February 8, 1999
A graduate student who is working with BROOK90 has asked me for some comments on the origin and history of the model, why I developed it, and what my intentions were. So here it is, in case anyone else is interested. This may be as close as I ever get to a scientific autobiography.
The roots of BROOK90 are deep. As a high schooler in 1954 I decided I wanted to get into forestry, and as a college student into forest soils and water. Two summers at the Hubbard Brook Experimental Forest in 1958 and 1959 (well before it became famous) turned into the only job of my career. Through the influence of Bob Pierce (my boss for most of my career), I became a graduate student in soil physics and micrometeorology at the University of Wisconsin, and in 1960 took my first computer programming course, in Fortran II. Only old-timers remember when running a program took several days from when you submitted the punch card deck at the computer center to when the printout came back telling you there was an error of one character on one card.
Energy balance and evapotranspiration were prime interests of Champ Tanner, my major professor, though he did not see how evaporation could be directly measured from forests because of likely severe fetch/advection problems. Although I made a couple of stabs at it later, I must say that I have never been successful in using a micrometeorological method to determine forest ET. Others have, but both instrumentation expense/difficulty and theoretical problems continue.
When I started full time work at Hubbard Brook in 1964, a major research area in forest hydrology was the amount and timing of streamflow increase produced by timber harvesting. Hubbard Brook has very tight gaged watersheds that provide excellent anuual ET values by difference between annual precipitation and annual streamflow. But how was the ET distributed seasonally, and how was it changed seasonally by harvest? As direct measurement in the steep terrain seemed impossible, I quickly turned to modeling as a more fruitful approach.
On December 10, 1965 my notebook entry is titled "Water balance program - Run 1", apparently the first results from what eventually became the BROOK model. This early work did not lead to publication, but the results are interesting. For monthly Watershed 1 streamflow (inches) at Hubbard Brook, using Thornthwaite PE for April through November:
This first model included 5% direct runoff, a first order lag in excess runoff (40%/day), an upper soil layer with one inch of available water evaporating at PE, and a lower soil layer of 4 inches available with AE reduced below PE by three different methods (only one shown above, not much difference). The simulated soil water agreed well with measurements from Colman soil moisture units. Pretty good results from a very simple model. As Thornthwaite had already shown, water budgeting is one of the easiest things to do in forest science and a simple model can't be too far off. No model is going to say it's wet when it's dry and vice versa.
I was concerned about the AE/PE relation and compared seven different methods at that time. One was what I called the Denmead method after Denmead and Shaw's classic paper. I wrote it as AE = min(PE, W/C) where W is available water and C is a constant. This began my interest in the concept that AE is the lesser of PE and a soil-water supply function, which culminated in a comparison of this simple relation with a complicated soil-plant-atmosphere model (Federer 1982). This concept has since found its way into land-surface components of GCM's.
Also in 1965 I made my first comparison of different potential evaporation estimates at Hubbard Brook, and used them to estimate monthly water budgets. In 1968 I wrote for Hubbard Brook that "the Penman method seems too high and the Thornthwaite method seems too low". Over 30 years later I am still comparing PE methods and finding that Penman is too high and Thornthwaite is too low almost everywhere (Federer et al 1996, Vorosmarty et al 1998). I tried to develop a simple PE method based on solar radiation alone, but we now know that vapor pressure deficit is more important than solar radiation in driving ET from forest canopies, and my effort never produced any publications.
Modeling was only part of my work over the next few years as I conducted field experiments in radiation, wind profiles, and snowmelt, and looked at streamflow recession. From 1970 to 1976 much of my effort was directed to field research on stomatal resistance and physical factors controlling it, using the newly developed diffusion porometers and pressure chambers. All of these efforts contributed in one way or another to the BROOK models. I feel strongly that modelers should also be field researchers, and that modelers should be familiar with the actual operation of the system components they are modeling.
On October 10, 1973 my notebook says "began work on revising H.B. watershed model". The equations for the first version of the BROOK model were written down by February 1974. The BROOK model reached publication in Federer and Lash (1978a, 1978b). This model was written in Fortran IV; by 1983 I had updated the code to Fortran-77 and decided to call the F-77 version BROOK2 to distinguish it from the Fortran IV version, though the algorithms were the same.
The field of hydrologic modeling was already about saturated by this time (pun intended), starting with the Stanford Watershed Model in the late 60's and followed in the 70's by numerous other models, so why did I add yet another one. Perhaps there are two types of people - those who are satisfied using someone else's tools, and those who think they can do it better. I guess I'm in the latter category. Other models just didn't do it the way I wanted it done. There was great debate then about whether a generalized super-model could be constructed to answer all questions, or whether each question required a separate model to answer it. The end result seems to be intermediate: to ask a question of several models and see if they agree or disagree. New knowledge awaits in the sources of any disagreements, and confidence arises when the results are similar.
