Creating a Site-Specific TMY3 Weather Input File for SAM

I really like SAM, and I’d like to use your PVWatts model (or possibly the components-based model) to inform my thinking about solar PV for our off-grid residential house in Hawaii (now in the early planning stages). In addition to optimizing energy capture, PV angle and orientation could affect our roof design, so I want to drive performance evaluation of any prospective system with actual onsite measured insolation data (e.g., seems like it's typically cloudier here in the afternoon than in the morning, which would argue for more of an eastward orientation than due south and maybe a slightly different roof pitch). I’m not sure that the nearby available TMY datasets for Hawaii (Hilo, Lihue) would reliably reflect our local microclimate up here on a windward hillside of the Big Island (in Paauilo).

To this end, I installed a weather station to collect hourly data when we moved here this July. I’ve got awhile to go before I’ll have a full year (8760 hours), but at this point I’ve got enough data that I want to start fooling around with analytical tools. I can create a preliminary test dataset in TMY format with 7 of the 9 input columns required in PVWATTS, but my little weather station doesn’t have any instrumental measurements for DNI or DHI, only for GHI (i.e., standard pyranometer).

So I guess I’ve got three questions about the required meteorological input file:

(1) Does PVWATTS need all three solar parameters to drive it, or can it work off of only GHI?

(2) If it needs all three, can you recommend a methodology to estimate DNI and DHI based on GHI? I note that somebody’s given that a shot in the past… excerpt from rredc.nrel.gov/solar/pubs/NSRDB/history.html... “Direct normal data for all stations were estimated using regression equations. Global horizontal and direct normal data for five stations (Albuquerque, New Mexico; Fort Hood, Texas; Livermore, California; Maynard, Massachusetts; and Raleigh, North Carolina) were used to develop regression equations to calculate direct normal values from global horizontal values (Randall and Whitson 1977). These few direct normal data were collected from 1974 to 1975, with the exception of Albuquerque (1961 to 1964). The regression equations were used to calculate all of the direct normal data for the 26 SOLMET stations for the entire period of record (16-1/2 to 24 years). Similar regression equations were used to calculate direct normal data for Typical Meteorological Year (TMY) data sets for the 222 ERSATZ stations (NCDC 1981). This brief historical summary of solar radiation measurements and data base developments for the United States reveals shortcomings and limitations that must be considered when using data from the NSRDB. Although the NSRDB has benefitted from improved data and improved models, the uncertainties attached to much of the data are still unacceptably high. The user may want to use the source and uncertainty flags to screen data to be used for critical computations and decisions.” ...but is there something more current?

(3) If no current methodology comes to mind, what do you think of the idea of -- for purposes of input to PVWatts -- calculating hourly max possible DNI (no clouds or haze) then adjusting it downward by some empirical factor derived from the ratio of measured GHI/max possible GHI?

Thanks so much!

Denning

Thanks, Paul, I'll give that a shot! One more question -- is the DNI column in a site-specific TMY3 dataset the calculated theoretical max incident on a surface normal to the sunshine, or is it data that reflects the actual degree of cloud cover in between?

Denning :-)

Aloha, Paul

I downloaded DISC, which is an enterable spreadsheet. Following the instructions, I did the following:
(1) pasted a year of my site-specific data (8760 hourly GHI data) into input column N;
(2) entered my Hawaii site location data (20N, 155W, time zone -10, 1000 mb pressure) in the appropriate cells in column A.

Got some weird results in the DNI output column U -- DNI way lower than my measured GHI. So I did the following:
(1) checked my input data (it seemed ok);
(2) tracked the cell equations backward from U to M (a convoluted path but nothing leapt out at me as being obviously erroneous);
(3) ran DISC on its existing sample GHI data (2 days of what looks to be a Philadelphia location), but changed the location data in the column A cells to reflect my Hawaii location.

That latter replicated the weird results -- DNI