1 1 Supporting information 1 2 Description of phenology models 3 For each model, the state of chilling (Sc) or forcing (Sf) is the time integral from p0 or t0 of 4 the rate of chilling (Rc) or forcing (Rf), which are functions of daily mean temperature 5 x(t). Chilling and forcing accumulate relative to base temperatures (Tbase, Tchill, Tforce, Tmax, 6 or Tmin) until a critical threshold (C* or F*) is reached. The date that the threshold is 7 reached corresponds to the predicted timing of the phenological metric of interest. 8 Chilling and forcing parameters are either fixed across all sites, or are modeled as a linear 9 function of the long-term mean annual temperature ( Ti ) of each site, i. 10 11 Spring Warming 1 12 ο ο 28.4 π₯(π‘) > 0 π π (π‘) = {1 + expβ‘(3.4 − 0.185 ∗ π₯(π‘)) 0 π₯(π‘) ≤ 0 13 ππ (π‘) = ∑ π π (π₯(π‘)) 14 15 16 Phenology metric is predicted to occur when Sf = F* π‘0 Spring Warming 2 17 28.4 π₯(π‘) > 0 π π (π‘) = {1 + expβ‘(3.4 − 0.185 ∗ π₯(π‘)) 0 π₯(π‘) ≤ 0 18 ππ (π‘) = ∑ π π (π₯(π‘)) 19 20 21 Phenology metric is predicted to occur when Sf = F* π0 Spring Warming 3 22 28.4 π₯(π‘) > 0 π π (π‘) = {1 + expβ‘(3.4 − 0.185 ∗ π₯(π‘)) 0 π₯(π‘) ≤ 0 23 ππ (π‘) = ∑ π π (π₯(π‘)) 24 25 26 π‘0 Phenology metric is predicted to occur when Sf = a Ti +b Spring Warming 4 ο ο 2 27 28.4 π₯(π‘) > 0 π π (π‘) = {1 + expβ‘(3.4 − 0.185 ∗ π₯(π‘)) 0 π₯(π‘) ≤ 0 28 ππ (π‘) = ∑ π π (π₯(π‘)) 29 30 31 Phenology metric is predicted to occur when Sf = a Ti +b π0 Parallel 1 32 28.4 ο ο π₯(π‘) > 0 π π (π‘) = {1 + expβ‘(3.4 − 0.185 ∗ π₯(π‘)) 0 π₯(π‘) ≤ 0 33 ππ (π‘) = ∑ π π (π₯(π‘)) π‘0 34 0 π₯(π‘) ≥ 10.4β‘or π₯(π‘) ≤ −3.4 π₯(π‘) + 3.4 −3.4β‘ < π₯(π‘) ≤ πopt π π (π‘) = πopt + 3.4 π₯(π‘) − 10.4 πopt < π₯(π‘) < 10.4 { πopt − 10.4 35 ππ (π‘) = ∑ ππ (π₯(π‘)) 36 37 38 π‘0 Phenology metric is predicted to occur when ππ (π‘) ≥ π ∗ expβ‘(π ∗ ππ (π‘)), where b < 0. Parallel 2 39 28.4 π₯(π‘) > 0 π π (π‘) = {1 + expβ‘(3.4 − 0.185 ∗ π₯(π‘)) 0 π₯(π‘) ≤ 0 40 ππ (π‘) = ∑ π π (π₯(π‘)) π0 41 0 π₯(π‘) ≥ 10.4β‘or π₯(π‘) ≤ −3.4 π₯(π‘) + 3.4 −3.4 < π₯(π‘) ≤ πopt π π (π‘) = πopt + 3.4 π₯(π‘) − 10.4 πopt < π₯(π‘) < 10.4 { πopt − 10.4 42 ππ (π‘) = ∑ ππ (π₯(π‘)) 43 44 45 Phenology metric is predicted to occur when ππ (π‘) ≥ π ∗ expβ‘(π ∗ ππ (π‘)), where b < 0. π0 46 Alternating 1 π₯(π‘) − πbase π π (π‘) = { 0 47 ππ (π‘) = ∑ π π (π₯(π‘)) π‘0 π₯(π‘) > πbase π₯(π‘) ≤ πbase 3 1 π₯(π‘) < πbase 0 π₯(π‘) ≥ πbase 48 π π (π‘) = β‘ { 49 ππ (π‘) = ∑ ππ (π₯(π‘)) 50 51 52 53 Phenology metric is predicted to occur when ππ (π‘) ≥ π + π ∗ expβ‘(π ∗ ππ (π‘)), where c < 0. π‘0 54 Alternating 2 π₯(π‘) − πbase π π (π‘) = { 0 55 ππ (π‘) = ∑ π π (π₯(π‘)) 56 1 π₯(π‘) < πbase π π (π‘) = β‘ { 0 π₯(π‘) ≥ πbase 57 ππ (π‘) = ∑ ππ (π₯(π‘)) 58 59 60 61 Phenology metric is predicted to occur when ππ (π‘) ≥ π + π ∗ expβ‘(π ∗ ππ (π‘)), where c < 0. π₯(π‘) > πbase π₯(π‘) ≤ πbase π0 π0 62 Sequential 1 π₯(π‘) − πforce π π (π‘) = { 0 63 ππ (π‘) = ∑ π π (π₯(π‘)) 64 1 π₯(π‘) < πchill π π (π‘) = β‘ { 0 π₯(π‘) ≥ πchill 65 ππ (π‘) = ∑ π π (π₯(π‘)) 66 67 68 69 Forcing summation (Sf ) begins at t1 when Sc = C*. Phenology metric is predicted to occur when