DAISY model simulations Directory UMM :Data Elmu:jurnal:A:Agricultural & Forest Meterology:Vol106.Issue3.2001:

P. van der Keur et al. Agricultural and Forest Meteorology 106 2001 215–231 223 The measurements of latent heat flux were evalu- ated by comparing to changes in water content mea- sured using the automated TDR station. Over periods with no rainfall, the difference from start to end in water content in the top 50 or 100 cm soil eight repli- cates of each probe length equals the amount of wa- ter lost as soil evaporation and crop transpiration to the atmosphere. We assume, in agreement to model predictions, that there is no significant drainage from the soil profile during such dry intervals. During the period 10–13 June 1997, the accumulated amount of water lost to the atmosphere according to the eddy correlation measurements was 11 mm, while the water content decline recorded by the 50 and 100 cm TDR probes corresponded to 11 and 13 mm of water, re- spectively. During the equally short dry period 18–20 June 1997, latent heat loss was equivalent to 9 mm of water, while the 50 and 100 cm TDR probes recorded water deficits of 9 and 12 mm, respectively. 9–23 July 1997 was a long dry spell. Observed water depletion during this period in the 0–50 and 0–100 cm soil depth was 49 and 69 mm, respectively. Accumulated latent heat loss during the dry spell was equivalent to 45 mm of water. Since the root zone of the wheat crop proba- bly exceeded 50 cm in July, and since we assume the TDR technique to be accurate in estimating relative changes in soil water content, the results indicate that eddy covariance estimates of latent heat flux could be underestimated in July.

