includes and extends information products generated under use case 1.
  3. Evaluate cropping systems: this use case comprises site and time specific ground surveys by CKW’s and farmer groups, supported by interpretations and spatial predictions of cropping system characteristics, e.g. crop types, management histories, productivity and market indicators, etc. It includes and extends information products generated under use cases 1,2,4 & 5 to evaluate total crop production flows and their uncertainties.
  4. Make (agro)-input recommendations: comprises computer supported optimization and decision-analyses for making agro-input recommendations by CKW’s based on site condition information and forecasts, expected crop yield responses and market information and including evaluations of prevailing risk-attitudes of (smallholder) farmers. It includes use cases 2 & 3 above.
  5. Evaluate outcomes: this comprises assessments of the recommendations generated under use case 4, including monitoring, evaluating and subsequently adjusting site-specific (agro)-input applications and potentially other management recommendations based on inputs from CKW’s, farmer groups and agro-input dealers.

In principle, the main use case described in this document can be implemented as computer software that helps to assess current site condition and/or use predictions on consistent data models (see: http://en.wikipedia.org/wiki/Data_model) based on free and open source computer software (see e.g.: http://www.r-project.org & http://grass.osgeo.org), and supported by largely automated analyses of remote sensing and market time series that will produce maps, monitoring and decision analysis products and services for the cropland areas of Africa. Once the system has been developed, it should be deployable via web and mobile services that would provide access to data and information products to CKW’s, farmer groups, agro-input dealers, land management policy-makers, and government agencies.

Site and time specific, agro-input decisions

Some of the most risky management decisions for smallholder farmers in Africa in terms of livelihood and business security involve:

A key attribute of these decisions is that there are uncertain variables involved that generate production risks for farmers, e.g. variations in base yields, responses to management, market prices and costs etc. The perceived and actual risks can lead to suboptimal use of farm resources.

Decision analyses with influence diagrams

An influence diagram is a compact graphical and mathematical representation of a decision problem. Influence diagrams are a generalization of a Bayesian network, in which both probabilistic inference problems and decision-making problems can be solved.

The single circles in Figure 2 below, represent uncertain quantities (random variables), double circles represent deterministic variables, squares represent decisions that are under the direct control of the decision-maker e.g., whether to plant maize or sorghum, and diamonds represent functions that are to be optimized e.g., Profit [Gain-Cost], Return on Investment, [ROI = (Gain – Cost) / Gain], or more generally Utilities, which take the risk attitudes of farmers into account.

The arrows between variables in the directed graph represent relationships between variables. Solid arrows represent conditioning relationships. Dashed arrows represent additional information flows, or so called, “no forgetting relationships” that express that a given decision is contingent on another or that the outcome of random events are known prior to making a particular decision.

Figure 2: Influence diagram for crop management decisions at a given site in time.

There are several advantages of explicitly quantifying uncertainty in decisions as opposed to deterministic cost-benefit analyses.  First, variables are represented as probability density functions, as opposed to e.g., agro-ecological zone averages. Averages are often highly misleading when statistical data about the nature of individual (sites) are deduced from inferences for the group to which those individual sites belong (see: http://en.wikipedia.org/wiki/Ecological_fallacy.

Secondly, explicitly representing uncertainty allows quantification of “risk” (the state of uncertainty where some of the possibilities involve losses, or other undesirable outcomes), critical to farmer and government concerns (e.g. based on food security, resilience, sustainability and crop failure considerations). Risks in investments can ultimately be expressed by one method: the ranges of uncertainty on the costs and benefits and probabilities on events that might affect them.

Thirdly, the expected monetary values (EMV’s) of acquiring further information can be calculated when the relevant uncertainties are represented. This places the focus on narrowing uncertainty sufficiently to make optimal decisions, and on how much should be spent of further measurement.