Sf = F* π₯(π‘) > πforce π₯(π‘) ≤ πforce π‘1 π‘1 π‘0 70 Sequential 2 π₯(π‘) − πforce π π (π‘) = { 0 71 ππ (π‘) = ∑ π π (π₯(π‘)) 72 1 π₯(π‘) < πchill π π (π‘) = β‘ { 0 π₯(π‘) ≥ πchill 73 ππ (π‘) = ∑ π π (π₯(π‘)) π₯(π‘) > πforce π₯(π‘) ≤ πforce π‘1 74 75 76 77 π‘1 π0 Forcing summation (Sf (t)) begins at t1 when Sc = C*. Phenology metric is predicted to occur when Sf = F* Sequential 3 4 π₯(π‘) − πforce 0 π₯(π‘) > πforce π₯(π‘) ≤ πforce 78 π π (π‘) = { 79 ππ (π‘) = ∑ π π (π₯(π‘)) 80 1 π₯(π‘) < πchill π π (π‘) = β‘ { 0 π₯(π‘) ≥ πchill 81 ππ (π‘) = ∑ π π (π₯(π‘)) 82 83 84 85 Forcing summation (Sf ) begins at t1 when Sc = C*. Phenology metric is predicted to occur when Sf = a Ti +b π‘1 π‘1 π0 86 Sequential 4 π₯(π‘) − πforce π πο ο (π‘) = { 0 87 ππ (π‘) = ∑ π π (π₯(π‘)) 88 1 π₯(π‘) < πchill π π (π‘) = β‘ { 0 π₯(π‘) ≥ πchill 89 ππ (π‘) = ∑ π π (π₯(π‘)) π₯(π‘) > πforce π₯(π‘) ≤ πforce π‘1 90 91 92 93 π‘1 π0 Forcing summation (Sf ) begins at t1 when Sc = a Ti +b. Phenology metric is predicted to occur when Sf = F* 94 Sequential 5 π₯(π‘) − πforce π π (π‘) = { 0 95 ππ (π‘) = ∑ π π (π₯(π‘)) 96 1 π₯(π‘) < πchill π π (π‘) = β‘ { 0 π₯(π‘) ≥ πchill 97 ππ (π‘) = ∑ π π (π₯(π‘)) π₯(π‘) > πforce ο ο π₯(π‘) ≤ πforce π‘1 98 99 100 101 102 103 104 105 π‘1 π0 Forcing summation (Sf ) begins at t1 when Sc = a Ti +b. Phenology metric is predicted to occur when Sf = c Ti +d. ο ο ο ο 5 106 Table S1: Summary of model parameters Model SW1 SW2 SW3 SW4 PAR1 PAR2 ALT1 ALT2 SEQ1 SEQ2 SEQ3 SEQ4 SEQ5 107 108 109 110 111 112 113 114 115 Parameters t0, F* p0, F* t0, a, b [F*] p0, a, b [F*] t0, Topt, a, b p0, Topt, a, b t0, Tref, a, b, c p0, Tref, a, b, c t0, Tforce, Tchill, F*, C* p0, Tforce, Tchill, F*, C* p0, Tforce, Tchill, C*, a, b [F*] p0, Tforce, Tchill, F*, a, b [C*] p0, Tforce, Tchill, a, b [F*], c, d [C*] 6 116 Table S2: Optimal parameters for each species based on out-of-sample testing results. Species codes are given in Table 2. Parameters 117 are given with associated 95% confidence intervals in parentheses. If a single numerical value j is listed in the table under F*, then F* 118 = j in the corresponding model. If two values j and k are listed, then πΉ ∗ = πΜ π + π, such that πΜ is mean annual temperature. In the first 119 parameter column (t0 / p0), values followed by a * indicate starting photoperiod (p0). Species Code ACRU ACSA FAGR PRSE BEPA POTR JUNI LITU QUAL QURU PhCam 120 121 t0 / p0* 51 (50, 53) 11.9* (11.8, 12.0) 12.5* (12.4, 12.7) 9.5* (9.3, 9.8) 12.7* (12.4,12.9) 12.1* (11.5, 13.2) 11.9* (11.6, 12.0) 11.6* (11.5, 11.6) 343 (339, 354) 77 (70, 87) 73 (70, 76) F* Tforce / Tref Topt -3.2 (-3.4, -2.3) a 288 (274, 312) b -0.02 (-0.02, -0.02) c 1.2 (1.0, 1.3) 168 (165, 172) 1155 (959, 1300) (-0.03, -0.02) 6.9 (6.0, 7.2) 105 (101, 108) 1408 (1212, 1653) (-0.04, -0.03) -5.7 (-8.6, -3.5) 24.0 (20.6, 26.9) 532 (424, 669) (-0.01, 0.02) 6.9 (4.5, 9.3) 112.6 (88.6, 130.1) 171.5 (153.6, 193.7) -0.03 135.7 (126.3, 148.1) 2.1 125.5 (-1.3, 5.5) (98.2, 146.1) 261.6 (237.2, 290.1) 150.1 (139.4, 158.1) -0.03 165.6 (152.0, 185.1) -0.006 7 122 Table S3: ΔAICc (difference between a particular model’s AICc