Pd/a Crsp Sixteenth Annual Technical Report

Recent work has been completed relating climatic and geographic factors to assess the suitability of particular agroecologic regions to aquaculture production. These studies were unable to compare the suitability of alternative land uses with aquacultural production. A study was therefore initiated to explore methods of generating terrestrial crop production estimates that: a) involve minimal use of complex simulation models and b) enable the use of biophysical input data likely to be available at the regional scale (e.g., monthly weather datasets). Such estimates are expected to assist regional-level decision makers to compare pond aquaculture with other types of farming systems. This work involved developing a framework to analyze and prioritize international development needs, and identifying and classifying indicators relating to sustainable development. Artificial neural networks were used to relate crop production to agricultural drivers. The Concurrent Decision-Making methodology appears to be a successful approach to facilitate stakeholder input into decision making and evaluation of alternatives intended to be used within group decision support tools. Development of a framework to assess international development needs and concomitant use of sustainable development indicators (SDI) should provide the target audience (i.e., international donor agencies, government organizations, and local groups) with a tool to examine where intervention would likely result in the greatest benefits. More specifically, such a tool can help to identify appropriate roles for aquaculture as well as other farming systems in disadvantaged communities. In this respect, the hierarchical structure of simulation models in the POND© software has provided a level of modeling (i.e., Level 1) that generates adequately accurate estimates of fish yield potential and associated resource needs for regional-scale analysis, as evidenced by their use in the studies cited above. One of the objectives of the current study was to explore methods of generating terrestrial crop production estimates that: a) involve minimal use of complex simulation models and b) enable the use of biophysical input data likely to be available at the regional scale (e.g., monthly weather datasets). Output from such methods is expected to assist regional-level decision makers to compare pond aquaculture with other types of farming systems. Our recent work in the area of computer tools for holistic, regional-scale planning also suggests that the following areas merit attention: a) a framework to analyze and prioritize international development needs and b) identification and classification of indicators relating to sustainable development. Development of a framework to assess international development needs and concomitant use of sustainable development indicators (SDI) should provide the target audience (i.e., international donor agencies, government organizations, and local groups) with a tool to examine where intervention would likely result in the greatest benefits. More specifically, such a tool can help to identify appropriate roles for aquaculture as well as other farming systems in disadvantaged communities. Thus, additional objectives of the work reported herein were to develop a framework for assessing SIXTEENTH ANNUAL TECHNICAL REPORT 186 international development needs, and to arrive at appropriate indicators of sustainable development that can be used for planning purposes. Work conducted to date in these two areas (Terrestrial Crop Performance Evaluation and Frameworks for Planning Sustainable Development) is presented below. TERRESTRIAL CROP PERFORMANCE EVALUATION In this study we are investigating the use of artificial neural networks (ANN) to estimate crop yields (CY), water requirements (WR), fertilizer requirements (FR), and grow-out period or time to harvest (TH). These variables represent output analogous to predicted data from simulation models. ANN is a relatively recent artificial intelligence technique well suited for pattern recognition problems. A major advantage of ANN is the speed at which predictions are arrived at, typically several orders of magnitude faster than multiple simulation model runs. Essentially, neural networks map input datasets (e.g., weather, soil, water, and management variables) to output data patterns (e.g., CY, WR, FR, and TH) such that if a “trained” ANN is presented with a new set of input data, it is able to accurately reproduce output variables. For evaluating crop performance by the use of ANN, one would ideally prefer to use measured input datasets together with output variables of interest from actual crop trials. Such data are difficult to come by—the alternate approach tested in this study was to use DSSAT (which has been extensively tested worldwide) as a means of generating synthetic output data with actual input weather, soil, and water datasets, together with likely management variable settings. Pilot runs have been made with DSSAT using soybeans as a test crop. The analysis used input datasets from locations in the state of Georgia. As previously indicated, a primary objective of this effort is to reduce the amount of input data because in most real world instances, it is necessary to work with sparse datasets. Consequently, input datasets for training the ANNs were substantially reduced by summarizing the daily weather datasets (min/max temperature, precipitation, and solar radiation) used in DSSAT in the form of monthly means. An additional variable included in the monthly summaries is photoperiod (day length) because this parameter strongly influences physiological responses of the different crops, particularly soybean cultivars. The approach of using monthly weather summaries is consistent with datasets that are typically used in regionalscale analysis by GIS (e.g., FAO, 1995). Other input variables used in the ANN included soil type, irrigation thresholds, photosensitivity coefficients, and planting dates. All of these inputs were used to train the ANN against desired outputs (i.e., CY, WR, FR, and TH) extracted from DSSATs summary output files. Preliminary results obtained using trained ANNs for a soybean cultivar planted either early or late in the season (Figure 1) suggest that predictions reasonably comparable to those obtained from DSSAT are possible, but with substantially reduced weather datasets. Relative errors were on average less than 10%. Following more extensive experimentation with several years of weather data, we plan to embed the trained neural networks in an expert system and apply it for estimating performance of different crops in a range of agroecological zones. Results from such analyses may be useful for regional decision makers to compare alternate farming systems, and to ultimately develop guidelines for land and water use management in different agroecological zones. FRAMEWORKS FOR PLANNING SUSTAINABLE DEVELOPMENT Strategic planning of development activities requires a systematic decision-making approach. One such approach, Concurrent Decision-Making methodology (CDM), intended to be used within group Decision Support System (DSS-decision tools for group meetings) has been outlined by Nath et al. (1998). The term “concurrent” indicates that all stakeholders present are actively involved in the phases of decision making. Concurrent Decision-Making Methodology CDM includes the following phases: 1. Identification and Selection of the Problem 2. Identification of Stakeholders 3. Problem Analysis 4. Goal Identification, Evaluation, and Specification 5. Generation of Solutions 6. Evaluation of Solutions 7. Selection of the Decision 8. Implementation and Monitoring These phases are briefly described below. The decision phases are listed numerically suggesting the approach is constrained by the need to move from one phase to the next in a linear manner. In reality, however, the process is iterative and somewhat fluid in that one is encouraged to step back to any of the earlier phases (or steps within a given phase) as needed. However, skipping to a future phase is strongly discouraged. 1. Identification and Selection of the Problem The first phase of the Concurrent Decision-Making methodology identifies and establishes problem objectives, which are stated in the form of a fairly general statement (e.g., degradation of the quality of life in a given region). This statement essentially provides some boundaries for the problem(s) to be addressed. 2. Identification of Stakeholders The second phase involves identification of stakeholders because diverse individuals and groups are potentially 0 50