Variable definitions

Site: represents a geographical location that can be defined on a map of the actual crop-production areas of Africa. For this use case we recommend that sites are defined and characterized on a standard 100-hectare grid for all of Africa. The relevant coordinate reference system (CRS) (http://trac.osgeo.org/proj) for the grid used by AfSIS is: +proj=laea +ellps=WGS84 +datum=WGS84 +lat_0=5 +lon_0=20.

Site time: represents the time at which the decision is to be made. The relevant time standard for AfSIS in this context is Unix/Posix time (http://en.wikipedia.org/wiki/Unix_time). Notably, decisions that are made at current versus future time (time period) will have outcome implications, because they either require forecasts of decision outcomes given the current situation, or not.

Site condition: this random variable represents the environmental conditions (prevailing weather, soil, biotic, terrain and geological material conditions = environmental state factors) that affect crop production and productivity in the absence of any specific management interventions. A common method for determining site condition is by relating crop yields to various site and time specific state factors.

Crop: this decision variable represents the choices between cropping systems or crop types of interest at a particular site and point in time.

Management: this decision variable represents the relevant management alternatives being evaluated. These can include but are not limited to agro-inputs (e.g. hybrid seed, fertilizer, lime, etc.).

Base yield: this random variable represents the yield of a particular cropping system at a site under typical management in the absence of the (new) management alternatives that are to be evaluated. That is, it typically represents crop yield (in kg/ha) under prevailing farmer’s practices.

Response: this random variable represents the crop yield response of a given management intervention relative to the base yield scenario.

Cost: this random variable is an indicator of the overall cost of the management intervention under consideration relative to the base yield scenario.

Price: this random variable is an indicator of location specific farm-gate for (the mix of) specific cropping system products.

Gain: represents the random variable Price × Response.

Utility (U): The overall level of satisfaction derived by farmers in response to Gain/Cost relationship that take the relevant uncertainties and prevailing risk attitudes into account.

Main use case description“Make (agro)-input recommendations”

Overarching goal: Support community knowledge workers (CKW’s) with near-real time information products and computer software to advise smallholder farmers in Africa on formulating potentially profitable and environmentally sustainable cropping system management options.

Primary actor/ system user: CKW.

Scope: Influence diagram based decision analysis and optimization software.

Stakeholder interests:

CKW(s) – want to provide feasible, profitable and environmentally sustainable land management options to support farm production goals. Depending on how good the provided advice is, this might be developed into a business/profit generating activity for CKW’s, e.g. on a “charge-for-performance” basis.

Farmer(s) – want to obtain realistic crop management options for improving farm profitability, returns on investment or utilities, which decrease livelihood and farm business risks over a given planning horizon.

(Agro)-input dealer(s) – want to match available agro-inputs to demands by farmers to increase agro-input business profits and returns on investments.

Policy-maker(s) – want to guide and achieve rational (economic, social, health & environmental) outcomes that are to the benefit of his/her constituents.

AfSIS staff – want to support stakeholder decisions with relevant data and information products. Ideally, AfSIS wants to devolve decision support in this area to national entities by providing computer workflow and software applications.

Precondition: CKW has access to a computer or a mobile device with an Internet connection.

Main success scenario:

  1. CKW registers a 100 ha site on AfSIS or e.g. EthioSIS optimization service.
  2. CKW obtains information on current and forecasted site conditions (e.g. weather, soils) via a web or mobile service.
  3. CKW evaluates current and potentially relevant cropping system and management options with farmers, including assessments of areas under different crops, commodity prices and management costs.
  4. Based on data inputs (3 above) the optimization software generates a report of recommendation alternatives and their utility and certainty equivalent outcomes.
  5. CKW registers, which alternatives are taken up (or rejected) by farmers on the service at a particular point in time, and these are logged into a database.
  6. System iterates from step 2 over site time.

M. Walsh & K. Shepherd (Africa Soil Information Service, http://africasoils.net)

Submit Comment

Your email address will not be published. Required fields are marked *