and the lowest AICc across all candidate models for a species) values 123 for models fit to species-specific NPN or PhenoCam data. Species codes are given in Table 2. The best model, based on Akaike’s 124 Information Criterion corrected for small samples (AICc) has ΔAICc = 0 and is indicated by bold type. 125 SW1 SW2 SW3 SW4 PAR1 ALT1 PAR2 ALT2 SEQ1 SEQ2 SEQ3 SEQ4 SEQ5 ACRU POTR PRSE ACSA FAGR JUNI LITU BEPA QUAL QURU PhCam 51.2 30.6 4.8 24.6 1.7 2.3 6.1 2.3 4.9 8.6 0.0 47.2 26.4 5.3 24.4 4.8 4.4 4.4 0.0 0.0 0.0 0.0 51.4 22.9 6.2 2.4 22.0 8.8 15.8 40.6 8.4 59.9 50.4 70.5 7.9 23.9 10.9 3.5 44.8 7.5 58.9 49.4 0.0 0.0 40.2 9.3 61.2 11.2 14.0 1.2 17.1 15.1 15.4 42.9 0.0 16.7 82.8 7.4 26.7 0.8 9.9 18.4 47.4 0.0 0.0 0.0 14.9 113.2 18.6 41.5 0.8 6.1 0.8 63.1 26.3 31.6 8.8 10.1 72.1 10.8 10.8 29.9 6.0 9.7 26.5 0.9 11.7 13.9 53.3 42.9 237.4 79.7 241.1 9.6 10.3 56.1 37.4 27.7 339.9 24.7 34.5 182.5 76.3 180.2 12.4 5.8 13.1 144.3 22.5 2.9 41.6 52.9 215.5 56.3 181.4 34.3 0.7 26.9 45.6 55.5 59.0 68.0 67.4 240.3 73.5 229.8 9.4 7.9 23.9 149.2 180.2 119.0 26.0 69.4 180.1 79.8 213.0 12.7 11.8 210.5 259.5 55.5 115.7 AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL Ph U R C U am 0.0 0.2 0.4 0.6 0.8 1.0 Pearson Correlation 0.0 0.2 0.4 0.6 0.8 1.0 Oak−Gum AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL Ph U R C U am (d) (e) Elm−Ash−Cottonwood (f) Aspen−Birch Pearson Correlation AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL Ph U R C U am Pearson Correlation 0.0 0.2 0.4 0.6 0.8 1.0 (b) 0.0 0.2 0.4 0.6 0.8 1.0 AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL Ph U R C U am Oak−Hickory 0.0 0.2 0.4 0.6 0.8 1.0 Pearson Correlation 0.0 0.2 0.4 0.6 0.8 1.0 Pearson Correlation (a) AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL Ph U R C U am AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL Ph U R C U am Pearson Correlation 8 Figure S1: Boxplots of Pearson correlation coefficients between MODIS estimated spring onset and model predicted spring onset across (a-e) pixels dominated by each forest type, according to 25 km FIA hexagon maps (see Figure 2) or (f) the entire study region. (c) Maple−Beech−Birch Total 2.0 2.0 3.0 3.0 AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL PhUR C U am AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL PhUR C U am AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL PhUR C U am 0.0 0.0 0.0 2.0 2.0 1.0 2.0 sobs spred 1.0 sobs spred 1.0 3.0 3.0 3.0 Oak−Gum 1.0 sobs spred 1.0 3.0 (d) (e) Elm−Ash−Cottonwood (f) Aspen−Birch 0.0 2.0 sobs spred 1.0 (b) AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL P h UR C U am 0.0 0.0 sobs spred (a) AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL P h UR C U am AC ACRU FA SA PRGR BESE PO PA T JU R LI N I Q TU U Q AL P h UR C U am sobs spred 9 Figure S2: Boxplots of model slope (calculated as quotient of standard deviation of MODIS estimated spring onset and standard deviation of predicted spring onset) across (a-e) pixels dominated by each forest type, according to 25 km FIA hexagon maps (see Figure 2) or (f) the entire study region. (c) Maple−Beech−Birch Total