Chris

We have also done exactly what you have suggested with the PALM (People and Landscapes Model). Several of the DSSAT crop models have been rewritten as objects in Delphi, with the rates calculation and integration of state variables separated and callable from outside the object. Any number of these objects can be run in parallel along with others for livestock, tree growth, weed growth, and human decision making, and in fact different instances of the same object can even be run simultaneously (e.g. to simulate two crops of wheat growing in separate fields, each with its own management regime). A number of Field objects (containing various combinations of the above objects) can be assembled to represent a 3D landscape (i.e. areal x depth) if necesary. The overall controlling routine (i.e. main model) calls the rates calculation routine of each object in turn, followed by the corresponding state integration routine of each object in turn, and also provides the route for communication between objects. In fact, the objects are more reactive agents, with this communication via KQML messaging. The structure allows the overall model to run a timestep at a time, after which variables within each object can be peeked at (actually a request for information by the peeking object) and and this information used as a basis for decision-making by the decision module. Thus, decisions can be made in 'real time' on the basis of the instantaneous states of these variables, rather than after each model had run as the old structure enforced.

Convesting the DSSAT models to this structure was not a big job once the architecture was worked out, although the system for routing and interpreting messages did take more time.

Papers on (a) the model and its architecture, and (b) on its application to study sustainability of different cropping systems, have just been submitted and are under review at the moment.

Robin Matthews
Reader in Biosystems Modelling
Cranfield University.

Note change in address after August 1, 2004:
============================================
Programme Leader, Landscape Change Programme
The Macaulay Institute
Craigiebuckler
Aberdeen AB15 8QH
UK
Tel: +44 (0) 1224 498200; Fax: +44 (0) 1224 311556; email: [log in to unmask]




-----Original Message-----
From: DSSAT - Crop Models and Applications on behalf of Chris G. Nicholas
Sent: Tue 6/8/2004 5:58 PM
To: [log in to unmask]
Subject:      Re: Modelling intercrops

Or, the next generation of DSSAT needs to be able to "single-step"; i.e. allow some external entity to "peek" at its state variables, perform cross-factor interaction calculations, and then "poke" the changed variables back into DSSAT for the next step. Other interesting aspects are how to put the interacting programs into "suspended animation" while performing such peeking and poking, but there are known checkpointing and gdb tricks to do this.

Obviously this is non-trivial, and would probably involve at least several weeks of work by someone familiar with DSSAT internals. But it would be interesting to be able to link 2 or more DSSAT models, and then have something external look at the individual LAIs, nitrogen and moisture uptakes, etc, and then go back into each and "poke" the next step.

One would want to write that in a general-purpose manner, such that one might be able to externally blend in other factors, such as pests, as well.

Indeed, if anyone out there has a companion computer science department writing GRID-based research proposals (UFL?), hooking up DSSAT using such a mechanism to the stuff of: http://www.uvm.edu/giee/SME3 would allow for modeling crop interactions with surface hydrology, herbavores, hill shading, etc in scalable, location-specific manner.

Chris