Modelling the effects of environmental conditions on apparent photosynthesis of Stipa bromoides by machine learning tools

Apparent leaf photosynthesis of the grass Stipa bromoides was measured in t he field in two sites of Northern Greece. For predicting apparent photosynt hesis from irradiance, temperature and relative air humidity data, we appli ed and compared two modelling approaches: ordinary statistical modelling an d automatic model construction based on machine learning procedures. Ordina ry statistical models were constructed based on background knowledge concer ning the response of photosynthesis to irradiance. A Michaelis-Menten type light saturation curve was selected among six candidate models and was exte nded to include air temperature effects. A bell-shaped function of temperat ure was substituted for the parameter describing the asymptotic maximum pho tosynthesis. The final model accounted for 67.3% of data variation and was further improved by splitting the data set by experimental site. Site-speci fic differences were detected regarding the half saturation constant for li ght and the optimal temperature for photosynthesis. Automatic model constru ction produced a number of regression trees that enabled a detailed but sim ple description of the way irradiance, temperature and relative humidity af fect photosynthesis. Photosynthesis increases with increasing irradiance, t emperature affects photosynthesis when irradiance is close to saturation le vels and relative humidity has an effect when both irradiance and temperatu re are high. There is a threshold value of relative humidity (at about 35%) , below which photosynthesis is independent of irradiance within the observ ed range, decreasing with increasing temperature when temperature is high ( > 25 degrees C) and increasing with increasing relative humidity when tempe rature is low (< 25 degrees C). The machine learning tools we used provide a very powerful modelling alternative to ordinary curve fitting methods. Th eir major advantages are the flexibility to select between accuracy and gen erality and their robustness against outliers and mixtures of differential responses. The models are transparent and easily interpreted. They seem to be able to handle quite complex dependencies among attributes, not requirin g prior expert knowledge. (C) 2000 Elsevier Science B.V. All rights reserve d.