The primary purpose of BROOK2 was to answer specific questions about the relationships between forest cover and the water budget; some of this was published in Federer and Lash (1978a). Somehow interest in BROOK2 slowly spread around the world, through students and faculty at the University of New Hampshire, through Hubbard Brook scientists, and through publications. Two things greatly fostered this interest - the model was free and its documentation was complete. It was also complicated enough to express the various processes separately but simple enough to understand. Over the next few years BROOK2 became used in North America primarily for teaching and in Europe primarily for research; I'm not quite sure why. Keith Eshleman developed a PC version at MIT, and Miller et al (1988) at UConn and UMass modified the model by adding parameter-estimating schemes and called it BROOK6. In 1989 Hans Keller organized a workshop for European users of BROOK2 where the focus was on ways for me to improve the model.
Through the 1970's, "forest hydrology" research transferred its focus from physical to chemical aspects of the system, largely influenced by Hubbard Brook. In the early 1980's the combination of my midlife crisis and Forest Service pressure drove me away from what I did best and into nitrogen processes and spruce decline, where a great deal of my effort led to minimal useful results.
Gradually the people using lysimeters to sample the chemistry of soil water and the people concerned about how that chemistry changed in the pathways to the stream started demanding quantitative information on soil water movement. The BROOK2 model, with its single-layer root zone, was inadequate for calculating fluxes past lysimeters, though some tried to use it that way. I was also unhappy with BROOK2 for other reasons: use of the temperature-only Hamon PE, empirical non-physical parameters for LAI-controlled processes, over-simplified soil-water pathways, and lack of generalization to other cover types. In addition I wanted to incorporate new knowledge of the importance of leaf resistance and the role of liquid-flow resistance in the soil and plant. Simultaneously with BROOK2 I had developed a detailed soil-plant-atmosphere model (Federer 1979) with a multi-layer soil and a big-leaf canopy to simulate measured behavior of leaf resistance and xylem potential (Federer 1980), but this model did not get incorporated into BROOK2. The significant, if not stupendous, use of BROOK2 worldwide gave me the incentive for going further.
In 1988 I got back into the modeling of water movement with some abortive work on a hillslope model using a generic model driver that I wrote. The attempt to put BROOK2 into the model driver didn't work, but the wheels had started turning again, and the name BROOK90 was first used in late 1989 even though I knew the new model wouldn't be done for a couple of years. The new number was chosen to avoid any possible conflict with any other BROOKn models that might be out there.
The major objective of BROOK90 was to produce a model of daily evaporation and soil water movement at a point that would work for all land surfaces at all times of year using a process-oriented approach with physically-meaningful parameters. Only enough water movement pathways were included to allow some comparison with measured streamflow where available. The complexities of hillslope hydrology and spatial distribution were omitted in order to focus on the details of the factors controlling evaporation. The model was designed to be a complex lumped-parameter model that could suggest by sensitivity analysis what should be included or excluded from simpler soil water submodels for models of plant growth, productivity, biogeography, and global hydrology.
BROOK90 Version 1 was a preliminary model distributed in July 1992 to a few interested people who just couldn't wait any longer. Versions 2F and 2Q were distributed starting January 23, 1993. John Aber convinced me during our work on the PnET model that QuickBasic on a PC was much easier for programming and debugging than Fortran. Once the QuickBasic was right, Cindy Veen translated it into Fortran-77 which could be used and modified anywhere by anyone. Although Version 2 was somewhat widely used, there were various areas of it I was dissatisfied with, and I continued working to finalize Version 3.1 just in time for my Forest Service retirement in January 1995. Once again both Fortran and QuickBasic versions were produced.
Versions 1 and 2 incorporated my old interest in comparing PE methods and had routines for many of the nine methods later compared by Federer et al (1996). In cooperation with Charlie Vörösmarty we separated the PE comparison work from BROOK90 and I decided to use only the Shuttleworth-Wallace PE in Version 3. We are continuing work now on using BROOK90 to suggest improvements in his Water Balance Model (WBM) a submodel of The Ecosystem Model (TEM).
Sometime in 1996 I began to get frustrated with the difficulty of working with the file editing and filename juggling required by Version 3.1 and decided to use VisualBasic to produce a Windows version. After a steep learning curve, Version 3.2 for Windows was ready for distribution in April 1997. After many years of working on the "guts" of the model (the algorithms) I had finally succumbed to the need for user-friendly "garbage" (the user interface), which has taken most of my limited available time since then. However, I hope that modelers do not forget that the garbage, which takes so much programming effort, is scientifically worthless; the guts should be where most of the energy goes.
So in many ways, BROOK90 is the product and summary of my whole research career. Its history is also a history of the tremendous increase over 40 years of our knowledge of how water moves through the soil-plant-atmosphere system.