4. DAISY model simulations

The model was set up to simulate on hourly basis from 1 June 1996 to 31 December 1997. Meteoro- logical forcings global radiation, air temperature, air humidity, precipitation and windspeed were retrieved from the RCF climate station using standard equip- ment. During the campaign period local forcing data were used whenever available as explained previously. The soil profile was partitioned in 20 compartments with discretization size varying from 2.5, 5 and 10 cm in the upper 75 cm to 10, 15, 20, 30 cm in the lower 75–200 cm. The hydraulic properties were parameter- ized following the Brooks and Corey 1964 model for soil water characteristics and the Mualem 1976 model for unsaturated conductivity. No soil profile measurements were conducted at the winter wheat site for laboratory analyses, so hydraulic properties were estimated from previously performed profile analyses from adjacent locations. Brooks and Corey parameters were determined from an average of three soil profiles at four depth intervals: 0–32.5, 32.5–52.5, 52.5–100, and 100–230 cm. In the absence of locally measured hydraulic data and aware of the fact that large spa- tial variations in soil physical characteristics probably occur, Brooks Corey parameters were adjusted to obtain good agreement between simulated and TDR measured soil moisture content for 0–0.2, 0–0.5 and 0–1.0 m. The saturated hydraulic conductivity is cal- culated as a logarithmic average of the three profiles for each horizon. Surface albedo for calculation of net radiation in Eq. 13 is estimated from measured in- coming and reflected radiation in 640–660 nm red using Skye SKR 1800 equipment during June, July and most of August. The mean value is 0.2 with a slight decreasing tendency from June to August. The winter wheat crop has been sampled several times during the growing season for crop development measurements like leaf area index, dry matter content and canopy height. LAI measurements for green, semigreen and yellow leaves by means of scanning in the laboratory have been supplemented by LAI2000 Li-Cor data for total LAI. Winter wheat was sown in September 1996 and harvested in August 1997. Application of both inorganic and organic pig slurry fertilizer are in accordance with recommended amounts Plantedi- rektoratet, 19971998. 4.1. Simulation results 4.1.1. Net radiation Simulated net radiation using the Brunt equation 14 and calculated relative sunshine duration using Eq. 15 for the period 13 June to 7 August yielded reasonably good agreement during day-time, whereas simulated night-time values generally were too low compared to the local net radiometer data REBS Q ∗ 7, REBS, Seattle, WA. Schelde et al. 1998 ob- served a similar bias for a bare soil in Denmark. Recently, significant discrepancies in estimates of R n between different 12 REBS net radiometers under different conditions were observed and analyzed by, e.g. Kustas et al. 1998, who found significant dif- ferences. Halldin and Lindroth 1992 compared six net radiometers and observed differences in output 224 P. van der Keur et al. Agricultural and Forest Meteorology 106 2001 215–231 Fig. 3. Net radiation estimated using the Brunt equation vs. mea- sured net radiation. ranging from 6 to 20. In this study the normal- ized root mean square error NRMSE defined as p R sim n − R obs n 2 n, i.e. RMSE, divided by the mean of R obs n is 0.36 for day- and night-time val- ues and 0.25 for day values only, defined by S i 10 W m −2 , for the period 13 June to 7 August on hourly basis. Sensitivity analyses suggest that the best agreement as compared to the Q ∗ 7 measurements NRSME = 0.31, Fig. 3 is obtained when the param- eter n sun in Eq. 14 is equal to zero, corresponding to full cloudiness, during night-time and estimated from Eq. 15 for day-time periods. The latter approach was therefore adopted for this study. 4.1.2. Crop development simulation DAISY simulated crop development was evaluated against measured leaf area index and dry matter con- tent. Modeled LAI showed a too fast development in May compared to measured green LAI GLAI as de- termined in the laboratory Fig. 4a, below. During June and July modeled LAI was slightly higher than measured GLAI. Simulated total dry matter content compared very well to measured total dry matter, i.e. green and dead material Fig. 4a, top. Sub samples of dry matter content fractioned after stem, leaf and ears Fig. 4b were also in good agreement with mod- eled data. Simulated nitrogen content not shown here appeared to be close to values sampled from a field nearby with the same crop and fertilizer treatment. Fig. 4. a Simulated — and measured d total winter wheat dry matter content and green LAI. b Simulated — and measured d winter wheat fractions. 4.1.3. Soil water modeling For the purpose of this study, which is to enable the DAISY model to be linked to remotely sensed data by adding a resistance network approach as previously described, modeled soil moisture content for the 0–20, 0–50 and 0–100 cm levels must satisfactorily match TDR measured soil water content with vertically in- serted probes Fig. 5. However, since the objective of P. van der Keur et al. Agricultural and Forest Meteorology 106 2001 215–231 225 Fig. 5. Volumetric soil moisture content measured by time domain reflectometry d and simulated by the DAISY model — for 10 June to 30 July 1997. this study is not modeling of soil water dynamics, it has not been attempted to optimize agreement between simulated and TDR measured soil moisture content for each horizontal level, nor to test model performance for other periods than the calibration period. 4.1.4. Energy balance modeling Simulation of energy fluxes was performed for the period of 13 June to 7 August, when eddy covariance data was available. Bare soil evaporation contribution to latent heat flux is assumed to be negligible under full canopy conditions, but constitutes an increasing part with decreasing LAI. The relative importance of the r min c parameter, amenable to RS data, is closer examined by substitution of r min c − 50, r min c and r min c + 50 in Eq. 24 for computation of canopy re- sistance r sc for subsequent use in Eq. 5 for simulation of latent heat flux from leaf surface to mean source height. This is demonstrated in Fig. 6 lower graph, where r sc , moderated by the environmental constraint functions in Eqs. 24–28, is modeled for 18 and 19 June. From Fig. 6 upper graph it is clear that given the same environmental constraints, the value of r min c , as potentially sensed by RS data, is important for a correct modeling of latent heat flux through r sc in Eq. 24. However, it must be borne in mind that a correct specification of stress functions, which may be site-specific to a high degree, are at least equally Fig. 6. Simulated — and measured d latent heat fluxes λE for different canopy resistance r sc values. Lower graph: r sc — as calculated from minimum canopy resistance, r min c in Eq. 24, r sc from r min c + 50 m and r sc from r min c − 50 . . Upper graph: largest simulated λE for smallest r sc value and vice versa. important in this respect. Fig. 7a and b show modeled latent heat flux and sensible heat flux against mea- sured eddy covariance data as well as r min c and canopy resistance r sc . At night stomata close and the canopy resistance may be very high, therefore night-time val- ues are truncated in Fig. 7a and b top. The periods 17–22 June and 17–22 July were chosen to illustrate differences in simulated energy fluxes when no water stress occurred and under stress, respectively, and both periods with sufficient fetch see Section 3.3. During the first period in June, it is clear that during day-time hours r sc is very close to r min c , indicating effectively no environmental stress, i.e. F i close to unity. After a prolonged drought in July Fig. 2, r sc becomes much higher than r min c during day-time hours, mainly as a result of an increasing regulation by the constraint functions, particularly the influence of F 4 in Eq. 28. Fig. 8 supports the notion that a larger discrepancy between measured and modeled λE occurs when soil water content in the root zone decreases. Observed latent heat flux as measured by the eddy covariance equipment is generally in good agreement with the modeled data in the first period with no water stress. In the second period from 17 to 22 July, energy fluxes are simulated less well, particularly when considering that measured latent heat flux is expected to be un- 226 P. van der Keur et al. Agricultural and Forest Meteorology 106 2001 215–231 Fig. 7. a Simulated — and measured d energy fluxes for 17–22 June. Upper graph: stressed . and unstressed - - - canopy resistance, r sc and r min c , respectively. Note that the very high stressed canopy resistance values during night-time are truncated. b Simulated — and measured d energy fluxes for 17–22 July. Upper graph: stressed . and unstressed - - - canopy resistance, r sc and r min c , respectively. Note that the very high stressed canopy resistance values during night-time are truncated. derestimated. The low simulated fluxes are due to an increased control by the environmental functions, as mentioned before. Although the hydraulic parameters in the F 4 function were derived from actual field data and therefore physically based, there is little doubt that this function overestimated the water stress effect and needs improving or an alternative parameterization. Fig. 8. Simulated — and measured d latent heat fluxes and simulated 0–50 cm soil moisture content, SMC - - -. The periods 17–22 June no water stress and 17–22 July water stress are represented by top and bottom graph, respectively. 4.2. Potential link to remote sensing data Correct estimation of minimum canopy conduc- tance is especially important when the DAISY model is incorporated into a distributed hydrological model for simulation of hydro-ecological processes at the landscape level. Remote sensing data is crucial in accomplishing this. The canopy resistance r sc is lin- early proportional to the upscaled unstressed canopy resistance r min c and an accurate estimation of the development of r min c is thus needed as well as a correct parameterization of the environmental con- straint functions, for modeling energy fluxes. It has been mentioned earlier that the r min c parameter can be estimated by means of vegetation indices and that could then lead to improved transpiration modeling provided that a more robust and physically based parameterization for constraint functions is obtained e.g. Jacobs, 1994; Monteith, 1995a,b; Baldocchi and Meyers, 1998. From the data presented in this study it is clear that transpiration modeling is sensitive to a correct r min c level for an agricultural crop during the growing season. Equally clear is the need for develop- ment of appropriate stress functions. However, within the interest sphere of distributed hydrologic modeling at the landscape level or larger, there is an obvious P. van der Keur et al. Agricultural and Forest Meteorology 106 2001 215–231 227 need to sense plant physiological development at this scale and use this information in simulation of surface energy fluxes.

5. Concluding remarks