Using crop growth simulation and a field experiment conducted from 1999 to 2005 at the USDA-ARS Conservation and Production Research Laboratory, Bushland, TX (35˚11'N, 102˚5'W; and 1170 m MSL) we evaluated row spacing and plant population effects on growth and yield of dryland cotton. A dryland wheat-cotton-fallow (W-Ctn-F) rotation that produces two crops in three years using, in part, precipitation stored during intervening 10-month idle/fallow periods was established on a 510 m long by 110 m wide nearly level Pullman clay loam [17] . We subdivided the 5.6 ha site into three 36 m wide strips that were sequentially assigned to wheat, cotton, and fallow after cotton so that each rotation phase appeared every year.

The wheat phase of our rotation began in September - October when soil water permitted establishment. The uniformly managed wheat (cv. TAM-110 Foundation Seed, College Station, TX) was planted using a high-clearance hoe opener grain drill at ~50 kg・ha⁻¹ sowing rate to achieve 200 plants・m⁻² in rows 0.3 m apart. No soil active herbicides were used to control growing season weeds because of potential residual effects on sensitive rotation crops, but spring broadleaf weeds were controlled as necessary by applying 0.56 kg a.i. ha⁻¹ 2, 4-D [(2, 4-dichlorophenoxy) acetic acid] before 1 May. This first phase of the W-Ctn-F rotation was typically machine harvested in July and then fallowed 10-months through winter until mid-May of the rotation second year. Weed control during the fallow after wheat and subsequent cotton crops was exclusively by on demand (3 - 4 operations) SM tillage to a 0.10-m depth using a 4.6-m-wide Richardson sweep-plow (Sunflower Mfg. Co., Inc., Beloit, KS). Annual mineralization and atmospheric deposition of N totals ~47 kg・ha⁻¹ for the Pullman soil [18] ; therefore, no fertilizer N was needed to achieve the expected dryland yields for wheat [19] and cotton lint [20] . We applied no P or K fertilizer because the Pullman clay mineralogy supplies sufficient K to meet crop demand [21] and dryland crop response to broadcast applied P fertilizer has been limited [22] [23] .

The cotton rotation-phase began after soil water and temperature suited seed germination with planting on May 16 - 24 except for a 2003 drought delay until 17 June. From 1999 until 2001 PM-183 (Cargill Research, Plainview, TX) was planted using a John Deer 750 no-till drill (Deere & Co., Moline, IL) in four 170-m long and 9-m wide main plots of randomly assigned 0.50 m and 0.76 m row width main plots replicated three times. Each main plot was subdivided into three 56-m long split plots with random 0.15, 0.10, and 0.075 m in-row seed spacing treatments. The resulting net seeding rates were approximately 13.1, 19.7, and 26.2 seeds・m⁻² for narrow, 0.5 m, rows or 8.7, 13.1, and 17.5 seeds・m⁻² for conventional, 0.76 m, row spacing. Growing season weeds were controlled with a pre-plant application of trifluralin [2, 6-dinitro-N, N-dipropyl-4-(triflu- oromethyl) benzenamine] at 0.7 kg a.i. ha⁻¹. After 2001, we changed to a glyphosate [N-(phosphonomethyl) glycine] tolerant cotton cultivar, PM-2145RR, for improved seasonal weed control with on demand applications of 0.56 kg a.i. ha⁻¹ glyphosate.

Beginning 2002, the higher population 0.076 m in-row spacing was discontinued and a 0.25 m ultra-narrow row width added. Cotton was then seeded using a Monosem NG Plus vacuum planter (Edwardsville, KS) in 170 m × 9 m main plots of four randomly assigned 0.15 m and 0.10 m in-row spacing main plots replicated three times. Each of these main plots were subdivided into three equal split plots with random 0.25, 0.5, and 0.76 m row widths. Net seeding rates per m² for the 0.15 and 0.10 m in-row spacing approximated 26.2 and 39.4 in the ultra-narrow rows, 13.1 and 19.7 for the 0.50 m narrow rows, and 8.7 and 13.1 for 0.76 m rows. Cotton was harvested at maturity in November followed by a ~10-month fallow phase before the rotation was repeated with wheat planting in late September or early October).

Final plant population, boll number, and lint yield were determined at maturity from 13.95 m² hand samples harvested from 9.1 m long rows centered in each plot. Bur cotton samples were ginned at the Texas A & M AgriLife Research and Extension Center at Lubbock and routine fiber quality parameters were quantified with a high volume instrument by the Fiber and Biopolymer Research Institute in Lubbock. Seasonal cotton water use was determined from the sum of the official location precipitation (0.75 km away) and the calculated difference in soil water at planting and harvest, assuming no contribution by upward flux from the ~100 m deep water table and likely negligible runon, runoff, or drainage during the summer growing season [17] . Soil water measurements were based on duplicate gravimetric soil cores taken to a 1.8 m depth and sampled in 0.3 m increments that were converted to volumetric soil water using previously measured soil density as described by [24] .

All possible combinations of in-row plant spacing and combination cultivar and row widths were replicated three times as 170 m long blocks to produce 36 (9 by 56 m) plots. We compared measured soil water and cotton growth and yield factors in response to cultivar, intra and inter-row spacing fixed effects according to a randomized complete block split plot design [25] using SAS mixed linear model ANOVA procedures [26] . Random effects were block and block × (row spacing) until 2002 when random effects were block and block × (plant spacing). We analyzed all dependent parameter responses to inter-row and intra-row spacing effects within years because of annually variable growing season conditions including air temperatures, and precipitation distribution and amount.

The accuracy of the computer crop growth simulation model GOSSYM ver. 4 [27] [28] to simulate plant responses under variable growing conditions is critical in evaluating the effects of cultural practices on cotton growth and yield. Although GOSSYM was previously validated for the Southern High Plains by [29] , yield of dryland cotton grown in 0.76 m rows at a 13 plants m⁻² population during four growing seasons from 1995 to 2000 at Bushland, TX were compared to modeled yields based on the recorded conditions. Problematic in-field crop establishment due to limited soil water delayed actual planting that, consequently, depressed dryland cotton growth during already short growing season. Eliminating any delayed planting bias reduced the number of available comparisons. Nevertheless, a paired t-test indicated model projected mean yields of 220 kg・ha⁻¹ did not differ significantly from the corresponding 15 % larger measured yields that averaged 253 kg・ha⁻¹. We regressed measured lint yields (n = 8) on the corresponding model calculated values (Figure 1) to produce a highly correlated, r = 0.92, relationship with a slope of 0.84 and RMSE of 41.2 kg・ha⁻¹. That is, observed yields exceeded the modeled estimates calculated for the recorded weather conditions with a difference of +15% at 200 kg・ha⁻¹, declining to practically the same for measured and modeled yields approaching 400 kg・ha⁻¹. Model projected yields shifted to a slight, 2% to 5%, overestimation when observed yield approached 500 to 600 kg・ha⁻¹. These results, although not 1:1, suggest that

modeled yields of dryland cotton that ranged from 300 to 500 kg・ha⁻¹ would objectively represent actual crop performance.

We, therefore, used the computer crop growth simulation model GOSSYM with long-term (1958-2000) weather records at Bushland, TX to expand the basis for comparing row width and plant spacing effects on dryland cotton despite annual climatic variability. All possible combinations of narrow, 0.38 m, or conventional, 0.76 m, row spacing and intra-row plant spacing of 75,100, and 150 mm for populations, in plants・m⁻², of 8.8, 13.1, and 17.5 (conventional) or 17.5, 26.3, and 35.1 (narrow) were simulated. For these conditions, we calculated the growth and yield of a stripper type cotton cultivar having an indeterminate growth habit similar to All-Tex Atlas (Levelland, TX) as described by the supplied variety file ST1 [29] . Weather records included the model required daily solar irradiance (MJ・m⁻²), maximum and minimum air temperature (˚C), precipitation (mm), and wind run (km) input. The soil profile was a 1.8-m deep Pullman clay loam described in two horizons (0.0 - 0.3 and 0.3 - 1.8 m) with soil bulk density, texture, and hydrologic properties based on field measurements [30] or the values supplied with the model for a Pullman soil at Lubbock, TX. The initial soil profile water content for all simulations was uniformly fixed at 100% of field capacity to optimize crop establishment and potential lint yield. No fertilizer N was applied because, as previously noted, annual atmospheric deposition and mineralization for a Pullman soil totaling ~47 kg N ha⁻¹ [18] is sufficient to achieve ~500 kg・ha⁻¹ lint yields [20] that might be expected for dryland cotton at Bushland. The effects of other nutrients on cotton growth are not simulated by GOSSYM and were not specified [31] . Simulations began on day of year (DOY) 131 or 2 wk before the target DOY 145 emergence date that follows 10 d after the recommended 15 May planting date for Bushland. Simulations continued until physiological maturity (100% open bolls) or a killing freeze, when lint yield and estimated water use or growing season ET were determined.

Using SAS mixed model ANOVA procedures [26] , we compared row spacing and plant spacing fixed effects on cotton growth and yield according to a factorial arrangement of a completely randomized design replicated by years as random effects. Those plant growth and yield factors were the simulated harvest height and node number, LAI at first open boll, days between bloom and cutout, yield and boll numbers on both a plant and spatial basis.

Plant height measured at harvest, shown in Table 1, reflected the growing

a. Means of effects for the same year followed by the same letter are not significantly different, P < 0.05, according to the Tukey significant difference test.

season conditions that may have been more nearly ideal in 2004 with precipitation that approached the long-term average and above average thermal energy. Despite receiving above average precipitation in 1999, plant height was less than during 2004 possibly because of below average thermal energy. Not surprisingly, the overall shortest plants grew during the drier years of 2000 and 2001 probably as a result of water deficit stress. Although decreasing the row spacing presents a way to increase plant population without using a more frequent plant spacing, our measured plant height for the 0.76 m row width was significantly greater than for the 0.25 m or 0.50 m row spacing in contrast to findings by [11] . Model calculated plant height averaged over 42 growing seasons for the narrow, 0.38 m, row spacing was about 70% of the significantly larger 0.50 m plant height for the 0.76 m row spacing (Table 2). Increased intra-row plant spacing that reduces population for a fixed row width has the potential to reduce competition for available water and resulted in often significantly larger plant heights for the 0.15

a. Means of effects for the same year followed by the same letter are not significantly different, P < 0.05, according to the Tukey significant difference test.

m plant spacing compared with the 0.075 m and, occasionally, 0.10 m spacing (Table 1). The modeled effects of in-row plant spacing on the 42 year average cotton plant height shown in Table 2, by contrast, did not vary significantly with plant spacing. Our data show that cotton plant height was generally depressed by narrow row spacing as previously reported [12] [14] [32] , but not by decreased plant spacing.

The observed boll number at harvest shown in Table 1 did not differ between row widths of 0.76 m and 0.50 m except for a 50% decrease during 2003. Cotton grown in the ultra-narrow, 0.25 m, row width consistently featured fewer bolls at harvest than for the 0.5 m row spacing and significantly fewer than with the conventional 0.76 m row spacing. The mean harvested boll number differed in favor of the 0.15 m plant spacing over the 0.075 m spacing during only 2 of the growing seasons possibly due to favorable above average precipitation in May. As was observed for model calculated plant height, the corresponding mean boll number for the conventional 0.76 m rows were significantly greater than for the 0.38 m row spacing (Table 2). Cotton boll number also increased significantly as plant spacing increased, suggesting that fewer plants in close proximity would tend to reduce water deficit stress and increase fruiting.

Leaf area index (LAI) at first open boll and the time elapsed between bloom and cutout reflect growing conditions promoting plant vigor. The model calculated LAI for the 0.38 m, narrow, row spacing was approximately 80% of that calculated for the overall larger plants for the 0.76 m, conventional, row spacing, but did not differ with plant spacing (Table 2). The modeled number of days from emergence to bloom did not vary with row or plant spacing (data not shown), but 0.38 m rows advanced cutout and reduced the time elapsed between bloom and cutout to about 9 days for narrow rows and >23 days for conventional rows. The period from bloom to cutout also decreased significantly from 17.8 d to fewer than 16 d as plant spacing was reduced from 0.15 m to <0.10 m, suggesting the lower plant population decreased water deficit stress and expanded the bloom to cutout period. This expanded period between bloom and cutout translates to greater fruiting opportunities and potential yield.

Our measured dryland lint yield (Table 3) reflected the cumulative rain and its growing season distribution and soil water use. Peak lint yield averaged 406

a. Means of effects for the same year followed by the same letter are not significantly different, P < 0.05, according to the Tukey significant difference test.

kg・ha⁻¹ in 1999 in response to >400 mm rain plus soil water for a total ET of 460 mm, but the 135 kg・ha⁻¹ mean yield low in 2000 followed seasonal rain totaling 139 mm and 70 mm soil water use that limited ET to a total of 210 mm (data not shown). The effects of row width and plant spacing on cotton lint yields were neutralized during these unusually wet or dry growing seasons. Measured lint yield for the conventional 0.76 m row spacing ranged from 15% to 120% larger than yields of the narrow, 0.50 m, row spacing that were significantly different during the 2001 and 2004 growing seasons. The 2004 mean 173 kg・ha⁻¹ yield for the ultra-narrow 0.25 m row spacing was about 50% of the lint yield for 0.76 m row spacing under dryland conditions; thus, differing from observations under irrigated or rain fed production [11] [12] [33] . By contrast, plant spacing produced no significant differences except for depressing the yield to 244 kg・ha⁻¹ compared with the average 307 kg・ha⁻¹ for the 0.10 m and 0.15 m plant spacing. Consistent with observed row spacing yield response, our modeled yield for the 0.38 m row spacing was ~50% of the 582 kg・ha⁻¹ yield calculated for the, 0.76 m, conventional row spacing and did not vary significantly with plant spacing (Table 2). Model calculated dryland yields for 0.76 m rows, plotted in Figure 3 with increasing observation exceedance probability, were consistently greater than the corresponding yields for 0.38 m rows. In fact, approximately 80% of those yields from 0.76 m row spacing exceeded the greatest yield of the 0.38 m row spacing although the modeled yields less than 300 kg・ha⁻¹ would be underestimated by as much as 50 kg・ha⁻¹. The modeled yield results also provide no compelling evidence that plant spacing influenced the lint yields under these dryland conditions as measured under dryland and irrigated conditions [10] . That is, the data show that row spacing and not plant spacing influenced lint yield.

Lint quality can be revealing of environmental stresses imposed on the crop and the effect of management practices to offset them. For example, [34] observed that uniformly irrigated cotton lint quality factors and yield were generally unaffected by row spacing. Cotton fibers are formed in two phases that begin with rapid cell elongation during the first three weeks after bloom that establishes fiber length followed by fiber wall thickening as cellulose deposition continues through maturation about 30 days later. Any environmental stress caused by, for example, nutrient or water deficiency typically results in shortened or improperly matured fine or overly thickened fibers.

Our measured primary fiber quality factors of length and micronaire in response to treatment row and plant spacing are listed in Table 4 by year with the level of significance, but no corroborating model fiber quality results were possible because GOSSYM does not determine those properties [35] [36] . Fiber length exceeded the grade minimum 20.6 mm, averaging 21.5 mm across the

a. Means of effects for the same year followed by the same letter are not significantly different, P < 0.05, according to the Tukey significant difference test.

combined row width and plant spacing treatments for both 1999 and 2000. The short fiber for 2000 was not surprising because during the July-September lint growth and maturation only 25 mm of precipitation was received during three events in July. For this period in 1999 rain exceeded 200 mm and would have been ample for lint growth, but the above average precipitation likely promoted use of soil nutrients including N vegetative growth in competition with nutrient demand for lint formation [36] . Precipitation during 2001 of 180 mm totaled less than almost 90% of the long-term May-September growing season precipitation and probably resulted in fiber length averaging 23.2 mm that was likewise unaffected by row or plant spacing treatments. Somewhat limited but better distributed precipitation in combination with available soil water resulted in significantly greater measured lint length for cotton grown in conventional 0.76 m wide rows compared with narrow or ultra-narrow row widths during 2003 and 2004. Although plant spacing resulted in no fiber length differences in 2003, lint length was greater with the 0.10 m plant spacing in 2004. The measured micronaire, as observed for fiber length, did not differ significantly among any row or plant spacing treatments during 1999-2001. In fact, fiber micronaire did not vary with plant spacing treatment during any year of the study. Favorable fiber micronaire values for the 0.76 conventional row spacing were significantly better than those of the narrow or ultra-narrow row spacing as was observed for fiber length. The lint quality data show that the growing conditions with the conventional 0.76 m row width improved fiber length and micronaire during two of the five growing seasons evaluated. The 0.15 - 0.075 m plant spacing treatments also did not affect fiber quality with the exception of length in 2004.

Length, micronaire, and other factors are combined to determine the lint price premium that we present in Table 5 as an integrated indicator of the fiber quality. No direct comparisons between years may be made because premium discounts were year specific, but lower price discounts after 2001 may reflect favorable June rain that supported fiber lengthening. No significant premium differences were identified during any year between the 0.50 and 0.76 m row spacing except in 2003 when planting was delayed until mid-June followed by a dry, ~7 mm rain, July. The conventional row spacing premium discount was significantly less than for the ultra-narrow, 0.25 m, rows during both 2003 in 2004. These data show that the ultra-narrow and narrow row spacing tended to have more heavily discounted price premiums than with conventional row spacing. Plant spacing resulted in no significant differences in price premiums during all five growing seasons from 1999 to 2004 because of negligible effects on lint quality factors.